test(core): stabilize shell exit-order test

The sequential PowerShell spawn/remove loop can cross Bun’s default 5-second test timeout on Windows. Give this process-heavy test the same explicit 15-second budget used by adjacent shell lifecycle coverage.
2026-08-27 20:16:17 +00:00 · 2026-08-22 13:14:03 +00:00
1167 changed files with 20447 additions and 90331 deletions
@@ -1,5 +0,0 @@
---
-"@opencode-ai/core": patch
---
-
-Propagate fatal Git ignore-check errors during index refresh so best-effort snapshot capture returns unavailable instead of treating the check as having no matches.
@@ -1,5 +0,0 @@
---
-"@opencode-ai/core": patch
---
-
-Rename the write, patch, and question tool formatting helpers from `toModelOutput` to `toModelContent` to match the result field they populate. Direct imports of these helpers must use the new name; generated content and declared machine output are unchanged.
@@ -1,5 +0,0 @@
---
-"@opencode-ai/core": minor
---
-
-Rename ModelResolver.supported to ModelResolver.hasPackage. Consumers of the old export must update the name; the predicate remains Boolean(model.package), checking only whether a catalog model declares a provider package, not whether it can be loaded. Default-model selection behavior is unchanged.
@@ -1,5 +0,0 @@
---
-"@opencode-ai/core": patch
---
-
-Include both paths of renamed files in new snapshot change lists so undo restores the original file instead of only deleting the renamed file.
@@ -1,10 +0,0 @@
---
-"@opencode-ai/core": patch
---
-
-Make the experimental portable shell scanner authoritative, with no Tree-sitter
-fallback. Scan common Bash and PowerShell control flow, heredocs, functions,
-expressions, quoting, and substitutions natively. Preserve existing redirect and
-declaration permission matching, and make PowerShell saved approvals cover the
-original command spelling. Parser failures remain visible without changing the
-permission engine. The default Tree-sitter path is unchanged.
@@ -1,5 +0,0 @@
---
-"@opencode-ai/core": patch
---
-
-Use parent directory names for SKILL.md IDs so individually downloaded skills do not overwrite each other.
@@ -1,6 +0,0 @@
---
-"@opencode-ai/client": patch
-"@opencode-ai/plugin": patch
---
-
-Add form reply and cancellation operations that reconcile terminal forms in the local TUI projection.
@@ -27,6 +27,7 @@ jobs:
        working-directory: packages/www
        run: bun run build
        env:
+          BLUME_ENV: ${{ github.ref_name == 'v2' && 'production' || 'dev' }}
          CLOUDFLARE_ENV: ${{ github.ref_name == 'v2' && 'production' || 'dev' }}

      - name: Deploy
@@ -135,16 +135,7 @@ jobs:

            const linkedIssues = result.repository.pullRequest.closingIssuesReferences.totalCount;

-            // GitHub only populates closingIssuesReferences when a PR targets the repository's
-            // default branch (dev). PRs targeting other branches like v2 always return totalCount 0.
-            // Fall back to checking the PR description for closing keywords (e.g. Closes #123).
-            const body = pr.body || '';
-            const issueMatch = body.match(/### Issue for this PR\s*\n([\s\S]*?)(?=###|$)/);
-            const issueContent = issueMatch ? issueMatch[1].trim() : body;
-            const hasBodyIssueRef = /(closes|fixes|resolves)\s+#\d+/i.test(issueContent) || /#\d+/.test(issueContent);
-            const hasLinkedIssue = linkedIssues > 0 || hasBodyIssueRef;
-
-            if (!hasLinkedIssue) {
+            if (linkedIssues === 0) {
              await addLabel('needs:issue');
              await comment('issue', `Thanks for your contribution!

@@ -91,7 +91,7 @@ jobs:

      - uses: ./.github/actions/setup-bun
        with:
-          bun-version: 1.4.0
+          bun-version: canary # Bun 1.4 until its stable release is published

      - name: Setup git committer
        id: committer
@@ -113,7 +113,7 @@ jobs:
        id: build
        run: ./packages/cli/script/build.ts ${{ (github.ref_name == 'beta' && '--sourcemaps') || '' }}
        env:
-          BUN_COMPILE_RELEASE: bun-v1.4.0
+          BUN_COMPILE_RELEASE: canary
          OPENCODE_VERSION: ${{ needs.version.outputs.version }}
          OPENCODE_RELEASE: ${{ needs.version.outputs.release }}
          GH_REPO: ${{ needs.version.outputs.repo }}
@@ -195,33 +195,9 @@ jobs:
          path: packages/cli/dist/cli-*
          if-no-files-found: error

-  build-node-app-archive:
-    needs: version
-    runs-on: blacksmith-4vcpu-ubuntu-2404
-    timeout-minutes: 30
-    if: github.repository == 'anomalyco/opencode'
-    steps:
-      - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0
-
-      - uses: ./.github/actions/setup-bun
-
-      - name: Build app archive
-        run: bun packages/cli/script/build-node.ts --app-archive-only --app-archive=.cache/app-archive.bin --skip-install
-        env:
-          OPENCODE_VERSION: ${{ needs.version.outputs.version }}
-          OPENCODE_RELEASE: ${{ needs.version.outputs.release }}
-
-      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
-        with:
-          name: opencode-node-app-archive
-          path: packages/cli/.cache/app-archive.bin
-          if-no-files-found: error
-
  build-node-cli:
-    needs:
-      - version
-      - build-node-app-archive
-    if: github.repository == 'anomalyco/opencode'
+    needs: version
+    if: github.repository == 'anomalyco/opencode' && false # Temporarily disabled
    strategy:
      fail-fast: false
      matrix:
@@ -234,7 +210,6 @@ jobs:
            host: macos-26
          - target: windows-arm64
            host: blacksmith-4vcpu-windows-2025
-            bun_install_flags: --cpu=*
          - target: windows-x64
            host: blacksmith-4vcpu-windows-2025
    runs-on: ${{ matrix.settings.host }}
@@ -246,19 +221,14 @@ jobs:

      - uses: ./.github/actions/setup-bun
        with:
-          install-flags: ${{ matrix.settings.bun_install_flags }}
+          install-flags: --os=* --cpu=*

      - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
        with:
          node-version: "26.4.0"

-      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
-        with:
-          name: opencode-node-app-archive
-          path: packages/cli/.cache
-
      - name: Build
-        run: bun packages/cli/script/build-node.ts --target=${{ matrix.settings.target }} --skip-install --outdir=dist/node --app-archive=.cache/app-archive.bin
+        run: bun packages/cli/script/build-node.ts --target=${{ matrix.settings.target }} --skip-install --outdir=dist/node
        env:
          OPENCODE_VERSION: ${{ needs.version.outputs.version }}
          OPENCODE_RELEASE: ${{ needs.version.outputs.release }}
@@ -573,7 +543,6 @@ jobs:
      - version
      - build-cli
      - sign-cli-macos
-      - build-node-app-archive
      - build-node-cli
      - sign-cli-windows
      - build-electron
@@ -22,36 +22,6 @@ env:
  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: true

 jobs:
-  affected:
-    name: affected packages
-    runs-on: blacksmith-4vcpu-ubuntu-2404
-    outputs:
-      app: ${{ steps.packages.outputs.app }}
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
-        with:
-          token: ${{ secrets.GITHUB_TOKEN }}
-          fetch-depth: 0
-
-      - name: Setup Bun
-        uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0
-        with:
-          bun-version-file: package.json
-
-      - name: Find affected packages
-        id: packages
-        env:
-          TURBO_SCM_BASE: ${{ github.event_name == 'pull_request' && format('{0}^1', github.sha) || github.event.before }}
-          TURBO_SCM_HEAD: ${{ github.sha }}
-        run: |
-          if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
-            echo "app=true" >> "$GITHUB_OUTPUT"
-            exit 0
-          fi
-          bun x turbo@2.10.2 ls --affected --filter=@opencode-ai/app --output=json > affected.json
-          bun -e 'const result = await Bun.file("affected.json").json(); console.log(`app=${result.packages.count > 0}`)' >> "$GITHUB_OUTPUT"
-
  unit:
    name: unit (${{ matrix.settings.name }})
    strategy:
@@ -71,7 +41,6 @@ jobs:
        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
        with:
          token: ${{ secrets.GITHUB_TOKEN }}
-          fetch-depth: 0

      - name: Setup Node
        uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
@@ -111,34 +80,9 @@ jobs:

      - name: Run unit tests
        timeout-minutes: 20
-        run: |
-          if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
-            GITHUB_ACTIONS=false bun turbo test
-            exit 0
-          fi
-          GITHUB_ACTIONS=false bun turbo test --affected
+        run: GITHUB_ACTIONS=false bun turbo test
        env:
          OPENCODE_EXPERIMENTAL_DISABLE_FILEWATCHER: ${{ runner.os == 'Windows' && 'true' || 'false' }}
-          TURBO_SCM_BASE: ${{ github.event_name == 'pull_request' && format('{0}^1', github.sha) || github.event.before }}
-          TURBO_SCM_HEAD: ${{ github.sha }}
-
-      - name: Verify published codemode package
-        if: runner.os == 'Linux'
-        working-directory: packages/codemode
-        run: bun run script/publish.ts --dry-run
-
-      - name: Verify packed workerd SDK
-        if: runner.os == 'Linux'
-        timeout-minutes: 15
-        run: |
-          if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
-            bun turbo verify:package --filter=@opencode-ai/sdk
-            exit 0
-          fi
-          bun turbo verify:package --affected --filter=@opencode-ai/sdk
-        env:
-          TURBO_SCM_BASE: ${{ github.event_name == 'pull_request' && format('{0}^1', github.sha) || github.event.before }}
-          TURBO_SCM_HEAD: ${{ github.sha }}

      - name: Verify compiled service lifecycle
        if: always()
@@ -178,7 +122,7 @@ jobs:

  e2e:
    name: e2e (${{ matrix.settings.name }})
-    needs: affected
+    if: github.ref_name != 'v2' && github.head_ref != 'v2'
    strategy:
      fail-fast: false
      matrix:
@@ -189,38 +133,32 @@ jobs:
            host: blacksmith-4vcpu-windows-2025
    runs-on: ${{ matrix.settings.host }}
    env:
-      E2E_ENABLED: ${{ needs.affected.outputs.app == 'true' && github.ref_name != 'v2' && github.head_ref != 'v2' }}
      PLAYWRIGHT_BROWSERS_PATH: ${{ github.workspace }}/.playwright-browsers
    defaults:
      run:
        shell: bash
    steps:
      - name: Checkout repository
-        if: env.E2E_ENABLED == 'true'
        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
        with:
          token: ${{ secrets.GITHUB_TOKEN }}

      - name: Setup Node
-        if: env.E2E_ENABLED == 'true'
        uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
        with:
          # Playwright 1.59 hangs while extracting Chromium with Node 24.16.
          node-version: "24.15"

      - name: Setup Bun
-        if: env.E2E_ENABLED == 'true'
        uses: ./.github/actions/setup-bun

      - name: Read Playwright version
-        if: env.E2E_ENABLED == 'true'
        id: playwright-version
        run: |
          version=$(node -e 'console.log(require("./package.json").workspaces.catalog["@playwright/test"])')
          echo "version=$version" >> "$GITHUB_OUTPUT"

      - name: Cache Playwright browsers
-        if: env.E2E_ENABLED == 'true'
        id: playwright-cache
        uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
        with:
@@ -228,30 +166,23 @@ jobs:
          key: ${{ runner.os }}-${{ runner.arch }}-playwright-${{ steps.playwright-version.outputs.version }}-chromium

      - name: Install Playwright system dependencies
-        if: env.E2E_ENABLED == 'true' && runner.os == 'Linux'
+        if: runner.os == 'Linux'
        working-directory: packages/app
        run: bunx playwright install-deps chromium

      - name: Install Playwright browsers
-        if: env.E2E_ENABLED == 'true' && steps.playwright-cache.outputs.cache-hit != 'true'
+        if: steps.playwright-cache.outputs.cache-hit != 'true'
        working-directory: packages/app
        run: bunx playwright install chromium

-      - name: Run app e2e tests against production build
-        if: env.E2E_ENABLED == 'true'
-        run: bun --cwd packages/app test:e2e:built
+      - name: Run app e2e tests
+        run: bun --cwd packages/app test:e2e:local
        env:
          CI: true
        timeout-minutes: 30

-      - name: Verify service worker precaching and upgrades
-        if: env.E2E_ENABLED == 'true'
-        working-directory: packages/app
-        run: bunx playwright test --config e2e/service-worker/playwright.config.ts
-        timeout-minutes: 5
-
      - name: Upload Playwright artifacts
-        if: always() && env.E2E_ENABLED == 'true'
+        if: always()
        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
        with:
          name: playwright-${{ matrix.settings.name }}-${{ github.run_attempt }}
@@ -2,7 +2,7 @@
 exact = true
 # Only install newly resolved package versions published at least 3 days ago.
 minimumReleaseAge = 259200
-minimumReleaseAgeExcludes = ["@ai-sdk/amazon-bedrock", "@ai-sdk/anthropic", "@brendonovich/vite-plugin-opencode", "@opencode-ai/sdk", "@opencode-ai/pty", "@opencode-ai/pty-darwin-arm64", "@opencode-ai/pty-darwin-x64", "@opencode-ai/pty-linux-arm64-gnu", "@opencode-ai/pty-linux-arm64-musl", "@opencode-ai/pty-linux-x64-gnu", "@opencode-ai/pty-linux-x64-musl", "@opentui/core", "@opentui/core-darwin-arm64", "@opentui/core-darwin-x64", "@opentui/core-linux-arm64", "@opentui/core-linux-arm64-musl", "@opentui/core-linux-x64", "@opentui/core-linux-x64-musl", "@opentui/core-win32-arm64", "@opentui/core-win32-x64", "@opentui/keymap", "@opentui/solid", "opentui-spinner", "gitlab-ai-provider", "opencode-gitlab-auth", "@ff-labs/fff-node", "@ff-labs/fff-bun", "@ff-labs/fff-bin-darwin-arm64", "@ff-labs/fff-bin-darwin-x64", "@ff-labs/fff-bin-linux-arm64-gnu", "@ff-labs/fff-bin-linux-arm64-musl", "@ff-labs/fff-bin-linux-x64-gnu", "@ff-labs/fff-bin-linux-x64-musl", "@ff-labs/fff-bin-win32-arm64", "@ff-labs/fff-bin-win32-x64", "@pierre/diffs", "@pierre/theming", "app-builder-lib", "dmg-builder", "electron-builder", "electron-publish", "blume"]
+minimumReleaseAgeExcludes = ["@ai-sdk/amazon-bedrock", "@ai-sdk/anthropic", "@opencode-ai/sdk", "@opentui/core", "@opentui/core-darwin-arm64", "@opentui/core-darwin-x64", "@opentui/core-linux-arm64", "@opentui/core-linux-arm64-musl", "@opentui/core-linux-x64", "@opentui/core-linux-x64-musl", "@opentui/core-win32-arm64", "@opentui/core-win32-x64", "@opentui/keymap", "@opentui/solid", "opentui-spinner", "gitlab-ai-provider", "opencode-gitlab-auth", "@ff-labs/fff-node", "@ff-labs/fff-bun", "@ff-labs/fff-bin-darwin-arm64", "@ff-labs/fff-bin-darwin-x64", "@ff-labs/fff-bin-linux-arm64-gnu", "@ff-labs/fff-bin-linux-arm64-musl", "@ff-labs/fff-bin-linux-x64-gnu", "@ff-labs/fff-bin-linux-x64-musl", "@ff-labs/fff-bin-win32-arm64", "@ff-labs/fff-bin-win32-x64", "@pierre/diffs", "@pierre/theming", "app-builder-lib", "dmg-builder", "electron-builder", "electron-publish", "blume"]

 [test]
 root = "./do-not-run-tests-from-root"
@@ -1,8 +1,8 @@
 {
  "nodeModules": {
-    "x86_64-linux": "sha256-agxF1Ab8P84tijHvw7jaG+9UZm/dJZhHw8CGNlQQqYc=",
-    "aarch64-linux": "sha256-Kg6CkhoEEsb+ZOGPN/rnNoirLdGSyQl6HU/Fy8t9Mqs=",
-    "aarch64-darwin": "sha256-SnxBpmcfPL2p3OYmSuG5Y/aFW8PeEOLhfQrxAs/53fs=",
-    "x86_64-darwin": "sha256-Dnxl1EcfiE7/rlHmPkko0BKc6hHVk6rQ/2h/PWuLfNo="
+    "x86_64-linux": "sha256-xRvq8FkSjn+1q+1wcab+jAmQJdo8lHF8OGnzU+xPLyI=",
+    "aarch64-linux": "sha256-uAwwtOz81LLimTqiDH6E1W65srSMLGk1QV0cUJYanj0=",
+    "aarch64-darwin": "sha256-F28kZvtYRb+ri5T8JoIG/VMK6J7ad5R3QivwJDLtmWA=",
+    "x86_64-darwin": "sha256-NWNoBjwQTTNAESdFWItT4UYc+fTLD1RIe/Ibbqtem7Q="
  }
 }
@@ -39,20 +39,20 @@
      "packages/stats/*"
    ],
    "catalog": {
-      "@effect/opentelemetry": "4.0.0-rc.111",
-      "@effect/platform-node": "4.0.0-rc.111",
-      "@effect/platform-node-shared": "4.0.0-rc.111",
-      "@effect/sql-sqlite-bun": "4.0.0-rc.111",
+      "@effect/opentelemetry": "4.0.0-rc.110",
+      "@effect/platform-node": "4.0.0-rc.110",
+      "@effect/platform-node-shared": "4.0.0-rc.110",
+      "@effect/sql-sqlite-bun": "4.0.0-rc.110",
      "@npmcli/arborist": "9.4.0",
      "@types/bun": "1.3.13",
      "@types/cross-spawn": "6.0.6",
      "@octokit/rest": "22.0.0",
      "@hono/standard-validator": "0.2.0",
      "@hono/zod-validator": "0.4.2",
-      "@opentui/core": "0.5.8",
-      "@opentui/keymap": "0.5.8",
-      "@opentui/solid": "0.5.8",
-      "@tanstack/solid-virtual": "3.13.37",
+      "@opentui/core": "0.5.6",
+      "@opentui/keymap": "0.5.6",
+      "@opentui/solid": "0.5.6",
+      "@tanstack/solid-virtual": "3.13.32",
      "@shikijs/stream": "4.2.0",
      "@standard-schema/spec": "1.1.0",
      "ulid": "3.0.1",
@@ -71,9 +71,9 @@
      "@tailwindcss/vite": "4.1.11",
      "diff": "8.0.2",
      "dompurify": "3.3.1",
-      "drizzle-kit": "1.0.0-rc.5-ab785fc",
-      "drizzle-orm": "1.0.0-rc.5-169397b",
-      "effect": "4.0.0-rc.111",
+      "drizzle-kit": "1.0.0-rc.2",
+      "drizzle-orm": "1.0.0-rc.2",
+      "effect": "4.0.0-rc.110",
      "ai": "6.0.168",
      "cross-spawn": "7.0.6",
      "hono": "4.10.7",
@@ -166,6 +166,7 @@
    "@npmcli/agent@4.0.2": "patches/@npmcli%2Fagent@4.0.2.patch",
    "@silvia-odwyer/photon-node@0.3.4": "patches/@silvia-odwyer%2Fphoton-node@0.3.4.patch",
    "@standard-community/standard-openapi@0.2.9": "patches/@standard-community%2Fstandard-openapi@0.2.9.patch",
+    "drizzle-orm@1.0.0-rc.2": "patches/drizzle-orm@1.0.0-rc.2.patch",
    "solid-js@1.9.10": "patches/solid-js@1.9.10.patch",
    "@ai-sdk/mistral@3.0.51": "patches/@ai-sdk%2Fmistral@3.0.51.patch",
    "gcp-metadata@8.1.2": "patches/gcp-metadata@8.1.2.patch",
@@ -173,7 +174,7 @@
    "@ai-sdk/google@3.0.73": "patches/@ai-sdk%2Fgoogle@3.0.73.patch",
    "@pierre/trees@1.0.0-beta.4": "patches/@pierre%2Ftrees@1.0.0-beta.4.patch",
    "@modelcontextprotocol/sdk@1.29.0": "patches/@modelcontextprotocol%2Fsdk@1.29.0.patch",
-    "@tanstack/virtual-core@3.17.8": "patches/@tanstack%2Fvirtual-core@3.17.8.patch",
+    "@tanstack/virtual-core@3.17.3": "patches/@tanstack%2Fvirtual-core@3.17.3.patch",
    "@ff-labs/fff-bun@0.10.5": "patches/@ff-labs%2Ffff-bun@0.10.5.patch"
  }
 }
@@ -19,14 +19,6 @@ Per-type constructors live on the type, not as top-level re-exports. Use `Messag
 - Use `testEffect(...)` from `test/lib/effect.ts` for tests requiring Effect layers.
 - Keep provider tests fixture-first. Live provider calls must stay behind `RECORD=true` and required API-key checks.

-## Errors
-
- `AIError` wraps a union of tagged reason errors. It stores only `reason`, derives `message` from the reason, and exposes the reason as its `cause`.
- Each reason owns its readable `message`, category-specific fields, and optional `body`, `http`, and underlying exception in `cause`.
- `reason.body` is the sole original-response or triggering-event payload field. Preserve original text before schema decoding removes fields; do not replace the complete event with only its nested error.
- `reason.http` describes an observed HTTP response with required `url`, `status`, and response `headers`. Do not invent status codes or derive a separate request ID from headers.
- Reclassification and transport recovery must preserve the reason's body, HTTP context, and underlying cause. Error `message` and `cause` are non-enumerable: copy them explicitly when constructing an enriched reason with its constructor or `AIErrorReason.make`.
-
 ## Architecture

 This package is an Effect Schema-first LLM core. The Schema classes in `src/schema/` are the canonical runtime data model. Convenience functions in `src/llm.ts` are thin constructors that return those same Schema class instances; they should improve callsites without creating a second model.
@@ -221,7 +213,7 @@ Errors must be expressed as `ToolFailure`. The runtime catches it and emits a `t
 - Input failed the `parameters` Schema.
 - The handler returned a `ToolFailure`.

-Provider-defined / hosted tools (Anthropic `web_search` / `code_execution` / `web_fetch`, OpenAI Responses `web_search_call` / `file_search_call` / `code_interpreter_call` / `mcp_call` / `image_generation_call` / `computer_use_call`) pass through the runtime untouched:
+Provider-defined / hosted tools (Anthropic `web_search` / `code_execution` / `web_fetch`, OpenAI Responses `web_search_call` / `file_search_call` / `code_interpreter_call` / `mcp_call` / `local_shell_call` / `image_generation_call` / `computer_use_call`) pass through the runtime untouched:

 - Routes surface the model's call as a `tool-call` event with `providerExecuted: true`, and the provider's result as a matching `tool-result` event with `providerExecuted: true`.
 - Callers detect `providerExecuted` on `tool-call` and **skip local dispatch** — no handler is invoked and no `tool-error` is raised for "unknown tool". The provider already executed it.
@@ -157,9 +157,9 @@ const PROVIDERS: ReadonlyArray<Provider> = [
    id: "togetherai",
    label: "TogetherAI",
    tier: "compatible",
-    note: "Native Together AI text/tool recorded tests",
-    vars: [{ name: "TOGETHER_API_KEY" }],
-    validate: (env) => validateBearer("https://api.together.xyz/v1/models", Redacted.make(env.TOGETHER_API_KEY)),
+    note: "Existing OpenAI-compatible text/tool recorded tests",
+    vars: [{ name: "TOGETHER_AI_API_KEY" }],
+    validate: (env) => validateBearer("https://api.together.xyz/v1/models", Redacted.make(env.TOGETHER_AI_API_KEY)),
  },
  {
    id: "minimax",
@@ -200,8 +200,8 @@ const PROVIDERS: ReadonlyArray<Provider> = [
  {
    id: "cerebras",
    label: "Cerebras",
-    tier: "compatible",
-    note: "Native Cerebras text/tool/tool-loop recorded tests",
+    tier: "optional",
+    note: "OpenAI-compatible bridge",
    vars: [{ name: "CEREBRAS_API_KEY" }],
    validate: (env) => validateBearer("https://api.cerebras.ai/v1/models", Redacted.make(env.CEREBRAS_API_KEY)),
  },
@@ -36,12 +36,7 @@ const resolve = (policy: CachePolicy | undefined): CachePolicyObject => {
 // Protocols whose wire format ignores inline cache markers (OpenAI's implicit
 // prefix caching, Gemini's implicit + out-of-band CachedContent). Skip the
 // whole policy pass for these — emitting hints would be harmless but pointless.
-const RESPECTS_INLINE_HINTS = new Set([
-  "anthropic-messages",
-  "google-vertex-messages",
-  "bedrock-converse",
-  "openrouter",
-])
+const RESPECTS_INLINE_HINTS = new Set(["anthropic-messages", "bedrock-converse", "openrouter"])

 const makeHint = (ttlSeconds: number | undefined): CacheHint =>
  ttlSeconds !== undefined ? new CacheHint({ type: "ephemeral", ttlSeconds }) : new CacheHint({ type: "ephemeral" })
@@ -1,8 +1,7 @@
 import { Context, Effect, Layer } from "effect"
 import { RequestExecutor } from "./route/executor.js"
-import { mergeHttpOptions, type AIError } from "./schema/index.js"
-import { sanitizeSurrogates } from "./utils/sanitize.js"
 import type { ImageOptions, ImageRequest, ImageRequestFor, ImageResponse } from "./image.js"
+import type { AIError } from "./schema/index.js"

 export type Execute = RequestExecutor.Interface["execute"]

@@ -27,18 +26,7 @@ export const layer: Layer.Layer<Service, never, RequestExecutor.Service> = Layer
  Effect.gen(function* () {
    const executor = yield* RequestExecutor.Service
    return Service.of({
-      generate: (request) =>
-        request.model.route.generate(
-          {
-            ...sanitizeSurrogates({
-              ...request,
-              model: undefined,
-              http: mergeHttpOptions(request.model.http, request.http),
-            }),
-            model: request.model,
-          },
-          executor.execute,
-        ),
+      generate: (request) => request.model.route.generate(request, executor.execute),
    })
  }),
 )
@@ -1,7 +1,7 @@
 import { Effect, Schema } from "effect"
 import {
  HttpOptions,
-  InvalidRequestError,
+  InvalidRequestReason,
  AIError,
  ModelID,
  ProviderID,
@@ -158,10 +158,9 @@ export function generate(input: ImageRequest | ImageRequestInput) {
    try: () => (input instanceof ImageRequest ? input : request(input)),
    catch: (error) =>
      new AIError({
-        reason: new InvalidRequestError({
-          message: error instanceof Error ? error.message : String(error),
-          cause: error,
-        }),
+        module: "Image",
+        method: "generate",
+        reason: new InvalidRequestReason({ message: error instanceof Error ? error.message : String(error) }),
      }),
  }).pipe(Effect.flatMap((request) => ImageClient.generate(request as unknown as ImageRequestFor<ImageOptions>)))
 }
@@ -3,7 +3,7 @@ import { LLMClient, Service } from "./route/client.js"
 import {
  GenerationOptions,
  HttpOptions,
-  InvalidProviderOutputError,
+  InvalidProviderOutputReason,
  AIError,
  LLMEvent,
  LLMRequest,
@@ -116,7 +116,9 @@ const runGenerateObject = Effect.fn("LLM.generateObject")(function* (
  )
  if (!call || !LLMEvent.is.toolCall(call))
    return yield* new AIError({
-      reason: new InvalidProviderOutputError({
+      module: "LLM",
+      method: "generateObject",
+      reason: new InvalidProviderOutputReason({
        message: `generateObject: model did not call the forced \`${GENERATE_OBJECT_TOOL_NAME}\` tool`,
      }),
    })
@@ -124,9 +126,10 @@ const runGenerateObject = Effect.fn("LLM.generateObject")(function* (
    Effect.mapError(
      (error) =>
        new AIError({
-          reason: new InvalidProviderOutputError({
+          module: "LLM",
+          method: "generateObject",
+          reason: new InvalidProviderOutputReason({
            message: `generateObject: tool input failed schema decode: ${error.message}`,
-            cause: error,
          }),
        }),
    ),
@@ -1,5 +1,4 @@
-import { Buffer } from "node:buffer"
-import { Effect, Option, Schema } from "effect"
+import { Effect, Schema } from "effect"
 import { Tool } from "@opencode-ai/schema/tool"
 import { Route } from "../route/client.js"
 import { Auth } from "../route/auth.js"
@@ -42,7 +41,6 @@ const SSE_EVENTS = new Set([
  "content_block_start",
  "content_block_delta",
  "content_block_stop",
-  "ping",
  "error",
 ])
 export const framing = Framing.sseEvents(SSE_EVENTS)
@@ -55,7 +53,7 @@ export type ThinkingInput =
  | {
      readonly type: "disabled"
    }
-  | ({ readonly type: "enabled"; readonly display?: "summarized" | "omitted" } & (
+  | ({ readonly type: "enabled" } & (
      | { readonly budgetTokens: number; readonly budget_tokens?: number }
      | { readonly budgetTokens?: number; readonly budget_tokens: number }
    ))
@@ -64,27 +62,6 @@ export interface OptionsInput {
  readonly [key: string]: unknown
  readonly thinking?: ThinkingInput
  readonly effort?: string
-  readonly service_tier?: "auto" | "standard_only"
-  readonly serviceTier?: "auto" | "standard_only"
-  // SDK Metadata:2649 {user_id?: string | null}
-  readonly metadata?: { readonly user_id?: string | null }
-  // SDK MessageCreateParamsContainer:2596 ContainerParams|string
-  readonly container?:
-    | string
-    | { readonly id?: string | null; readonly skills?: ReadonlyArray<Record<string, unknown>> | null }
-  readonly inference_geo?: string | null
-  readonly inferenceGeo?: string | null
-  readonly cache_control?: { readonly type: "ephemeral"; readonly ttl?: "5m" | "1h" }
-  readonly cacheControl?: { readonly type: "ephemeral"; readonly ttl?: "5m" | "1h" }
-  // SDK OutputConfig:2684 {effort, format: JSONOutputFormat}
-  readonly output_config?: {
-    readonly effort?: string | null
-    readonly format?: { readonly type: "json_schema"; readonly schema: Record<string, unknown> } | null
-  }
-  readonly outputConfig?: {
-    readonly effort?: string | null
-    readonly format?: { readonly type: "json_schema"; readonly schema: Record<string, unknown> } | null
-  }
 }

 export type ProviderOptionsInput = OptionsInput
@@ -104,68 +81,32 @@ const AnthropicTextBlock = Schema.Struct({
 })
 type AnthropicTextBlock = Schema.Schema.Type<typeof AnthropicTextBlock>

-// SDK: Base64ImageSource:201 {type:"base64", media_type:"image/jpeg"|... , data}, URLImageSource:3817 {type:"url", url}, FileImageSource:2350 {type:"file", file_id}
-// SDK: ImageBlockParam:2356 {source: Base64|URL|File, cache_control, transformations:2381 {oversized_image?}}
-const AnthropicBase64ImageSource = Schema.Struct({
-  type: Schema.tag("base64"),
-  media_type: Schema.String,
-  data: Schema.String,
-})
-const AnthropicURLImageSource = Schema.Struct({ type: Schema.tag("url"), url: Schema.String })
-const AnthropicFileImageSource = Schema.Struct({ type: Schema.tag("file"), file_id: Schema.String })
-const AnthropicImageSource = Schema.Union([
-  AnthropicBase64ImageSource,
-  AnthropicURLImageSource,
-  AnthropicFileImageSource,
-])
-const AnthropicImageTransformations = Schema.Struct({
-  oversized_image: Schema.optional(Schema.Literals(["downsize", "error"])),
-})
-
 const AnthropicImageBlock = Schema.Struct({
  type: Schema.tag("image"),
-  source: AnthropicImageSource,
+  source: Schema.Struct({
+    type: Schema.tag("base64"),
+    media_type: Schema.String,
+    data: Schema.String,
+  }),
  cache_control: Schema.optional(AnthropicCacheControl),
-  transformations: Schema.optional(AnthropicImageTransformations),
 })
 type AnthropicImageBlock = Schema.Schema.Type<typeof AnthropicImageBlock>

-// SDK: Base64PDFSource:209 {type:"base64", media_type:"application/pdf", data}, PlainTextSource:2716 {type:"text", media_type:"text/plain", data},
-// SDK: URLPDFSource:3823 {type:"url", url}, FileDocumentSource:2344 {type:"file", file_id}, ContentBlockSource:2266 {type:"content", content}
-// SDK: DocumentBlockParam:2297 {source: 5-way union, cache_control, citations, context, title}
-const AnthropicBase64PDFSource = Schema.Struct({
-  type: Schema.tag("base64"),
-  media_type: Schema.Literal("application/pdf"),
-  data: Schema.String,
-})
-const AnthropicPlainTextSource = Schema.Struct({
-  type: Schema.tag("text"),
-  media_type: Schema.Literal("text/plain"),
-  data: Schema.String,
-})
-const AnthropicURLPDFSource = Schema.Struct({ type: Schema.tag("url"), url: Schema.String })
-const AnthropicFileDocumentSource = Schema.Struct({ type: Schema.tag("file"), file_id: Schema.String })
-const AnthropicDocumentSource = Schema.Union([
-  AnthropicBase64PDFSource,
-  AnthropicPlainTextSource,
-  AnthropicURLPDFSource,
-  AnthropicFileDocumentSource,
-])
-
 const AnthropicDocumentBlock = Schema.Struct({
  type: Schema.tag("document"),
-  source: AnthropicDocumentSource,
+  source: Schema.Struct({
+    type: Schema.tag("base64"),
+    media_type: Schema.Literal("application/pdf"),
+    data: Schema.String,
+  }),
  cache_control: Schema.optional(AnthropicCacheControl),
-  title: Schema.optional(Schema.String),
-  context: Schema.optional(Schema.String),
-  citations: Schema.optional(Schema.Struct({ enabled: Schema.Boolean })),
 })
 type AnthropicDocumentBlock = Schema.Schema.Type<typeof AnthropicDocumentBlock>

 const AnthropicThinkingBlock = Schema.Struct({
  type: Schema.tag("thinking"),
  thinking: Schema.String,
-  signature: Schema.String,
+  signature: Schema.optional(Schema.String),
  cache_control: Schema.optional(AnthropicCacheControl),
 })

@@ -263,22 +204,14 @@ const AnthropicTool = Schema.Struct({
 type AnthropicTool = Schema.Schema.Type<typeof AnthropicTool>

 const AnthropicToolChoice = Schema.Union([
-  Schema.Struct({
-    type: Schema.Literals(["auto", "any", "none"]),
-    disable_parallel_tool_use: Schema.optional(Schema.Boolean),
-  }),
-  Schema.Struct({
-    type: Schema.tag("tool"),
-    name: Schema.String,
-    disable_parallel_tool_use: Schema.optional(Schema.Boolean),
-  }),
+  Schema.Struct({ type: Schema.Literals(["auto", "any", "none"]) }),
+  Schema.Struct({ type: Schema.tag("tool"), name: Schema.String }),
 ])

 const AnthropicThinking = Schema.Union([
  Schema.Struct({
    type: Schema.tag("enabled"),
    budget_tokens: Schema.Number,
-    display: Schema.optional(Schema.Literals(["summarized", "omitted"])),
  }),
  Schema.Struct({
    type: Schema.tag("adaptive"),
@@ -289,28 +222,10 @@ const AnthropicThinking = Schema.Union([
  }),
 ])

-// SDK OutputConfig:2684 {effort?: "low"|"medium"|"high"|"xhigh"|"max"|null, format?: JSONOutputFormat:2399}
-const AnthropicJsonOutputFormat = Schema.Struct({
-  type: Schema.Literal("json_schema"),
-  schema: JsonObject,
-})
 const AnthropicOutputConfig = Schema.Struct({
  effort: Schema.optional(Schema.String),
-  format: Schema.optional(Schema.NullOr(AnthropicJsonOutputFormat)),
 })

-// SDK Metadata:2649 {user_id?: string|null}
-const AnthropicMetadata = Schema.Struct({ user_id: optionalNull(Schema.String) })
-
-// SDK MessageCreateParamsContainer:2596 ContainerParams|string; ContainerParams:2172 {id?, skills?}
-const AnthropicContainer = Schema.Union([
-  Schema.String,
-  Schema.Struct({
-    id: optionalNull(Schema.String),
-    skills: optionalNull(Schema.Array(JsonObject)),
-  }),
-])
-
 const AnthropicBodyFields = {
  model: Schema.String,
  system: optionalArray(AnthropicTextBlock),
@@ -325,12 +240,6 @@ const AnthropicBodyFields = {
  stop_sequences: optionalArray(Schema.String),
  thinking: Schema.optional(AnthropicThinking),
  output_config: Schema.optional(AnthropicOutputConfig),
-  // SDK top-level passthrough: cache_control:4638, container:4643, inference_geo:4649, metadata:4654, service_tier:4670
-  cache_control: Schema.optional(AnthropicCacheControl),
-  container: Schema.optional(Schema.NullOr(AnthropicContainer)),
-  inference_geo: Schema.optional(Schema.NullOr(Schema.String)),
-  metadata: Schema.optional(AnthropicMetadata),
-  service_tier: Schema.optional(Schema.Literals(["auto", "standard_only"])),
 }
 export const AnthropicMessagesBody = Schema.Struct(AnthropicBodyFields)
 export type AnthropicMessagesBody = Schema.Schema.Type<typeof AnthropicMessagesBody>
@@ -373,8 +282,6 @@ const AnthropicStreamBlock = Schema.Struct({
  tool_use_id: Schema.optional(Schema.String),
  content: Schema.optional(Schema.Unknown),
 })
-type AnthropicStreamBlock = Schema.Schema.Type<typeof AnthropicStreamBlock>
-const decodeAnthropicStreamBlock = Schema.decodeUnknownOption(AnthropicStreamBlock)

 const AnthropicStreamDelta = Schema.Struct({
  type: Schema.optional(Schema.String),
@@ -385,15 +292,13 @@ const AnthropicStreamDelta = Schema.Struct({
  stop_reason: optionalNull(Schema.String),
  stop_sequence: optionalNull(Schema.String),
 })
-type AnthropicStreamDelta = Schema.Schema.Type<typeof AnthropicStreamDelta>
-const decodeAnthropicStreamDelta = Schema.decodeUnknownOption(AnthropicStreamDelta)

 const AnthropicEvent = Schema.Struct({
  type: Schema.String,
  index: Schema.optional(Schema.Number),
  message: Schema.optional(Schema.Struct({ usage: Schema.optional(AnthropicUsage) })),
-  content_block: Schema.optional(Schema.Unknown),
-  delta: Schema.optional(Schema.Unknown),
+  content_block: Schema.optional(AnthropicStreamBlock),
+  delta: Schema.optional(AnthropicStreamDelta),
  usage: Schema.optional(AnthropicUsage),
  // `type` and `message` are both required per Anthropic's spec, but
  // OpenAI-compatible proxies and gateway translations occasionally drop one
@@ -406,7 +311,6 @@ const AnthropicEvent = Schema.Struct({
 type AnthropicEvent = Schema.Schema.Type<typeof AnthropicEvent>

 interface ParserState {
-  readonly providerMetadataKey: string
  readonly tools: ToolStream.State<number>
  readonly reasoningSignatures: Readonly<Record<number, string>>
  readonly usage?: Usage
@@ -441,18 +345,18 @@ const cacheControl = (breakpoints: Cache.Breakpoints, cache: CacheHint | undefin
  return Cache.ttlBucket(cache.ttlSeconds) === "1h" ? EPHEMERAL_1H : EPHEMERAL_5M
 }

-const providerMetadata = (key: string, metadata: Record<string, unknown>): ProviderMetadata => ({ [key]: metadata })
+const anthropicMetadata = (metadata: Record<string, unknown>): ProviderMetadata => ({ anthropic: metadata })

-const signatureFromMetadata = (metadata: ProviderMetadata | undefined, key: string): string | undefined => {
-  const provider = metadata?.[key]
-  if (!ProviderShared.isRecord(provider)) return undefined
-  return typeof provider.signature === "string" ? provider.signature : undefined
+const signatureFromMetadata = (metadata: ProviderMetadata | undefined): string | undefined => {
+  const anthropic = metadata?.anthropic
+  if (!ProviderShared.isRecord(anthropic)) return undefined
+  return typeof anthropic.signature === "string" ? anthropic.signature : undefined
 }

-const redactedDataFromMetadata = (metadata: ProviderMetadata | undefined, key: string): string | undefined => {
-  const provider = metadata?.[key]
-  if (!ProviderShared.isRecord(provider)) return undefined
-  return typeof provider.redactedData === "string" ? provider.redactedData : undefined
+const redactedDataFromMetadata = (metadata: ProviderMetadata | undefined): string | undefined => {
+  const anthropic = metadata?.anthropic
+  if (!ProviderShared.isRecord(anthropic)) return undefined
+  return typeof anthropic.redactedData === "string" ? anthropic.redactedData : undefined
 }

 const lowerTool = (breakpoints: Cache.Breakpoints, tool: ToolDefinition, inputSchema: JsonSchema): AnthropicTool => ({
@@ -464,26 +368,10 @@ const lowerTool = (breakpoints: Cache.Breakpoints, tool: ToolDefinition, inputSc

 const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>) =>
  ProviderShared.matchToolChoice("Anthropic Messages", toolChoice, {
-    auto: () => ({
-      type: "auto" as const,
-      ...(toolChoice.disableParallelToolUse === undefined
-        ? {}
-        : { disable_parallel_tool_use: toolChoice.disableParallelToolUse }),
-    }),
+    auto: () => ({ type: "auto" as const }),
    none: () => ({ type: "none" as const }),
-    required: () => ({
-      type: "any" as const,
-      ...(toolChoice.disableParallelToolUse === undefined
-        ? {}
-        : { disable_parallel_tool_use: toolChoice.disableParallelToolUse }),
-    }),
-    tool: (name) => ({
-      type: "tool" as const,
-      name,
-      ...(toolChoice.disableParallelToolUse === undefined
-        ? {}
-        : { disable_parallel_tool_use: toolChoice.disableParallelToolUse }),
-    }),
+    required: () => ({ type: "any" as const }),
+    tool: (name) => ({ type: "tool" as const, name }),
  })

 const scrubToolCallID = (id: string) => id.replace(/[^a-zA-Z0-9_-]/g, "_")
@@ -512,167 +400,17 @@ const serverToolResultType = (name: string): AnthropicServerToolResultType | und
  return undefined
 }

-const lowerServerToolResult = Effect.fn("AnthropicMessages.lowerServerToolResult")(function* (
-  part: ToolResultPart,
-  providerMetadataKey: string,
-) {
+const lowerServerToolResult = Effect.fn("AnthropicMessages.lowerServerToolResult")(function* (part: ToolResultPart) {
  const wireType = serverToolResultType(part.name)
  if (!wireType)
    return yield* invalid(`Anthropic Messages does not know how to round-trip server tool result for ${part.name}`)
  // Prefer the provider-owned replay payload; fall back to the result value for
  // histories constructed directly from provider events.
-  const payload = part.providerMetadata?.[providerMetadataKey]?.["result"] ?? part.result.value
-  return {
-    type: wireType,
-    tool_use_id: scrubToolCallID(part.id),
-    content: payload,
-  } satisfies AnthropicServerToolResultBlock
+  const payload = part.providerMetadata?.anthropic?.["result"] ?? part.result.value
+  return { type: wireType, tool_use_id: scrubToolCallID(part.id), content: payload } satisfies AnthropicServerToolResultBlock
 })

-const fileIdFromMetadata = (metadata: MediaPart["metadata"]): string | undefined => {
-  if (!ProviderShared.isRecord(metadata)) return undefined
-  const anthropic = metadata.anthropic
-  if (ProviderShared.isRecord(anthropic)) {
-    if (typeof anthropic.file_id === "string") return anthropic.file_id
-    if (typeof anthropic.fileId === "string") return anthropic.fileId
-  }
-  if (typeof metadata.file_id === "string") return metadata.file_id
-  if (typeof metadata.fileId === "string") return metadata.fileId
-  return undefined
-}
-
-const transformationsFromMetadata = (
-  metadata: MediaPart["metadata"],
-): AnthropicImageBlock["transformations"] | undefined => {
-  if (!ProviderShared.isRecord(metadata)) return undefined
-  const anthropic = ProviderShared.isRecord(metadata.anthropic) ? metadata.anthropic : undefined
-  const raw = anthropic?.transformations ?? metadata.transformations
-  if (ProviderShared.isRecord(raw)) {
-    const value = raw.oversized_image
-    if (value === "downsize" || value === "error") return { oversized_image: value }
-  }
-  if (anthropic && (anthropic.oversized_image === "downsize" || anthropic.oversized_image === "error"))
-    return { oversized_image: anthropic.oversized_image }
-  return undefined
-}
-
-const documentTitleFromPart = (part: MediaPart): string | undefined => {
-  if (ProviderShared.isRecord(part.metadata)) {
-    const anthropic = part.metadata.anthropic
-    if (ProviderShared.isRecord(anthropic) && typeof anthropic.title === "string") return anthropic.title
-    if (typeof part.metadata.title === "string") return part.metadata.title
-  }
-  if (typeof part.filename === "string" && part.filename.length > 0) return part.filename
-  return undefined
-}
-
-const documentContextFromMetadata = (metadata: MediaPart["metadata"]): string | undefined => {
-  if (!ProviderShared.isRecord(metadata)) return undefined
-  const anthropic = ProviderShared.isRecord(metadata.anthropic) ? metadata.anthropic : undefined
-  if (anthropic && typeof anthropic.context === "string") return anthropic.context
-  if (typeof metadata.context === "string") return metadata.context
-  return undefined
-}
-
-const citationsFromMetadata = (metadata: MediaPart["metadata"]): AnthropicDocumentBlock["citations"] | undefined => {
-  if (!ProviderShared.isRecord(metadata)) return undefined
-  const raw = ProviderShared.isRecord(metadata.anthropic)
-    ? (metadata.anthropic.citations ?? metadata.citations)
-    : metadata.citations
-  if (ProviderShared.isRecord(raw) && typeof raw.enabled === "boolean") return { enabled: raw.enabled }
-  return undefined
-}
-
-const isHttpUrl = (value: string) => /^https?:\/\//i.test(value.trim())
-
-const lowerMedia = Effect.fn("AnthropicMessages.lowerMedia")(function* (
-  part: MediaPart,
-  breakpoints?: Cache.Breakpoints,
-) {
-  const mime = part.mediaType.toLowerCase()
-  const cacheControlValue = breakpoints ? cacheControl(breakpoints, part.cache) : undefined
-  const fileId = fileIdFromMetadata(part.metadata)
-
-  // SDK file sources: FileImageSource:2350 / FileDocumentSource:2344 {type:"file", file_id}
-  if (fileId) {
-    if (mime.startsWith("image/"))
-      return {
-        type: "image" as const,
-        source: { type: "file" as const, file_id: fileId },
-        ...(cacheControlValue === undefined ? {} : { cache_control: cacheControlValue }),
-        ...(transformationsFromMetadata(part.metadata) === undefined
-          ? {}
-          : { transformations: transformationsFromMetadata(part.metadata)! }),
-      } satisfies AnthropicImageBlock
-    return {
-      type: "document" as const,
-      source: { type: "file" as const, file_id: fileId },
-      ...(cacheControlValue === undefined ? {} : { cache_control: cacheControlValue }),
-      ...(documentTitleFromPart(part) === undefined ? {} : { title: documentTitleFromPart(part)! }),
-      ...(documentContextFromMetadata(part.metadata) === undefined
-        ? {}
-        : { context: documentContextFromMetadata(part.metadata)! }),
-      ...(citationsFromMetadata(part.metadata) === undefined
-        ? {}
-        : { citations: citationsFromMetadata(part.metadata)! }),
-    } satisfies AnthropicDocumentBlock
-  }
-
-  const rawString = typeof part.data === "string" ? part.data.trim() : undefined
-  // SDK URL sources: URLImageSource:3817 / URLPDFSource:3823 {type:"url", url}
-  if (rawString && isHttpUrl(rawString) && !rawString.startsWith("data:")) {
-    if (mime.startsWith("image/"))
-      return {
-        type: "image" as const,
-        source: { type: "url" as const, url: rawString },
-        ...(cacheControlValue === undefined ? {} : { cache_control: cacheControlValue }),
-        ...(transformationsFromMetadata(part.metadata) === undefined
-          ? {}
-          : { transformations: transformationsFromMetadata(part.metadata)! }),
-      } satisfies AnthropicImageBlock
-    if (mime === "application/pdf")
-      return {
-        type: "document" as const,
-        source: { type: "url" as const, url: rawString },
-        ...(cacheControlValue === undefined ? {} : { cache_control: cacheControlValue }),
-        ...(documentTitleFromPart(part) === undefined ? {} : { title: documentTitleFromPart(part)! }),
-        ...(documentContextFromMetadata(part.metadata) === undefined
-          ? {}
-          : { context: documentContextFromMetadata(part.metadata)! }),
-        ...(citationsFromMetadata(part.metadata) === undefined
-          ? {}
-          : { citations: citationsFromMetadata(part.metadata)! }),
-      } satisfies AnthropicDocumentBlock
-  }
-
-  // SDK PlainTextSource:2716 {type:"text", media_type:"text/plain", data}
-  if (mime === "text/plain") {
-    const textData =
-      typeof part.data !== "string"
-        ? Buffer.from(part.data).toString("utf8")
-        : part.data.startsWith("data:")
-          ? (() => {
-              const comma = part.data.indexOf(",")
-              const payload = comma >= 0 ? part.data.slice(comma + 1) : part.data
-              return part.data.includes(";base64")
-                ? Buffer.from(payload, "base64").toString("utf8")
-                : decodeURIComponent(payload)
-            })()
-          : part.data
-    return {
-      type: "document" as const,
-      source: { type: "text" as const, media_type: "text/plain" as const, data: textData },
-      ...(cacheControlValue === undefined ? {} : { cache_control: cacheControlValue }),
-      ...(documentTitleFromPart(part) === undefined ? {} : { title: documentTitleFromPart(part)! }),
-      ...(documentContextFromMetadata(part.metadata) === undefined
-        ? {}
-        : { context: documentContextFromMetadata(part.metadata)! }),
-      ...(citationsFromMetadata(part.metadata) === undefined
-        ? {}
-        : { citations: citationsFromMetadata(part.metadata)! }),
-    } satisfies AnthropicDocumentBlock
-  }
-
+const lowerMedia = Effect.fn("AnthropicMessages.lowerMedia")(function* (part: MediaPart) {
  const media = ProviderShared.normalizeMedia(part)
  if (media.mime === "application/pdf")
    return {
@@ -682,14 +420,6 @@ const lowerMedia = Effect.fn("AnthropicMessages.lowerMedia")(function* (
        media_type: "application/pdf" as const,
        data: media.base64,
      },
-      ...(cacheControlValue === undefined ? {} : { cache_control: cacheControlValue }),
-      ...(documentTitleFromPart(part) === undefined ? {} : { title: documentTitleFromPart(part)! }),
-      ...(documentContextFromMetadata(part.metadata) === undefined
-        ? {}
-        : { context: documentContextFromMetadata(part.metadata)! }),
-      ...(citationsFromMetadata(part.metadata) === undefined
-        ? {}
-        : { citations: citationsFromMetadata(part.metadata)! }),
    } satisfies AnthropicDocumentBlock
  if (!media.mime.startsWith("image/"))
    return yield* invalid(`Anthropic Messages does not support media type ${part.mediaType}`)
@@ -700,10 +430,6 @@ const lowerMedia = Effect.fn("AnthropicMessages.lowerMedia")(function* (
      media_type: media.mime,
      data: media.base64,
    },
-    ...(cacheControlValue === undefined ? {} : { cache_control: cacheControlValue }),
-    ...(transformationsFromMetadata(part.metadata) === undefined
-      ? {}
-      : { transformations: transformationsFromMetadata(part.metadata)! }),
  } satisfies AnthropicImageBlock
 })

@@ -723,25 +449,6 @@ const lowerToolResultContent = Effect.fnUntraced(function* (part: ToolResultPart
  return yield* Effect.forEach(content, lowerToolResultContentItem)
 })

-const requireThinkingSignature = (request: LLMRequest) => {
-  if (request.model.compatibility?.requireSignature !== undefined) return request.model.compatibility.requireSignature
-  const provider = request.model.provider.toLowerCase()
-  const model = request.model.id.toLowerCase()
-  const baseURL = (request.model.route.endpoint.baseURL ?? "").toLowerCase()
-  if (
-    provider === "kimi-for-coding" ||
-    provider === "moonshotai" ||
-    provider === "moonshotai-cn" ||
-    model.startsWith("kimi-") ||
-    baseURL.includes("api.kimi.com/coding") ||
-    baseURL.includes("api.moonshot.ai/anthropic") ||
-    baseURL.includes("api.moonshot.cn/anthropic")
-  )
-    return false
-  if (provider.includes("xiaomi") || model.includes("mimo") || baseURL.includes("xiaomimimo.com")) return false
-  return true
-}
-
 // Mid-conversation system messages became available with Opus 4.8 and version
 // 5 of the other supported Claude families. Treat later family versions as
 // compatible without assuming that every Anthropic Messages model is Claude.
@@ -761,12 +468,9 @@ const endsInServerToolUse = (message: LLMRequest["messages"][number]) => {
  return message.role === "assistant" && last?.type === "tool-call" && last.providerExecuted === true
 }

-const canUseNativeSystemUpdate = (request: LLMRequest, index: number) => {
-  const previous = request.messages[index - 1]
-  const next = request.messages[index + 1]
-  // Vertex currently rejects/404s for a system message after local tool results,
-  // so fold it into the user tool-result turn across continuations and history.
-  if (request.model.route.id === "google-vertex-messages" && previous?.role === "tool") return false
+const canUseNativeSystemUpdate = (messages: LLMRequest["messages"], index: number) => {
+  const previous = messages[index - 1]
+  const next = messages[index + 1]
  return (
    previous !== undefined &&
    previous.role !== "system" &&
@@ -808,13 +512,12 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
  breakpoints: Cache.Breakpoints,
 ) {
  const messages: AnthropicMessage[] = []
-  const providerMetadataKey = request.model.route.providerMetadataKey ?? String(request.model.provider)

  for (const [index, message] of request.messages.entries()) {
    if (message.role === "system") {
      if (splitsLocalToolResults(request.messages, index))
        return yield* invalid("Anthropic Messages system updates cannot split a local tool call from its tool result")
-      if (supportsNativeSystemUpdates(request) && canUseNativeSystemUpdate(request, index)) {
+      if (supportsNativeSystemUpdates(request) && canUseNativeSystemUpdate(request.messages, index)) {
        messages.push(yield* lowerNativeSystemUpdate(message, breakpoints))
        continue
      }
@@ -835,7 +538,7 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
          continue
        }
        if (part.type === "media") {
-          content.push(yield* lowerMedia(part, breakpoints))
+          content.push(yield* lowerMedia(part))
          continue
        }
        return yield* ProviderShared.unsupportedContent("Anthropic Messages", "user", ["text", "media"])
@@ -852,30 +555,15 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
          continue
        }
        if (part.type === "reasoning") {
-          // A signature marks visible thinking; only signature-less parts carrying
-          // redactedData round-trip as opaque redacted_thinking blocks.
-          const signature = part.encrypted ?? signatureFromMetadata(part.providerMetadata, providerMetadataKey)
-          const redactedData = redactedDataFromMetadata(part.providerMetadata, providerMetadataKey)
+          // Mirrors Vercel's @ai-sdk/anthropic: a signature marks visible
+          // thinking; only signature-less parts carrying redactedData
+          // round-trip as opaque redacted_thinking blocks.
+          const signature = part.encrypted ?? signatureFromMetadata(part.providerMetadata)
+          const redactedData = redactedDataFromMetadata(part.providerMetadata)
          if (signature === undefined && redactedData !== undefined) {
            content.push({ type: "redacted_thinking", data: redactedData })
            continue
          }
-          if (typeof signature !== "string" || signature.trim().length === 0) {
-            if (part.text.trim().length === 0) continue
-            if (!requireThinkingSignature(request)) {
-              content.push({ type: "thinking", thinking: part.text, signature: "" })
-              continue
-            }
-            // Without a signature this cannot be a valid thinking block per
-            // the SDK ThinkingBlockParam:3217 — demote to text so the
-            // conversation remains sendable.
-            content.push({
-              type: "text",
-              text: part.text,
-              cache_control: cacheControl(breakpoints, part.cache),
-            })
-            continue
-          }
          content.push({ type: "thinking", thinking: part.text, signature })
          continue
        }
@@ -884,7 +572,7 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
          continue
        }
        if (part.type === "tool-result" && part.providerExecuted) {
-          content.push(yield* lowerServerToolResult(part, providerMetadataKey))
+          content.push(yield* lowerServerToolResult(part))
          continue
        }
        return yield* invalid(
@@ -917,76 +605,24 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
 })

 const resolveOptions = Effect.fn("AnthropicMessages.resolveOptions")(function* (request: LLMRequest) {
-  const input = request.providerOptions as Record<string, unknown> | undefined
-  const rawServiceTier =
-    (input as Record<string, unknown> | undefined)?.service_tier ??
-    (input as Record<string, unknown> | undefined)?.serviceTier
-  const service_tier =
-    rawServiceTier === "auto" || rawServiceTier === "standard_only"
-      ? (rawServiceTier as "auto" | "standard_only")
-      : undefined
-  const rawMetadata = (input as Record<string, unknown> | undefined)?.metadata
-  const metadata =
-    ProviderShared.isRecord(rawMetadata) && (typeof rawMetadata.user_id === "string" || rawMetadata.user_id === null)
-      ? { user_id: rawMetadata.user_id as string | null }
-      : undefined
-  const container =
-    typeof (input as Record<string, unknown> | undefined)?.container === "string" ||
-    ProviderShared.isRecord((input as Record<string, unknown> | undefined)?.container)
-      ? ((input as Record<string, unknown>).container as
-          | string
-          | { id?: string | null; skills?: ReadonlyArray<Record<string, unknown>> | null })
-      : undefined
-  const rawInferenceGeo =
-    (input as Record<string, unknown> | undefined)?.inference_geo ??
-    (input as Record<string, unknown> | undefined)?.inferenceGeo
-  const inference_geo = typeof rawInferenceGeo === "string" ? rawInferenceGeo : undefined
-  const rawCacheControl =
-    (input as Record<string, unknown> | undefined)?.cache_control ??
-    (input as Record<string, unknown> | undefined)?.cacheControl
-  const cache_control =
-    ProviderShared.isRecord(rawCacheControl) && rawCacheControl.type === "ephemeral"
-      ? (rawCacheControl as { type: "ephemeral"; ttl?: "5m" | "1h" })
-      : undefined
-  const rawOutputConfig =
-    (input as Record<string, unknown> | undefined)?.output_config ??
-    (input as Record<string, unknown> | undefined)?.outputConfig
-  const outputConfigEffort =
-    typeof (input as Record<string, unknown> | undefined)?.effort === "string"
-      ? ((input as Record<string, unknown>).effort as string)
-      : ProviderShared.isRecord(rawOutputConfig) && typeof rawOutputConfig.effort === "string"
-        ? (rawOutputConfig.effort as string)
-        : undefined
-  const outputConfigFormat =
-    ProviderShared.isRecord(rawOutputConfig) && ProviderShared.isRecord(rawOutputConfig.format)
-      ? (rawOutputConfig.format as { type: "json_schema"; schema: Record<string, unknown> })
-      : undefined
-  const output_config =
-    outputConfigEffort === undefined && outputConfigFormat === undefined
-      ? undefined
-      : {
-          ...(outputConfigEffort === undefined ? {} : { effort: outputConfigEffort }),
-          ...(outputConfigFormat === undefined ? {} : { format: outputConfigFormat }),
-        }
+  const input = request.providerOptions
  return {
    thinking: yield* resolveThinking(input?.thinking),
-    effort: outputConfigEffort,
-    output_config,
-    service_tier,
-    metadata,
-    container,
-    inference_geo,
-    cache_control,
+    effort: typeof input?.effort === "string" ? input.effort : undefined,
  }
 })

 const resolveThinking = Effect.fn("AnthropicMessages.resolveThinking")(function* (input: unknown) {
  if (!ProviderShared.isRecord(input)) return undefined
-  const display =
-    input.display === "summarized" || input.display === "omitted"
-      ? (input.display as "summarized" | "omitted")
-      : undefined
-  if (input.type === "adaptive") return { type: "adaptive" as const, ...(display === undefined ? {} : { display }) }
+  if (input.type === "adaptive") {
+    const display =
+      input.display === "summarized"
+        ? ("summarized" as const)
+        : input.display === "omitted"
+          ? ("omitted" as const)
+          : undefined
+    return { type: "adaptive" as const, ...(display === undefined ? {} : { display }) }
+  }
  if (input.type === "disabled") return { type: "disabled" as const }
  if (input.type !== "enabled") return undefined
  const budget =
@@ -997,7 +633,7 @@ const resolveThinking = Effect.fn("AnthropicMessages.resolveThinking")(function*
        : undefined
  if (budget === undefined)
    return yield* ProviderShared.invalidRequest("Anthropic thinking provider option requires budgetTokens")
-  return { type: "enabled" as const, budget_tokens: budget, ...(display === undefined ? {} : { display }) }
+  return { type: "enabled" as const, budget_tokens: budget }
 })

 const fromRequest = Effect.fn("AnthropicMessages.fromRequest")(function* (request: LLMRequest) {
@@ -1047,13 +683,7 @@ const fromRequest = Effect.fn("AnthropicMessages.fromRequest")(function* (reques
    top_k: generation?.topK,
    stop_sequences: generation?.stop,
    thinking: options.thinking,
-    output_config: options.output_config,
-    // top-level passthrough per SDK MessageCreateParamsBase:4638,4643,4649,4654,4670
-    cache_control: options.cache_control,
-    container: options.container,
-    inference_geo: options.inference_geo,
-    metadata: options.metadata,
-    service_tier: options.service_tier,
+    output_config: options.effort === undefined ? undefined : { effort: options.effort },
  }
 })

@@ -1074,7 +704,7 @@ const mapFinishReason = (reason: string | null | undefined): FinishReason => {
 // inclusive `inputTokens` the rest of the contract expects. Extended
 // thinking tokens are included in `output_tokens`; newer responses also
 // expose that subset through `output_tokens_details.thinking_tokens`.
-const mapUsage = (usage: AnthropicUsage | undefined, providerMetadataKey: string): Usage | undefined => {
+const mapUsage = (usage: AnthropicUsage | undefined): Usage | undefined => {
  if (!usage) return undefined
  const nonCached = usage.input_tokens ?? undefined
  const cacheRead = usage.cache_read_input_tokens ?? undefined
@@ -1088,7 +718,7 @@ const mapUsage = (usage: AnthropicUsage | undefined, providerMetadataKey: string
    cacheWriteInputTokens: cacheWrite,
    reasoningTokens: usage.output_tokens_details?.thinking_tokens,
    totalTokens: ProviderShared.totalTokens(inputTokens, usage.output_tokens, undefined),
-    providerMetadata: { [providerMetadataKey]: usage },
+    providerMetadata: { anthropic: usage },
  })
 }

@@ -1097,7 +727,7 @@ const mapUsage = (usage: AnthropicUsage | undefined, providerMetadataKey: string
 // field prefers `right` when defined, falls back to `left`. `inputTokens` is
 // recomputed from the merged breakdown so the inclusive total stays
 // consistent with `nonCached + cacheRead + cacheWrite`.
-const mergeUsage = (left: Usage | undefined, right: Usage | undefined, providerMetadataKey: string) => {
+const mergeUsage = (left: Usage | undefined, right: Usage | undefined) => {
  if (!left) return right
  if (!right) return left
  const nonCachedInputTokens = right.nonCachedInputTokens ?? left.nonCachedInputTokens
@@ -1115,9 +745,7 @@ const mergeUsage = (left: Usage | undefined, right: Usage | undefined, providerM
    reasoningTokens,
    totalTokens: ProviderShared.totalTokens(inputTokens, outputTokens, undefined),
    providerMetadata: {
-      [providerMetadataKey]:
-        mergeJsonRecords(left.providerMetadata?.[providerMetadataKey], right.providerMetadata?.[providerMetadataKey]) ??
-        {},
+      anthropic: mergeJsonRecords(left.providerMetadata?.["anthropic"], right.providerMetadata?.["anthropic"]) ?? {},
    },
  })
 }
@@ -1135,7 +763,7 @@ const SERVER_TOOL_RESULT_NAMES: Record<AnthropicServerToolResultType, string> =

 const isServerToolResultType = (type: string): type is AnthropicServerToolResultType => type in SERVER_TOOL_RESULT_NAMES

-const serverToolResultEvent = (block: AnthropicStreamBlock, providerMetadataKey: string): LLMEvent | undefined => {
+const serverToolResultEvent = (block: NonNullable<AnthropicEvent["content_block"]>): LLMEvent | undefined => {
  if (!block.type || !isServerToolResultType(block.type)) return undefined
  const errorPayload =
    typeof block.content === "object" && block.content !== null && "type" in block.content
@@ -1149,7 +777,7 @@ const serverToolResultEvent = (block: AnthropicStreamBlock, providerMetadataKey:
    providerExecuted: true,
    // The complete payload is irreducible provider replay state: subsequent
    // stateless requests must round-trip the typed result block verbatim.
-    providerMetadata: providerMetadata(providerMetadataKey, { blockType: block.type, result: block.content }),
+    providerMetadata: anthropicMetadata({ blockType: block.type, result: block.content }),
  })
 }

@@ -1158,19 +786,15 @@ type StepResult = readonly [ParserState, ReadonlyArray<LLMEvent>]
 const NO_EVENTS: StepResult["1"] = []

 const onMessageStart = (state: ParserState, event: AnthropicEvent): StepResult => {
-  const usage = mapUsage(event.message?.usage, state.providerMetadataKey)
-  return [usage ? { ...state, usage: mergeUsage(state.usage, usage, state.providerMetadataKey) } : state, NO_EVENTS]
+  const usage = mapUsage(event.message?.usage)
+  return [usage ? { ...state, usage: mergeUsage(state.usage, usage) } : state, NO_EVENTS]
 }

-const onContentBlockStart = (
-  state: ParserState,
-  event: AnthropicEvent & { readonly content_block: AnthropicStreamBlock },
-): StepResult => {
+const onContentBlockStart = (state: ParserState, event: AnthropicEvent): StepResult => {
  const block = event.content_block
  if (!block) return [state, NO_EVENTS]

-  if (block.type === "tool_use" || block.type === "server_tool_use") {
-    if (event.index === undefined || !block.id) return [state, NO_EVENTS]
+  if ((block.type === "tool_use" || block.type === "server_tool_use") && event.index !== undefined) {
    const events: LLMEvent[] = []
    const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
    return [
@@ -1178,7 +802,7 @@ const onContentBlockStart = (
        ...state,
        lifecycle,
        tools: ToolStream.start(state.tools, event.index, {
-          id: block.id,
+          id: block.id ?? String(event.index),
          name: block.name ?? "",
          input:
            block.input !== undefined && (!ProviderShared.isRecord(block.input) || Object.keys(block.input).length > 0)
@@ -1190,7 +814,7 @@ const onContentBlockStart = (
      [
        ...events,
        LLMEvent.toolInputStart({
-          id: block.id,
+          id: block.id ?? String(event.index),
          name: block.name ?? "",
          providerExecuted: block.type === "server_tool_use" ? true : undefined,
        }),
@@ -1211,16 +835,14 @@ const onContentBlockStart = (
  if (block.type === "thinking" && block.thinking !== undefined) {
    const events: LLMEvent[] = []
    const id = `reasoning-${event.index ?? 0}`
-    const metadata =
-      block.signature === undefined
-        ? undefined
-        : providerMetadata(state.providerMetadataKey, { signature: block.signature })
-    const lifecycle = Lifecycle.reasoningStart(state.lifecycle, events, id, metadata)
+    const providerMetadata =
+      block.signature === undefined ? undefined : anthropicMetadata({ signature: block.signature })
+    const lifecycle = Lifecycle.reasoningStart(state.lifecycle, events, id, providerMetadata)
    return [
      {
        ...state,
        lifecycle: block.thinking
-          ? Lifecycle.reasoningDelta(lifecycle, events, id, block.thinking, metadata)
+          ? Lifecycle.reasoningDelta(lifecycle, events, id, block.thinking, providerMetadata)
          : lifecycle,
        reasoningSignatures:
          event.index === undefined || block.signature === undefined
@@ -1243,14 +865,14 @@ const onContentBlockStart = (
          state.lifecycle,
          events,
          `reasoning-${event.index ?? 0}`,
-          providerMetadata(state.providerMetadataKey, { redactedData: block.data }),
+          anthropicMetadata({ redactedData: block.data }),
        ),
      },
      events,
    ]
  }

-  const result = serverToolResultEvent(block, state.providerMetadataKey)
+  const result = serverToolResultEvent(block)
  if (!result) return [state, NO_EVENTS]
  const events: LLMEvent[] = []
  return [{ ...state, lifecycle: Lifecycle.stepStart(state.lifecycle, events) }, [...events, result]]
@@ -1258,12 +880,11 @@ const onContentBlockStart = (

 const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(function* (
  state: ParserState,
-  event: AnthropicEvent & { readonly delta: AnthropicStreamDelta },
+  event: AnthropicEvent,
 ) {
  const delta = event.delta

  if (delta?.type === "text_delta" && delta.text) {
-    if (!state.lifecycle.text.has(`text-${event.index ?? 0}`)) return [state, NO_EVENTS] satisfies StepResult
    const events: LLMEvent[] = []
    return [
      { ...state, lifecycle: Lifecycle.textDelta(state.lifecycle, events, `text-${event.index ?? 0}`, delta.text) },
@@ -1272,7 +893,6 @@ const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(f
  }

  if (delta?.type === "thinking_delta" && delta.thinking) {
-    if (!state.lifecycle.reasoning.has(`reasoning-${event.index ?? 0}`)) return [state, NO_EVENTS] satisfies StepResult
    const events: LLMEvent[] = []
    return [
      {
@@ -1285,7 +905,6 @@ const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(f

  if (delta?.type === "signature_delta" && delta.signature) {
    const index = event.index ?? 0
-    if (!state.lifecycle.reasoning.has(`reasoning-${index}`)) return [state, NO_EVENTS] satisfies StepResult
    return [
      {
        ...state,
@@ -1330,7 +949,7 @@ const onContentBlockStop = Effect.fn("AnthropicMessages.onContentBlockStop")(fun
        Lifecycle.textEnd(state.lifecycle, events, `text-${event.index}`),
        events,
        `reasoning-${event.index}`,
-        signature === undefined ? undefined : providerMetadata(state.providerMetadataKey, { signature }),
+        signature === undefined ? undefined : anthropicMetadata({ signature }),
      )
  events.push(...resultEvents)
  const reasoningSignatures = { ...state.reasoningSignatures }
@@ -1338,11 +957,8 @@ const onContentBlockStop = Effect.fn("AnthropicMessages.onContentBlockStop")(fun
  return [{ ...state, lifecycle, tools: result.tools, reasoningSignatures }, events] satisfies StepResult
 })

-const onMessageDelta = (
-  state: ParserState,
-  event: AnthropicEvent & { readonly delta?: AnthropicStreamDelta },
-): StepResult => {
-  const usage = mergeUsage(state.usage, mapUsage(event.usage, state.providerMetadataKey), state.providerMetadataKey)
+const onMessageDelta = (state: ParserState, event: AnthropicEvent): StepResult => {
+  const usage = mergeUsage(state.usage, mapUsage(event.usage))
  return [
    {
      ...state,
@@ -1355,7 +971,7 @@ const onMessageDelta = (
        providerMetadata:
          event.delta?.stop_sequence === null || event.delta?.stop_sequence === undefined
            ? undefined
-            : providerMetadata(state.providerMetadataKey, { stopSequence: event.delta.stop_sequence }),
+            : anthropicMetadata({ stopSequence: event.delta.stop_sequence }),
      },
    },
    NO_EVENTS,
@@ -1387,80 +1003,19 @@ const providerErrorMessage = (event: AnthropicEvent): string => {
  return message || type || "Anthropic Messages stream error"
 }

-const onError = (event: AnthropicEvent) => {
-  const message = providerErrorMessage(event)
-  const body = ProviderShared.encodeJson(event)
-  return Effect.fail(
-    new AIError({
-      reason: classifyProviderFailure({ message, rawBody: body }),
-    }),
-  )
-}
-
-const isKnownStreamBlockType = (type: string) =>
-  type === "text" ||
-  type === "thinking" ||
-  type === "redacted_thinking" ||
-  type === "tool_use" ||
-  type === "server_tool_use" ||
-  isServerToolResultType(type)
-
-const isKnownStreamDeltaType = (type: string) =>
-  type === "text_delta" || type === "thinking_delta" || type === "signature_delta" || type === "input_json_delta"
-
-const invalidStreamEvent = (event: AnthropicEvent) =>
-  Effect.fail(
-    ProviderShared.eventError(
-      ADAPTER,
-      "Invalid anthropic/anthropic-messages stream event",
-      ProviderShared.encodeJson(event),
-    ),
-  )
+const onError = (event: AnthropicEvent) =>
+  new AIError({
+    module: ADAPTER,
+    method: "stream",
+    reason: classifyProviderFailure({ message: providerErrorMessage(event), code: event.error?.type }),
+  })

 const step = (state: ParserState, event: AnthropicEvent) => {
-  if (!SSE_EVENTS.has(event.type)) return Effect.succeed<StepResult>([state, NO_EVENTS])
-  if (
-    event.type !== "content_block_start" &&
-    event.content_block !== undefined &&
-    Option.isNone(decodeAnthropicStreamBlock(event.content_block))
-  )
-    return invalidStreamEvent(event)
-  if (
-    event.type !== "content_block_delta" &&
-    event.delta !== undefined &&
-    Option.isNone(decodeAnthropicStreamDelta(event.delta))
-  )
-    return invalidStreamEvent(event)
  if (event.type === "message_start") return Effect.succeed(onMessageStart(state, event))
-  if (event.type === "content_block_start") {
-    if (!ProviderShared.isRecord(event.content_block) || typeof event.content_block.type !== "string")
-      return invalidStreamEvent(event)
-    if (!isKnownStreamBlockType(event.content_block.type)) return Effect.succeed<StepResult>([state, NO_EVENTS])
-    const decoded = decodeAnthropicStreamBlock(event.content_block)
-    if (Option.isNone(decoded)) return invalidStreamEvent(event)
-    const block = decoded.value
-    if (block.type === "tool_use" || block.type === "server_tool_use") {
-      if (event.index === undefined)
-        return Effect.fail(ProviderShared.eventError(ADAPTER, `Anthropic ${block.type} missing index`))
-      if (!block.id)
-        return Effect.fail(ProviderShared.eventError(ADAPTER, `Anthropic tool_use missing id at index ${event.index}`))
-    }
-    return Effect.succeed(onContentBlockStart(state, { ...event, content_block: block }))
-  }
-  if (event.type === "content_block_delta") {
-    if (!ProviderShared.isRecord(event.delta)) return invalidStreamEvent(event)
-    if (typeof event.delta.type === "string" && !isKnownStreamDeltaType(event.delta.type))
-      return Effect.succeed<StepResult>([state, NO_EVENTS])
-    const decoded = decodeAnthropicStreamDelta(event.delta)
-    if (Option.isNone(decoded)) return invalidStreamEvent(event)
-    return onContentBlockDelta(state, { ...event, delta: decoded.value })
-  }
+  if (event.type === "content_block_start") return Effect.succeed(onContentBlockStart(state, event))
+  if (event.type === "content_block_delta") return onContentBlockDelta(state, event)
  if (event.type === "content_block_stop") return onContentBlockStop(state, event)
-  if (event.type === "message_delta") {
-    const decoded = decodeAnthropicStreamDelta(event.delta)
-    if (Option.isNone(decoded)) return invalidStreamEvent(event)
-    return Effect.succeed(onMessageDelta(state, { ...event, delta: decoded.value }))
-  }
+  if (event.type === "message_delta") return Effect.succeed(onMessageDelta(state, event))
  if (event.type === "message_stop") return onMessageStop(state)
  if (event.type === "error") return onError(event)
  return Effect.succeed<StepResult>([state, NO_EVENTS])
@@ -1482,8 +1037,7 @@ export const protocol = Protocol.make({
  },
  stream: {
    event: Protocol.jsonEvent(AnthropicEvent),
-    initial: (request) => ({
-      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
+    initial: () => ({
      tools: ToolStream.empty<number>(),
      reasoningSignatures: {},
      lifecycle: Lifecycle.initial(),
@@ -1497,9 +1051,7 @@ export const route = Route.make({
  provider: "anthropic",
  providerMetadataKey: "anthropic",
  protocol,
-  endpoint: Endpoint.path((input) => (input.request.model.provider === "anthropic" ? `${PATH}?beta=true` : PATH), {
-    baseURL: DEFAULT_BASE_URL,
-  }),
+  endpoint: Endpoint.path(PATH, { baseURL: DEFAULT_BASE_URL }),
  auth: Auth.none,
  framing,
  headers: () => ({ "anthropic-version": "2023-06-01" }),
@@ -212,7 +212,11 @@ const BedrockEvent = Schema.Struct({
      metrics: Schema.optional(Schema.Unknown),
    }),
  ),
-  exception: Schema.optional(Schema.Struct({ type: Schema.String, details: BedrockStreamException })),
+  internalServerException: Schema.optional(BedrockStreamException),
+  modelStreamErrorException: Schema.optional(BedrockStreamException),
+  validationException: Schema.optional(BedrockStreamException),
+  throttlingException: Schema.optional(BedrockStreamException),
+  serviceUnavailableException: Schema.optional(BedrockStreamException),
 })
 type BedrockEvent = Schema.Schema.Type<typeof BedrockEvent>

@@ -258,21 +262,19 @@ const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>) =>
    tool: (name) => ({ tool: { name } }) as const,
  })

-const providerMetadata = (key: string, metadata: Record<string, unknown>): ProviderMetadata => ({ [key]: metadata })
+const bedrockMetadata = (metadata: Record<string, unknown>): ProviderMetadata => ({ bedrock: metadata })

-const reasoningSignature = (part: ReasoningPart, providerMetadataKey: string) => {
-  const metadata = part.providerMetadata?.[providerMetadataKey]
+const reasoningSignature = (part: ReasoningPart) => {
+  const bedrock = part.providerMetadata?.bedrock
  return (
    part.encrypted ??
-    (ProviderShared.isRecord(metadata) && typeof metadata.signature === "string" ? metadata.signature : undefined)
+    (ProviderShared.isRecord(bedrock) && typeof bedrock.signature === "string" ? bedrock.signature : undefined)
  )
 }

-const reasoningRedactedData = (part: ReasoningPart, providerMetadataKey: string) => {
-  const metadata = part.providerMetadata?.[providerMetadataKey]
-  return ProviderShared.isRecord(metadata) && typeof metadata.redactedData === "string"
-    ? metadata.redactedData
-    : undefined
+const reasoningRedactedData = (part: ReasoningPart) => {
+  const bedrock = part.providerMetadata?.bedrock
+  return ProviderShared.isRecord(bedrock) && typeof bedrock.redactedData === "string" ? bedrock.redactedData : undefined
 }

 const lowerToolCall = (part: ToolCallPart): BedrockToolUseBlock => ({
@@ -320,7 +322,6 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (
  breakpoints: BedrockCache.Breakpoints,
 ) {
  const messages: BedrockMessage[] = []
-  const providerMetadataKey = request.model.route.providerMetadataKey ?? String(request.model.provider)

  for (const message of request.messages) {
    if (message.role === "system") {
@@ -368,8 +369,8 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (
          continue
        }
        if (part.type === "reasoning") {
-          const signature = reasoningSignature(part, providerMetadataKey)
-          const redactedData = reasoningRedactedData(part, providerMetadataKey)
+          const signature = reasoningSignature(part)
+          const redactedData = reasoningRedactedData(part)
          if (signature === undefined && redactedData !== undefined) {
            content.push({ reasoningContent: { redactedContent: redactedData } })
            continue
@@ -469,7 +470,7 @@ const mapFinishReason = (reason: string): FinishReason => {

 // AWS reports inputTokens separately from cache reads and writes.
 // Bedrock does not break reasoning out of outputTokens for current models.
-const mapUsage = (usage: BedrockUsageSchema | undefined, providerMetadataKey: string): Usage | undefined => {
+const mapUsage = (usage: BedrockUsageSchema | undefined): Usage | undefined => {
  if (!usage) return undefined
  const inputTokens = ProviderShared.sumTokens(
    usage.inputTokens,
@@ -483,12 +484,11 @@ const mapUsage = (usage: BedrockUsageSchema | undefined, providerMetadataKey: st
    cacheReadInputTokens: usage.cacheReadInputTokens,
    cacheWriteInputTokens: usage.cacheWriteInputTokens,
    totalTokens: ProviderShared.totalTokens(inputTokens, usage.outputTokens, usage.totalTokens),
-    providerMetadata: { [providerMetadataKey]: usage },
+    providerMetadata: { bedrock: usage },
  })
 }

 interface ParserState {
-  readonly providerMetadataKey: string
  readonly tools: ToolStream.State<number>
  // Bedrock splits the finish into `messageStop` (carries `stopReason`) and
  // `metadata` (carries usage). Hold the terminal event in state so `onHalt`
@@ -545,14 +545,20 @@ const step = (state: ParserState, event: BedrockEvent) =>
      const reasoning = event.contentBlockDelta.delta.reasoningContent
      const events: LLMEvent[] = []
      const redactedData = reasoning.redactedContent ?? reasoning.data
-      const metadata = reasoning.signature
-        ? providerMetadata(state.providerMetadataKey, { signature: reasoning.signature })
+      const providerMetadata = reasoning.signature
+        ? bedrockMetadata({ signature: reasoning.signature })
        : redactedData !== undefined
-          ? providerMetadata(state.providerMetadataKey, { redactedData })
+          ? bedrockMetadata({ redactedData })
          : undefined
      const lifecycle =
-        reasoning.text !== undefined || metadata !== undefined
-          ? Lifecycle.reasoningDelta(state.lifecycle, events, `reasoning-${index}`, reasoning.text ?? "", metadata)
+        reasoning.text !== undefined || providerMetadata !== undefined
+          ? Lifecycle.reasoningDelta(
+              state.lifecycle,
+              events,
+              `reasoning-${index}`,
+              reasoning.text ?? "",
+              providerMetadata,
+            )
          : state.lifecycle
      return [
        {
@@ -594,7 +600,7 @@ const step = (state: ParserState, event: BedrockEvent) =>
            events,
            `reasoning-${index}`,
            state.reasoningSignatures[index]
-              ? providerMetadata(state.providerMetadataKey, { signature: state.reasoningSignatures[index] })
+              ? bedrockMetadata({ signature: state.reasoningSignatures[index] })
              : undefined,
          )
      events.push(...resultEvents)
@@ -631,7 +637,7 @@ const step = (state: ParserState, event: BedrockEvent) =>
    }

    if (event.metadata) {
-      const usage = mapUsage(event.metadata.usage, state.providerMetadataKey) ?? state.pendingFinish?.usage
+      const usage = mapUsage(event.metadata.usage) ?? state.pendingFinish?.usage
      return [
        {
          ...state,
@@ -644,14 +650,22 @@ const step = (state: ParserState, event: BedrockEvent) =>
      ] as const
    }

-    if (event.exception) {
-      const message =
-        event.exception.details.message ?? event.exception.details.originalMessage ?? "Bedrock Converse stream error"
-      const body = ProviderShared.encodeJson(event)
+    const exception = (
+      [
+        ["internalServerException", event.internalServerException],
+        ["modelStreamErrorException", event.modelStreamErrorException],
+        ["serviceUnavailableException", event.serviceUnavailableException],
+        ["throttlingException", event.throttlingException],
+        ["validationException", event.validationException],
+      ] as const
+    ).find((entry) => entry[1] !== undefined)
+    if (exception) {
      return yield* new AIError({
+        module: ADAPTER,
+        method: "stream",
        reason: classifyProviderFailure({
-          message,
-          rawBody: body,
+          message: exception[1]?.message ?? exception[1]?.originalMessage ?? "Bedrock Converse stream error",
+          code: exception[0],
        }),
      })
    }
@@ -694,8 +708,7 @@ export const protocol = Protocol.make({
  },
  stream: {
    event: BedrockEvent,
-    initial: (request) => ({
-      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
+    initial: () => ({
      tools: ToolStream.empty<number>(),
      pendingFinish: undefined,
      hasToolCalls: false,
@@ -703,7 +716,7 @@ export const protocol = Protocol.make({
      reasoningSignatures: {},
    }),
    step,
-    onHalt: (state) => Effect.succeed(onHalt(state)),
+    onHalt,
  },
 })

@@ -1,7 +1,6 @@
 import { EventStreamCodec } from "@smithy/eventstream-codec"
 import { fromUtf8, toUtf8 } from "@smithy/util-utf8"
-import { Effect, Encoding, Stream } from "effect"
-import { AIError, AIErrorReason } from "../schema/index.js"
+import { Effect, Stream } from "effect"
 import { Framing } from "../route/framing.js"
 import { ProviderShared } from "./shared.js"

@@ -50,14 +49,10 @@ const consumeFrames = (route: string) => (state: FrameBufferState, chunk: Uint8A
            `Failed to decode Bedrock Converse event-stream frame: ${
              error instanceof Error ? error.message : String(error)
            }`,
-            Encoding.encodeBase64(view.subarray(0, totalLength)),
-            error,
          ),
      })
      cursor = { buffer: cursor.buffer, offset: cursor.offset + totalLength }

-      const payload = utf8.decode(decoded.body)
-      const body = ProviderShared.encodeJson({ headers: decoded.headers, body: payload })
      const messageType = decoded.headers[":message-type"]?.value
      if (messageType === "error") {
        const code = decoded.headers[":error-code"]?.value
@@ -66,7 +61,6 @@ const consumeFrames = (route: string) => (state: FrameBufferState, chunk: Uint8A
          route,
          [code, message].filter((value): value is string => typeof value === "string").join(": ") ||
            "Bedrock Converse event-stream error",
-          body,
        )
      }
      const eventType =
@@ -76,6 +70,7 @@ const consumeFrames = (route: string) => (state: FrameBufferState, chunk: Uint8A
            ? decoded.headers[":exception-type"]?.value
            : undefined
      if (typeof eventType !== "string") continue
+      const payload = utf8.decode(decoded.body)
      if (!payload) continue
      // The AWS event stream pads short payloads with a `p` field. Drop it
      // before handing the object to the chunk schema. JSON decode goes
@@ -85,21 +80,9 @@ const consumeFrames = (route: string) => (state: FrameBufferState, chunk: Uint8A
        route,
        payload,
        "Failed to parse Bedrock Converse event-stream payload",
-      ).pipe(
-        Effect.mapError(
-          (error) =>
-            new AIError({
-              reason: AIErrorReason.make({ ...error.reason, message: error.message, cause: error.reason.cause, body }),
-            }),
-        ),
      )) as Record<string, unknown>
      delete parsed.p
-      out.push({
-        ...(messageType === "exception"
-          ? { exception: { type: eventType, details: parsed } }
-          : { [eventType]: parsed }),
-        rawBody: body,
-      })
+      out.push({ [eventType]: parsed })
    }
    return [cursor, out] as const
  })
@@ -112,7 +95,6 @@ const consumeFrames = (route: string) => (state: FrameBufferState, chunk: Uint8A
 */
 export const framing = (route: string): Framing.Definition<object> => ({
  id: "aws-event-stream",
-  body: (frame) => ("rawBody" in frame && typeof frame.rawBody === "string" ? frame.rawBody : undefined),
  frame: (bytes) => bytes.pipe(Stream.mapAccumEffect(() => initialFrameBuffer, consumeFrames(route))),
 })

@@ -1,4 +1,4 @@
-import { Effect, Option, Schema } from "effect"
+import { Effect, Schema } from "effect"
 import { Tool } from "@opencode-ai/schema/tool"
 import { Route } from "../route/client.js"
 import { Auth } from "../route/auth.js"
@@ -6,7 +6,6 @@ import { Endpoint } from "../route/endpoint.js"
 import { Framing } from "../route/framing.js"
 import { Protocol } from "../route/protocol.js"
 import {
-  AIError,
  LLMEvent,
  Usage,
  type FinishReason,
@@ -18,8 +17,7 @@ import {
  type ToolCallPart,
  type ToolDefinition,
 } from "../schema/index.js"
-import { classifyProviderFailure } from "../provider-error.js"
-import { JsonObject, optionalArray, optionalNull, ProviderShared } from "./shared.js"
+import { JsonObject, optionalArray, ProviderShared } from "./shared.js"
 import { GeminiToolSchema } from "./utils/gemini-tool-schema.js"
 import { Lifecycle } from "./utils/lifecycle.js"
 import { ToolSchemaProjection } from "./utils/tool-schema.js"
@@ -43,13 +41,6 @@ const requiresThoughtSignatureFallback = (modelID: string) => {
 // so their tool-result attachments lower as a separate user turn instead.
 const routesLegacyToolMedia = (modelID: string) => /gemini-2[.-]5(?:[.-]|$)/i.test(modelID)

-// Blacklist: Gemini 1.x/2.x ignore or reject explicit function call ids.
-// Every other model id (Gemini 3+, gemma, anything unrecognized) gets them.
-const omitsFunctionCallIds = (modelID: string) => {
-  const match = /^gemini(?:-live)?-(\d+)/i.exec(modelID)
-  return match !== null && Number(match[1]) < 3
-}
-
 export interface OptionsInput {
  readonly [key: string]: unknown
  readonly cachedContent?: string
@@ -84,15 +75,10 @@ export type ProviderOptionsInput = OptionsInput
 // =============================================================================
 // Request Body Schema
 // =============================================================================
-// Gemini is known to send explicit `null` for optional streaming fields
-// (usage counts, flags, whole subtrees), so every response-side optional uses
-// `optionalNull` instead of bare `Schema.optional`. The same part/content
-// schemas lower the outbound request body; encoding drops `undefined` keys,
-// so the shared schemas stay safe there.
 const GeminiTextPart = Schema.Struct({
  text: Schema.String,
-  thought: optionalNull(Schema.Boolean),
-  thoughtSignature: optionalNull(Schema.String),
+  thought: Schema.optional(Schema.Boolean),
+  thoughtSignature: Schema.optional(Schema.String),
 })

 const GeminiInlineDataPart = Schema.Struct({
@@ -105,11 +91,11 @@ type GeminiInlineDataPart = Schema.Schema.Type<typeof GeminiInlineDataPart>

 const GeminiFunctionCallPart = Schema.Struct({
  functionCall: Schema.Struct({
-    id: optionalNull(Schema.String),
+    id: Schema.optional(Schema.String),
    name: Schema.String,
    args: Schema.optional(Schema.Unknown),
  }),
-  thoughtSignature: optionalNull(Schema.String),
+  thoughtSignature: Schema.optional(Schema.String),
 })

 const GeminiFunctionResponsePart = Schema.Struct({
@@ -127,19 +113,13 @@ const GeminiContentPart = Schema.Union([
  GeminiFunctionCallPart,
  GeminiFunctionResponsePart,
 ])
-const decodeGeminiContentPart = Schema.decodeUnknownOption(GeminiContentPart)

 const GeminiContent = Schema.Struct({
-  role: optionalNull(Schema.Literals(["user", "model"])),
-  parts: optionalNull(Schema.Array(GeminiContentPart)),
+  role: Schema.Literals(["user", "model"]),
+  parts: Schema.Array(GeminiContentPart),
 })
 type GeminiContent = Schema.Schema.Type<typeof GeminiContent>

-const GeminiResponseContent = Schema.Struct({
-  role: optionalNull(Schema.Literals(["user", "model"])),
-  parts: optionalNull(Schema.Array(Schema.Unknown)),
-})
-
 const GeminiSystemInstruction = Schema.Struct({
  parts: Schema.Array(Schema.Struct({ text: Schema.String })),
 })
@@ -199,40 +179,37 @@ const GeminiBody = Schema.Struct(GeminiBodyFields)
 export type GeminiBody = Schema.Schema.Type<typeof GeminiBody>

 const GeminiUsage = Schema.Struct({
-  cachedContentTokenCount: optionalNull(Schema.Number),
-  thoughtsTokenCount: optionalNull(Schema.Number),
-  promptTokenCount: optionalNull(Schema.Number),
-  candidatesTokenCount: optionalNull(Schema.Number),
-  totalTokenCount: optionalNull(Schema.Number),
+  cachedContentTokenCount: Schema.optional(Schema.Number),
+  thoughtsTokenCount: Schema.optional(Schema.Number),
+  promptTokenCount: Schema.optional(Schema.Number),
+  candidatesTokenCount: Schema.optional(Schema.Number),
+  totalTokenCount: Schema.optional(Schema.Number),
 })
 type GeminiUsage = Schema.Schema.Type<typeof GeminiUsage>

 const GeminiCandidate = Schema.Struct({
-  content: optionalNull(GeminiResponseContent),
-  finishReason: optionalNull(Schema.String),
+  content: Schema.optional(GeminiContent),
+  finishReason: Schema.optional(Schema.String),
 })

 const GeminiPromptFeedback = Schema.StructWithRest(
  Schema.Struct({
-    blockReason: optionalNull(Schema.String),
-    blockReasonMessage: optionalNull(Schema.String),
-    safetyRatings: optionalNull(Schema.Unknown),
+    blockReason: Schema.optional(Schema.String),
+    blockReasonMessage: Schema.optional(Schema.String),
+    safetyRatings: Schema.optional(Schema.Unknown),
  }),
  [Schema.Record(Schema.String, Schema.Unknown)],
 )
 type GeminiPromptFeedback = Schema.Schema.Type<typeof GeminiPromptFeedback>

 const GeminiEvent = Schema.Struct({
-  error: Schema.optional(Schema.Unknown),
-  candidates: optionalNull(Schema.Array(GeminiCandidate)),
-  promptFeedback: optionalNull(GeminiPromptFeedback),
-  usageMetadata: optionalNull(GeminiUsage),
+  candidates: optionalArray(GeminiCandidate),
+  promptFeedback: Schema.optional(GeminiPromptFeedback),
+  usageMetadata: Schema.optional(GeminiUsage),
 })
 type GeminiEvent = Schema.Schema.Type<typeof GeminiEvent>

 interface ParserState {
-  readonly route: string
-  readonly providerMetadataKey: string
  readonly finishReason?: string
  readonly hasToolCalls: boolean
  readonly promptFeedback?: GeminiPromptFeedback
@@ -240,7 +217,6 @@ interface ParserState {
  readonly lifecycle: Lifecycle.State
  readonly reasoningSignature?: string
  readonly textSignature?: string
-  readonly seenCallIds?: ReadonlySet<string>
 }

 // =============================================================================
@@ -289,24 +265,29 @@ const lowerUserPart = Effect.fn("Gemini.lowerUserPart")(function* (part: TextPar
  return { inlineData: { mimeType: media.mime, data: media.base64 } }
 })

-const providerMetadata = (key: string, metadata: Record<string, unknown>): ProviderMetadata => ({ [key]: metadata })
+const googleMetadata = (metadata: Record<string, unknown>): ProviderMetadata => ({ google: metadata })

-const thoughtSignature = (metadata: ProviderMetadata | undefined, key: string) => {
-  const value = metadata?.[key]
-  return ProviderShared.isRecord(value) && typeof value.thoughtSignature === "string"
-    ? value.thoughtSignature
+const thoughtSignature = (providerMetadata: ProviderMetadata | undefined) => {
+  const google = providerMetadata?.google
+  return ProviderShared.isRecord(google) && typeof google.thoughtSignature === "string"
+    ? google.thoughtSignature
    : undefined
 }

-const lowerToolCall = (part: ToolCallPart, omitIds: boolean, metadataKey: string) => ({
-  functionCall: { ...(omitIds ? {} : { id: part.id }), name: part.name, args: part.input },
-  thoughtSignature: thoughtSignature(part.providerMetadata, metadataKey),
+const functionCallId = (providerMetadata: ProviderMetadata | undefined) => {
+  const google = providerMetadata?.google
+  return ProviderShared.isRecord(google) && typeof google.functionCallId === "string"
+    ? google.functionCallId
+    : undefined
+}
+
+const lowerToolCall = (part: ToolCallPart) => ({
+  functionCall: { id: functionCallId(part.providerMetadata), name: part.name, args: part.input },
+  thoughtSignature: thoughtSignature(part.providerMetadata),
 })

 const lowerMessages = Effect.fn("Gemini.lowerMessages")(function* (request: LLMRequest) {
  const contents: GeminiContent[] = []
-  const metadataKey = request.model.route.providerMetadataKey ?? String(request.model.provider)
-  const omitCallIds = omitsFunctionCallIds(request.model.id)
  const legacyToolMedia = routesLegacyToolMedia(request.model.id)
  let pendingMedia: GeminiInlineDataPart[] | undefined
  const flushMedia = () => {
@@ -322,8 +303,8 @@ const lowerMessages = Effect.fn("Gemini.lowerMessages")(function* (request: LLMR
      const previous = contents.at(-1)
      // Gemini rejects a continuation whose function-response turn carries extra
      // parts, so an update after a tool result starts its own user turn.
-      if (previous?.role === "user" && !(previous.parts ?? []).some((item) => "functionResponse" in item))
-        contents[contents.length - 1] = { role: "user", parts: [...(previous.parts ?? []), { text: part.text }] }
+      if (previous?.role === "user" && !previous.parts.some((item) => "functionResponse" in item))
+        contents[contents.length - 1] = { role: "user", parts: [...previous.parts, { text: part.text }] }
      else contents.push({ role: "user", parts: [{ text: part.text }] })
      continue
    }
@@ -347,19 +328,15 @@ const lowerMessages = Effect.fn("Gemini.lowerMessages")(function* (request: LLMR
        if (!ProviderShared.supportsContent(part, ["text", "reasoning", "tool-call"]))
          return yield* ProviderShared.unsupportedContent("Gemini", "assistant", ["text", "reasoning", "tool-call"])
        if (part.type === "text") {
-          parts.push({ text: part.text, thoughtSignature: thoughtSignature(part.providerMetadata, metadataKey) })
+          parts.push({ text: part.text, thoughtSignature: thoughtSignature(part.providerMetadata) })
          continue
        }
        if (part.type === "reasoning") {
-          parts.push({
-            text: part.text,
-            thought: true,
-            thoughtSignature: thoughtSignature(part.providerMetadata, metadataKey),
-          })
+          parts.push({ text: part.text, thought: true, thoughtSignature: thoughtSignature(part.providerMetadata) })
          continue
        }
        if (part.type === "tool-call") {
-          const lowered = lowerToolCall(part, omitCallIds, metadataKey)
+          const lowered = lowerToolCall(part)
          const signature = lowered.thoughtSignature
          parts.push({
            ...lowered,
@@ -384,7 +361,7 @@ const lowerMessages = Effect.fn("Gemini.lowerMessages")(function* (request: LLMR
      if (part.result.type !== "content") {
        parts.push({
          functionResponse: {
-            ...(omitCallIds ? {} : { id: part.id }),
+            id: functionCallId(part.providerMetadata),
            name: part.name,
            response: {
              name: part.name,
@@ -405,7 +382,7 @@ const lowerMessages = Effect.fn("Gemini.lowerMessages")(function* (request: LLMR
      if (legacyToolMedia && media.length > 0) (pendingMedia ??= []).push(...media)
      parts.push({
        functionResponse: {
-          ...(omitCallIds ? {} : { id: part.id }),
+          id: functionCallId(part.providerMetadata),
          name: part.name,
          response: {
            name: part.name,
@@ -418,8 +395,8 @@ const lowerMessages = Effect.fn("Gemini.lowerMessages")(function* (request: LLMR
    // Gemini requires every response to a parallel call batch in one user turn,
    // so consecutive tool results join the open function-response turn.
    const previous = contents.at(-1)
-    if (previous?.role === "user" && (previous.parts ?? []).some((item) => "functionResponse" in item))
-      contents[contents.length - 1] = { role: "user", parts: [...(previous.parts ?? []), ...parts] }
+    if (previous?.role === "user" && previous.parts.some((item) => "functionResponse" in item))
+      contents[contents.length - 1] = { role: "user", parts: [...previous.parts, ...parts] }
    else contents.push({ role: "user", parts })
  }

@@ -507,28 +484,24 @@ const fromRequest = Effect.fn("Gemini.fromRequest")(function* (request: LLMReque
 // `cachedContentTokenCount` subset. `candidatesTokenCount` is *exclusive*
 // of `thoughtsTokenCount` — visible-only, not a total — so we sum the two
 // to produce the inclusive `outputTokens` the rest of the contract expects.
-const mapUsage = (usage: GeminiUsage | undefined, metadataKey: string) => {
+const mapUsage = (usage: GeminiUsage | undefined) => {
  if (!usage) return undefined
-  // Explicit provider nulls decode as `null`; normalize to `undefined` so the
-  // token arithmetic below treats them like absent counts.
-  const promptTokens = usage.promptTokenCount ?? undefined
-  const cached = usage.cachedContentTokenCount ?? undefined
-  const thoughts = usage.thoughtsTokenCount ?? undefined
-  const visible = usage.candidatesTokenCount ?? undefined
-  const nonCached = ProviderShared.subtractTokens(promptTokens, cached)
+  const cached = usage.cachedContentTokenCount
+  const nonCached = ProviderShared.subtractTokens(usage.promptTokenCount, cached)
  // `candidatesTokenCount` is visible-only; sum with thoughts to produce the
  // inclusive `outputTokens` the contract expects. Only compute the total
  // when the visible component is reported — otherwise we'd fabricate an
  // inclusive number from a partial breakdown.
-  const outputTokens = visible !== undefined ? visible + (thoughts ?? 0) : undefined
+  const outputTokens =
+    usage.candidatesTokenCount !== undefined ? usage.candidatesTokenCount + (usage.thoughtsTokenCount ?? 0) : undefined
  return new Usage({
-    inputTokens: promptTokens,
+    inputTokens: usage.promptTokenCount,
    outputTokens,
    nonCachedInputTokens: nonCached,
    cacheReadInputTokens: cached,
-    reasoningTokens: thoughts,
-    totalTokens: ProviderShared.totalTokens(promptTokens, outputTokens, usage.totalTokenCount ?? undefined),
-    providerMetadata: providerMetadata(metadataKey, usage),
+    reasoningTokens: usage.thoughtsTokenCount,
+    totalTokens: ProviderShared.totalTokens(usage.promptTokenCount, outputTokens, usage.totalTokenCount),
+    providerMetadata: { google: usage },
  })
 }

@@ -562,10 +535,7 @@ const mapFinishReason = (finishReason: string | undefined, hasToolCalls: boolean
 }

 const finish = (state: ParserState): ReadonlyArray<LLMEvent> => {
-  // `?? undefined` normalizes an explicit `null` blockReason back to absent so
-  // the "nothing to finish" check below keeps its meaning.
-  const promptBlockReason =
-    state.finishReason === undefined ? (state.promptFeedback?.blockReason ?? undefined) : undefined
+  const promptBlockReason = state.finishReason === undefined ? state.promptFeedback?.blockReason : undefined
  const finishReason = state.finishReason ?? promptBlockReason
  if (finishReason === undefined && state.usage === undefined) return []

@@ -576,15 +546,10 @@ const finish = (state: ParserState): ReadonlyArray<LLMEvent> => {
      lifecycle,
      events,
      "reasoning-0",
-      providerMetadata(state.providerMetadataKey, { thoughtSignature: state.reasoningSignature }),
+      googleMetadata({ thoughtSignature: state.reasoningSignature }),
    )
  if (state.textSignature !== undefined)
-    lifecycle = Lifecycle.textEnd(
-      lifecycle,
-      events,
-      "text-0",
-      providerMetadata(state.providerMetadataKey, { thoughtSignature: state.textSignature }),
-    )
+    lifecycle = Lifecycle.textEnd(lifecycle, events, "text-0", googleMetadata({ thoughtSignature: state.textSignature }))
  Lifecycle.finish(lifecycle, events, {
    reason: {
      normalized:
@@ -593,32 +558,16 @@ const finish = (state: ParserState): ReadonlyArray<LLMEvent> => {
    },
    usage: state.usage,
    providerMetadata:
-      state.promptFeedback === undefined
-        ? undefined
-        : providerMetadata(state.providerMetadataKey, { promptFeedback: state.promptFeedback }),
+      state.promptFeedback === undefined ? undefined : googleMetadata({ promptFeedback: state.promptFeedback }),
  })
  return events
 }

 const step = (state: ParserState, event: GeminiEvent) => {
-  if (ProviderShared.isRecord(event.error) && typeof event.error.message === "string") {
-    const body = ProviderShared.encodeJson(event)
-    return Effect.fail(
-      new AIError({
-        reason: classifyProviderFailure({
-          message: event.error.message,
-          status: typeof event.error.code === "number" ? event.error.code : undefined,
-          rawBody: body,
-        }),
-      }),
-    )
-  }
  const nextState = {
    ...state,
    promptFeedback: event.promptFeedback ?? state.promptFeedback,
-    usage: event.usageMetadata
-      ? (mapUsage(event.usageMetadata, state.providerMetadataKey) ?? state.usage)
-      : state.usage,
+    usage: event.usageMetadata ? (mapUsage(event.usageMetadata) ?? state.usage) : state.usage,
  }
  const candidate = event.candidates?.[0]
  if (!candidate?.content)
@@ -632,24 +581,8 @@ const step = (state: ParserState, event: GeminiEvent) => {
  let lifecycle = nextState.lifecycle
  let reasoningSignature = nextState.reasoningSignature
  let textSignature = nextState.textSignature
-  // Supplier ids must be tracked across chunks of the same response, not just within one event's parts.
-  const seenCallIds = new Set(nextState.seenCallIds)

-  for (const input of candidate.content.parts ?? []) {
-    if (
-      ProviderShared.isRecord(input) &&
-      !("text" in input) &&
-      !("inlineData" in input) &&
-      !("functionCall" in input) &&
-      !("functionResponse" in input)
-    )
-      continue
-    const decoded = decodeGeminiContentPart(input)
-    if (Option.isNone(decoded))
-      return Effect.fail(
-        ProviderShared.eventError(ADAPTER, `Invalid ${state.route} stream event`, ProviderShared.encodeJson(event)),
-      )
-    const part = decoded.value
+  for (const part of candidate.content.parts) {
    const signature = "thoughtSignature" in part && part.thoughtSignature ? part.thoughtSignature : undefined
    // Gemini attaches replay signatures to thought parts, visible text, or function calls;
    // each block kind must retain the signature attached to its own parts.
@@ -662,7 +595,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
          events,
          "reasoning-0",
          part.text,
-          signature ? providerMetadata(state.providerMetadataKey, { thoughtSignature: signature }) : undefined,
+          signature ? googleMetadata({ thoughtSignature: signature }) : undefined,
        )
        continue
      }
@@ -670,16 +603,14 @@ const step = (state: ParserState, event: GeminiEvent) => {
        lifecycle,
        events,
        "reasoning-0",
-        reasoningSignature
-          ? providerMetadata(state.providerMetadataKey, { thoughtSignature: reasoningSignature })
-          : undefined,
+        reasoningSignature ? googleMetadata({ thoughtSignature: reasoningSignature }) : undefined,
      )
      lifecycle = Lifecycle.textDelta(
        lifecycle,
        events,
        "text-0",
        part.text,
-        textSignature ? providerMetadata(state.providerMetadataKey, { thoughtSignature: textSignature }) : undefined,
+        textSignature ? googleMetadata({ thoughtSignature: textSignature }) : undefined,
      )
      textSignature = undefined
      continue
@@ -687,21 +618,18 @@ const step = (state: ParserState, event: GeminiEvent) => {

    if ("functionCall" in part) {
      const input = part.functionCall.args === undefined ? {} : part.functionCall.args
-      // Gemini 2.0+ supplies a unique function call ID on the part; when omitted (e.g. Gemini 1.5),
+      // Gemini 2.0+ and Vertex supply a unique function call ID on the part; when omitted (e.g. Gemini 1.5),
      // generate a globally unique ID rather than a per-request counter to prevent cross-request collisions in downstream registries.
-      // A repeated supplier id would replay as two identical calls, so only the first occurrence keeps it.
-      // A `null` supplier id normalizes to absent so the generated-id fallback applies.
-      const supplied = part.functionCall.id ?? undefined
-      const duplicate = supplied !== undefined && seenCallIds.has(supplied)
-      if (supplied !== undefined) seenCallIds.add(supplied)
-      const id = supplied !== undefined && !duplicate ? supplied : `tool_${crypto.randomUUID().replaceAll("-", "")}`
+      const id = part.functionCall.id ?? `tool_${crypto.randomUUID().replaceAll("-", "")}`
+      const metadata = {
+        ...(part.functionCall.id === undefined ? {} : { functionCallId: part.functionCall.id }),
+        ...(part.thoughtSignature === undefined ? {} : { thoughtSignature: part.thoughtSignature }),
+      }
      lifecycle = Lifecycle.reasoningEnd(
        lifecycle,
        events,
        "reasoning-0",
-        reasoningSignature
-          ? providerMetadata(state.providerMetadataKey, { thoughtSignature: reasoningSignature })
-          : undefined,
+        reasoningSignature ? googleMetadata({ thoughtSignature: reasoningSignature }) : undefined,
      )
      lifecycle = Lifecycle.stepStart(lifecycle, events)
      events.push(
@@ -709,9 +637,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
          id,
          name: part.functionCall.name,
          input,
-          providerMetadata: part.thoughtSignature
-            ? providerMetadata(state.providerMetadataKey, { thoughtSignature: part.thoughtSignature })
-            : undefined,
+          providerMetadata: Object.keys(metadata).length > 0 ? googleMetadata(metadata) : undefined,
        }),
      )
      hasToolCalls = true
@@ -725,7 +651,6 @@ const step = (state: ParserState, event: GeminiEvent) => {
      lifecycle,
      reasoningSignature,
      textSignature,
-      seenCallIds,
      finishReason: candidate.finishReason ?? nextState.finishReason,
    },
    events,
@@ -747,14 +672,9 @@ export const protocol = Protocol.make({
  },
  stream: {
    event: Protocol.jsonEvent(GeminiEvent),
-    initial: (request) => ({
-      route: `${request.model.provider}/${request.model.route.id}`,
-      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
-      hasToolCalls: false,
-      lifecycle: Lifecycle.initial(),
-    }),
+    initial: () => ({ hasToolCalls: false, lifecycle: Lifecycle.initial() }),
    step,
-    onHalt: (state) => Effect.succeed(finish(state)),
+    onHalt: finish,
  },
 })

@@ -9,7 +9,15 @@ import {
  type ImageRoute,
 } from "../image.js"
 import { Auth, type Definition as AuthDefinition } from "../route/auth.js"
-import { AIError, Usage, mergeHttpOptions, mergeJsonRecords, type HttpOptions } from "../schema/index.js"
+import {
+  InvalidProviderOutputReason,
+  AIError,
+  Usage,
+  mergeHttpOptions,
+  mergeJsonRecords,
+  type HttpOptions,
+  type ProviderMetadata,
+} from "../schema/index.js"
 import { ProviderShared } from "./shared.js"
 import { ImageInputs } from "./utils/image-input.js"

@@ -116,6 +124,13 @@ const nativeOptions = (options: GoogleImageOptions | undefined) => {
  )
 }

+const invalidOutput = (message: string, providerMetadata?: ProviderMetadata) =>
+  new AIError({
+    module: ADAPTER,
+    method: "generate",
+    reason: new InvalidProviderOutputReason({ message, route: ADAPTER, providerMetadata }),
+  })
+
 const applyQuery = (url: string, query: Record<string, string> | undefined) => {
  if (!query) return url
  const next = new URL(url)
@@ -154,9 +169,11 @@ export const model = (input: ModelInput) => {
          HttpClientRequest.bodyText(text, "application/json"),
        ),
      )
-      const output = yield* ProviderShared.imageResponse(ADAPTER, "Google Images", response)
-      const decoded = yield* Schema.decodeUnknownEffect(Schema.fromJsonString(GoogleImageResponse))(output.body).pipe(
-        Effect.mapError((cause) => output.invalid("Google Images returned an invalid response", cause)),
+      const payload = yield* response.json.pipe(
+        Effect.mapError(() => invalidOutput("Failed to read the Google Images response")),
+      )
+      const decoded = yield* Schema.decodeUnknownEffect(GoogleImageResponse)(payload).pipe(
+        Effect.mapError(() => invalidOutput("Google Images returned an invalid response")),
      )
      const candidates = decoded.candidates ?? []
      const candidateMetadata = candidates.map((candidate, candidateIndex) => ({
@@ -191,10 +208,9 @@ export const model = (input: ModelInput) => {
      )
      const images = yield* Effect.forEach(encoded, (item) =>
        Effect.fromResult(Encoding.decodeBase64(item.inlineData.data)).pipe(
-          Effect.mapError((cause) =>
-            output.invalid(
+          Effect.mapError(() =>
+            invalidOutput(
              `Google Images candidate ${item.candidateIndex} part ${item.partIndex} contains invalid base64 data`,
-              cause,
            ),
          ),
          Effect.map(
@@ -221,10 +237,16 @@ export const model = (input: ModelInput) => {
        const finishReasons = candidates.flatMap((candidate) =>
          candidate.finishReason === undefined ? [] : [candidate.finishReason],
        )
-        return yield* output.invalid(
+        return yield* invalidOutput(
          `Google Images returned no final images${
            finishReasons.length === 0 ? "" : ` (finish reasons: ${finishReasons.join(", ")})`
-          }; inspect body for prompt feedback and candidate details`,
+          }; inspect reason.providerMetadata.google for prompt feedback and candidate details`,
+          {
+            google: {
+              promptFeedback: decoded.promptFeedback,
+              candidates: candidateMetadata,
+            },
+          },
        )
      }
      const usage = decoded.usageMetadata
@@ -268,11 +290,12 @@ const googleImagePart = (image: ImageInput): Effect.Effect<Record<string, unknow
    return Effect.succeed({ inlineData: { mimeType: image.mediaType, data: Encoding.encodeBase64(image.data) } })
  if (image.type === "file-uri") return Effect.succeed({ fileData: { mimeType: image.mediaType, fileUri: image.uri } })
  if (image.type === "url")
-    return ImageInputs.decodeDataUrl(image.url).pipe(
+    return ImageInputs.decodeDataUrl(image.url, ADAPTER).pipe(
      Effect.flatMap((decoded) => {
        if (decoded === undefined)
          return Effect.fail(
            ImageInputs.invalid(
+              ADAPTER,
              "Google generateContent does not fetch public image URLs; use bytes, a data URL, or a Gemini file URI",
            ),
          )
@@ -282,7 +305,7 @@ const googleImagePart = (image: ImageInput): Effect.Effect<Record<string, unknow
      }),
    )
  return Effect.fail(
-    ImageInputs.invalid("Google generateContent requires Gemini file URIs rather than provider file IDs"),
+    ImageInputs.invalid(ADAPTER, "Google generateContent requires Gemini file URIs rather than provider file IDs"),
  )
 }

@@ -8,4 +8,3 @@ export * as OpenAICompatibleResponses from "./openai-compatible-responses.js"
 export * as OpenAIResponses from "./openai-responses.js"
 export * as OpenResponses from "./open-responses.js"
 export * as OpenResponsesChannel from "./open-responses-channel.js"
-export * as XAIResponses from "./xai-responses.js"
@@ -1,7 +1,6 @@
 import { Effect, Schema, Stream } from "effect"
 import { Headers } from "effect/unstable/http"
 import { Framing } from "../route/framing.js"
-import type { HttpContext } from "../schema/index.js"
 import {
  HttpTransport,
  WebSocketTransport,
@@ -11,7 +10,6 @@ import {
 } from "../route/transport/index.js"
 import * as ProviderShared from "./shared.js"
 import { OpenResponses } from "./open-responses.js"
-import { OpenResponsesContinuation } from "./open-responses-continuation.js"

 const WebSocketResponseCreate = Schema.StructWithRest(Schema.Struct({ type: Schema.tag("response.create") }), [
  Schema.Record(Schema.String, Schema.Unknown),
@@ -27,6 +25,11 @@ export interface Options {
  readonly enabled?: (url: string) => boolean
  readonly url?: (url: string) => string
  readonly headers?: (headers: Headers.Headers) => Headers.Headers
+  readonly driver?: (input: {
+    readonly request: Readonly<Record<string, unknown>>
+    readonly message: string
+    readonly base: WebSocketChannelDriver
+  }) => WebSocketChannelDriver
 }

 export interface Prepared {
@@ -61,8 +64,8 @@ const driver = (options: Options, body: string): WebSocketChannelDriver => {
    observe: (_create, frame) =>
      Effect.gen(function* () {
        const event = yield* decodeEvent(frame).pipe(
-          Effect.mapError((cause) =>
-            ProviderShared.eventError(options.id, `Invalid ${options.name} WebSocket event`, frame, cause),
+          Effect.mapError(() =>
+            ProviderShared.eventError(options.id, `Invalid ${options.name} WebSocket event`, frame),
          ),
        )
        if (terminal)
@@ -74,13 +77,13 @@ const driver = (options: Options, body: string): WebSocketChannelDriver => {
        if (event.type === "error") {
          terminal = true
          yield* OpenResponses.decodeKnownErrorEvent(event).pipe(
-            Effect.mapError((cause) =>
-              ProviderShared.eventError(options.id, `${options.name} returned a malformed error event`, frame, cause),
+            Effect.mapError(() =>
+              ProviderShared.eventError(options.id, `${options.name} returned a malformed error event`, frame),
            ),
          )
          return {
            type: "provider-failure",
-            error: OpenResponses.providerFailure(event, `${options.name} stream error`, frame),
+            error: OpenResponses.providerFailure(options.id, event, `${options.name} stream error`),
          }
        }
        if (event.type === "response.failed") {
@@ -93,7 +96,7 @@ const driver = (options: Options, body: string): WebSocketChannelDriver => {
            )
          return {
            type: "provider-failure",
-            error: OpenResponses.providerFailure(event, `${options.name} response failed`, frame),
+            error: OpenResponses.providerFailure(options.id, event, `${options.name} response failed`),
          }
        }
        if (event.type === "response.created") {
@@ -113,8 +116,6 @@ const driver = (options: Options, body: string): WebSocketChannelDriver => {
          responseID = created
          return { type: "frame", frame }
        }
-        // Keepalives carry no response state and may arrive before response.created.
-        if (event.type === "keepalive") return { type: "frame", frame }
        if (!responseID)
          return yield* ProviderShared.eventError(
            options.id,
@@ -157,13 +158,7 @@ export const transport = <Body>(options: Options): Transport<Body, Prepared, str
                  url: yield* WebSocketTransport.toWebSocketUrl(options.url?.(parts.url) ?? parts.url),
                  headers,
                  rotateAfterMs: options.rotateAfterMs,
-                  driver: OpenResponsesContinuation.driver({
-                    id: options.id,
-                    name: options.name,
-                    request: create.request,
-                    message: create.message,
-                    base,
-                  }),
+                  driver: options.driver?.({ request: create.request, message: create.message, base }) ?? base,
                }
              })
            : undefined
@@ -176,37 +171,23 @@ export const transport = <Body>(options: Options): Transport<Body, Prepared, str
          channel,
        }
      }),
-    execute: (prepared, request, runtime, executeOptions) =>
-      Effect.gen(function* () {
-        if (!executeOptions?.webSocket || !prepared.channel) return yield* http.execute(prepared.http, request, runtime)
-        let fallbackHttp: HttpContext | undefined
-        const exchange: WebSocketChannelExchange = {
-          id: request.id ?? "request",
-          connect: {
-            url: prepared.channel.url,
-            headers: prepared.channel.headers,
-            rotateAfterMs: prepared.channel.rotateAfterMs,
-          },
-          fallback: () =>
-            Stream.unwrap(
-              http.execute(prepared.http, request, runtime).pipe(
-                Effect.map((execution) => {
-                  fallbackHttp = execution.http
-                  return execution.frames
-                }),
-              ),
-            ),
-          driver: prepared.channel.driver,
-        }
-        const execution = yield* executeOptions.webSocket.execute(exchange)
-        return {
-          frames: execution.frames,
-          complete: execution.complete,
-          get http() {
-            return fallbackHttp ?? execution.http
-          },
-        }
-      }),
+    execute: (prepared, request, runtime, executeOptions) => {
+      if (!executeOptions?.webSocket || !prepared.channel) return http.execute(prepared.http, request, runtime)
+      const exchange: WebSocketChannelExchange = {
+        id: request.id ?? "request",
+        connect: {
+          url: prepared.channel.url,
+          headers: prepared.channel.headers,
+          rotateAfterMs: prepared.channel.rotateAfterMs,
+        },
+        fallback: () =>
+          Stream.unwrap(
+            http.execute(prepared.http, request, runtime).pipe(Effect.map((execution) => execution.frames)),
+          ),
+        driver: prepared.channel.driver,
+      }
+      return executeOptions.webSocket.execute(exchange)
+    },
  }
 }

@@ -5,7 +5,6 @@ import { Protocol } from "../route/protocol.js"
 import {
  AIError,
  LLMEvent,
-  ProviderInternalError,
  Usage,
  type FinishReason,
  type JsonSchema,
@@ -43,12 +42,8 @@ const OpenResponsesInputImage = Schema.Struct({
 const OpenResponsesInputFile = Schema.Struct({
  type: Schema.tag("input_file"),
  filename: Schema.String,
-  file_data: Schema.optional(Schema.String),
-  file_url: Schema.optional(Schema.String),
-})
-const OpenResponsesInputVideo = Schema.Struct({
-  type: Schema.tag("input_video"),
-  video_url: Schema.String,
+  file_data: Schema.String,
+  mime_type: Schema.optional(Schema.String),
 })
 const MediaInput = Schema.Union([OpenResponsesInputImage, OpenResponsesInputFile])
 export type MediaInput = Schema.Schema.Type<typeof MediaInput>
@@ -59,14 +54,9 @@ const OpenResponsesOutputText = Schema.Struct({
  text: Schema.String,
 })

-export const MessagePhase = Schema.NullOr(Schema.Literals(["commentary", "final_answer"]))
+export const MessagePhase = Schema.Literals(["commentary", "final_answer"])
 type MessagePhase = Schema.Schema.Type<typeof MessagePhase>

-const messagePhase = (value: unknown): MessagePhase | undefined => {
-  if (value === null || value === "commentary" || value === "final_answer") return value
-  return undefined
-}
-
 const OpenResponsesReasoningSummaryText = Schema.Struct({
  type: Schema.tag("summary_text"),
  text: Schema.String,
@@ -79,60 +69,10 @@ const OpenResponsesReasoningItem = Schema.Struct({
  encrypted_content: optionalNull(Schema.String),
 })

-const OpenResponsesWebSearchCall = Schema.StructWithRest(
-  Schema.Struct({
-    type: Schema.tag("web_search_call"),
-    id: Schema.String,
-    status: Schema.optional(Schema.String),
-    action: optionalNull(JsonObject),
-  }),
-  [JsonObject],
-)
-
-const OpenResponsesFileSearchCall = Schema.StructWithRest(
-  Schema.Struct({
-    type: Schema.tag("file_search_call"),
-    id: Schema.String,
-    status: Schema.optional(Schema.String),
-    queries: Schema.optional(Schema.Array(Schema.String)),
-    results: optionalNull(Schema.Array(JsonObject)),
-  }),
-  [JsonObject],
-)
-
-const OpenResponsesCodeInterpreterCall = Schema.StructWithRest(
-  Schema.Struct({
-    type: Schema.tag("code_interpreter_call"),
-    id: Schema.String,
-    status: Schema.optional(Schema.String),
-    code: optionalNull(Schema.String),
-    container_id: optionalNull(Schema.String),
-    outputs: optionalNull(Schema.Array(JsonObject)),
-  }),
-  [JsonObject],
-)
-
-const OpenResponsesMCPCall = Schema.StructWithRest(
-  Schema.Struct({
-    type: Schema.tag("mcp_call"),
-    id: Schema.String,
-    status: Schema.optional(Schema.String),
-    server_label: Schema.optional(Schema.String),
-    name: Schema.optional(Schema.String),
-    arguments: Schema.optional(Schema.String),
-    output: optionalNull(Schema.String),
-    error: Schema.optional(Schema.Unknown),
-  }),
-  [JsonObject],
-)
-
-export const HostedToolItem = Schema.Union([
-  OpenResponsesWebSearchCall,
-  OpenResponsesFileSearchCall,
-  OpenResponsesCodeInterpreterCall,
-  OpenResponsesMCPCall,
-])
-export type HostedToolItem = Schema.Schema.Type<typeof HostedToolItem>
+const OpenResponsesItemReference = Schema.Struct({
+  type: Schema.tag("item_reference"),
+  id: Schema.String,
+})

 // `function_call_output.output` accepts either a plain string or an ordered
 // array of content items so tools can return images and files in addition to text.
@@ -141,7 +81,6 @@ const OpenResponsesFunctionCallOutputContent = Schema.Union([
  OpenResponsesInputText,
  OpenResponsesInputImage,
  OpenResponsesInputFile,
-  OpenResponsesInputVideo,
 ])

 const OpenResponsesFunctionCallOutput = Schema.Union([
@@ -161,6 +100,7 @@ export const InputItem = Schema.Union([
    phase: Schema.optionalKey(MessagePhase),
  }),
  OpenResponsesReasoningItem,
+  OpenResponsesItemReference,
  Schema.Struct({
    type: Schema.tag("function_call"),
    id: Schema.optionalKey(Schema.String),
@@ -173,17 +113,10 @@ export const InputItem = Schema.Union([
    call_id: Schema.String,
    output: OpenResponsesFunctionCallOutput,
  }),
-  HostedToolItem,
 ])
 type OpenResponsesInputItem = Schema.Schema.Type<typeof InputItem>
-export type ExtendedHostedToolItem = {
-  readonly type: string
-  readonly id: string
-  readonly [key: string]: unknown
-}
 type LoweredInputItem =
  | OpenResponsesInputItem
-  | ExtendedHostedToolItem
  | {
      readonly type: "message"
      readonly id?: string
@@ -196,7 +129,7 @@ type LoweredInputItem =
 // multiple streamed summary parts into the same item before flushing.
 type OpenResponsesReasoningInput = {
  type: "reasoning"
-  id?: string
+  id: string
  summary: Array<{ type: "summary_text"; text: string }>
  encrypted_content?: string | null
 }
@@ -304,7 +237,6 @@ const OpenResponsesErrorPayload = Schema.Struct({
  message: optionalNull(Schema.String),
  param: optionalNull(Schema.String),
 })
-type OpenResponsesErrorPayload = Schema.Schema.Type<typeof OpenResponsesErrorPayload>

 const WebSocketErrorHeader = Schema.Union([Schema.String, Schema.Number, Schema.Boolean])
 export const WebSocketErrorEvent = Schema.StructWithRest(
@@ -341,20 +273,16 @@ export const Event = Schema.StructWithRest(
  Schema.Struct({
    type: Schema.String,
    delta: Schema.optional(Schema.String),
-    arguments: Schema.optional(Schema.String),
    text: Schema.optional(Schema.String),
    item_id: Schema.optional(Schema.String),
-    output_index: Schema.optional(Schema.Number),
    summary_index: Schema.optional(Schema.Number),
-    // OutputItemAdded/Done permit a null item in the Open Responses OpenAPI schema.
-    item: optionalNull(StreamItem),
+    item: Schema.optional(StreamItem),
    response: Schema.optional(
      Schema.StructWithRest(
        Schema.Struct({
          id: Schema.optional(Schema.String),
          service_tier: optionalNull(Schema.String),
          incomplete_details: optionalNull(Schema.Struct({ reason: Schema.optional(Schema.String) })),
-          output: Schema.optional(Schema.Array(StreamItem)),
          usage: optionalNull(OpenResponsesUsage),
          error: optionalNull(OpenResponsesErrorPayload),
        }),
@@ -373,6 +301,20 @@ export const Event = Schema.StructWithRest(
 )
 export type Event = Schema.Schema.Type<typeof Event>

+const RefusalEvent = Schema.Union([
+  Schema.Struct({
+    type: Schema.tag("response.refusal.delta"),
+    item_id: Schema.String,
+    delta: Schema.String,
+  }),
+  Schema.Struct({
+    type: Schema.tag("response.refusal.done"),
+    item_id: Schema.String,
+    refusal: Schema.String,
+  }),
+])
+const isRefusalEvent = Schema.is(RefusalEvent)
+
 export interface Extension {
  readonly id: string
  readonly name: string
@@ -381,7 +323,7 @@ export interface Extension {
    readonly media: ProviderShared.NormalizedMedia
    readonly request: LLMRequest
  }) => MediaInput | undefined
-  readonly lowerHostedToolItem?: (item: unknown) => ExtendedHostedToolItem | undefined
+  readonly messagePhase?: (value: unknown) => MessagePhase | null | undefined
 }

 const BASE: Extension = { id: ADAPTER, name: NAME }
@@ -393,10 +335,11 @@ export interface ParserState {
  readonly tools: ToolStream.State<string>
  readonly hasFunctionCall: boolean
  readonly lifecycle: Lifecycle.State
-  readonly outputItems: Readonly<Record<number, string>>
  readonly messageItems: ReadonlySet<string>
+  readonly messagePhase: (value: unknown) => MessagePhase | null | undefined
  readonly messagePhases: Readonly<Record<string, MessagePhase | null>>
  readonly reasoningItems: Readonly<Record<string, ReasoningStreamItem>>
+  readonly store: boolean | undefined
 }

 type ReasoningSummaryStatus = "active" | "can-conclude" | "concluded"
@@ -407,10 +350,6 @@ interface ReasoningStreamItem {
  // strings, but typing the map as `Record<number, ...>` documents intent
  // and matches the wire field.
  readonly summaryParts: Readonly<Record<number, ReasoningSummaryStatus>>
-  // Summary indexes that received at least one streamed delta. The `:0` block
-  // is started eagerly when the item opens, so block existence cannot tell
-  // whether a `.done` final would duplicate streamed text.
-  readonly deltaIndexes: ReadonlySet<number>
 }

 // =============================================================================
@@ -427,7 +366,7 @@ export const lowerTool = Effect.fn("OpenResponses.lowerTool")(function* (
    type: "function" as const,
    name: tool.name,
    description: tool.description,
-    parameters: inputSchema,
+    parameters: ToolSchemaProjection.responses(inputSchema),
    // The common tool definition does not currently express Responses strict-schema policy.
    strict: false,
  }
@@ -441,20 +380,18 @@ export const lowerToolChoice = (protocolName: string, toolChoice: NonNullable<LL
    tool: (toolName) => ({ type: "function" as const, name: toolName }),
  })

-// Server-issued item ids need a nonempty prefix and suffix, but the prefix is
-// provider-defined and does not necessarily identify the item's semantic type.
 const itemID = (providerMetadata: ProviderMetadata | undefined, providerMetadataKey: string) => {
  const metadata = providerMetadata?.[providerMetadataKey]
-  if (!ProviderShared.isRecord(metadata) || typeof metadata.itemId !== "string") return undefined
-  const separator = metadata.itemId.indexOf("_")
-  return separator > 0 && separator < metadata.itemId.length - 1 ? metadata.itemId : undefined
+  return ProviderShared.isRecord(metadata) && typeof metadata.itemId === "string" && metadata.itemId.length > 0
+    ? metadata.itemId
+    : undefined
 }

 const lowerToolCall = (part: ToolCallPart, providerMetadataKey: string): OpenResponsesInputItem => {
  const id = itemID(part.providerMetadata, providerMetadataKey)
  return {
    type: "function_call",
-    ...(id === undefined ? {} : { id }),
+    ...(id ? { id } : {}),
    call_id: part.id,
    name: part.name,
    arguments: ProviderShared.encodeJson(part.input),
@@ -463,43 +400,40 @@ const lowerToolCall = (part: ToolCallPart, providerMetadataKey: string): OpenRes

 const lowerReasoning = (part: ReasoningPart, providerMetadataKey: string): OpenResponsesReasoningInput | undefined => {
  const metadata = part.providerMetadata?.[providerMetadataKey]
-  if (!ProviderShared.isRecord(metadata)) return undefined
  const id = itemID(part.providerMetadata, providerMetadataKey)
+  if (!ProviderShared.isRecord(metadata) || !id) return undefined
  const encryptedContent =
    typeof metadata.reasoningEncryptedContent === "string" || metadata.reasoningEncryptedContent === null
      ? metadata.reasoningEncryptedContent
      : undefined
  return {
    type: "reasoning",
-    ...(id === undefined ? {} : { id }),
+    id,
    summary: part.text.length > 0 ? [{ type: "summary_text", text: part.text }] : [],
    encrypted_content: encryptedContent,
  }
 }

+const hostedToolItemID = (part: ToolResultPart, providerMetadataKey: string) => {
+  return itemID(part.providerMetadata, providerMetadataKey)
+}
+
 const lowerMedia = Effect.fn("OpenResponses.lowerMedia")(function* (
  part: MediaPart,
  request: LLMRequest,
  extension: Extension,
-  target: "message" | "tool-result",
 ) {
  const media = ProviderShared.normalizeMedia(part)
  const extended = extension.lowerMedia?.({ part, media, request })
  if (extended) return extended
-  const url =
-    typeof part.data === "string" && (part.data.startsWith("https://") || part.data.startsWith("http://"))
-      ? part.data
-      : undefined
  if (!media.mime.startsWith("image/")) {
-    if (target === "tool-result" && media.mime.startsWith("video/"))
-      return { type: "input_video" as const, video_url: url ?? media.dataUrl }
    return {
      type: "input_file" as const,
      filename: part.filename ?? (media.mime === "application/pdf" ? "document.pdf" : "file"),
-      ...(url ? { file_url: url } : { file_data: media.dataUrl }),
+      file_data: media.dataUrl,
    }
  }
-  return { type: "input_image" as const, image_url: url ?? media.dataUrl }
+  return { type: "input_image" as const, image_url: media.dataUrl }
 })

 const lowerUserContent = Effect.fnUntraced(function* (
@@ -508,17 +442,10 @@ const lowerUserContent = Effect.fnUntraced(function* (
  extension: Extension,
 ) {
  if (part.type === "text") return { type: "input_text" as const, text: part.text }
-  if (part.type === "media") return yield* lowerMessageMedia(part, request, extension)
+  if (part.type === "media") return yield* lowerMedia(part, request, extension)
  return yield* ProviderShared.unsupportedContent(extension.name, "user", ["text", "media"])
 })

-const lowerMessageMedia = Effect.fnUntraced(function* (part: MediaPart, request: LLMRequest, extension: Extension) {
-  const lowered = yield* lowerMedia(part, request, extension, "message")
-  if (lowered.type === "input_video")
-    return yield* ProviderShared.invalidRequest(`${extension.name} user messages do not support input_video`)
-  return lowered
-})
-
 // Tool results may carry structured text, images, and files. Keep media as provider-native
 // content instead of JSON-stringifying base64 into a prompt string.
 const lowerToolResultContentItem = Effect.fnUntraced(function* (
@@ -531,20 +458,6 @@ const lowerToolResultContentItem = Effect.fnUntraced(function* (
    { type: "media", mediaType: item.mime, data: item.uri, filename: item.name },
    request,
    extension,
-    "tool-result",
-  )
-})
-
-const lowerHostedToolResultContentItem = Effect.fnUntraced(function* (
-  item: Content,
-  request: LLMRequest,
-  extension: Extension,
-) {
-  if (item.type === "text") return { type: "input_text" as const, text: item.text }
-  return yield* lowerMessageMedia(
-    { type: "media", mediaType: item.mime, data: item.uri, filename: item.name },
-    request,
-    extension,
  )
 })

@@ -562,7 +475,10 @@ const lowerToolResultOutput = Effect.fnUntraced(function* (
 })

 const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (request: LLMRequest, extension: Extension) {
-  const input: LoweredInputItem[] = []
+  const system: LoweredInputItem[] =
+    request.system.length === 0 ? [] : [{ role: "system", content: ProviderShared.joinText(request.system) }]
+  const input: LoweredInputItem[] = [...system]
+  const store = OpenResponsesOptions.resolve(request).store
  const providerMetadataKey = request.model.route.providerMetadataKey ?? "openresponses"

  for (const message of request.messages) {
@@ -585,7 +501,8 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
    if (message.role === "assistant") {
      const content: TextPart[] = []
      const reasoningItems: Record<string, OpenResponsesReasoningInput> = {}
-      const hostedToolItems = new Set<string>()
+      const reasoningReferences = new Set<string>()
+      const hostedToolReferences = new Set<string>()
      const flushText = () => {
        if (content.length === 0) return
        const groups = content.reduce<
@@ -593,7 +510,7 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
        >((groups, part) => {
          const metadata = part.providerMetadata?.[providerMetadataKey]
          const id = itemID(part.providerMetadata, providerMetadataKey)
-          const phase = ProviderShared.isRecord(metadata) ? messagePhase(metadata.phase) : undefined
+          const phase = ProviderShared.isRecord(metadata) ? messagePhase(metadata.phase, extension) : undefined
          const group = groups.at(-1)
          if (group && group.id === id && group.phase === phase) group.parts.push(part)
          else groups.push({ id, phase, parts: [part] })
@@ -619,14 +536,19 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
          flushText()
          const reasoning = lowerReasoning(part, providerMetadataKey)
          if (!reasoning) continue
-          const existing = reasoning.id === undefined ? undefined : reasoningItems[reasoning.id]
+          if (store !== false) {
+            if (!reasoningReferences.has(reasoning.id)) input.push({ type: "item_reference", id: reasoning.id })
+            reasoningReferences.add(reasoning.id)
+            continue
+          }
+          const existing = reasoningItems[reasoning.id]
          if (existing) {
            existing.summary.push(...reasoning.summary)
            if (typeof reasoning.encrypted_content === "string")
              existing.encrypted_content = reasoning.encrypted_content
            continue
          }
-          if (reasoning.id !== undefined) reasoningItems[reasoning.id] = reasoning
+          reasoningItems[reasoning.id] = reasoning
          input.push(reasoning)
          continue
        }
@@ -638,30 +560,17 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
        }
        if (part.type === "tool-result" && part.providerExecuted === true) {
          flushText()
-          const id = itemID(part.providerMetadata, providerMetadataKey)
-          const hosted =
-            part.result.type !== "json"
-              ? undefined
-              : Schema.is(HostedToolItem)(part.result.value)
-                ? part.result.value
-                : extension.lowerHostedToolItem?.(part.result.value)
-          if (id !== undefined && hosted?.id === id) {
-            if (!hostedToolItems.has(id)) {
-              input.push(hosted)
-              hostedToolItems.add(id)
-            }
-            continue
+          const itemID = hostedToolItemID(part, providerMetadataKey)
+          if (store !== false && itemID && !hostedToolReferences.has(itemID))
+            input.push({ type: "item_reference", id: itemID })
+          if (store === false && part.result.type === "content") {
+            const content: ReadonlyArray<Content> = part.result.value
+            input.push({
+              role: "user",
+              content: yield* Effect.forEach(content, (item) => lowerToolResultContentItem(item, request, extension)),
+            })
          }
-          const content: ReadonlyArray<Content> =
-            part.result.type === "content"
-              ? part.result.value
-              : [{ type: "text", text: ProviderShared.toolResultText(part) }]
-          input.push({
-            role: "user",
-            content: yield* Effect.forEach(content, (item) =>
-              lowerHostedToolResultContentItem(item, request, extension),
-            ),
-          })
+          if (itemID) hostedToolReferences.add(itemID)
          continue
        }
        return yield* ProviderShared.unsupportedContent(extension.name, "assistant", [
@@ -686,16 +595,20 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
    }
  }

-  return input
+  // With store:false, Responses APIs only accept previous reasoning items when the
+  // complete item has encrypted state. Summary blocks for one item may carry
+  // that state only on the last block, so filter after they have been joined.
+  return store === false
+    ? input.filter(
+        (item) => !("type" in item) || item.type !== "reasoning" || typeof item.encrypted_content === "string",
+      )
+    : input
 })

 const lowerOptions = (request: LLMRequest) => {
  const options = OpenResponsesOptions.resolve(request)
-  const instructions = ProviderShared.joinText(request.system)
-  const cacheKey = ProviderShared.promptCacheKey(request)
-  const parallelToolCalls = resolveParallelToolCalls(request)
  return {
-    ...(instructions ? { instructions } : {}),
+    ...(options.instructions ? { instructions: options.instructions } : {}),
    ...(options.store !== undefined ? { store: options.store } : {}),
    ...(options.metadata ? { metadata: options.metadata } : {}),
    ...(options.safetyIdentifier ? { safety_identifier: options.safetyIdentifier } : {}),
@@ -703,7 +616,7 @@ const lowerOptions = (request: LLMRequest) => {
      ? { stream_options: { include_obfuscation: options.streamOptions.includeObfuscation } }
      : {}),
    ...(options.topLogprobs !== undefined ? { top_logprobs: options.topLogprobs } : {}),
-    ...(cacheKey ? { prompt_cache_key: cacheKey } : {}),
+    ...(request.promptCacheKey ? { prompt_cache_key: request.promptCacheKey } : {}),
    ...(options.include ? { include: options.include } : {}),
    ...(options.reasoningEffort || options.reasoningSummary
      ? { reasoning: { effort: options.reasoningEffort, summary: options.reasoningSummary } }
@@ -711,18 +624,11 @@ const lowerOptions = (request: LLMRequest) => {
    ...(options.textVerbosity ? { text: { verbosity: options.textVerbosity } } : {}),
    ...(options.serviceTier ? { service_tier: options.serviceTier } : {}),
    ...(options.maxToolCalls !== undefined ? { max_tool_calls: options.maxToolCalls } : {}),
-    ...(parallelToolCalls !== undefined ? { parallel_tool_calls: parallelToolCalls } : {}),
+    ...(options.parallelToolCalls !== undefined ? { parallel_tool_calls: options.parallelToolCalls } : {}),
    ...(options.truncation ? { truncation: options.truncation } : {}),
  }
 }

-export const resolveParallelToolCalls = (request: LLMRequest) => {
-  const configured = OpenResponsesOptions.resolve(request).parallelToolCalls
-  if (configured !== undefined) return configured
-  const disabled = request.toolChoice?.disableParallelToolUse
-  return disabled === undefined ? undefined : !disabled
-}
-
 const allowedToolChoice = (request: LLMRequest) => {
  const allowed = OpenResponsesOptions.resolve(request).allowedTools
  if (!allowed) return undefined
@@ -813,7 +719,7 @@ export const providerMetadata = (state: ParserState, metadata: Record<string, un
 })

 const isReasoningItem = (item: StreamItem): item is StreamItem & { type: "reasoning"; id: string } =>
-  item.type === "reasoning" && typeof item.id === "string"
+  item.type === "reasoning" && typeof item.id === "string" && item.id.length > 0

 export type StepResult = readonly [ParserState, ReadonlyArray<LLMEvent>]

@@ -826,7 +732,7 @@ const TERMINAL_TYPES = new Set(["error", "response.completed", "response.incompl
 export const terminal = (event: Event) => TERMINAL_TYPES.has(event.type)

 const onOutputTextDelta = (state: ParserState, event: Event, id: string): StepResult => {
-  if (!event.delta || !state.messageItems.has(id)) return [state, NO_EVENTS]
+  if (!event.delta) return [state, NO_EVENTS]
  const events: LLMEvent[] = []
  const phase = state.messagePhases[id]
  const metadata = providerMetadata(state, { itemId: id, ...(phase === undefined ? {} : { phase }) })
@@ -843,37 +749,21 @@ const onOutputTextDone = (state: ParserState, event: Event, id: string): StepRes
  return [{ ...state, lifecycle: Lifecycle.textEnd(state.lifecycle, events, id) }, events]
 }

-export const outputItemID = (state: ParserState, event: Event) =>
-  event.output_index === undefined ? event.item_id : (state.outputItems[event.output_index] ?? event.item_id)
-
 export const onReasoningDelta = (state: ParserState, event: Event, itemID: string): StepResult => {
-  const item = state.reasoningItems[itemID]
-  if (!event.delta || !item) return [state, NO_EVENTS]
-  const index = event.summary_index ?? 0
+  if (!event.delta) return [state, NO_EVENTS]
  const events: LLMEvent[] = []
+  const id =
+    event.summary_index !== undefined || state.reasoningItems[itemID] ? `${itemID}:${event.summary_index ?? 0}` : itemID
  return [
    {
      ...state,
-      lifecycle: Lifecycle.reasoningDelta(state.lifecycle, events, `${itemID}:${index}`, event.delta),
-      reasoningItems: {
-        ...state.reasoningItems,
-        [itemID]: { ...item, deltaIndexes: new Set([...item.deltaIndexes, index]) },
-      },
+      lifecycle: Lifecycle.reasoningDelta(state.lifecycle, events, id, event.delta),
    },
    events,
  ]
 }

-// Some compatible gateways emit a reasoning final without streaming any
-// deltas, mirroring `response.output_text.done`. Reconcile the complete text
-// as a single delta unless that summary index already streamed one.
-export const onReasoningDone = (state: ParserState, event: Event, itemID: string): StepResult => {
-  const item = state.reasoningItems[itemID]
-  if (!item || typeof event.text !== "string") return [state, NO_EVENTS]
-  const index = event.summary_index ?? 0
-  if (item.deltaIndexes.has(index)) return [state, NO_EVENTS]
-  return onReasoningDelta(state, { ...event, delta: event.text }, itemID)
-}
+export const onReasoningDone = (state: ParserState, _event: Event): StepResult => [state, NO_EVENTS]

 const reasoningMetadata = (state: ParserState, item: StreamItem & { id: string }) =>
  providerMetadata(state, { itemId: item.id, reasoningEncryptedContent: item.encrypted_content ?? null })
@@ -892,17 +782,18 @@ const reasoningMetadata = (state: ParserState, item: StreamItem & { id: string }
 // best-effort, not guaranteed.
 const onOutputItemAdded = (state: ParserState, event: Event): StepResult => {
  const item = event.item
-  if (item?.type === "message" && item.id !== undefined) {
-    const phase = messagePhase(item.phase)
+  if (item?.type === "message" && item.id)
    return [
      {
        ...state,
        messageItems: new Set([...state.messageItems, item.id]),
-        messagePhases: phase === undefined ? state.messagePhases : { ...state.messagePhases, [item.id]: phase },
+        messagePhases: (() => {
+          const phase = state.messagePhase(item.phase)
+          return phase === undefined ? state.messagePhases : { ...state.messagePhases, [item.id]: phase }
+        })(),
      },
      NO_EVENTS,
    ]
-  }
  if (item && isReasoningItem(item)) {
    const events: LLMEvent[] = []
    return [
@@ -911,41 +802,57 @@ const onOutputItemAdded = (state: ParserState, event: Event): StepResult => {
        lifecycle: Lifecycle.reasoningStart(state.lifecycle, events, `${item.id}:0`, reasoningMetadata(state, item)),
        reasoningItems: {
          ...state.reasoningItems,
-          [item.id]: {
-            encryptedContent: item.encrypted_content,
-            summaryParts: { 0: "active" },
-            deltaIndexes: new Set(),
-          },
+          [item.id]: { encryptedContent: item.encrypted_content, summaryParts: { 0: "active" } },
        },
      },
      events,
    ]
  }
-  if (item?.type !== "function_call" || !item.call_id) return [state, NO_EVENTS]
-  const id = item.id ?? item.call_id
-  const metadata = item.id !== undefined ? providerMetadata(state, { itemId: item.id }) : undefined
+  if (item?.type !== "function_call" || !item.id) return [state, NO_EVENTS]
+  const metadata = providerMetadata(state, { itemId: item.id })
  const events: LLMEvent[] = []
  const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
  return [
    {
      ...state,
      lifecycle,
-      tools: ToolStream.start(state.tools, id, {
-        id: item.call_id,
+      tools: ToolStream.start(state.tools, item.id, {
+        id: item.call_id ?? item.id,
        name: item.name ?? "",
        input: item.arguments ?? "",
        providerMetadata: metadata,
      }),
    },
-    [...events, LLMEvent.toolInputStart({ id: item.call_id, name: item.name ?? "", providerMetadata: metadata })],
+    [
+      ...events,
+      LLMEvent.toolInputStart({ id: item.call_id ?? item.id, name: item.name ?? "", providerMetadata: metadata }),
+    ],
  ]
 }

 const onReasoningSummaryPartAdded = (state: ParserState, event: Event): StepResult => {
-  if (event.item_id === undefined || event.summary_index === undefined) return [state, NO_EVENTS]
-  const item = state.reasoningItems[event.item_id]
-  if (!item) return [state, NO_EVENTS]
-  if (event.summary_index === 0) return [state, NO_EVENTS]
+  if (!event.item_id || event.summary_index === undefined) return [state, NO_EVENTS]
+  const item = state.reasoningItems[event.item_id] ?? { encryptedContent: undefined, summaryParts: {} }
+  if (event.summary_index === 0) {
+    if (state.reasoningItems[event.item_id]) return [state, NO_EVENTS]
+    const events: LLMEvent[] = []
+    return [
+      {
+        ...state,
+        lifecycle: Lifecycle.reasoningStart(
+          state.lifecycle,
+          events,
+          `${event.item_id}:0`,
+          providerMetadata(state, { itemId: event.item_id, reasoningEncryptedContent: null }),
+        ),
+        reasoningItems: {
+          ...state.reasoningItems,
+          [event.item_id]: { ...item, summaryParts: { 0: "active" } },
+        },
+      },
+      events,
+    ]
+  }

  const events: LLMEvent[] = []
  const closed = Object.entries(item.summaryParts)
@@ -989,24 +896,34 @@ const onReasoningSummaryPartAdded = (state: ParserState, event: Event): StepResu
 }

 const onReasoningSummaryPartDone = (state: ParserState, event: Event): StepResult => {
-  if (event.item_id === undefined || event.summary_index === undefined) return [state, NO_EVENTS]
+  if (!event.item_id || event.summary_index === undefined) return [state, NO_EVENTS]
  const item = state.reasoningItems[event.item_id]
  if (!item) return [state, NO_EVENTS]
+  const events: LLMEvent[] = []
  return [
    {
      ...state,
+      lifecycle:
+        state.store !== false
+          ? Lifecycle.reasoningEnd(
+              state.lifecycle,
+              events,
+              `${event.item_id}:${event.summary_index}`,
+              providerMetadata(state, { itemId: event.item_id }),
+            )
+          : state.lifecycle,
      reasoningItems: {
        ...state.reasoningItems,
        [event.item_id]: {
          ...item,
          summaryParts: {
            ...item.summaryParts,
-            [event.summary_index]: "can-conclude",
+            [event.summary_index]: state.store !== false ? "concluded" : "can-conclude",
          },
        },
      },
    },
-    NO_EVENTS,
+    events,
  ]
 }

@@ -1014,24 +931,12 @@ const onFunctionCallArgumentsDelta = Effect.fn("OpenResponses.onFunctionCallArgu
  state: ParserState,
  event: Event,
 ) {
-  if (event.item_id === undefined) return [state, NO_EVENTS] satisfies StepResult
-  const tool = state.tools[event.item_id]
-  if (!tool) return [state, NO_EVENTS] satisfies StepResult
-  const final = event.type === "response.function_call_arguments.done" ? event.arguments : undefined
-  if (event.type === "response.function_call_arguments.done" && final === undefined)
-    return [state, NO_EVENTS] satisfies StepResult
-  if (final !== undefined && !final.startsWith(tool.input))
-    return [
-      { ...state, tools: ToolStream.start(state.tools, event.item_id, { ...tool, input: final }) },
-      NO_EVENTS,
-    ] satisfies StepResult
-  const delta = final === undefined ? event.delta : final.slice(tool.input.length)
-  if (!delta) return [state, NO_EVENTS] satisfies StepResult
+  if (!event.item_id || !event.delta) return [state, NO_EVENTS] satisfies StepResult
  const result = ToolStream.appendExisting(
    state.id,
    state.tools,
    event.item_id,
-    delta,
+    event.delta,
    `${state.name} tool argument delta is missing its tool call`,
  )
  if (ToolStream.isError(result)) return yield* result
@@ -1045,8 +950,8 @@ const onOutputItemDone = Effect.fn("OpenResponses.onOutputItemDone")(function* (
  const item = event.item
  if (!item) return [state, NO_EVENTS] satisfies StepResult

-  if (item.type === "message" && item.id !== undefined) {
-    const itemPhase = messagePhase(item.phase)
+  if (item.type === "message" && item.id) {
+    const itemPhase = state.messagePhase(item.phase)
    const phase = itemPhase === undefined ? state.messagePhases[item.id] : itemPhase
    const events: LLMEvent[] = []
    const messageItems = new Set(state.messageItems)
@@ -1069,19 +974,18 @@ const onOutputItemDone = Effect.fn("OpenResponses.onOutputItemDone")(function* (
  }

  if (item.type === "function_call") {
-    if (!item.call_id || !item.name) return [state, NO_EVENTS] satisfies StepResult
-    const id = item.id ?? item.call_id
-    const tools = state.tools[id]
+    if (!item.id || !item.call_id || !item.name) return [state, NO_EVENTS] satisfies StepResult
+    const tools = state.tools[item.id]
      ? state.tools
-      : ToolStream.start(state.tools, id, {
+      : ToolStream.start(state.tools, item.id, {
          id: item.call_id,
          name: item.name,
-          providerMetadata: item.id !== undefined ? providerMetadata(state, { itemId: item.id }) : undefined,
+          providerMetadata: providerMetadata(state, { itemId: item.id }),
        })
    const result =
      item.arguments === undefined
-        ? yield* ToolStream.finish(state.id, tools, id)
-        : yield* ToolStream.finishWithInput(state.id, tools, id, item.arguments)
+        ? yield* ToolStream.finish(state.id, tools, item.id)
+        : yield* ToolStream.finishWithInput(state.id, tools, item.id, item.arguments)
    const events: LLMEvent[] = []
    const resultEvents = result.events ?? []
    const lifecycle = resultEvents.length ? Lifecycle.stepStart(state.lifecycle, events) : state.lifecycle
@@ -1129,94 +1033,66 @@ const onOutputItemDone = Effect.fn("OpenResponses.onOutputItemDone")(function* (
 })

 const onResponseFinish = Effect.fn("OpenResponses.onResponseFinish")(function* (state: ParserState, event: Event) {
-  const reconciled =
-    event.type === "response.completed"
-      ? yield* Effect.reduce(
-          event.response?.output ?? [],
-          () => [state, NO_EVENTS] satisfies StepResult,
-          ([current, events], item) => {
-            const id = item.id ?? (item.type === "function_call" ? item.call_id : undefined)
-            if (
-              id === undefined ||
-              ((item.type !== "function_call" || !current.tools[id]) &&
-                (item.type !== "reasoning" || !current.reasoningItems[id]))
-            )
-              return Effect.succeed([current, events] satisfies StepResult)
-            return onOutputItemDone(current, { type: "response.output_item.done", item }).pipe(
-              Effect.map(([next, emitted]) => [next, [...events, ...emitted]] satisfies StepResult),
-            )
-          },
-        )
-      : ([state, NO_EVENTS] satisfies StepResult)
-  const current = reconciled[0]
  // Some compatible providers omit output_item.done even after completing the response.
  const pending =
    event.type === "response.completed"
-      ? yield* ToolStream.finishAll(current.id, current.tools)
-      : { tools: current.tools, events: NO_EVENTS }
-  const events: LLMEvent[] = [...reconciled[1], ...pending.events]
+      ? yield* ToolStream.finishAll(state.id, state.tools)
+      : { tools: state.tools, events: NO_EVENTS }
+  const events: LLMEvent[] = [...pending.events]
  const hasFunctionCall =
    pending.events.some((event) => LLMEvent.is.toolCall(event) || LLMEvent.is.toolInputError(event)) ||
-    current.hasFunctionCall
-  const lifecycle = Lifecycle.finish(current.lifecycle, events, {
+    state.hasFunctionCall
+  const lifecycle = Lifecycle.finish(state.lifecycle, events, {
    reason: {
      normalized: mapFinishReason(event, hasFunctionCall),
      raw: event.response?.incomplete_details?.reason,
    },
-    usage: mapUsage(event.response?.usage, current.providerMetadataKey),
+    usage: mapUsage(event.response?.usage, state.providerMetadataKey),
    providerMetadata:
      event.response?.id || event.response?.service_tier
-        ? providerMetadata(current, {
+        ? providerMetadata(state, {
            responseId: event.response.id,
            serviceTier: event.response.service_tier,
          })
        : undefined,
  })
-  return [{ ...current, lifecycle, hasFunctionCall, tools: pending.tools }, events] satisfies StepResult
+  return [{ ...state, lifecycle, hasFunctionCall, tools: pending.tools }, events] satisfies StepResult
 })

-// Build the prettiest summary available from whatever the provider supplied.
+// Build a single human-readable message from whatever the provider supplied.
 // When both code and message are present, prefix the code so consumers see
 // the failure mode (e.g. `rate_limit_exceeded: Slow down`) instead of just
 // the bare message — production rate limits and context-length failures used
-// to be indistinguishable from generic stream drops. Returns undefined when
-// the payload carries no usable summary.
-const providerErrorMessage = (event: Event, nested: OpenResponsesErrorPayload | undefined): string | undefined => {
+// to be indistinguishable from generic stream drops.
+const providerErrorMessage = (event: Event, fallback: string): string => {
+  const nested = event.error ?? event.response?.error ?? undefined
  const message = event.message || nested?.message || undefined
  const code = event.code || nested?.code || undefined
  if (message && code) return `${code}: ${message}`
-  return message || code
+  return message || code || fallback
 }

-export const providerFailure = (event: Event, fallback: string, body = ProviderShared.encodeJson(event)) => {
-  const nested = event.error ?? event.response?.error ?? undefined
-  const summary = providerErrorMessage(event, nested)
-  const message = summary ?? (body === "{}" ? fallback : body)
+export const providerFailure = (id: string, event: Event, fallback: string) => {
+  const code = event.code || event.error?.code || event.response?.error?.code || undefined
+  const message = providerErrorMessage(event, fallback)
  const status =
    typeof event.status === "number"
      ? event.status
      : typeof event.status_code === "number"
        ? event.status_code
        : undefined
-  const reason =
-    event.type === "error" &&
-    event.error === undefined &&
-    event.response === undefined &&
-    summary === undefined &&
-    status === undefined
-      ? new ProviderInternalError({ message, body })
-      : classifyProviderFailure({ message, status, rawBody: body })
-  return new AIError({ reason })
+  return new AIError({
+    module: id,
+    method: "stream",
+    reason: classifyProviderFailure({ message, code, status }),
+  })
 }

-export const step = (state: ParserState, input: Event) => {
-  // The OpenAPI requires string IDs but imposes no minLength; empty is not missing.
-  const event =
-    input.item_id !== undefined && outputItemID(state, input) !== input.item_id
-      ? { ...input, item_id: outputItemID(state, input) }
-      : input
+const providerError = (state: ParserState, event: Event, fallback: string) => providerFailure(state.id, event, fallback)
+
+export const step = (state: ParserState, event: Event) => {
  if (event.type === "response.output_text.delta" || event.type === "response.output_text.done") {
-    if (event.item_id === undefined) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
+    if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
    return Effect.succeed(
      event.type === "response.output_text.delta"
        ? onOutputTextDelta(state, event, event.item_id)
@@ -1224,70 +1100,46 @@ export const step = (state: ParserState, input: Event) => {
    )
  }
  if (event.type === "response.refusal.delta" || event.type === "response.refusal.done") {
-    const value = event.type === "response.refusal.delta" ? event.delta : event.refusal
-    if (event.item_id === undefined || typeof value !== "string")
-      return ProviderShared.eventError(state.id, `${event.type} is malformed`)
+    if (!isRefusalEvent(event)) return ProviderShared.eventError(state.id, `${event.type} is malformed`)
    return Effect.succeed(
      event.type === "response.refusal.delta"
        ? onOutputTextDelta(state, event, event.item_id)
-        : onOutputTextDone(state, { ...event, text: value }, event.item_id),
+        : onOutputTextDone(state, { ...event, text: event.refusal }, event.item_id),
    )
  }
  if (event.type === "response.reasoning.delta" || event.type === "response.reasoning_summary_text.delta") {
-    if (event.item_id === undefined) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
+    if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
    return Effect.succeed(onReasoningDelta(state, event, event.item_id))
  }
-  if (
-    event.type === "response.reasoning.done" ||
-    event.type === "response.reasoning_summary_text.done" ||
-    event.type === "response.reasoning_text.done"
-  ) {
-    if (event.item_id === undefined) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
-    return Effect.succeed(onReasoningDone(state, event, event.item_id))
+  if (event.type === "response.reasoning.done" || event.type === "response.reasoning_summary_text.done") {
+    if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
+    return Effect.succeed(onReasoningDone(state, event))
  }
  if (event.type === "response.reasoning_summary_part.added")
-    return event.item_id !== undefined
+    return event.item_id
      ? Effect.succeed(onReasoningSummaryPartAdded(state, event))
      : ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
  if (event.type === "response.reasoning_summary_part.done")
-    return event.item_id !== undefined
+    return event.item_id
      ? Effect.succeed(onReasoningSummaryPartDone(state, event))
      : ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
  if (event.type === "response.output_item.added") {
-    if (event.item?.type === "message" && event.item.id === undefined)
+    if (event.item?.type === "message" && !event.item.id)
      return ProviderShared.eventError(state.id, `${event.type} message is missing id`)
-    const id = event.item?.id ?? (event.item?.type === "function_call" ? event.item.call_id : undefined)
-    return Effect.succeed(
-      onOutputItemAdded(
-        event.output_index !== undefined && id !== undefined
-          ? { ...state, outputItems: { ...state.outputItems, [event.output_index]: id } }
-          : state,
-        event,
-      ),
-    )
+    return Effect.succeed(onOutputItemAdded(state, event))
  }
-  if (event.type === "response.function_call_arguments.delta" || event.type === "response.function_call_arguments.done")
-    return event.item_id !== undefined
-      ? onFunctionCallArgumentsDelta(state, event)
-      : ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
+  if (event.type === "response.function_call_arguments.delta") return onFunctionCallArgumentsDelta(state, event)
  if (event.type === "response.output_item.done") {
-    if (event.item?.type === "message" && event.item.id === undefined)
+    if (event.item?.type === "message" && !event.item.id)
      return ProviderShared.eventError(state.id, `${event.type} message is missing id`)
    return onOutputItemDone(state, event)
  }
  if (event.type === "response.completed" || event.type === "response.incomplete") return onResponseFinish(state, event)
-  if (event.type === "response.failed") return providerFailure(event, `${state.name} response failed`)
+  if (event.type === "response.failed") return providerError(state, event, `${state.name} response failed`)
  if (event.type === "error")
    return decodeKnownErrorEvent(event).pipe(
-      Effect.mapError((cause) =>
-        ProviderShared.eventError(
-          state.id,
-          `${state.name} returned a malformed error event`,
-          ProviderShared.encodeJson(event),
-          cause,
-        ),
-      ),
-      Effect.flatMap(() => providerFailure(event, `${state.name} stream error`)),
+      Effect.mapError(() => ProviderShared.eventError(state.id, `${state.name} returned a malformed error event`)),
+      Effect.flatMap(() => providerError(state, event, `${state.name} stream error`)),
    )
  return Effect.succeed<StepResult>([state, NO_EVENTS])
 }
@@ -1306,12 +1158,18 @@ export const initial = (request: LLMRequest, extension: Extension = BASE): Parse
  hasFunctionCall: false,
  tools: ToolStream.empty<string>(),
  lifecycle: Lifecycle.initial(),
-  outputItems: {},
  messageItems: new Set<string>(),
+  messagePhase: (value) => messagePhase(value, extension),
  messagePhases: {},
  reasoningItems: {},
+  store: OpenResponsesOptions.resolve(request).store,
 })

+const messagePhase = (value: unknown, extension: Extension): MessagePhase | null | undefined => {
+  if (value === "commentary" || value === "final_answer") return value
+  return extension.messagePhase?.(value)
+}
+
 export const protocol = Protocol.make({
  id: ADAPTER,
  body: {
@@ -7,11 +7,7 @@ import { HttpTransport } from "../route/transport/index.js"
 import { Protocol } from "../route/protocol.js"
 import {
  AIError,
-  AIErrorReason,
-  InvalidProviderOutputError,
  LLMEvent,
-  ProviderInternalError,
-  UnknownProviderError,
  Usage,
  type FinishReason,
  type FinishReasonDetails,
@@ -55,12 +51,7 @@ const OpenAIChatFunction = Schema.Struct({

 const OpenAIChatTool = Schema.Struct({
  type: Schema.tag("function"),
-  function: Schema.Struct({
-    name: Schema.String,
-    description: Schema.String,
-    parameters: JsonObject,
-    strict: Schema.optional(Schema.Boolean),
-  }),
+  function: OpenAIChatFunction,
  cache_control: Schema.optional(OpenAIChatCacheControl),
 })
 type OpenAIChatTool = Schema.Schema.Type<typeof OpenAIChatTool>
@@ -142,7 +133,6 @@ export const bodyFields = {
  store: Schema.optional(Schema.Boolean),
  prompt_cache_key: Schema.optional(Schema.String),
  reasoning_effort: Schema.optional(OpenAIOptions.OpenAIReasoningEffort),
-  tool_stream: Schema.optional(Schema.Boolean),
  max_completion_tokens: Schema.optional(Schema.Number),
  max_tokens: Schema.optional(Schema.Number),
  temperature: Schema.optional(Schema.Number),
@@ -166,9 +156,6 @@ const OpenAIChatUsage = Schema.StructWithRest(
    prompt_tokens: optionalNull(Schema.Number),
    completion_tokens: optionalNull(Schema.Number),
    total_tokens: optionalNull(Schema.Number),
-    // Zai reports cache hits as top-level `cached_tokens`; DeepSeek uses `prompt_cache_hit_tokens`.
-    cached_tokens: optionalNull(Schema.Number),
-    prompt_cache_hit_tokens: optionalNull(Schema.Number),
    prompt_tokens_details: optionalNull(
      Schema.StructWithRest(
        Schema.Struct({
@@ -217,33 +204,22 @@ const OpenAIChatDelta = Schema.StructWithRest(
  [Schema.Record(Schema.String, Schema.Unknown)],
 )

-const OpenAIChatChoice = Schema.StructWithRest(
-  Schema.Struct({
-    delta: optionalNull(OpenAIChatDelta),
-    finish_reason: optionalNull(Schema.String),
-    native_finish_reason: optionalNull(Schema.String),
-    // Moonshot streams usage on `choice.usage` instead of top-level `usage`.
-    usage: optionalNull(OpenAIChatUsage),
-  }),
-  [Schema.Record(Schema.String, Schema.Unknown)],
-)
+const OpenAIChatChoice = Schema.Struct({
+  delta: optionalNull(OpenAIChatDelta),
+  finish_reason: optionalNull(Schema.String),
+  native_finish_reason: optionalNull(Schema.String),
+})

-const OpenAIChatError = Schema.StructWithRest(
-  Schema.Struct({
-    code: optionalNull(Schema.Union([Schema.String, Schema.Number])),
-    message: Schema.String,
-  }),
-  [Schema.Record(Schema.String, Schema.Unknown)],
-)
+const OpenAIChatError = Schema.Struct({
+  code: optionalNull(Schema.Union([Schema.String, Schema.Number])),
+  message: Schema.String,
+})

-export const OpenAIChatEvent = Schema.StructWithRest(
-  Schema.Struct({
-    choices: optionalNull(Schema.Array(OpenAIChatChoice)),
-    usage: optionalNull(OpenAIChatUsage),
-    error: optionalNull(OpenAIChatError),
-  }),
-  [Schema.Record(Schema.String, Schema.Unknown)],
-)
+export const OpenAIChatEvent = Schema.Struct({
+  choices: optionalNull(Schema.Array(OpenAIChatChoice)),
+  usage: optionalNull(OpenAIChatUsage),
+  error: optionalNull(OpenAIChatError),
+})
 export type OpenAIChatEvent = Schema.Schema.Type<typeof OpenAIChatEvent>
 type OpenAIChatRequestMessage = LLMRequest["messages"][number]

@@ -254,7 +230,6 @@ interface PendingToolDelta {
 }

 export interface ParserState {
-  readonly providerMetadataKey: string
  readonly tools: ToolStream.State<number>
  readonly pendingTools: Partial<Record<number, PendingToolDelta>>
  readonly toolCallEvents: ReadonlyArray<LLMEvent>
@@ -267,7 +242,6 @@ export interface ParserState {
  readonly reasoningEmitted: boolean
  readonly latestToolIndex?: number
  readonly nextToolIndex: number
-  readonly requireFinishReason: boolean
 }

 // =============================================================================
@@ -280,21 +254,14 @@ interface LoweringOptions {
  readonly cacheControl?: (
    cache: CacheHint | undefined,
  ) => Schema.Schema.Type<typeof OpenAIChatCacheControl> | undefined
-  readonly toolCallID?: (id: string) => string
 }

-const lowerTool = (
-  tool: ToolDefinition,
-  inputSchema: JsonSchema,
-  options: LoweringOptions,
-  supportsStrictMode: boolean,
-): OpenAIChatTool => ({
+const lowerTool = (tool: ToolDefinition, inputSchema: JsonSchema, options: LoweringOptions): OpenAIChatTool => ({
  type: "function",
  function: {
    name: tool.name,
    description: tool.description,
-    parameters: inputSchema,
-    ...(supportsStrictMode ? { strict: false } : {}),
+    parameters: ToolSchemaProjection.openAI(inputSchema),
  },
  cache_control: options.cacheControl?.(tool.cache),
 })
@@ -307,8 +274,8 @@ const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>) =>
    tool: (name) => ({ type: "function" as const, function: { name } }),
  })

-const lowerToolCall = (part: ToolCallPart, options: LoweringOptions): OpenAIChatAssistantToolCall => ({
-  id: options.toolCallID?.(part.id) ?? part.id,
+const lowerToolCall = (part: ToolCallPart): OpenAIChatAssistantToolCall => ({
+  id: part.id,
  type: "function",
  function: {
    name: part.name,
@@ -326,18 +293,17 @@ const lowerMedia = Effect.fn("OpenAIChat.lowerMedia")(function* (part: MediaPart
 const openAICompatibleReasoningContent = (native: unknown) =>
  isRecord(native) && typeof native.reasoning_content === "string" ? native.reasoning_content : undefined

-const reasoningField = (part: ReasoningPart, providerMetadataKey: string) => {
-  const field = part.providerMetadata?.[providerMetadataKey]?.reasoningField
+const reasoningField = (part: ReasoningPart) => {
+  const field = part.providerMetadata?.openai?.reasoningField
  return typeof field === "string" ? field : undefined
 }

-const reasoningDetails = (parts: ReadonlyArray<ReasoningPart>, native: unknown, providerMetadataKey: string) => {
+const reasoningDetails = (parts: ReadonlyArray<ReasoningPart>, native: unknown) => {
  const observed = parts.flatMap((part) => {
-    const details = part.providerMetadata?.[providerMetadataKey]?.reasoningDetails
+    const details = part.providerMetadata?.openai?.reasoningDetails
    return Array.isArray(details) ? details : []
  })
-  if (parts.some((part) => Array.isArray(part.providerMetadata?.[providerMetadataKey]?.reasoningDetails)))
-    return observed
+  if (parts.some((part) => Array.isArray(part.providerMetadata?.openai?.reasoningDetails))) return observed
  if (isRecord(native) && Array.isArray(native.reasoning_details)) return native.reasoning_details
 }

@@ -367,9 +333,8 @@ const lowerUserMessage = Effect.fn("OpenAIChat.lowerUserMessage")(function* (

 const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(function* (
  message: OpenAIChatRequestMessage,
-  configuredField: string | undefined,
-  requireReasoning: boolean,
-  options: LoweringOptions & { readonly providerMetadataKey: string },
+  configuredField?: string,
+  options: LoweringOptions = {},
 ) {
  const content: TextPart[] = []
  const reasoning: ReasoningPart[] = []
@@ -386,31 +351,25 @@ const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(func
      continue
    }
    if (part.type === "tool-call") {
-      toolCalls.push(lowerToolCall(part, options))
+      toolCalls.push(lowerToolCall(part))
      continue
    }
  }
  const text = reasoning.map((part) => part.text).join("")
-  const details = reasoningDetails(reasoning, message.native?.openaiCompatible, options.providerMetadataKey)
-  const observedField = reasoning
-    .map((part) => reasoningField(part, options.providerMetadataKey))
-    .find((value) => value !== undefined)
+  const details = reasoningDetails(reasoning, message.native?.openaiCompatible)
+  const observedField = reasoning.map(reasoningField).find((value) => value !== undefined)
  const nativeReasoning = openAICompatibleReasoningContent(message.native?.openaiCompatible)
-  const fullyStructured = reasoning.every((part) =>
-    Array.isArray(part.providerMetadata?.[options.providerMetadataKey]?.reasoningDetails),
-  )
+  const fullyStructured = reasoning.every((part) => Array.isArray(part.providerMetadata?.openai?.reasoningDetails))
  const field = (() => {
-    if (configuredField !== undefined && (requireReasoning || reasoning.length > 0 || nativeReasoning !== undefined))
-      return configuredField
-    if (reasoning.length === 0) return requireReasoning ? "reasoning_content" : undefined
+    if (configuredField !== undefined) return configuredField
+    if (reasoning.length === 0) return undefined
    if (observedField !== undefined) return observedField
    if (nativeReasoning !== undefined) return "reasoning_content"
-    if (!fullyStructured || requireReasoning) return "reasoning_content"
+    if (!fullyStructured) return "reasoning_content"
  })()
  const reasoningText = (() => {
-    if (configuredField !== undefined)
-      return reasoning.length === 0 ? (nativeReasoning ?? (requireReasoning ? "" : undefined)) : text
-    if (reasoning.length === 0) return nativeReasoning ?? (requireReasoning ? "" : undefined)
+    if (configuredField !== undefined) return reasoning.length === 0 ? (nativeReasoning ?? "") : text
+    if (reasoning.length === 0) return nativeReasoning
    return text
  })()
  const cached = message.content.findLast((part) => "cache" in part && part.cache !== undefined)
@@ -438,7 +397,7 @@ const lowerToolMessages = Effect.fn("OpenAIChat.lowerToolMessages")(function* (
    if (part.result.type !== "content") {
      messages.push({
        role: "tool",
-        tool_call_id: options.toolCallID?.(part.id) ?? part.id,
+        tool_call_id: part.id,
        content: ProviderShared.toolResultText(part),
        cache_control: options.cacheControl?.(part.cache),
      })
@@ -448,7 +407,7 @@ const lowerToolMessages = Effect.fn("OpenAIChat.lowerToolMessages")(function* (
    const text = content.filter((item) => item.type === "text").map((item) => item.text)
    messages.push({
      role: "tool",
-      tool_call_id: options.toolCallID?.(part.id) ?? part.id,
+      tool_call_id: part.id,
      content: text.join("\n"),
      cache_control: options.cacheControl?.(part.cache),
    })
@@ -464,13 +423,11 @@ const lowerToolMessages = Effect.fn("OpenAIChat.lowerToolMessages")(function* (

 const lowerMessage = Effect.fn("OpenAIChat.lowerMessage")(function* (
  message: OpenAIChatRequestMessage,
-  reasoningField: string | undefined,
-  requireReasoning: boolean,
-  options: LoweringOptions & { readonly providerMetadataKey: string },
+  reasoningField?: string,
+  options: LoweringOptions = {},
 ) {
  if (message.role === "user") return [yield* lowerUserMessage(message, options)]
-  if (message.role === "assistant")
-    return [yield* lowerAssistantMessage(message, reasoningField, requireReasoning, options)]
+  if (message.role === "assistant") return [yield* lowerAssistantMessage(message, reasoningField, options)]
  return (yield* lowerToolMessages(message, options)).messages
 })

@@ -491,43 +448,12 @@ const lowerMessages = Effect.fn("OpenAIChat.lowerMessages")(function* (request:
          ]
        : [{ role: "system", content: ProviderShared.joinText(request.system) }]
  const messages = [...system]
-  const modelID = request.model.id.toLowerCase()
-  const requireReasoning =
-    request.model.compatibility?.requireReasoning ??
-    (request.model.compatibility?.reasoningField !== undefined ||
-      request.model.provider === "deepseek" ||
-      request.model.route.endpoint.baseURL?.toLowerCase().includes("deepseek.com") ||
-      modelID.includes("deepseek"))
-  const reasoningField = request.model.compatibility?.reasoningField
-  const mistral = ["mistral", "devstral", "codestral", "pixtral", "mixtral"].some((family) => modelID.includes(family))
-  const lowering = {
-    ...options,
-    providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
-    toolCallID: (id: string) => {
-      if (mistral)
-        return id
-          .replace(/[^a-zA-Z0-9]/g, "")
-          .slice(0, 9)
-          .padEnd(9, "0")
-      if (modelID.includes("claude")) return id.replace(/[^a-zA-Z0-9_-]/g, "_")
-      if (request.model.provider === "openai" || request.model.provider === "azure" || modelID.startsWith("openai/"))
-        return id.slice(0, 40)
-      return id
-    },
-  }
-  const requireAssistantAfterTool = request.model.compatibility?.requireAssistantAfterTool ?? mistral
-  const bridgeTools = () => {
-    if (requireAssistantAfterTool && messages.at(-1)?.role === "tool")
-      messages.push({ role: "assistant", content: "Done." })
-  }
  const pendingImages: Array<Schema.Schema.Type<typeof OpenAIChatUserContent>> = []
  const flushImages = () => {
    if (pendingImages.length === 0) return
-    bridgeTools()
    messages.push({ role: "user", content: pendingImages.splice(0) })
  }
  for (const message of request.messages) {
-    if (message.role === "user") bridgeTools()
    if (message.role === "system") {
      const part = yield* ProviderShared.wrappedSystemUpdate("OpenAI Chat", message)
      if (pendingImages.length > 0) {
@@ -570,152 +496,24 @@ const lowerMessages = Effect.fn("OpenAIChat.lowerMessages")(function* (request:
        )
      continue
    }
-    if (
-      message.role === "assistant" &&
-      message.content.every((part) => part.type === "text" && part.text.trim() === "")
-    )
-      continue
    if (message.role === "tool") {
-      const lowered = yield* lowerToolMessages(message, lowering)
+      const lowered = yield* lowerToolMessages(message, options)
      messages.push(...lowered.messages)
      pendingImages.push(...lowered.images)
      continue
    }
    flushImages()
-    messages.push(...(yield* lowerMessage(message, reasoningField, requireReasoning, lowering)))
+    messages.push(...(yield* lowerMessage(message, request.model.compatibility?.reasoningField, options)))
  }
  flushImages()
  return messages
 })

-// Anthropic via LiteLLM and Amazon Bedrock require `tools` to be present
-// whenever the conversation history contains tool calls/results. Send an
-// explicit empty array when we have history but no active tools.
-const hasToolHistory = (messages: ReadonlyArray<LLMRequest["messages"][number]>) => {
-  for (const message of messages) {
-    if (message.role === "tool") return true
-    if (message.role === "assistant" && message.content.some((part) => part.type === "tool-call")) return true
-  }
-  return false
-}
-
-// Derive `max_tokens` vs `max_completion_tokens` from provider/baseURL when
-// explicit `compatibility.maxTokensField` is not set. Aligned with
-// models.dev provider naming: DeepSeek, Moonshot AI, Together AI, ZAI
-// (Zhipu + Coding Plan variants), Nvidia, Cerebras, Chutes, etc. still
-// require `max_tokens`.
-const detectMaxTokensField = (
-  provider: string,
-  baseURL: string | undefined,
-): "max_tokens" | "max_completion_tokens" => {
-  const p = provider.toLowerCase()
-  const url = (baseURL ?? "").toLowerCase()
-  if (
-    p === "deepseek" ||
-    url.includes("deepseek.com") ||
-    p === "moonshotai" ||
-    url.includes("api.moonshot.ai") ||
-    p === "togetherai" ||
-    url.includes("api.together.") ||
-    p === "zai" ||
-    p === "zai-coding-plan" ||
-    p === "zhipuai" ||
-    p === "zhipuai-coding-plan" ||
-    url.includes("api.z.ai") ||
-    url.includes("open.bigmodel.cn") ||
-    p === "nvidia" ||
-    url.includes("integrate.api.nvidia.com") ||
-    p === "cerebras" ||
-    url.includes("cerebras.ai") ||
-    url.includes("llm.chutes.ai") ||
-    p === "chutes" ||
-    p === "cloudflare-ai-gateway" ||
-    url.includes("gateway.ai.cloudflare.com") ||
-    p === "cloudflare-workers-ai" ||
-    url.includes("api.cloudflare.com")
-  )
-    return "max_tokens"
-  return "max_completion_tokens"
-}
-
-const detectSupportsStore = (provider: string, baseURL: string | undefined): boolean => {
-  const p = provider.toLowerCase()
-  const url = (baseURL ?? "").toLowerCase()
-  const isNvidia = p === "nvidia" || url.includes("integrate.api.nvidia.com")
-  const isMoonshot = p === "moonshotai" || p === "moonshotai-cn" || url.includes("api.moonshot.")
-  const isTogether = p === "togetherai" || p === "together" || url.includes("api.together.")
-  const isZai =
-    p === "zai" ||
-    p === "zai-coding-plan" ||
-    p === "zhipuai" ||
-    p === "zhipuai-coding-plan" ||
-    url.includes("api.z.ai") ||
-    url.includes("open.bigmodel.cn")
-  const isDeepSeek = p === "deepseek" || url.includes("deepseek.com")
-  const isCerebras = p === "cerebras" || url.includes("cerebras.ai")
-  const isXai = p === "xai" || url.includes("api.x.ai")
-  const isChutes = p === "chutes" || url.includes("chutes.ai")
-  const isCloudflareWorkersAI = p === "cloudflare-workers-ai" || url.includes("api.cloudflare.com")
-  const isCloudflareAiGateway = p === "cloudflare-ai-gateway" || url.includes("gateway.ai.cloudflare.com")
-  const isVercelAiGateway =
-    p === "vercel-ai-gateway" || url.includes("ai-gateway.vercel.sh") || url.includes("vercel.sh")
-  const isAntLing = p === "ant-ling" || url.includes("api.ant-ling.com")
-  const isOpencode = p === "opencode" || url.includes("opencode.ai")
-  const isNonStandard =
-    isNvidia ||
-    isCerebras ||
-    isXai ||
-    isTogether ||
-    isChutes ||
-    isDeepSeek ||
-    isZai ||
-    isMoonshot ||
-    isOpencode ||
-    isCloudflareWorkersAI ||
-    isCloudflareAiGateway ||
-    isVercelAiGateway ||
-    isAntLing
-  return !isNonStandard
-}
-
-const detectSupportsUsageInStreaming = (): boolean => true
-
-const detectSupportsStrictMode = (provider: string, baseURL: string | undefined): boolean => {
-  const p = provider.toLowerCase()
-  const url = (baseURL ?? "").toLowerCase()
-  const isMoonshot = p === "moonshotai" || p === "moonshotai-cn" || url.includes("api.moonshot.")
-  const isTogether = p === "togetherai" || p === "together" || url.includes("api.together.")
-  const isCloudflareAiGateway = p === "cloudflare-ai-gateway" || url.includes("gateway.ai.cloudflare.com")
-  const isNvidia = p === "nvidia" || url.includes("integrate.api.nvidia.com")
-  return !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia
-}
-
-const detectZaiToolStream = (provider: string, baseURL: string | undefined, modelID: string): boolean => {
-  const p = provider.toLowerCase()
-  const url = (baseURL ?? "").toLowerCase()
-  const isZai =
-    p === "zai" ||
-    p === "zai-coding-plan" ||
-    p === "zhipuai" ||
-    p === "zhipuai-coding-plan" ||
-    url.includes("api.z.ai") ||
-    url.includes("open.bigmodel.cn")
-  if (!isZai) return false
-  const id = modelID.toLowerCase()
-  if (id === "glm-4.5" || id === "glm-4.5-air" || id === "glm-4.5-flash" || id === "glm-4.5v") return false
-  return true
-}
-
-const lowerOptions = (request: LLMRequest, supportsStore: boolean) => {
+const lowerOptions = (request: LLMRequest) => {
  const options = OpenAIOptions.resolve(request)
-  const cacheKey = ProviderShared.promptCacheKey(request)
  return {
-    ...(supportsStore && options.store !== undefined ? { store: options.store } : {}),
-    // For providers that support `store`, ensure stateless `store:false` is sent
-    // even when no explicit `providerOptions.store` was supplied, mirroring the
-    // native OpenAI Chat default. Non-standard providers omit `store` entirely.
-    ...(supportsStore && options.store === undefined ? { store: false } : {}),
-    ...(cacheKey ? { prompt_cache_key: cacheKey } : {}),
+    ...(options.store !== undefined ? { store: options.store } : {}),
+    ...(request.promptCacheKey ? { prompt_cache_key: request.promptCacheKey } : {}),
    ...(options.reasoningEffort ? { reasoning_effort: options.reasoningEffort } : {}),
  }
 }
@@ -733,39 +531,23 @@ export const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (
    )
  const generation = request.generation
  const toolSchemaCompatibility = request.model.compatibility?.toolSchema
-  const provider = String(request.model.provider)
-  const baseURL = request.model.route.endpoint.baseURL
-  const detectedMaxTokensField = detectMaxTokensField(provider, baseURL)
-  const maxTokensField = request.model.compatibility?.maxTokensField ?? detectedMaxTokensField
-  const supportsStore = request.model.compatibility?.supportsStore ?? detectSupportsStore(provider, baseURL)
-  const supportsUsageInStreaming =
-    request.model.compatibility?.supportsUsageInStreaming ?? detectSupportsUsageInStreaming()
-  const supportsStrictMode =
-    request.model.compatibility?.supportsStrictMode ?? detectSupportsStrictMode(provider, baseURL)
-  const zaiToolStream =
-    request.model.compatibility?.zaiToolStream ?? detectZaiToolStream(provider, baseURL, request.model.id)
-  const hasHistory = hasToolHistory(request.messages)
-  const hasActiveTools = request.tools.length > 0
+  const maxTokensField = request.model.compatibility?.maxTokensField ?? "max_tokens"
  return {
    model: request.model.id,
    messages: yield* lowerMessages(request, options),
    tools:
      request.tools.length === 0
-        ? hasHistory
-          ? []
-          : undefined
+        ? undefined
        : request.tools.map((tool) =>
            lowerTool(
              tool,
              ToolSchemaProjection.modelCompatibility(tool.inputSchema, toolSchemaCompatibility),
              options,
-              supportsStrictMode,
            ),
          ),
    tool_choice: request.toolChoice ? yield* lowerToolChoice(request.toolChoice) : undefined,
    stream: true as const,
-    ...(supportsUsageInStreaming ? { stream_options: { include_usage: true } } : {}),
-    ...(zaiToolStream && hasActiveTools ? { tool_stream: true } : {}),
+    stream_options: { include_usage: true },
    ...(maxTokensField === "max_completion_tokens"
      ? { max_completion_tokens: generation?.maxTokens }
      : { max_tokens: generation?.maxTokens }),
@@ -775,7 +557,7 @@ export const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (
    presence_penalty: generation?.presencePenalty,
    seed: generation?.seed,
    stop: generation?.stop,
-    ...lowerOptions(request, supportsStore),
+    ...lowerOptions(request),
  }
 })

@@ -785,53 +567,25 @@ export const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (
 // Streaming parsers are small state machines: every event returns a new state
 // plus the common `LLMEvent`s produced by that event. Tool calls are accumulated
 // because OpenAI streams JSON arguments across multiple deltas.
-const mapFinishReason = Effect.fn("OpenAIChat.mapFinishReason")(function* (event: OpenAIChatEvent, reason: string) {
-  switch (reason) {
-    case "error":
-      return yield* new AIError({
-        reason: new UnknownProviderError({
-          message: "Provider reported an error (finish_reason: error)",
-          body: ProviderShared.encodeJson(event),
-        }),
-      })
-    case "network_error":
-      return yield* new AIError({
-        reason: new ProviderInternalError({
-          message: "Provider reported a network error (finish_reason: network_error)",
-          body: ProviderShared.encodeJson(event),
-        }),
-      })
-    case "stop":
-    case "end":
-      return "stop" as const
-    case "length":
-      return "length" as const
-    case "content_filter":
-      return "content-filter" as const
-    case "function_call":
-    case "tool_calls":
-      return "tool-calls" as const
-    default:
-      return "unknown" as const
-  }
-})
+const mapFinishReason = (reason: string | null | undefined): FinishReason => {
+  if (reason === "stop") return "stop"
+  if (reason === "length") return "length"
+  if (reason === "content_filter") return "content-filter"
+  if (reason === "function_call" || reason === "tool_calls") return "tool-calls"
+  if (reason === "error") return "error"
+  return "unknown"
+}

 // OpenAI Chat reports `prompt_tokens` (inclusive total) with a
 // cached-read and cache-write subsets, and `completion_tokens` (inclusive
 // total) with a `reasoning_tokens` subset. We pass the inclusive totals
 // through and derive the non-cached breakdown so the `AI.Usage` contract is
 // satisfied on both sides.
-// Providers differ on cache-hit location: OpenAI uses
-// `prompt_tokens_details.cached_tokens`, DeepSeek uses
-// `prompt_cache_hit_tokens`, and Zai uses top-level `cached_tokens`.
-const mapUsage = (usage: OpenAIChatEvent["usage"], providerMetadataKey: string): Usage | undefined => {
+const mapUsage = (usage: OpenAIChatEvent["usage"]): Usage | undefined => {
  if (!usage) return undefined
  const input = usage.prompt_tokens ?? undefined
  const output = usage.completion_tokens ?? undefined
-  const cached = (usage.prompt_tokens_details?.cached_tokens ??
-    (usage as { prompt_cache_hit_tokens?: number | null }).prompt_cache_hit_tokens ??
-    (usage as { cached_tokens?: number | null }).cached_tokens ??
-    undefined) as number | undefined
+  const cached = usage.prompt_tokens_details?.cached_tokens ?? undefined
  const cacheWrite = usage.prompt_tokens_details?.cache_write_tokens ?? undefined
  const reasoning = usage.completion_tokens_details?.reasoning_tokens ?? undefined
  const nonCached = ProviderShared.subtractTokens(input, ProviderShared.sumTokens(cached, cacheWrite))
@@ -843,7 +597,7 @@ const mapUsage = (usage: OpenAIChatEvent["usage"], providerMetadataKey: string):
    cacheWriteInputTokens: cacheWrite,
    reasoningTokens: reasoning,
    totalTokens: ProviderShared.totalTokens(input, output, usage.total_tokens ?? undefined),
-    providerMetadata: { [providerMetadataKey]: usage },
+    providerMetadata: { openai: usage },
  })
 }

@@ -917,12 +671,8 @@ const conflictingReasoningTextDetails = (previous: Record<string, unknown>, curr
 const conflictingDetailValue = (previous: unknown, current: unknown) =>
  previous !== undefined && previous !== null && current !== undefined && current !== null && previous !== current

-const reasoningMetadata = (
-  providerMetadataKey: string,
-  field: ParserState["reasoningField"],
-  details?: ReadonlyArray<unknown>,
-) => ({
-  [providerMetadataKey]: {
+const reasoningMetadata = (field: ParserState["reasoningField"], details?: ReadonlyArray<unknown>) => ({
+  openai: {
    ...(field ? { reasoningField: field } : {}),
    ...(details ? { reasoningDetails: details } : {}),
  },
@@ -930,32 +680,24 @@ const reasoningMetadata = (

 const step = (state: ParserState, event: OpenAIChatEvent) =>
  Effect.gen(function* () {
-    if (event.error) {
-      const body = ProviderShared.encodeJson(event)
+    if (event.error)
      return yield* new AIError({
+        module: ADAPTER,
+        method: "stream",
        reason: classifyProviderFailure({
          message: event.error.message,
+          code: event.error.code === undefined || event.error.code === null ? undefined : String(event.error.code),
          status: typeof event.error.code === "number" ? event.error.code : undefined,
-          rawBody: body,
        }),
      })
-    }
    const events: LLMEvent[] = []
+    const usage = mapUsage(event.usage) ?? state.usage
    const choice = event.choices?.[0]
-    // Moonshot (and a few other OpenAI-compatible providers) attach usage to
-    // `choice.usage` instead of the top-level `usage` field.
-    const choiceUsage = (choice as unknown as { usage?: OpenAIChatEvent["usage"] })?.usage
-    const usage =
-      mapUsage(event.usage, state.providerMetadataKey) ??
-      (choiceUsage ? mapUsage(choiceUsage, state.providerMetadataKey) : undefined) ??
-      state.usage
    const rawFinishReason = choice?.finish_reason
-    const finishReason = rawFinishReason
-      ? {
-          normalized: yield* mapFinishReason(event, rawFinishReason),
-          raw: choice?.native_finish_reason ?? rawFinishReason,
-        }
-      : state.finishReason
+    const finishReason =
+      rawFinishReason !== undefined && rawFinishReason !== null
+        ? { normalized: mapFinishReason(rawFinishReason), raw: choice?.native_finish_reason ?? rawFinishReason }
+        : state.finishReason
    const delta = choice?.delta
    const toolDeltas = delta?.tool_calls ?? []
    let tools = state.tools
@@ -974,11 +716,7 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
      toolDeltas.some((tool) => Boolean(tool.id) || Boolean(tool.function?.name) || Boolean(tool.function?.arguments))
    if (state.finishReason !== undefined) {
      if (hasLateContent)
-        return yield* ProviderShared.eventError(
-          ADAPTER,
-          "OpenAI Chat received content after the finish reason",
-          ProviderShared.encodeJson(event),
-        )
+        return yield* ProviderShared.eventError(ADAPTER, "OpenAI Chat received content after the finish reason")
      return [{ ...state, usage }, events] as const
    }

@@ -986,7 +724,7 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
    const detailDelta = Array.isArray(delta?.reasoning_details) ? delta.reasoning_details : undefined
    if (detailDelta !== undefined) appendReasoningDetails(state.reasoningDetails, detailDelta)
    const reasoningDetailsObserved = state.reasoningDetailsObserved || detailDelta !== undefined
-    const deltaMetadata = reasoningMetadata(state.providerMetadataKey, reasoningField)
+    const deltaMetadata = reasoningMetadata(reasoningField)
    const text = detailDelta?.length ? (detailText(detailDelta) ?? reasoning?.text) : reasoning?.text
    if (text !== undefined) lifecycle = Lifecycle.reasoningDelta(lifecycle, events, "reasoning-0", text, deltaMetadata)
    else if (
@@ -1002,11 +740,7 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
        lifecycle,
        events,
        "reasoning-0",
-        reasoningMetadata(
-          state.providerMetadataKey,
-          reasoningField,
-          reasoningDetailsObserved ? state.reasoningDetails : undefined,
-        ),
+        reasoningMetadata(reasoningField, reasoningDetailsObserved ? state.reasoningDetails : undefined),
      )
      lifecycle = Lifecycle.textDelta(lifecycle, events, "text-0", delta.content)
    }
@@ -1016,11 +750,7 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
        lifecycle,
        events,
        "reasoning-0",
-        reasoningMetadata(
-          state.providerMetadataKey,
-          reasoningField,
-          reasoningDetailsObserved ? state.reasoningDetails : undefined,
-        ),
+        reasoningMetadata(reasoningField, reasoningDetailsObserved ? state.reasoningDetails : undefined),
      )
      lifecycle = Lifecycle.textDelta(lifecycle, events, "text-0", delta.refusal)
    }
@@ -1058,26 +788,14 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
        { id: id || undefined, name: name || undefined, text },
        "OpenAI Chat tool call delta is missing id or name",
      )
-      if (ToolStream.isError(result))
-        return yield* new AIError({
-          reason: AIErrorReason.make({
-            ...result.reason,
-            message: result.message,
-            cause: result.reason.cause,
-            body: ProviderShared.encodeJson(event),
-          }),
-        })
+      if (ToolStream.isError(result)) return yield* result
      tools = result.tools
      if (result.events.length) lifecycle = Lifecycle.stepStart(lifecycle, events)
      events.push(...result.events)
    }

    if (finishReason !== undefined && state.finishReason === undefined && Object.keys(pendingTools).length > 0)
-      return yield* ProviderShared.eventError(
-        ADAPTER,
-        "OpenAI Chat tool call delta is missing id or name",
-        ProviderShared.encodeJson(event),
-      )
+      return yield* ProviderShared.eventError(ADAPTER, "OpenAI Chat tool call delta is missing id or name")

    // Finalize accumulated tool inputs eagerly when finish_reason arrives so
    // valid calls and malformed local calls settle independently.
@@ -1088,7 +806,6 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>

    return [
      {
-        providerMetadataKey: state.providerMetadataKey,
        tools: finished?.tools ?? tools,
        pendingTools,
        toolCallEvents: finished?.events ?? state.toolCallEvents,
@@ -1101,25 +818,16 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
        reasoningEmitted,
        latestToolIndex,
        nextToolIndex,
-        requireFinishReason: state.requireFinishReason,
      },
      events,
    ] as const
  })

-const finishEvents = Effect.fn("OpenAIChat.finishEvents")(function* (state: ParserState) {
-  if (state.finishReason === undefined && state.requireFinishReason)
-    return yield* new AIError({
-      reason: new InvalidProviderOutputError({
-        message: "OpenAI Chat stream ended without finish_reason",
-        classification: "incomplete-stream",
-        route: ADAPTER,
-      }),
-    })
+const finishEvents = (state: ParserState): ReadonlyArray<LLMEvent> => {
  const events: LLMEvent[] = []
  const toolCallEvents =
    state.finishReason === undefined && Object.keys(state.tools).length > 0
-      ? (yield* ToolStream.finishAll(ADAPTER, state.tools)).events
+      ? Effect.runSync(ToolStream.finishAll(ADAPTER, state.tools)).events
      : state.toolCallEvents
  const hasToolCalls = toolCallEvents.length > 0
  const reason = state.finishReason
@@ -1128,27 +836,21 @@ const finishEvents = Effect.fn("OpenAIChat.finishEvents")(function* (state: Pars
        normalized:
          state.finishReason.normalized === "stop" && hasToolCalls ? "tool-calls" : state.finishReason.normalized,
      }
-    : { normalized: hasToolCalls ? ("tool-calls" as const) : ("stop" as const) }
+    : { normalized: hasToolCalls ? ("tool-calls" as const) : ("unknown" as const) }
  const metadata = reasoningMetadata(
-    state.providerMetadataKey,
    state.reasoningField,
    state.reasoningDetailsObserved ? state.reasoningDetails : undefined,
  )
  const started =
    state.reasoningDetailsObserved && !state.reasoningEmitted
-      ? Lifecycle.reasoningStart(
-          state.lifecycle,
-          events,
-          "reasoning-0",
-          reasoningMetadata(state.providerMetadataKey, state.reasoningField),
-        )
+      ? Lifecycle.reasoningStart(state.lifecycle, events, "reasoning-0", reasoningMetadata(state.reasoningField))
      : state.lifecycle
  const ended = Lifecycle.reasoningEnd(started, events, "reasoning-0", metadata)
  const lifecycle = toolCallEvents.length ? Lifecycle.stepStart(ended, events) : ended
  events.push(...toolCallEvents)
  Lifecycle.finish(lifecycle, events, { reason, usage: state.usage })
  return events
-})
+}

 // =============================================================================
 // Protocol And OpenAI Route
@@ -1168,7 +870,6 @@ export const protocol = Protocol.make({
  stream: {
    event: Protocol.jsonEvent(OpenAIChatEvent),
    initial: (request) => ({
-      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
      tools: ToolStream.empty<number>(),
      pendingTools: {},
      toolCallEvents: [],
@@ -1178,7 +879,6 @@ export const protocol = Protocol.make({
      reasoningDetailsObserved: false,
      reasoningEmitted: false,
      nextToolIndex: 0,
-      requireFinishReason: request.model.compatibility?.requireFinishReason ?? true,
    }),
    step,
    onHalt: finishEvents,
@@ -17,7 +17,6 @@ export const route = Route.make({
  protocol: OpenResponses.protocol,
  endpoint: Endpoint.path(OpenResponses.PATH),
  transport: OpenResponses.httpTransport,
-  defaults: { providerOptions: { store: false, include: ["reasoning.encrypted_content"] } },
 })

 export * as OpenAICompatibleResponses from "./openai-compatible-responses.js"
@@ -9,7 +9,14 @@ import {
  type ImageRoute,
 } from "../image.js"
 import { Auth, type Definition as AuthDefinition } from "../route/auth.js"
-import { Usage, mergeHttpOptions, mergeJsonRecords, type HttpOptions } from "../schema/index.js"
+import {
+  InvalidProviderOutputReason,
+  AIError,
+  Usage,
+  mergeHttpOptions,
+  mergeJsonRecords,
+  type HttpOptions,
+} from "../schema/index.js"
 import { ProviderShared } from "./shared.js"
 import { ImageInputs } from "./utils/image-input.js"
 import { OpenAIImage } from "./utils/openai-image.js"
@@ -77,6 +84,13 @@ const nativeOptions = (options: OpenAIImageOptions | undefined) => {
  }
 }

+const invalidOutput = (message: string) =>
+  new AIError({
+    module: ADAPTER,
+    method: "generate",
+    reason: new InvalidProviderOutputReason({ message, route: ADAPTER }),
+  })
+
 const applyQuery = (url: string, query: Record<string, string> | undefined) => {
  if (!query) return url
  const next = new URL(url)
@@ -90,12 +104,12 @@ export const model = (input: ModelInput) => {
    generate: Effect.fn("OpenAIImages.generate")(function* (request: ImageRequestFor<OpenAIImageOptions>, execute) {
      const mask = request.options?.mask
      if (mask !== undefined && (request.images?.length ?? 0) === 0)
-        return yield* ImageInputs.invalid("An OpenAI image mask requires at least one input image")
+        return yield* ImageInputs.invalid(ADAPTER, "An OpenAI image mask requires at least one input image")
      const http = mergeHttpOptions(request.model.http, request.http)
      const sourceImages = request.images ?? []
      const multipartImages = yield* Effect.forEach(sourceImages, (image) => {
        if (image.type === "bytes") return Effect.succeed({ data: image.data, mediaType: image.mediaType })
-        if (image.type === "url") return ImageInputs.decodeDataUrl(image.url)
+        if (image.type === "url") return ImageInputs.decodeDataUrl(image.url, ADAPTER)
        return Effect.undefined
      })
      const multipartMask =
@@ -104,7 +118,7 @@ export const model = (input: ModelInput) => {
          : mask.type === "bytes"
            ? { data: mask.data, mediaType: mask.mediaType }
            : mask.type === "url"
-              ? yield* ImageInputs.decodeDataUrl(mask.url)
+              ? yield* ImageInputs.decodeDataUrl(mask.url, ADAPTER)
              : undefined
      const useMultipart =
        sourceImages.length > 0 &&
@@ -147,7 +161,7 @@ export const model = (input: ModelInput) => {
        return undefined
      })
      if (references.some((image) => image === undefined))
-        return yield* ImageInputs.invalid("OpenAI Images accepts image URLs, data URLs, bytes, and file IDs")
+        return yield* ImageInputs.invalid(ADAPTER, "OpenAI Images accepts image URLs, data URLs, bytes, and file IDs")
      const maskReference =
        mask === undefined
          ? undefined
@@ -159,7 +173,7 @@ export const model = (input: ModelInput) => {
                ? { file_id: mask.id }
                : undefined
      if (mask !== undefined && maskReference === undefined)
-        return yield* ImageInputs.invalid("OpenAI Images accepts masks as URLs, data URLs, bytes, or file IDs")
+        return yield* ImageInputs.invalid(ADAPTER, "OpenAI Images accepts masks as URLs, data URLs, bytes, or file IDs")
      const requestBody = mergeJsonRecords(
        {
          model: request.model.id,
@@ -195,9 +209,11 @@ const parseResponse = Effect.fn("OpenAIImages.parseResponse")(function* (
  options: OpenAIImageOptions | undefined,
  overlay: Record<string, unknown> | undefined,
 ) {
-  const output = yield* ProviderShared.imageResponse(ADAPTER, "OpenAI Images", response)
-  const decoded = yield* Schema.decodeUnknownEffect(Schema.fromJsonString(OpenAIImageResponse))(output.body).pipe(
-    Effect.mapError((cause) => output.invalid("OpenAI Images returned an invalid response", cause)),
+  const payload = yield* response.json.pipe(
+    Effect.mapError(() => invalidOutput("Failed to read the OpenAI Images response")),
+  )
+  const decoded = yield* Schema.decodeUnknownEffect(OpenAIImageResponse)(payload).pipe(
+    Effect.mapError(() => invalidOutput("OpenAI Images returned an invalid response")),
  )
  const requestBody = mergeJsonRecords(nativeOptions(options), overlay)
  const format =
@@ -205,7 +221,7 @@ const parseResponse = Effect.fn("OpenAIImages.parseResponse")(function* (
  const images = yield* Effect.forEach(decoded.data, (item, index) => {
    if (item.b64_json)
      return Effect.fromResult(Encoding.decodeBase64(item.b64_json)).pipe(
-        Effect.mapError((cause) => output.invalid(`OpenAI Images result ${index} contains invalid base64 data`, cause)),
+        Effect.mapError(() => invalidOutput(`OpenAI Images result ${index} contains invalid base64 data`)),
        Effect.map(
          (data) =>
            new GeneratedImage({
@@ -225,9 +241,9 @@ const parseResponse = Effect.fn("OpenAIImages.parseResponse")(function* (
            item.revised_prompt === undefined ? undefined : { openai: { revisedPrompt: item.revised_prompt } },
        }),
      )
-    return Effect.fail(output.invalid(`OpenAI Images result ${index} has neither image data nor a URL`))
+    return Effect.fail(invalidOutput(`OpenAI Images result ${index} has neither image data nor a URL`))
  })
-  if (images.length === 0) return yield* output.invalid("OpenAI Images returned no images")
+  if (images.length === 0) return yield* invalidOutput("OpenAI Images returned no images")
  return new ImageResponse({
    images,
    usage:
@@ -1,10 +1,10 @@
-import { AIError, TransportError } from "../schema/index.js"
+import { AIError, TransportReason } from "../schema/index.js"
 import type { ChannelCheckpoint, ChannelObservation, WebSocketChannelDriver } from "../route/transport/index.js"
 import { Effect, Option, Schema } from "effect"
 import * as ProviderShared from "./shared.js"
 import { OpenResponses } from "./open-responses.js"

-const PROTOCOL = "open-responses.websocket.v1"
+const PROTOCOL = "openai-responses.websocket.v1"
 const VERSION = 1
 const decodeEvent = Schema.decodeUnknownEffect(OpenResponses.protocol.stream.event)

@@ -99,17 +99,17 @@ const incremental = (
 const code = (event: OpenResponses.Event) => event.code || event.error?.code || event.response?.error?.code || undefined

 const rejected = (
+  input: DriverInput,
  observation: Extract<ChannelObservation, { readonly type: "provider-failure" }>,
  recovery: "retry-full" | "rotate-and-retry-full",
 ): ChannelObservation => ({
  type: "rejected",
  recovery,
  error: new AIError({
-    reason: new TransportError({
+    module: input.id,
+    method: "stream",
+    reason: new TransportReason({
      message: observation.error.message,
-      body: observation.error.reason.body,
-      http: observation.error.reason.http,
-      cause: observation.error.reason.cause,
      transport: "websocket",
      operation: "read",
      phase: "receive",
@@ -137,16 +137,15 @@ export const driver = (input: DriverInput): WebSocketChannelDriver => {
    observe: (create, frame) =>
      Effect.gen(function* () {
        const event = yield* decodeEvent(frame).pipe(
-          Effect.mapError((cause) =>
-            ProviderShared.eventError(input.id, `Invalid ${input.name} WebSocket event`, frame, cause),
-          ),
+          Effect.mapError(() => ProviderShared.eventError(input.id, `Invalid ${input.name} WebSocket event`, frame)),
        )
        const observation = yield* input.base.observe(create, frame)
        if (event.type === "response.output_item.done" && event.item) output.push(event.item)
        if (observation.type === "provider-failure") {
          const rejection = code(event)
-          if (rejection === "previous_response_not_found") return rejected(observation, "retry-full")
-          if (rejection === "websocket_connection_limit_reached") return rejected(observation, "rotate-and-retry-full")
+          if (rejection === "previous_response_not_found") return rejected(input, observation, "retry-full")
+          if (rejection === "websocket_connection_limit_reached")
+            return rejected(input, observation, "rotate-and-retry-full")
        }
        if (observation.type !== "completed") return observation
        const responseID = event.response?.id
@@ -155,16 +154,11 @@ export const driver = (input: DriverInput): WebSocketChannelDriver => {
          ...observation,
          checkpoint: {
            protocol: PROTOCOL,
-            value: {
-              version: VERSION,
-              responseID,
-              request,
-              output: event.response?.output ? [...event.response.output] : output.slice(),
-            } satisfies CheckpointValue,
+            value: { version: VERSION, responseID, request, output: output.slice() } satisfies CheckpointValue,
          },
        }
      }),
  }
 }

-export const OpenResponsesContinuation = { driver } as const
+export const OpenAIResponsesChannel = { driver } as const
@@ -5,13 +5,14 @@ import { Auth } from "../route/auth.js"
 import { Endpoint } from "../route/endpoint.js"
 import { Protocol } from "../route/protocol.js"
 import { HttpTransport } from "../route/transport/index.js"
-import { LLMRequest, type JsonSchema, type ToolDefinition } from "../schema/index.js"
+import { LLMEvent, LLMRequest, type JsonSchema, type ToolDefinition } from "../schema/index.js"
 import { OpenResponses } from "./open-responses.js"
-import { JsonObject, optionalArray, optionalNull, ProviderShared } from "./shared.js"
+import { optionalArray, ProviderShared } from "./shared.js"
+import { Lifecycle } from "./utils/lifecycle.js"
 import { OpenAIImage } from "./utils/openai-image.js"
-import { ResponsesHostedTools } from "./utils/responses-hosted-tools.js"
 import { ToolSchemaProjection } from "./utils/tool-schema.js"
-import { OpenResponsesChannel } from "./open-responses-channel.js"
+import { OpenResponsesChannel, type Options } from "./open-responses-channel.js"
+import { OpenAIResponsesChannel } from "./openai-responses-channel.js"

 const ADAPTER = "openai-responses"
 const NAME = "OpenAI Responses"
@@ -32,40 +33,6 @@ const OpenAIResponsesImageGenerationTool = Schema.Struct({
  size: Schema.optional(OpenAIImage.Size),
 })

-const OpenAIResponsesHostedToolItem = Schema.Union([
-  Schema.StructWithRest(
-    Schema.Struct({
-      type: Schema.tag("computer_call"),
-      id: Schema.String,
-      status: Schema.optional(Schema.String),
-      call_id: Schema.optional(Schema.String),
-      action: optionalNull(JsonObject),
-      pending_safety_checks: Schema.optional(Schema.Array(JsonObject)),
-    }),
-    [JsonObject],
-  ),
-  Schema.StructWithRest(
-    Schema.Struct({
-      type: Schema.tag("web_search_preview_call"),
-      id: Schema.String,
-      status: Schema.optional(Schema.String),
-      action: optionalNull(JsonObject),
-    }),
-    [JsonObject],
-  ),
-  Schema.StructWithRest(
-    Schema.Struct({
-      type: Schema.tag("image_generation_call"),
-      id: Schema.String,
-      status: Schema.optional(Schema.String),
-      result: optionalNull(Schema.String),
-      output_format: Schema.optional(Schema.Literals(["png", "jpeg", "webp"])),
-      revised_prompt: optionalNull(Schema.String),
-    }),
-    [JsonObject],
-  ),
-])
-
 const OpenAIResponsesTools = Schema.Union([OpenResponses.Tool, OpenAIResponsesImageGenerationTool])

 const OpenAIResponsesToolChoice = Schema.Union([
@@ -73,9 +40,20 @@ const OpenAIResponsesToolChoice = Schema.Union([
  Schema.Struct({ type: Schema.tag("image_generation") }),
 ])

+const OpenAIResponsesInputItem = Schema.Union([
+  Schema.Struct({
+    type: Schema.tag("message"),
+    id: Schema.optionalKey(Schema.String),
+    role: Schema.tag("assistant"),
+    content: Schema.Array(Schema.Struct({ type: Schema.tag("output_text"), text: Schema.String })),
+    phase: Schema.optionalKey(Schema.NullOr(OpenResponses.MessagePhase)),
+  }),
+  OpenResponses.InputItem,
+])
+
 const OpenAIResponsesCoreFields = {
  ...OpenResponses.coreFields,
-  input: Schema.Array(Schema.Union([OpenResponses.InputItem, OpenAIResponsesHostedToolItem])),
+  input: Schema.Array(OpenAIResponsesInputItem),
  tools: optionalArray(OpenAIResponsesTools),
  tool_choice: Schema.optional(OpenAIResponsesToolChoice),
 }
@@ -89,7 +67,16 @@ export type OpenAIResponsesBody = Schema.Schema.Type<typeof OpenAIResponsesBody>
 const extension = {
  id: ADAPTER,
  name: NAME,
-  lowerHostedToolItem: (item: unknown) => (Schema.is(OpenAIResponsesHostedToolItem)(item) ? item : undefined),
+  messagePhase: (value: unknown) => (value === null ? null : undefined),
+  lowerMedia: ({ part, media, request }) => {
+    if (request.model.provider !== "xai" || media.mime !== "application/pdf") return undefined
+    return {
+      type: "input_file",
+      filename: part.filename ?? "document.pdf",
+      file_data: media.base64,
+      mime_type: media.mime,
+    }
+  },
 } satisfies OpenResponses.Extension

 const nativeImageToolInput = (tool: ToolDefinition) => {
@@ -122,18 +109,14 @@ const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>, tool
        : { type: "function" as const, name },
  })

-const decodeBody = ProviderShared.validateWith(Schema.decodeUnknownEffect(OpenAIResponsesBody))
-
 const fromRequest = Effect.fn("OpenAIResponses.fromRequest")(function* (request: LLMRequest) {
  const body = yield* OpenResponses.fromRequestWithExtension(
    LLMRequest.update(request, { tools: [], toolChoice: undefined }),
    extension,
  )
  const toolSchemaCompatibility = request.model.compatibility?.toolSchema
-  const parallelToolCalls = OpenResponses.resolveParallelToolCalls(request)
-  return yield* decodeBody({
+  return {
    ...body,
-    ...(parallelToolCalls === undefined ? {} : { parallel_tool_calls: parallelToolCalls }),
    tools:
      request.tools.length === 0
        ? undefined
@@ -142,16 +125,53 @@ const fromRequest = Effect.fn("OpenAIResponses.fromRequest")(function* (request:
          ),
    tool_choice:
      body.tool_choice ?? (request.toolChoice ? yield* lowerToolChoice(request.toolChoice, request.tools) : undefined),
-  })
+  } satisfies OpenAIResponsesBody
 })

-const hostedToolResult = Effect.fn("OpenAIResponses.hostedToolResult")(function* (item: ResponsesHostedTools.Item) {
+type HostedToolData = OpenResponses.StreamItem & {
+  readonly id: string
+  readonly status?: string
+  readonly action?: unknown
+  readonly queries?: unknown
+  readonly results?: unknown
+  readonly code?: string
+  readonly container_id?: string
+  readonly outputs?: unknown
+  readonly server_label?: string
+  readonly output?: unknown
+  readonly result?: string
+  readonly output_format?: "png" | "jpeg" | "webp"
+  readonly error?: unknown
+}
+
+const HOSTED_TOOLS = {
+  web_search_call: { name: "web_search", input: (item) => item.action ?? {} },
+  web_search_preview_call: { name: "web_search_preview", input: (item) => item.action ?? {} },
+  file_search_call: { name: "file_search", input: (item) => ({ queries: item.queries ?? [] }) },
+  code_interpreter_call: {
+    name: "code_interpreter",
+    input: (item) => ({ code: item.code, container_id: item.container_id }),
+  },
+  computer_use_call: { name: "computer_use", input: (item) => item.action ?? {} },
+  image_generation_call: { name: "image_generation", input: () => ({}) },
+  mcp_call: {
+    name: "mcp",
+    input: (item) => ({ server_label: item.server_label, name: item.name, arguments: item.arguments }),
+  },
+  local_shell_call: { name: "local_shell", input: (item) => item.action ?? {} },
+} as const satisfies Record<string, { readonly name: string; readonly input: (item: HostedToolData) => unknown }>
+
+type HostedToolType = keyof typeof HOSTED_TOOLS
+type HostedToolItem = HostedToolData & { readonly type: HostedToolType }
+
+const isHostedToolItem = (item: OpenResponses.StreamItem): item is HostedToolItem =>
+  item.type in HOSTED_TOOLS && typeof item.id === "string" && item.id.length > 0
+
+const hostedToolResult = Effect.fn("OpenAIResponses.hostedToolResult")(function* (item: HostedToolItem) {
  const isError = item.error !== undefined && item.error !== null
  if (item.type === "image_generation_call" && item.result) {
    yield* Effect.fromResult(Encoding.decodeBase64(item.result)).pipe(
-      Effect.mapError((cause) =>
-        ProviderShared.eventError(ADAPTER, "OpenAI Responses returned invalid image base64", undefined, cause),
-      ),
+      Effect.mapError(() => ProviderShared.eventError(ADAPTER, "OpenAI Responses returned invalid image base64")),
    )
    const format = item.output_format ?? "png"
    return {
@@ -168,31 +188,44 @@ const hostedToolResult = Effect.fn("OpenAIResponses.hostedToolResult")(function*
  return isError ? { type: "error" as const, value: item.error } : { type: "json" as const, value: item }
 })

-const HOSTED_TOOLS = {
-  web_search_call: { name: "web_search", input: (item) => item.action ?? {} },
-  web_search_preview_call: { name: "web_search_preview", input: (item) => item.action ?? {} },
-  file_search_call: { name: "file_search", input: (item) => ({ queries: item.queries ?? [] }) },
-  code_interpreter_call: {
-    name: "code_interpreter",
-    input: (item) => ({ code: item.code, container_id: item.container_id }),
-  },
-  computer_call: { name: "computer_use", input: (item) => item.action ?? {} },
-  image_generation_call: { name: "image_generation", input: () => ({}), result: hostedToolResult },
-  mcp_call: {
-    name: "mcp",
-    input: (item) => ({ server_label: item.server_label, name: item.name, arguments: item.arguments }),
-  },
-} as const satisfies ResponsesHostedTools.Definitions
+const onHostedToolDone = Effect.fn("OpenAIResponses.onHostedToolDone")(function* (
+  state: OpenResponses.ParserState,
+  item: HostedToolItem,
+) {
+  const tool = HOSTED_TOOLS[item.type]
+  const providerMetadata = OpenResponses.providerMetadata(state, { itemId: item.id })
+  const events: LLMEvent[] = []
+  const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
+  events.push(
+    LLMEvent.toolCall({
+      id: item.id,
+      name: tool.name,
+      input: tool.input(item),
+      providerExecuted: true,
+      providerMetadata,
+    }),
+    LLMEvent.toolResult({
+      id: item.id,
+      name: tool.name,
+      result: yield* hostedToolResult(item),
+      providerExecuted: true,
+      providerMetadata,
+    }),
+  )
+  return [{ ...state, lifecycle }, events] satisfies OpenResponses.StepResult
+})

 const step = (state: OpenResponses.ParserState, event: OpenResponses.Event) => {
-  if (event.type === "response.reasoning_text.delta")
-    return event.item_id !== undefined
-      ? Effect.succeed(
-          OpenResponses.onReasoningDelta(state, event, OpenResponses.outputItemID(state, event) ?? event.item_id),
-        )
+  if (event.type === "response.reasoning_text.delta" || event.type === "response.reasoning_summary.delta")
+    return event.item_id
+      ? Effect.succeed(OpenResponses.onReasoningDelta(state, event, event.item_id))
      : ProviderShared.eventError(ADAPTER, `${event.type} is missing item_id`)
-  if (event.type === "response.output_item.done" && event.item && ResponsesHostedTools.isItem(event.item, HOSTED_TOOLS))
-    return ResponsesHostedTools.onDone(state, event.item, HOSTED_TOOLS)
+  if (event.type === "response.reasoning_text.done" || event.type === "response.reasoning_summary.done")
+    return event.item_id
+      ? Effect.succeed(OpenResponses.onReasoningDone(state, event))
+      : ProviderShared.eventError(ADAPTER, `${event.type} is missing item_id`)
+  if (event.type === "response.output_item.done" && event.item && isHostedToolItem(event.item))
+    return onHostedToolDone(state, event.item)
  return OpenResponses.step(state, event)
 }

@@ -214,7 +247,11 @@ const endpoint = Endpoint.path<OpenAIResponsesBody>(PATH, { baseURL: DEFAULT_BAS
 const auth = Auth.none

 export const httpTransport = HttpTransport.sseJson.with<OpenAIResponsesBody>()
-export const channelTransport = OpenResponsesChannel.transport<OpenAIResponsesBody>
+export const channelTransport = (options: Omit<Options, "driver">) =>
+  OpenResponsesChannel.transport<OpenAIResponsesBody>({
+    ...options,
+    driver: (input) => OpenAIResponsesChannel.driver({ id: options.id, name: options.name, ...input }),
+  })
 export const transport = channelTransport({
  id: ADAPTER,
  name: NAME,
@@ -230,7 +267,7 @@ export const route = Route.make({
  endpoint,
  auth,
  transport,
-  defaults: { providerOptions: { store: false, include: ["reasoning.encrypted_content"] } },
+  defaults: { providerOptions: { store: false } },
 })

 export * as OpenAIResponses from "./openai-responses.js"
@@ -2,12 +2,11 @@ import { Buffer } from "node:buffer"
 import { Tool } from "@opencode-ai/schema/tool"
 import { Effect, Schema, Stream } from "effect"
 import * as Sse from "effect/unstable/encoding/Sse"
-import { Headers, HttpClientRequest, HttpClientResponse } from "effect/unstable/http"
+import { Headers, HttpClientRequest } from "effect/unstable/http"
 import {
-  InvalidProviderOutputError,
-  InvalidRequestError,
+  InvalidProviderOutputReason,
+  InvalidRequestReason,
  AIError,
-  HttpContext,
  type ContentPart,
  type LLMRequest,
  type MediaPart,
@@ -25,17 +24,6 @@ export const JsonObject = Schema.Record(Schema.String, Schema.Unknown)
 export const optionalArray = <const S extends Schema.Top>(schema: S) => Schema.optional(Schema.Array(schema))
 export const optionalNull = <const S extends Schema.Top>(schema: S) => Schema.optional(Schema.NullOr(schema))

-export const OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH = 64
-
-// OpenAI limits `prompt_cache_key` to 64 chars; DeepSeek and Zai inherit the same
-// limit via their OpenAI-compatible APIs. Clamp with unicode-aware slicing.
-export const promptCacheKey = (request: LLMRequest): string | undefined => {
-  if (request.cache === "none" || request.promptCacheKey === undefined) return undefined
-  const chars = Array.from(request.promptCacheKey)
-  if (chars.length <= OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH) return request.promptCacheKey
-  return chars.slice(0, OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH).join("")
-}
-
 /**
 * Streaming tool-call accumulator. Adapters that build a tool call across
 * multiple `tool-input-delta` chunks store the partial JSON input string here
@@ -97,15 +85,17 @@ export const sumTokens = (...values: ReadonlyArray<number | undefined>): number
  return values.reduce((acc: number, value) => acc + (value ?? 0), 0)
 }

-export const eventError = (route: string, message: string, body?: string, cause?: unknown) =>
+export const eventError = (route: string, message: string, raw?: string) =>
  new AIError({
-    reason: new InvalidProviderOutputError({ route, message, body, cause }),
+    module: "ProviderShared",
+    method: "stream",
+    reason: new InvalidProviderOutputReason({ route, message, raw }),
  })

 export const parseJson = (route: string, input: string, message: string) =>
  Effect.try({
    try: () => decodeJson(input),
-    catch: (cause) => eventError(route, message, input, cause),
+    catch: () => eventError(route, message, input),
  })

 /**
@@ -209,9 +199,10 @@ export const errorText = (error: unknown) => {
 * `framing` step for Server-Sent Events. Decodes UTF-8, runs the SSE channel
 * decoder, optionally filters named events, and drops empty / `[DONE]`
 * keep-alive events so the protocol event schema sees one JSON string per
- * element. Retry control events are ignored without interrupting the stream.
- * Decoder failures become provider output errors so the public error channel
- * stays `AIError`.
+ * element. The SSE channel emits a
+ * `Retry` control event on its error channel; we drop it here (we don't
+ * implement client-driven retries). Decoder failures become provider output
+ * errors so the public error channel stays `AIError`.
 */
 export const sseFraming = (
  bytes: Stream.Stream<Uint8Array, AIError>,
@@ -219,23 +210,9 @@ export const sseFraming = (
 ): Stream.Stream<string, AIError> =>
  bytes.pipe(
    Stream.decodeText(),
-    Stream.mapAccumEffect(
-      () => {
-        const output: Sse.Event[] = []
-        return {
-          output,
-          parser: Sse.makeParser((event) => {
-            if (event._tag === "Event") output.push(event)
-          }),
-        }
-      },
-      (state, chunk) =>
-        Effect.gen(function* () {
-          const error = state.parser.feed(chunk)
-          if (error) return yield* eventError("sse", error.message, chunk, error)
-          return [state, state.output.splice(0)] as const
-        }),
-    ),
+    Stream.pipeThroughChannel(Sse.decode()),
+    Stream.catchTag("Retry", () => Stream.empty),
+    Stream.catchTag("SseError", (error) => Stream.fail(eventError("sse", error.message))),
    Stream.filter(
      (event) =>
        (events === undefined || events.has(event.event)) &&
@@ -248,39 +225,13 @@ export const sseFraming = (
 /**
 * Canonical invalid-request constructor shared by protocol lowering.
 */
-export const invalidRequest = (message: string, cause?: unknown) =>
+export const invalidRequest = (message: string) =>
  new AIError({
-    reason: new InvalidRequestError({ message, cause }),
+    module: "ProviderShared",
+    method: "request",
+    reason: new InvalidRequestReason({ message }),
  })

-export const imageResponse = Effect.fn("ProviderShared.imageResponse")(function* (
-  route: string,
-  name: string,
-  response: HttpClientResponse.HttpClientResponse,
-) {
-  const http = new HttpContext({ url: response.request.url, status: response.status, headers: response.headers })
-  const body = yield* response.text.pipe(
-    Effect.mapError(
-      (cause) =>
-        new AIError({
-          reason: new InvalidProviderOutputError({
-            route,
-            message: `Failed to read the ${name} response`,
-            http,
-            cause,
-          }),
-        }),
-    ),
-  )
-  return {
-    body,
-    invalid: (message: string, cause?: unknown) =>
-      new AIError({
-        reason: new InvalidProviderOutputError({ route, message, body, http, cause }),
-      }),
-  }
-})
-
 export const matchToolChoice = <Auto, None, Required, Tool>(
  route: string,
  toolChoice: NonNullable<LLMRequest["toolChoice"]>,
@@ -327,7 +278,7 @@ export const unsupportedContent = (
 export const validateWith =
  <A, I, E extends { readonly message: string }>(decode: (input: I) => Effect.Effect<A, E>) =>
  (payload: I) =>
-    decode(payload).pipe(Effect.mapError((error) => invalidRequest(error.message, error)))
+    decode(payload).pipe(Effect.mapError((error) => invalidRequest(error.message)))

 /**
 * Build an HTTP POST with a JSON body. Sets `content-type: application/json`
@@ -1,10 +1,12 @@
 import { Effect, Encoding } from "effect"
 import type { ImageInput } from "../../image.js"
-import { InvalidRequestError, AIError } from "../../schema/index.js"
+import { InvalidRequestReason, AIError } from "../../schema/index.js"

-const invalid = (message: string, cause?: unknown) =>
+const invalid = (module: string, message: string) =>
  new AIError({
-    reason: new InvalidRequestError({ message, cause }),
+    module,
+    method: "generate",
+    reason: new InvalidRequestReason({ message }),
  })

 export const dataUrl = (input: Extract<ImageInput, { readonly type: "bytes" }>) =>
@@ -12,12 +14,13 @@ export const dataUrl = (input: Extract<ImageInput, { readonly type: "bytes" }>)

 export const decodeDataUrl = (
  url: string,
+  module: string,
 ): Effect.Effect<{ readonly mediaType: string; readonly data: Uint8Array } | undefined, AIError> => {
  if (!url.startsWith("data:")) return Effect.undefined
  const match = /^data:([^;,]+);base64,(.*)$/s.exec(url)
-  if (!match) return Effect.fail(invalid("Image data URLs must contain a MIME type and base64 data"))
+  if (!match) return Effect.fail(invalid(module, "Image data URLs must contain a MIME type and base64 data"))
  return Effect.fromResult(Encoding.decodeBase64(match[2])).pipe(
-    Effect.mapError((cause) => invalid("Image data URL contains invalid base64 data", cause)),
+    Effect.mapError(() => invalid(module, "Image data URL contains invalid base64 data")),
    Effect.map((data) => ({ mediaType: match[1], data })),
  )
 }
@@ -28,10 +28,7 @@ export const ResponseIncludables = [
 export type ResponseIncludable = (typeof ResponseIncludables)[number] | (string & {})

 export const ServiceTiers = ["auto", "default", "flex", "priority"] as const
-export type ServiceTier = (typeof ServiceTiers)[number] | (string & {})
-export const ServiceTier = Schema.declare<ServiceTier>((value): value is ServiceTier => typeof value === "string", {
-  title: "ServiceTier",
-})
+export type ServiceTier = (typeof ServiceTiers)[number]

 export const Truncations = ["auto", "disabled"] as const
 export type Truncation = (typeof Truncations)[number]
@@ -41,7 +38,7 @@ export const ResponseIncludableSchema = Schema.declare<ResponseIncludable>(
  (value): value is ResponseIncludable => typeof value === "string",
  { title: "ResponseIncludable" },
 )
-export const ServiceTierSchema = ServiceTier
+export const ServiceTierSchema = Schema.Literals(ServiceTiers)
 export const TruncationSchema = Schema.Literals(Truncations)

 export const AllowedTools = Schema.Struct({
@@ -55,6 +52,7 @@ export const StreamOptions = Schema.Struct({
 })

 export const Options = Schema.Struct({
+  instructions: Schema.optional(Schema.String),
  store: Schema.optional(Schema.Boolean),
  metadata: Schema.optional(Schema.Record(Schema.String, Schema.String)),
  safetyIdentifier: Schema.optional(Schema.String),
@@ -9,8 +9,8 @@ export type OpenAITextVerbosity = OpenResponsesOptions.TextVerbosity
 // in lockstep with `openai-node/src/resources/responses/responses.ts`.
 export const OpenAIResponseIncludables = OpenResponsesOptions.ResponseIncludables
 export type OpenAIResponseIncludable = OpenResponsesOptions.ResponseIncludable
-export const OpenAIServiceTiers = [...OpenResponsesOptions.ServiceTiers, "scale"] as const
-export type OpenAIServiceTier = (typeof OpenAIServiceTiers)[number] | (string & {})
+export const OpenAIServiceTiers = OpenResponsesOptions.ServiceTiers
+export type OpenAIServiceTier = OpenResponsesOptions.ServiceTier

 export const OpenAIReasoningEffort = OpenResponsesOptions.ReasoningEffort
 export const OpenAITextVerbosity = OpenResponsesOptions.TextVerbosity
@@ -1,64 +0,0 @@
-/*
- * Adapted from partial-json by the Promplate Dev Team:
- * https://github.com/promplate/partial-json-parser-js/blob/main/src/options.ts
- * Licensed under the MIT License; see partial-json.ts for the complete notice.
- */
-
-/**
- * allow partial strings like `"hello \u12` to be parsed as `"hello `
- */
-export const STR = 0b000000001
-
-/**
- * allow partial numbers like `123.` to be parsed as `123`
- */
-export const NUM = 0b000000010
-
-/**
- * allow partial arrays like `[1, 2,` to be parsed as `[1, 2]`
- */
-export const ARR = 0b000000100
-
-/**
- * allow partial objects like `{"a": 1, "b":` to be parsed as `{"a": 1}`
- */
-export const OBJ = 0b000001000
-
-/**
- * allow `nu` to be parsed as `null`
- */
-export const NULL = 0b000010000
-
-/**
- * allow `tr` to be parsed as `true`, and `fa` to be parsed as `false`
- */
-export const BOOL = 0b000100000
-
-/**
- * allow `Na` to be parsed as `NaN`
- */
-export const NAN = 0b001000000
-
-/**
- * allow `Inf` to be parsed as `Infinity`
- */
-export const INFINITY = 0b010000000
-
-/**
- * allow `-Inf` to be parsed as `-Infinity`
- */
-export const _INFINITY = 0b100000000
-
-export const INF = INFINITY | _INFINITY
-export const SPECIAL = NULL | BOOL | INF | NAN
-export const ATOM = STR | NUM | SPECIAL
-export const COLLECTION = ARR | OBJ
-export const ALL = ATOM | COLLECTION
-
-/**
- * Control what types you allow to be partially parsed.
- * The default is to allow all types to be partially parsed, which in most cases is the best option.
- */
-export const Allow = { STR, NUM, ARR, OBJ, NULL, BOOL, NAN, INFINITY, _INFINITY, INF, SPECIAL, ATOM, COLLECTION, ALL }
-
-export default Allow
@@ -1,282 +0,0 @@
-/*
- * Adapted from partial-json by the Promplate Dev Team:
- * https://github.com/promplate/partial-json-parser-js
- *
- * MIT License
- *
- * Copyright (c) 2023 Promplate Dev Team
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-
-import { Schema } from "effect"
-import { Allow } from "./partial-json-options.js"
-export * from "./partial-json-options.js"
-
-export class PartialJSON extends Error {}
-export class MalformedJSON extends Error {}
-
-const decodeJson = Schema.decodeUnknownSync(Schema.fromJsonString(Schema.Unknown))
-
-/** Parse complete or incomplete JSON, restricted by the supplied partial-value flags. */
-export function parseJSON(jsonString: string, allowPartial = Allow.ALL): unknown {
-  if (typeof jsonString !== "string") throw new TypeError(`expecting str, got ${typeof jsonString}`)
-  const input = jsonString.trim()
-  if (!input) throw new Error(`${jsonString} is empty`)
-  try {
-    return decodeJson(input)
-  } catch {}
-
-  const repaired = repairJSON(input)
-  if (repaired !== input) {
-    try {
-      return decodeJson(repaired)
-    } catch {}
-  }
-
-  try {
-    return _parseJSON(input, allowPartial)
-  } catch (error) {
-    if (repaired !== input) return _parseJSON(repaired, allowPartial)
-    throw error
-  }
-}
-
-const repairJSON = (input: string) => {
-  let repaired = ""
-  let quoted = false
-
-  for (let index = 0; index < input.length; index++) {
-    const character = input[index]
-    if (!quoted) {
-      repaired += character
-      if (character === '"') quoted = true
-      continue
-    }
-
-    if (character === '"') {
-      repaired += character
-      quoted = false
-      continue
-    }
-
-    if (character === "\\") {
-      const next = input[index + 1]
-      if (next === "u" && /^[0-9a-fA-F]{4}$/.test(input.slice(index + 2, index + 6))) {
-        repaired += input.slice(index, index + 6)
-        index += 5
-        continue
-      }
-      if (next !== undefined && '"\\/bfnrtu'.includes(next)) {
-        repaired += `\\${next}`
-        index++
-        continue
-      }
-      repaired += "\\\\"
-      continue
-    }
-
-    const code = character.charCodeAt(0)
-    repaired += code <= 0x1f ? `\\u${code.toString(16).padStart(4, "0")}` : character
-  }
-
-  return repaired
-}
-
-const _parseJSON = (jsonString: string, allow: number) => {
-  const length = jsonString.length
-  let index = 0
-
-  const markPartialJSON = (message: string): never => {
-    throw new PartialJSON(`${message} at position ${index}`)
-  }
-
-  const throwMalformedError = (message: string): never => {
-    throw new MalformedJSON(`${message} at position ${index}`)
-  }
-
-  const parseAny = (): unknown => {
-    skipBlank()
-    if (index >= length) markPartialJSON("Unexpected end of input")
-    if (jsonString[index] === '"') return parseStr()
-    if (jsonString[index] === "{") return parseObj()
-    if (jsonString[index] === "[") return parseArr()
-    if (
-      jsonString.substring(index, index + 4) === "null" ||
-      (Allow.NULL & allow && length - index < 4 && "null".startsWith(jsonString.substring(index)))
-    ) {
-      index += 4
-      return null
-    }
-    if (
-      jsonString.substring(index, index + 4) === "true" ||
-      (Allow.BOOL & allow && length - index < 4 && "true".startsWith(jsonString.substring(index)))
-    ) {
-      index += 4
-      return true
-    }
-    if (
-      jsonString.substring(index, index + 5) === "false" ||
-      (Allow.BOOL & allow && length - index < 5 && "false".startsWith(jsonString.substring(index)))
-    ) {
-      index += 5
-      return false
-    }
-    if (
-      jsonString.substring(index, index + 8) === "Infinity" ||
-      (Allow.INFINITY & allow && length - index < 8 && "Infinity".startsWith(jsonString.substring(index)))
-    ) {
-      index += 8
-      return Infinity
-    }
-    if (
-      jsonString.substring(index, index + 9) === "-Infinity" ||
-      (Allow._INFINITY & allow &&
-        1 < length - index &&
-        length - index < 9 &&
-        "-Infinity".startsWith(jsonString.substring(index)))
-    ) {
-      index += 9
-      return -Infinity
-    }
-    if (
-      jsonString.substring(index, index + 3) === "NaN" ||
-      (Allow.NAN & allow && length - index < 3 && "NaN".startsWith(jsonString.substring(index)))
-    ) {
-      index += 3
-      return NaN
-    }
-    return parseNum()
-  }
-
-  const parseStr = (): string => {
-    const start = index
-    let escape = false
-    index++
-    while (index < length && (jsonString[index] !== '"' || (escape && jsonString[index - 1] === "\\"))) {
-      escape = jsonString[index] === "\\" ? !escape : false
-      index++
-    }
-    if (jsonString.charAt(index) === '"') {
-      try {
-        return decodeJson(jsonString.substring(start, ++index - Number(escape))) as string
-      } catch (error) {
-        throwMalformedError(String(error))
-      }
-    }
-    if (Allow.STR & allow) {
-      try {
-        return decodeJson(`${jsonString.substring(start, index - Number(escape))}"`) as string
-      } catch {
-        return decodeJson(`${jsonString.substring(start, jsonString.lastIndexOf("\\"))}"`) as string
-      }
-    }
-    return markPartialJSON("Unterminated string literal")
-  }
-
-  const parseObj = (): Record<string, unknown> => {
-    index++
-    skipBlank()
-    const object: Record<string, unknown> = {}
-    try {
-      while (jsonString[index] !== "}") {
-        skipBlank()
-        if (index >= length && Allow.OBJ & allow) return object
-        const key = parseStr()
-        skipBlank()
-        index++
-        try {
-          Object.defineProperty(object, key, {
-            value: parseAny(),
-            enumerable: true,
-            configurable: true,
-            writable: true,
-          })
-        } catch (error) {
-          if (Allow.OBJ & allow) return object
-          throw error
-        }
-        skipBlank()
-        if (jsonString[index] === ",") index++
-      }
-    } catch {
-      if (Allow.OBJ & allow) return object
-      return markPartialJSON("Expected '}' at end of object")
-    }
-    index++
-    return object
-  }
-
-  const parseArr = (): unknown[] => {
-    index++
-    const array: unknown[] = []
-    try {
-      while (jsonString[index] !== "]") {
-        array.push(parseAny())
-        skipBlank()
-        if (jsonString[index] === ",") index++
-      }
-    } catch {
-      if (Allow.ARR & allow) return array
-      return markPartialJSON("Expected ']' at end of array")
-    }
-    index++
-    return array
-  }
-
-  const parseNum = (): unknown => {
-    if (index === 0) {
-      if (jsonString === "-") throwMalformedError("Not sure what '-' is")
-      try {
-        return decodeJson(jsonString)
-      } catch (error) {
-        if (Allow.NUM & allow) {
-          try {
-            return decodeJson(jsonString.substring(0, jsonString.lastIndexOf("e")))
-          } catch {}
-        }
-        throwMalformedError(String(error))
-      }
-    }
-
-    const start = index
-    if (jsonString[index] === "-") index++
-    while (jsonString[index] && !",]}".includes(jsonString[index])) index++
-    if (index === length && !(Allow.NUM & allow)) markPartialJSON("Unterminated number literal")
-
-    try {
-      return decodeJson(jsonString.substring(start, index))
-    } catch (error) {
-      if (jsonString.substring(start, index) === "-") markPartialJSON("Not sure what '-' is")
-      try {
-        return decodeJson(jsonString.substring(start, jsonString.lastIndexOf("e")))
-      } catch {
-        throwMalformedError(String(error))
-      }
-    }
-  }
-
-  const skipBlank = () => {
-    while (index < length && " \n\r\t".includes(jsonString[index])) index++
-  }
-
-  return parseAny()
-}
-
-export const parse = parseJSON
@@ -1,68 +0,0 @@
-import { Effect } from "effect"
-import { LLMEvent, type AIError, type ToolResultPart } from "../../schema/index.js"
-import { OpenResponses } from "../open-responses.js"
-import { Lifecycle } from "./lifecycle.js"
-
-export type Item = OpenResponses.StreamItem & {
-  readonly id: string
-  readonly status?: string
-  readonly action?: unknown
-  readonly queries?: unknown
-  readonly results?: unknown
-  readonly code?: string
-  readonly container_id?: string
-  readonly outputs?: unknown
-  readonly server_label?: string
-  readonly output?: unknown
-  readonly result?: string
-  readonly output_format?: "png" | "jpeg" | "webp"
-  readonly error?: unknown
-}
-
-export interface Definition {
-  readonly name: string
-  readonly input: (item: Item) => unknown
-  readonly result?: (item: Item) => Effect.Effect<ToolResultPart["result"], AIError>
-}
-
-export type Definitions = Readonly<Record<string, Definition>>
-
-export const isItem = <Tools extends Definitions>(item: OpenResponses.StreamItem, tools: Tools): item is Item =>
-  item.type in tools && typeof item.id === "string" && item.id.length > 0
-
-export const onDone: (
-  state: OpenResponses.ParserState,
-  item: Item,
-  tools: Definitions,
-) => Effect.Effect<OpenResponses.StepResult, AIError> = Effect.fn("ResponsesHostedTools.onDone")(
-  function* (state, item, tools) {
-    const tool = tools[item.type]
-    if (!tool) return [state, []] satisfies OpenResponses.StepResult
-    const providerMetadata = OpenResponses.providerMetadata(state, { itemId: item.id })
-    const events: LLMEvent[] = []
-    const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
-    events.push(
-      LLMEvent.toolCall({
-        id: item.id,
-        name: tool.name,
-        input: tool.input(item),
-        providerExecuted: true,
-        providerMetadata,
-      }),
-      LLMEvent.toolResult({
-        id: item.id,
-        name: tool.name,
-        result: tool.result
-          ? yield* tool.result(item)
-          : item.error !== undefined && item.error !== null
-            ? { type: "error", value: item.error }
-            : { type: "json", value: item },
-        providerExecuted: true,
-        providerMetadata,
-      }),
-    )
-    return [{ ...state, lifecycle }, events] satisfies OpenResponses.StepResult
-  },
-)
-
-export * as ResponsesHostedTools from "./responses-hosted-tools.js"
@@ -2,6 +2,20 @@ import type { JsonSchema, LanguageModelToolSchemaCompatibility } from "../../sch
 import { isRecord } from "../../utils/record.js"
 import { GeminiToolSchema } from "./gemini-tool-schema.js"

+const removeNullSchemas = (value: unknown): unknown => {
+  if (Array.isArray(value)) return value.map(removeNullSchemas)
+  if (!isRecord(value)) return value
+  const fields = Object.fromEntries(
+    Object.entries(value)
+      .filter(([key]) => key !== "anyOf")
+      .map(([key, field]) => [key, removeNullSchemas(field)]),
+  )
+  if (!Array.isArray(value.anyOf)) return fields
+  const variants = value.anyOf.filter((variant) => !isRecord(variant) || variant.type !== "null").map(removeNullSchemas)
+  if (variants.length === 1 && isRecord(variants[0])) return { ...fields, ...variants[0] }
+  return { ...fields, anyOf: variants }
+}
+
 const tupleItemsSchema = (items: ReadonlyArray<unknown>) => {
  const projected = items.map(moonshotNode)
  if (projected.length === 0) return {}
@@ -31,7 +45,24 @@ const moonshot = (schema: JsonSchema): JsonSchema => {
  return isRecord(projected) ? projected : {}
 }

-const openAI = (schema: JsonSchema): JsonSchema => schema
+const openAI = (schema: JsonSchema): JsonSchema => {
+  const variants = Array.isArray(schema.anyOf) ? schema.anyOf.filter(isRecord) : []
+  const flattened =
+    variants.length === 0
+      ? { ...schema, type: "object" }
+      : {
+          ...Object.fromEntries(Object.entries(schema).filter(([key]) => key !== "anyOf")),
+          type: "object",
+          properties: variants.reduce(
+            (properties, variant) => ({ ...(isRecord(variant.properties) ? variant.properties : {}), ...properties }),
+            {},
+          ),
+          additionalProperties: false,
+        }
+  const normalized = removeNullSchemas(flattened)
+  return isRecord(normalized) ? normalized : { type: "object" }
+}
+
 const responses = openAI

 const gemini = (schema: JsonSchema): JsonSchema => GeminiToolSchema.convert(schema) ?? {}
@@ -1,10 +1,8 @@
-import { Effect, Option } from "effect"
-import { AIError, LLMEvent, type ProviderMetadata, type ToolCall } from "../../schema/index.js"
+import { Effect } from "effect"
+import { AIError, LLMEvent, type ProviderMetadata, type ToolCall, type ToolInputError } from "../../schema/index.js"
 import { eventError, parseToolInput, type ToolAccumulator } from "../shared.js"
-import { parse } from "./partial-json.js"

 type StreamKey = string | number
-const parsePartialInput = Option.liftThrowable(parse)

 /**
 * One pending streamed tool call. Providers emit the tool identity and JSON
@@ -64,39 +62,38 @@ const inputDelta = (tool: PendingTool, text: string) =>
    id: tool.id,
    name: tool.name,
    text,
-    input: Option.getOrElse(parsePartialInput(tool.input), () => ({})),
  })

 const toolCall = (route: string, tool: PendingTool, inputOverride?: string) => {
  const raw = inputOverride ?? tool.input
  return parseToolInput(route, tool.name, raw).pipe(
+    Effect.map((input): ToolCall | ToolInputError =>
+      LLMEvent.toolCall({
+        id: tool.id,
+        name: tool.name,
+        input,
+        providerExecuted: tool.providerExecuted ? true : undefined,
+        providerMetadata: tool.providerMetadata,
+      }),
+    ),
    Effect.catch((error) =>
      tool.providerExecuted
        ? Effect.fail(error)
        : Effect.succeed(
-            Option.getOrElse(
-              Option.map(parsePartialInput(raw), (input) => input ?? {}),
-              () => ({}),
-            ),
+            LLMEvent.toolInputError({
+              id: tool.id,
+              name: tool.name,
+              raw,
+            }),
          ),
    ),
-    Effect.map(
-      (input): ToolCall =>
-        LLMEvent.toolCall({
-          id: tool.id,
-          name: tool.name,
-          input,
-          providerExecuted: tool.providerExecuted ? true : undefined,
-          providerMetadata: tool.providerMetadata,
-        }),
-    ),
  )
 }

-const finishEvents = (tool: PendingTool, event: ToolCall): ReadonlyArray<LLMEvent> => [
-  LLMEvent.toolInputEnd({ id: tool.id, name: tool.name, providerMetadata: tool.providerMetadata }),
-  event,
-]
+const finishEvents = (tool: PendingTool, event: ToolCall | ToolInputError): ReadonlyArray<LLMEvent> =>
+  event.type === "tool-input-error"
+    ? [event]
+    : [LLMEvent.toolInputEnd({ id: tool.id, name: tool.name, providerMetadata: tool.providerMetadata }), event]

 /** Store the updated tool and produce the optional public delta event. */
 const appendTool = <K extends StreamKey>(
@@ -179,7 +176,7 @@ export const appendExisting = <K extends StreamKey>(

 /**
 * Finalize one pending tool call: parse the accumulated raw JSON, remove it
- * from state, and recover incomplete local arguments when needed.
+ * from state, and return either a call or a non-executable local input error.
 * Missing keys are a no-op because some providers emit stop events for
 * non-tool content blocks.
 */
@@ -2,7 +2,14 @@ import { Effect, Encoding, Schema } from "effect"
 import { Headers, HttpClientRequest } from "effect/unstable/http"
 import { GeneratedImage, ImageModel, ImageResponse, type ImageRequestFor, type ImageRoute } from "../image.js"
 import { Auth, type Definition as AuthDefinition } from "../route/auth.js"
-import { Usage, mergeHttpOptions, mergeJsonRecords, type HttpOptions } from "../schema/index.js"
+import {
+  InvalidProviderOutputReason,
+  AIError,
+  Usage,
+  mergeHttpOptions,
+  mergeJsonRecords,
+  type HttpOptions,
+} from "../schema/index.js"
 import { ProviderShared, optionalNull } from "./shared.js"
 import { ImageInputs } from "./utils/image-input.js"

@@ -87,6 +94,13 @@ const nativeOptions = (options: XAIImageOptions | undefined) => {
  }
 }

+const invalidOutput = (message: string) =>
+  new AIError({
+    module: ADAPTER,
+    method: "generate",
+    reason: new InvalidProviderOutputReason({ message, route: ADAPTER }),
+  })
+
 const applyQuery = (url: string, query: Record<string, string> | undefined) => {
  if (!query) return url
  const next = new URL(url)
@@ -106,7 +120,7 @@ export const model = (input: ModelInput) => {
        return undefined
      })
      if (imageReferences.some((image) => image === undefined))
-        return yield* ImageInputs.invalid("xAI Images accepts image URLs, data URLs, bytes, and file IDs")
+        return yield* ImageInputs.invalid(ADAPTER, "xAI Images accepts image URLs, data URLs, bytes, and file IDs")
      const requestBody = mergeJsonRecords(
        {
          model: request.model.id,
@@ -135,17 +149,17 @@ export const model = (input: ModelInput) => {
          HttpClientRequest.bodyText(text, "application/json"),
        ),
      )
-      const output = yield* ProviderShared.imageResponse(ADAPTER, "xAI Images", response)
-      const decoded = yield* Schema.decodeUnknownEffect(Schema.fromJsonString(XAIImageResponse))(output.body).pipe(
-        Effect.mapError((cause) => output.invalid("xAI Images returned an invalid response", cause)),
+      const payload = yield* response.json.pipe(
+        Effect.mapError(() => invalidOutput("Failed to read the xAI Images response")),
+      )
+      const decoded = yield* Schema.decodeUnknownEffect(XAIImageResponse)(payload).pipe(
+        Effect.mapError(() => invalidOutput("xAI Images returned an invalid response")),
      )
      const images = yield* Effect.forEach(decoded.data, (item, index) => {
        const mediaType = item.mime_type ?? "application/octet-stream"
        if (item.b64_json)
          return Effect.fromResult(Encoding.decodeBase64(item.b64_json)).pipe(
-            Effect.mapError((cause) =>
-              output.invalid(`xAI Images result ${index} contains invalid base64 data`, cause),
-            ),
+            Effect.mapError(() => invalidOutput(`xAI Images result ${index} contains invalid base64 data`)),
            Effect.map(
              (data) =>
                new GeneratedImage({
@@ -169,9 +183,9 @@ export const model = (input: ModelInput) => {
                  : { xai: { revisedPrompt: item.revised_prompt } },
            }),
          )
-        return Effect.fail(output.invalid(`xAI Images result ${index} has neither image data nor a URL`))
+        return Effect.fail(invalidOutput(`xAI Images result ${index} has neither image data nor a URL`))
      })
-      if (images.length === 0) return yield* output.invalid("xAI Images returned no images")
+      if (images.length === 0) return yield* invalidOutput("xAI Images returned no images")
      const usage = ProviderShared.isRecord(decoded.usage) ? decoded.usage : undefined
      return new ImageResponse({
        images,
@@ -1,87 +0,0 @@
-import { Effect, Schema } from "effect"
-import { Protocol } from "../route/protocol.js"
-import type { LLMRequest } from "../schema/index.js"
-import { OpenResponses } from "./open-responses.js"
-import { JsonObject, optionalNull, ProviderShared } from "./shared.js"
-import { ResponsesHostedTools } from "./utils/responses-hosted-tools.js"
-
-const ADAPTER = "xai-responses"
-const NAME = "xAI Responses"
-
-const XAIResponsesHostedToolItem = Schema.Union([
-  Schema.StructWithRest(
-    Schema.Struct({
-      type: Schema.tag("x_search_call"),
-      id: Schema.String,
-      status: Schema.optional(Schema.String),
-      action: optionalNull(JsonObject),
-    }),
-    [JsonObject],
-  ),
-  Schema.StructWithRest(
-    Schema.Struct({
-      type: Schema.tag("image_generation_call"),
-      id: Schema.String,
-      status: Schema.optional(Schema.String),
-      result: Schema.optional(Schema.Unknown),
-      error: Schema.optional(Schema.Unknown),
-    }),
-    [JsonObject],
-  ),
-])
-
-const XAIResponsesBody = Schema.Struct({
-  ...OpenResponses.coreFields,
-  input: Schema.Array(Schema.Union([OpenResponses.InputItem, XAIResponsesHostedToolItem])),
-  stream: Schema.Literal(true),
-})
-
-const extension = {
-  id: ADAPTER,
-  name: NAME,
-  lowerHostedToolItem: (item: unknown) => (Schema.is(XAIResponsesHostedToolItem)(item) ? item : undefined),
-} satisfies OpenResponses.Extension
-
-const decodeBody = ProviderShared.validateWith(Schema.decodeUnknownEffect(XAIResponsesBody))
-const fromRequest = Effect.fn("XAIResponses.fromRequest")(function* (request: LLMRequest) {
-  return yield* decodeBody(yield* OpenResponses.fromRequestWithExtension(request, extension))
-})
-
-const HOSTED_TOOLS = {
-  web_search_call: { name: "web_search", input: (item) => item.action ?? {} },
-  x_search_call: { name: "x_search", input: (item) => item.action ?? {} },
-  file_search_call: { name: "file_search", input: (item) => ({ queries: item.queries ?? [] }) },
-  code_interpreter_call: {
-    name: "code_interpreter",
-    input: (item) => ({ code: item.code, container_id: item.container_id }),
-  },
-  image_generation_call: { name: "image_generation", input: () => ({}) },
-  mcp_call: {
-    name: "mcp",
-    input: (item) => ({ server_label: item.server_label, name: item.name, arguments: item.arguments }),
-  },
-} as const satisfies ResponsesHostedTools.Definitions
-
-// Grok speaks the standard Responses reasoning dialect (`reasoning_summary_text.*`,
-// handled by the baseline); only its hosted tool vocabulary differs.
-const step = (state: OpenResponses.ParserState, event: OpenResponses.Event) => {
-  if (event.type === "response.output_item.done" && event.item && ResponsesHostedTools.isItem(event.item, HOSTED_TOOLS))
-    return ResponsesHostedTools.onDone(state, event.item, HOSTED_TOOLS)
-  return OpenResponses.step(state, event)
-}
-
-export const protocol = Protocol.make({
-  id: ADAPTER,
-  body: {
-    schema: XAIResponsesBody,
-    from: fromRequest,
-  },
-  stream: {
-    event: OpenResponses.protocol.stream.event,
-    initial: (request) => OpenResponses.initial(request, extension),
-    step,
-    terminal: OpenResponses.terminal,
-  },
-})
-
-export * as XAIResponses from "./xai-responses.js"
@@ -2,7 +2,13 @@ import { Effect, Schema } from "effect"
 import { Headers, HttpClientRequest } from "effect/unstable/http"
 import { GeneratedImage, ImageModel, ImageResponse, type ImageRequestFor, type ImageRoute } from "../image.js"
 import { Auth, type Definition as AuthDefinition } from "../route/auth.js"
-import { mergeHttpOptions, mergeJsonRecords, type HttpOptions } from "../schema/index.js"
+import {
+  InvalidProviderOutputReason,
+  AIError,
+  mergeHttpOptions,
+  mergeJsonRecords,
+  type HttpOptions,
+} from "../schema/index.js"
 import { ProviderShared } from "./shared.js"
 import { ImageInputs } from "./utils/image-input.js"

@@ -57,6 +63,13 @@ const nativeOptions = (options: ZAIImageOptions | undefined) => {
  }
 }

+const invalidOutput = (message: string) =>
+  new AIError({
+    module: ADAPTER,
+    method: "generate",
+    reason: new InvalidProviderOutputReason({ message, route: ADAPTER }),
+  })
+
 const applyQuery = (url: string, query: Record<string, string> | undefined) => {
  if (!query) return url
  const next = new URL(url)
@@ -69,7 +82,7 @@ export const model = (input: ModelInput) => {
    id: ADAPTER,
    generate: Effect.fn("ZAIImages.generate")(function* (request: ImageRequestFor<ZAIImageOptions>, execute) {
      if ((request.images?.length ?? 0) > 0)
-        return yield* ImageInputs.invalid("Z.ai hosted image generation does not support image inputs")
+        return yield* ImageInputs.invalid(ADAPTER, "Z.ai hosted image generation does not support image inputs")
      const http = mergeHttpOptions(request.model.http, request.http)
      const requestBody = mergeJsonRecords(
        { model: request.model.id, prompt: request.prompt },
@@ -91,11 +104,13 @@ export const model = (input: ModelInput) => {
          HttpClientRequest.bodyText(text, "application/json"),
        ),
      )
-      const output = yield* ProviderShared.imageResponse(ADAPTER, "Z.ai Images", response)
-      const decoded = yield* Schema.decodeUnknownEffect(Schema.fromJsonString(ZAIImageResponse))(output.body).pipe(
-        Effect.mapError((cause) => output.invalid("Z.ai Images returned an invalid response", cause)),
+      const payload = yield* response.json.pipe(
+        Effect.mapError(() => invalidOutput("Failed to read the Z.ai Images response")),
      )
-      if (decoded.data.length === 0) return yield* output.invalid("Z.ai Images returned no images")
+      const decoded = yield* Schema.decodeUnknownEffect(ZAIImageResponse)(payload).pipe(
+        Effect.mapError(() => invalidOutput("Z.ai Images returned an invalid response")),
+      )
+      if (decoded.data.length === 0) return yield* invalidOutput("Z.ai Images returned no images")
      return new ImageResponse({
        images: decoded.data.map(
          (item) =>
@@ -1,16 +1,17 @@
 import { Option, Schema } from "effect"
 import {
-  AuthenticationError,
-  ContentPolicyError,
-  InvalidRequestError,
+  AuthenticationReason,
+  ContentPolicyReason,
+  InvalidRequestReason,
  AIError,
  ProviderErrorEvent,
-  ProviderInternalError,
-  QuotaExceededError,
-  RateLimitError,
-  UnknownProviderError,
+  ProviderInternalReason,
+  QuotaExceededReason,
+  RateLimitReason,
+  UnknownProviderReason,
  type HttpContext,
  type HttpRateLimitDetails,
+  type ProviderMetadata,
 } from "./schema/index.js"

 const patterns = [
@@ -39,16 +40,15 @@ const patterns = [
  /model_context_window_exceeded/i,
  /too many tokens/i,
  /token limit exceeded/i,
-  /request_too_large/i,
 ]

-const payloadPatterns = [/request entity too large/i, /payload too large/i, /request too large/i]
+const payloadPatterns = [/request_too_large/i, /request entity too large/i, /payload too large/i, /request too large/i]

 const exclusions = [/^(throttling error|service unavailable):/i, /rate limit/i, /too many requests/i]

 export const isContextOverflow = (message: string) =>
  !exclusions.some((pattern) => pattern.test(message)) &&
-  (patterns.some((pattern) => pattern.test(message)) || /^4(?:00|13)\s*(status code)?\s*\(no body\)/i.test(message))
+  (patterns.some((pattern) => pattern.test(message)) || /^400\s*(status code)?\s*\(no body\)/i.test(message))

 export const isPayloadTooLarge = (message: string) => payloadPatterns.some((pattern) => pattern.test(message))

@@ -79,96 +79,86 @@ const NETWORK_ERROR_TEXT = /network[-_\s]error/i
 export interface ProviderFailure {
  readonly message: string
  readonly status?: number | undefined
-  // Raw wire payload, scanned for failure signals (codes, overflow phrases)
-  // that the summary message does not carry. Not shown to users.
-  readonly rawBody?: string | undefined
-  // Some SDKs supply parsed error data separately from the original response text.
-  readonly data?: unknown
-  readonly http?: HttpContext | undefined
-  readonly cause?: unknown
+  readonly code?: string | undefined
  readonly retryAfterMs?: number | undefined
  readonly rateLimit?: HttpRateLimitDetails | undefined
+  readonly http?: HttpContext | undefined
+  readonly providerMetadata?: ProviderMetadata | undefined
 }

 // Keep HTTP failures and provider-reported stream failures on one typed path so
 // session retry policy never needs provider-specific string matching.
 export function classifyProviderFailure(input: ProviderFailure): AIError["reason"] {
-  const details = { message: input.message, body: input.rawBody, http: input.http, cause: input.cause }
-  const body = input.rawBody ?? ""
-  const codes = [...providerCodes(input.data), ...providerCodes(body), ...providerCodes(input.message)].map((code) =>
-    code.toLowerCase(),
-  )
-  // Scan the raw payload too so signals missing from the summary message
-  // (e.g. overflow phrases nested in a JSON error body) still classify.
-  const text = [input.message, body].filter((value) => value.length > 0).join("\n")
+  const body = input.http?.body ?? ""
+  const codes = [input.code, ...providerCodes(body), ...providerCodes(input.message)]
+    .filter((code): code is string => code !== undefined)
+    .map((code) => code.toLowerCase())
+  const text = body || input.message
+  const common = { message: input.message, providerMetadata: input.providerMetadata, http: input.http }
  const clientScoped = input.status === undefined || (input.status >= 400 && input.status < 500)

  if (
    clientScoped &&
    (codes.includes("context_length_exceeded") ||
      codes.includes("model_context_window_exceeded") ||
-      codes.includes("request_too_large") ||
      isContextOverflow(text))
  )
-    return new InvalidRequestError({ ...details, classification: "context-overflow" })
+    return new InvalidRequestReason({ ...common, classification: "context-overflow" })
  if (input.status === 413 || isPayloadTooLarge(text))
-    return new InvalidRequestError({ ...details, classification: "payload-too-large" })
-  if (CONTENT_POLICY_TEXT.test(text)) return new ContentPolicyError(details)
+    return new InvalidRequestReason({ ...common, classification: "payload-too-large" })
+  if (CONTENT_POLICY_TEXT.test(text)) return new ContentPolicyReason(common)
  if (codes.some((code) => QUOTA_CODES.has(code)) || (input.status === 429 && QUOTA_TEXT.test(text)))
-    return new QuotaExceededError(details)
-  if (input.status === 401) return new AuthenticationError({ ...details, kind: "invalid" })
-  if (input.status === 403) return new AuthenticationError({ ...details, kind: "insufficient-permissions" })
-  if (codes.includes("authentication_error")) return new AuthenticationError({ ...details, kind: "invalid" })
+    return new QuotaExceededReason(common)
+  if (input.status === 401) return new AuthenticationReason({ ...common, kind: "invalid" })
+  if (input.status === 403) return new AuthenticationReason({ ...common, kind: "insufficient-permissions" })
+  if (codes.includes("authentication_error")) return new AuthenticationReason({ ...common, kind: "invalid" })
  if (codes.includes("permission_error"))
-    return new AuthenticationError({ ...details, kind: "insufficient-permissions" })
+    return new AuthenticationReason({ ...common, kind: "insufficient-permissions" })
  if (
    codes.some((code) => code.includes("rate_limit") || code === "too_many_requests" || code === "throttlingexception")
  )
-    return new RateLimitError({
-      ...details,
+    return new RateLimitReason({
+      ...common,
      retryAfterMs: input.retryAfterMs,
      rateLimit: input.rateLimit,
    })
  if (RATE_LIMIT_TEXT.test(text))
-    return new RateLimitError({
-      ...details,
+    return new RateLimitReason({
+      ...common,
      retryAfterMs: input.retryAfterMs,
      rateLimit: input.rateLimit,
    })
-  if (NETWORK_ERROR_TEXT.test(text)) return new ProviderInternalError(details)
+  if (NETWORK_ERROR_TEXT.test(text)) return new ProviderInternalReason({ ...common, status: input.status })
  if (codes.some((code) => SERVER_CODES.has(code) || code.includes("exhausted") || code.includes("unavailable")))
-    return new ProviderInternalError({
-      ...details,
+    return new ProviderInternalReason({
+      ...common,
+      status: input.status,
      retryAfterMs: input.retryAfterMs,
    })
  if (input.status === 429) {
-    return new RateLimitError({
-      ...details,
+    return new RateLimitReason({
+      ...common,
      retryAfterMs: input.retryAfterMs,
      rateLimit: input.rateLimit,
    })
  }
  if (input.status === 408 || input.status === 409 || (input.status !== undefined && input.status >= 500))
-    return new ProviderInternalError({
-      ...details,
+    return new ProviderInternalReason({
+      ...common,
+      status: input.status,
      retryAfterMs: input.retryAfterMs,
    })
-  if (codes.some((code) => INVALID_REQUEST_CODES.has(code))) return new InvalidRequestError(details)
+  if (codes.some((code) => INVALID_REQUEST_CODES.has(code))) return new InvalidRequestReason(common)
  if (input.status === 400 || input.status === 404 || input.status === 413 || input.status === 422)
-    return new InvalidRequestError(details)
-  return new UnknownProviderError(details)
+    return new InvalidRequestReason(common)
+  return new UnknownProviderReason({ ...common, status: input.status })
 }

-function providerCodes(value: unknown) {
-  const decoded = typeof value === "string" ? Option.getOrUndefined(decodeJson(value)) : value
+function providerCodes(value: string) {
+  const decoded = Option.getOrUndefined(decodeJson(value))
  if (!isRecord(decoded)) return []
  const error = isRecord(decoded.error) ? decoded.error : undefined
-  const response = isRecord(decoded.response) ? decoded.response : undefined
-  const responseError = response && isRecord(response.error) ? response.error : undefined
-  const exception = isRecord(decoded.exception) ? decoded.exception : undefined
-  return [decoded.code, error?.code, error?.type, error?.status, responseError?.code, exception?.type].filter(
-    (value): value is string => typeof value === "string",
-  )
+  return [decoded.code, error?.code, error?.type].filter((value): value is string => typeof value === "string")
 }

 function isRecord(value: unknown): value is Record<string, unknown> {
@@ -1,5 +1,5 @@
 import { Auth } from "../route/auth.js"
-import { Route, type RouteDefaultsInput } from "../route/client.js"
+import type { Route as RouteDef, RouteDefaultsInput } from "../route/client.js"
 import type { ProviderPackage } from "../provider-package.js"
 import { OpenAIChat } from "../protocols/openai-chat.js"
 import { OpenAIResponses } from "../protocols/openai-responses.js"
@@ -23,30 +23,22 @@ export interface Settings extends ProviderPackage.Settings {
  readonly baseURL?: string
  readonly credentials?: Credentials
  readonly region?: string
-  readonly topP?: number
  readonly providerOptions?: OpenAIProviderOptionsInput
 }

-const responsesRoute = Route.make({
+const responsesRoute = OpenAIResponses.route.with({
  id: "bedrock-mantle-responses",
  provider: id,
-  providerMetadataKey: "mantle",
-  protocol: OpenAIResponses.protocol,
-  endpoint: OpenAIResponses.route.endpoint,
-  auth: OpenAIResponses.route.auth,
-  transport: OpenAIResponses.httpTransport,
-  defaults: OpenAIResponses.route.defaults,
 })

 const chatRoute = OpenAIChat.route.with({
  id: "bedrock-mantle-chat",
  provider: id,
-  providerMetadataKey: "mantle",
 })

 export const routes = [responsesRoute, chatRoute]

-const configuredRoute = <Body, Prepared>(route: Route<Body, Prepared>, input: Config) => {
+const configuredRoute = <Body, Prepared>(route: RouteDef<Body, Prepared>, input: Config) => {
  const region = input.region ?? input.credentials?.region ?? "us-east-1"
  const credentials = input.credentials === undefined ? undefined : { ...input.credentials, region }
  return route.with({
@@ -78,7 +70,7 @@ export const configure = (input: Config = {}) => {

  return {
    id,
-    model: responses,
+    model: chat,
    chat,
    responses,
    configure,
@@ -96,7 +88,6 @@ const config = (settings: Settings): Config => {
    apiKey: settings.auth === "sigv4" ? undefined : settings.apiKey,
    baseURL: settings.baseURL,
    credentials: settings.credentials,
-    generation: settings.topP === undefined ? undefined : { topP: settings.topP },
    headers: settings.headers === undefined ? undefined : { ...settings.headers },
    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
    providerOptions: settings.providerOptions,
@@ -112,4 +103,4 @@ export const responsesModel: ProviderPackage.Definition<Settings, OpenAIProvider
  modelID,
  settings,
 ) => configure(config(settings)).responses(modelID)
-export const model = responsesModel
+export const model = chatModel
@@ -35,7 +35,6 @@ const configuredRoute = (input: Config) => {
  return BedrockConverse.route.with({
    ...rest,
    provider: id,
-    providerMetadataKey: "bedrock",
    endpoint: { baseURL: baseURL ?? bedrockBaseURL(resolvedRegion) },
    auth: apiKey === undefined ? BedrockConverse.sigV4Auth(credentials) : Auth.bearer(apiKey),
  })
@@ -1,2 +1,2 @@
-export { responsesModel as model } from "../amazon-bedrock-mantle.js"
+export { chatModel as model } from "../amazon-bedrock-mantle.js"
 export type { Settings } from "../amazon-bedrock-mantle.js"
@@ -1,58 +0,0 @@
-import type { ProviderPackage } from "../provider-package.js"
-import { OpenAICompatibleChat } from "../protocols/openai-compatible-chat.js"
-import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
-import type { RouteDefaultsInput } from "../route/client.js"
-import { ProviderID, type ModelID } from "../schema/index.js"
-import { profiles } from "./openai-compatible-profile.js"
-import type { OpenAIProviderOptionsInput } from "./openai-options.js"
-
-export const id = ProviderID.make("cerebras")
-
-export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
-  ProviderAuthOption<"optional"> & {
-    readonly baseURL?: string
-    readonly providerOptions?: OpenAIProviderOptionsInput
-  }
-
-export interface Settings extends ProviderPackage.Settings {
-  readonly apiKey?: string
-  readonly baseURL?: string
-  readonly providerOptions?: OpenAIProviderOptionsInput
-}
-
-export const route = OpenAICompatibleChat.route.with({
-  id: "cerebras-chat",
-  provider: id,
-  endpoint: { baseURL: profiles.cerebras.baseURL },
-})
-
-export const routes = [route]
-
-export const configure = (input: LanguageModelOptions = {}) => {
-  const { apiKey: _apiKey, auth: _auth, baseURL, ...defaults } = input
-  const configured = route.with({
-    ...defaults,
-    endpoint: { baseURL: baseURL ?? profiles.cerebras.baseURL },
-    auth: AuthOptions.bearer(input, "CEREBRAS_API_KEY"),
-  })
-  return {
-    id,
-    model: (modelID: string | ModelID) =>
-      configured.model<OpenAIProviderOptionsInput>({
-        id: modelID,
-        compatibility: { maxTokensField: "max_tokens", reasoningField: "reasoning", supportsStore: false },
-      }),
-    configure,
-  }
-}
-
-export const provider = configure()
-
-export const model: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (modelID, settings) =>
-  configure({
-    apiKey: settings.apiKey,
-    baseURL: settings.baseURL,
-    headers: settings.headers === undefined ? undefined : { ...settings.headers },
-    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    providerOptions: settings.providerOptions,
-  }).model(modelID)
@@ -1,61 +0,0 @@
-import type { ProviderPackage } from "../provider-package.js"
-import { OpenAICompatibleChat } from "../protocols/openai-compatible-chat.js"
-import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
-import type { RouteDefaultsInput } from "../route/client.js"
-import { ProviderID, type ModelID } from "../schema/index.js"
-import { profiles } from "./openai-compatible-profile.js"
-import type { OpenAIProviderOptionsInput } from "./openai-options.js"
-
-export const id = ProviderID.make("deepinfra")
-
-export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
-  ProviderAuthOption<"optional"> & {
-    readonly baseURL?: string
-    readonly providerOptions?: OpenAIProviderOptionsInput
-  }
-
-export interface Settings extends ProviderPackage.Settings {
-  readonly apiKey?: string
-  readonly baseURL?: string
-  readonly providerOptions?: OpenAIProviderOptionsInput
-}
-
-export const route = OpenAICompatibleChat.route.with({
-  id: "deepinfra-chat",
-  provider: id,
-  endpoint: { baseURL: profiles.deepinfra.baseURL },
-})
-
-export const routes = [route]
-
-export const configure = (input: LanguageModelOptions = {}) => {
-  const { apiKey: _apiKey, auth: _auth, baseURL, ...defaults } = input
-  const root = baseURL?.replace(/\/+$/, "")
-  const configured = route.with({
-    ...defaults,
-    endpoint: {
-      baseURL: root === undefined ? profiles.deepinfra.baseURL : root.endsWith("/openai") ? root : `${root}/openai`,
-    },
-    auth: AuthOptions.bearer(input, "DEEPINFRA_API_KEY"),
-  })
-  return {
-    id,
-    model: (modelID: string | ModelID) =>
-      configured.model<OpenAIProviderOptionsInput>({
-        id: modelID,
-        compatibility: { maxTokensField: "max_tokens", reasoningField: "reasoning_content", supportsStore: false },
-      }),
-    configure,
-  }
-}
-
-export const provider = configure()
-
-export const model: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (modelID, settings) =>
-  configure({
-    apiKey: settings.apiKey,
-    baseURL: settings.baseURL,
-    headers: settings.headers === undefined ? undefined : { ...settings.headers },
-    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    providerOptions: settings.providerOptions,
-  }).model(modelID)
@@ -27,7 +27,6 @@ export interface Settings extends ProviderPackage.Settings {
 const route = OpenAICompatibleChat.route.with({
  id: "google-vertex-chat",
  provider: id,
-  providerMetadataKey: "vertex",
 })

 export const routes = [route]
@@ -27,7 +27,6 @@ export interface Settings extends ProviderPackage.Settings {
 const route = OpenAICompatibleResponses.route.with({
  id: "google-vertex-responses",
  provider: id,
-  providerMetadataKey: "vertex",
  providerOptions: { store: false },
 })

@@ -38,23 +38,13 @@ export type Settings = ProviderPackage.Settings &

 const fromRequest = Effect.fn("GoogleVertex.fromRequest")(function* (request: LLMRequest) {
  const body = yield* Gemini.protocol.body.from(request)
-  // Vertex's native REST schema rejects `id` on FunctionCall/FunctionResponse parts with HTTP 400,
-  // unlike AI Studio, so history minted there cannot be lowered verbatim.
-  const contents = body.contents.map((content) => ({
-    ...content,
-    parts: (content.parts ?? []).map((part) => {
-      if ("functionCall" in part) return { ...part, functionCall: { ...part.functionCall, id: undefined } }
-      if ("functionResponse" in part) return { ...part, functionResponse: { ...part.functionResponse, id: undefined } }
-      return part
-    }),
-  }))
  const value = request.providerOptions?.labels
  const labels = ProviderShared.isRecord(value)
    ? Object.fromEntries(
        Object.entries(value).filter((entry): entry is [string, string] => typeof entry[1] === "string"),
      )
    : undefined
-  return { ...body, contents, labels }
+  return { ...body, labels }
 })

 const protocol = {
@@ -68,7 +58,7 @@ const protocol = {
 const route = Route.make({
  id: "google-vertex-gemini",
  provider: id,
-  providerMetadataKey: "vertex",
+  providerMetadataKey: "google",
  protocol,
  endpoint: Endpoint.path(({ request }) => {
    const model = String(request.model.id)
@@ -1,116 +0,0 @@
-import { Effect, Schema } from "effect"
-import type { ProviderPackage } from "../provider-package.js"
-import { OpenAIChat } from "../protocols/openai-chat.js"
-import { ProviderShared } from "../protocols/shared.js"
-import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
-import { Route, type RouteDefaultsInput } from "../route/client.js"
-import { Endpoint } from "../route/endpoint.js"
-import { Framing } from "../route/framing.js"
-import { Protocol } from "../route/protocol.js"
-import { ProviderID, type ModelID, type LLMRequest } from "../schema/index.js"
-import { profiles } from "./openai-compatible-profile.js"
-import type { OpenAIProviderOptionsInput } from "./openai-options.js"
-
-export const id = ProviderID.make("groq")
-
-export type ProviderOptions = Pick<OpenAIProviderOptionsInput, "reasoningEffort"> & {
-  /** Controls visible reasoning on GPT-OSS; other models always use parsed reasoning. */
-  readonly includeReasoning?: boolean
-  readonly parallelToolCalls?: boolean
-  readonly serviceTier?: "on_demand" | "flex" | "auto" | "performance" | (string & {})
-  readonly user?: string
-}
-
-export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
-  ProviderAuthOption<"optional"> & {
-    readonly baseURL?: string
-    readonly providerOptions?: ProviderOptions
-  }
-
-export interface Settings extends ProviderPackage.Settings {
-  readonly apiKey?: string
-  readonly baseURL?: string
-  readonly providerOptions?: ProviderOptions
-}
-
-const Options = Schema.Struct({
-  includeReasoning: Schema.optional(Schema.Boolean),
-  parallelToolCalls: Schema.optional(Schema.Boolean),
-  serviceTier: Schema.optional(Schema.String),
-  user: Schema.optional(Schema.String),
-})
-
-export const protocol = Protocol.make({
-  id: "groq-chat",
-  body: {
-    schema: Schema.Struct({
-      ...OpenAIChat.bodyFields,
-      reasoning_format: Schema.optional(Schema.Literal("parsed")),
-      include_reasoning: Schema.optional(Schema.Boolean),
-      parallel_tool_calls: Schema.optional(Schema.Boolean),
-      service_tier: Schema.optional(Schema.String),
-      user: Schema.optional(Schema.String),
-    }),
-    from: Effect.fn("Groq.fromRequest")(function* (request: LLMRequest) {
-      const options = yield* ProviderShared.validateWith(Schema.decodeUnknownEffect(Options))(
-        request.providerOptions ?? {},
-      )
-      const gptOSS = request.model.id.startsWith("openai/gpt-oss-")
-      return {
-        ...(yield* OpenAIChat.fromRequest(request)),
-        reasoning_format: gptOSS ? undefined : ("parsed" as const),
-        include_reasoning: gptOSS ? options.includeReasoning : undefined,
-        parallel_tool_calls: options.parallelToolCalls,
-        service_tier: options.serviceTier,
-        user: options.user,
-      }
-    }),
-  },
-  stream: OpenAIChat.protocol.stream,
-})
-
-export const route = Route.make({
-  id: "groq-chat",
-  provider: id,
-  providerMetadataKey: "openai",
-  protocol,
-  endpoint: Endpoint.path("/chat/completions", { baseURL: profiles.groq.baseURL }),
-  framing: Framing.sse,
-})
-
-export const configure = (input: LanguageModelOptions = {}) => {
-  const { apiKey: _apiKey, auth: _auth, baseURL, ...defaults } = input
-  const configured = route.with({
-    ...defaults,
-    endpoint: { baseURL: baseURL ?? profiles.groq.baseURL },
-    auth: AuthOptions.bearer(input, "GROQ_API_KEY"),
-  })
-  return {
-    id,
-    model: (modelID: string | ModelID) =>
-      configured.model<ProviderOptions>({
-        id: modelID,
-        compatibility: {
-          maxTokensField: "max_completion_tokens",
-          reasoningField: "reasoning",
-          requireReasoning: false,
-          supportsStore: false,
-          supportsStrictMode: false,
-        },
-      }),
-    configure,
-  }
-}
-
-export const provider = configure()
-
-export const model: ProviderPackage.Definition<Settings, ProviderOptions>["model"] = (modelID, settings) =>
-  configure({
-    apiKey: settings.apiKey,
-    baseURL: settings.baseURL,
-    headers: settings.headers === undefined ? undefined : { ...settings.headers },
-    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    providerOptions: settings.providerOptions,
-  }).model(modelID)
-
-export * as Groq from "./groq.js"
@@ -3,20 +3,16 @@ export * as AnthropicCompatible from "./anthropic-compatible.js"
 export * as AmazonBedrock from "./amazon-bedrock.js"
 export * as AmazonBedrockMantle from "./amazon-bedrock-mantle.js"
 export * as Azure from "./azure.js"
-export * as Cerebras from "./cerebras.js"
 export * as Cloudflare from "./cloudflare.js"
 export { CloudflareAIGateway, CloudflareWorkersAI } from "./cloudflare.js"
-export * as DeepInfra from "./deepinfra.js"
 export * as Google from "./google.js"
 export * as GoogleVertex from "./google-vertex.js"
 export * as GoogleVertexChat from "./google-vertex-chat.js"
 export * as GoogleVertexMessages from "./google-vertex-messages.js"
 export * as GoogleVertexResponses from "./google-vertex-responses.js"
-export * as Groq from "./groq.js"
 export * as OpenAI from "./openai.js"
 export * as OpenAICompatible from "./openai-compatible.js"
 export * as OpenAICompatibleResponses from "./openai-compatible-responses.js"
 export * as OpenRouter from "./openrouter.js"
-export * as TogetherAI from "./togetherai.js"
 export * as XAI from "./xai.js"
 export * as ZAI from "./zai.js"
@@ -1,13 +1,9 @@
 import { mergeProviderOptions, type ProviderOptions } from "../schema/index.js"
-import type { OpenAIServiceTier } from "../protocols/utils/openai-options.js"
-import type { Options } from "../protocols/utils/open-responses-options.js"
+import type { OpenResponsesOptionsInput } from "./open-responses-options.js"

 export type { OpenAIResponseIncludable, OpenAIServiceTier } from "../protocols/utils/openai-options.js"

-export type OpenAIOptionsInput = Omit<Options, "serviceTier"> & {
-  readonly serviceTier?: OpenAIServiceTier
-  readonly [key: string]: unknown
-}
+export type OpenAIOptionsInput = OpenResponsesOptionsInput

 export type OpenAIProviderOptionsInput = OpenAIOptionsInput

@@ -119,6 +119,7 @@ export const protocol = Protocol.make({
            ...body,
            messages,
            ...bodyOptions(request.providerOptions),
+            ...(request.promptCacheKey ? { prompt_cache_key: request.promptCacheKey } : {}),
          } as OpenRouterBody
        }),
      ),
@@ -164,7 +165,6 @@ const bodyOptions = (input: unknown) => {
 export const route = Route.make({
  id: ADAPTER,
  provider: profile.provider,
-  providerMetadataKey: "openrouter",
  protocol,
  endpoint: Endpoint.path("/chat/completions", { baseURL: profile.baseURL }),
  framing: Framing.sse,
@@ -1,58 +0,0 @@
-import type { ProviderPackage } from "../provider-package.js"
-import { OpenAICompatibleChat } from "../protocols/openai-compatible-chat.js"
-import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
-import type { RouteDefaultsInput } from "../route/client.js"
-import { ProviderID, type ModelID } from "../schema/index.js"
-import { profiles } from "./openai-compatible-profile.js"
-import type { OpenAIProviderOptionsInput } from "./openai-options.js"
-
-export const id = ProviderID.make("togetherai")
-
-export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
-  ProviderAuthOption<"optional"> & {
-    readonly baseURL?: string
-    readonly providerOptions?: OpenAIProviderOptionsInput
-  }
-
-export interface Settings extends ProviderPackage.Settings {
-  readonly apiKey?: string
-  readonly baseURL?: string
-  readonly providerOptions?: OpenAIProviderOptionsInput
-}
-
-export const route = OpenAICompatibleChat.route.with({
-  id: "togetherai-chat",
-  provider: id,
-  endpoint: { baseURL: profiles.togetherai.baseURL },
-})
-
-export const routes = [route]
-
-export const configure = (input: LanguageModelOptions = {}) => {
-  const { apiKey: _apiKey, auth: _auth, baseURL, ...defaults } = input
-  const configured = route.with({
-    ...defaults,
-    endpoint: { baseURL: baseURL ?? profiles.togetherai.baseURL },
-    auth: AuthOptions.bearer(input, ["TOGETHER_API_KEY", "TOGETHER_AI_API_KEY"]),
-  })
-  return {
-    id,
-    model: (modelID: string | ModelID) =>
-      configured.model<OpenAIProviderOptionsInput>({
-        id: modelID,
-        compatibility: { maxTokensField: "max_tokens", supportsStore: false, supportsStrictMode: false },
-      }),
-    configure,
-  }
-}
-
-export const provider = configure()
-
-export const model: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (modelID, settings) =>
-  configure({
-    apiKey: settings.apiKey,
-    baseURL: settings.baseURL,
-    headers: settings.headers === undefined ? undefined : { ...settings.headers },
-    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    providerOptions: settings.providerOptions,
-  }).model(modelID)
@@ -5,8 +5,7 @@ import { HttpOptions, ProviderID, type ModelID } from "../schema/index.js"
 import * as OpenAICompatibleProfiles from "./openai-compatible-profile.js"
 import * as OpenAICompatibleChat from "../protocols/openai-compatible-chat.js"
 import * as OpenAIChat from "../protocols/openai-chat.js"
-import { OpenResponsesChannel } from "../protocols/open-responses-channel.js"
-import { XAIResponses } from "../protocols/xai-responses.js"
+import * as OpenAIResponses from "../protocols/openai-responses.js"
 import { XAIImages } from "../protocols/xai-images.js"
 import type { OpenAIOptionsInput } from "./openai-options.js"
 import type { ProviderPackage } from "../provider-package.js"
@@ -35,14 +34,14 @@ const responsesRoute = Route.make({
  id: "openai-responses",
  provider: id,
  providerMetadataKey: "xai",
-  protocol: XAIResponses.protocol,
+  protocol: OpenAIResponses.protocol,
  endpoint: Endpoint.path("/responses", { baseURL: OpenAICompatibleProfiles.profiles.xai.baseURL }),
-  transport: OpenResponsesChannel.transport({
+  transport: OpenAIResponses.channelTransport({
    id: "openai-responses",
    name: "xAI Responses",
    rotateAfterMs: RESPONSES_WEBSOCKET_ROTATE_AFTER_MS,
  }),
-  defaults: { providerOptions: { store: false, include: ["reasoning.encrypted_content"] } },
+  defaults: { providerOptions: { store: false } },
 })

 const chatRoute = Route.make({
@@ -1,6 +1,6 @@
 import { Config, Effect, Redacted } from "effect"
 import { Headers } from "effect/unstable/http"
-import { AuthenticationError, InvalidRequestError, AIError, type HttpOptions } from "../schema/index.js"
+import { AuthenticationReason, InvalidRequestReason, AIError, type HttpOptions } from "../schema/index.js"

 export class MissingCredentialError extends Error {
  readonly _tag = "MissingCredentialError"
@@ -137,10 +137,12 @@ export function bearerHeader(name: string, source?: Secret | Credential) {
 const toAIError = (error: AuthError): AIError => {
  if (error instanceof MissingCredentialError || error instanceof Config.ConfigError) {
    return new AIError({
+      module: "Auth",
+      method: "apply",
      reason:
        error instanceof MissingCredentialError
-          ? new AuthenticationError({ message: error.message, cause: error, kind: "missing" })
-          : new InvalidRequestError({ message: `Failed to resolve auth config: ${error.message}`, cause: error }),
+          ? new AuthenticationReason({ message: error.message, kind: "missing" })
+          : new InvalidRequestReason({ message: `Failed to resolve auth config: ${error.message}` }),
    })
  }
  return error
@@ -7,19 +7,17 @@ import { HttpTransport } from "./transport/index.js"
 import type { HttpMiddleware, Transport, TransportRuntime, WebSocketChannelExecutor } from "./transport/index.js"
 import type { Protocol } from "./protocol.js"
 import { applyCachePolicy } from "../cache-policy.js"
-import { sanitizeSurrogates } from "../utils/sanitize.js"
 import * as ProviderShared from "../protocols/shared.js"
 import type { ProtocolID, ProviderOptions } from "../schema/index.js"
 import {
  AIError,
-  AIErrorReason,
  GenerationOptions,
  HttpOptions,
  LLMRequest,
  LLMResponse,
  LanguageModel,
  LLMEvent,
-  InvalidProviderOutputError,
+  InvalidProviderOutputReason,
  ProviderID,
  mergeGenerationOptions,
  mergeHttpOptions,
@@ -90,7 +88,6 @@ export interface RouteDefaultsInput {
 export interface RoutePatch<Body, Prepared> extends RouteDefaultsInput {
  readonly id?: string
  readonly provider?: string | ProviderID
-  readonly providerMetadataKey?: string
  readonly auth?: Auth.Definition
  readonly transport?: Transport<Body, Prepared, unknown>
  readonly endpoint?: EndpointPatch<Body>
@@ -228,14 +225,16 @@ export interface MakeTransportInput<Body, Prepared, Frame, Event, State> {
 const streamError = (route: string, message: string, cause: Cause.Cause<unknown>) => {
  const failed = cause.reasons.find(Cause.isFailReason)?.error
  if (failed instanceof AIError) return failed
-  return ProviderShared.eventError(route, message, undefined, cause)
+  return ProviderShared.eventError(route, message, Cause.pretty(cause))
 }

 const incompleteStreamError = (route: string) =>
  new AIError({
-    reason: new InvalidProviderOutputError({
-      message: "The provider response ended unexpectedly.",
+    module: "LLMClient",
+    method: "stream",
+    reason: new InvalidProviderOutputReason({
      classification: "incomplete-stream",
+      message: "The provider response ended unexpectedly.",
      route,
    }),
  })
@@ -264,12 +263,11 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
  const decodeEventEffect = Schema.decodeUnknownEffect(protocol.stream.event)
  const decodeEvent = (route: string) => (frame: Frame) =>
    decodeEventEffect(frame).pipe(
-      Effect.mapError((cause) =>
+      Effect.mapError(() =>
        ProviderShared.eventError(
          input.id,
          `Invalid ${route} stream event`,
          typeof frame === "string" ? frame : ProviderShared.encodeJson(frame),
-          cause,
        ),
      ),
    )
@@ -290,16 +288,11 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
      defaults: routeInput.defaults ?? {},
      body: protocol.body,
      with: (patch: RoutePatch<Body, Prepared>) => {
-        const { id, provider, providerMetadataKey, auth, transport, endpoint, ...defaults } = patch
+        const { id, provider, auth, transport, endpoint, ...defaults } = patch
        return build({
          ...routeInput,
          id: id ?? routeInput.id,
          provider: provider ?? routeInput.provider,
-          providerMetadataKey:
-            providerMetadataKey ??
-            (provider !== undefined && String(provider) !== String(routeInput.provider)
-              ? String(provider)
-              : routeInput.providerMetadataKey),
          auth: auth ?? routeInput.auth,
          endpoint: endpoint ? Endpoint.merge(routeInput.endpoint, endpoint) : routeInput.endpoint,
          transport: (transport as Transport<Body, Prepared, Frame> | undefined) ?? routeInput.transport,
@@ -324,74 +317,18 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
        return Stream.unwrap(
          routeInput.transport.execute(prepared, request, runtime, options).pipe(
            Effect.map((execution) => {
-              const terminal = protocol.stream.terminal
-              // Preserve assembled inputs; replace only serialized event fallbacks with their original wire data.
-              const frameError =
-                (frame: Frame, event: Frame | Event = frame) =>
-                (error: AIError) =>
-                  new AIError({
-                    reason: AIErrorReason.make({
-                      ...error.reason,
-                      message: error.reason.message,
-                      cause: error.reason.cause,
-                      body:
-                        error.reason.body !== undefined && error.reason.body !== ProviderShared.encodeJson(event)
-                          ? error.reason.body
-                          : (execution.body?.(frame) ??
-                            (typeof frame === "string" ? frame : ProviderShared.encodeJson(frame))),
-                    }),
-                  })
              const events = execution.frames.pipe(
-                Stream.mapEffect((frame) =>
-                  decodeEvent(route)(frame).pipe(
-                    Effect.catchCause((cause) =>
-                      Effect.fail(streamError(route, `Failed to decode ${route} event`, cause)),
-                    ),
-                    Effect.map((event) => ({ event, frame })),
-                    Effect.mapError(frameError(frame)),
-                  ),
-                ),
-                terminal ? Stream.takeUntil(({ event }) => terminal(event)) : (stream) => stream,
+                Stream.mapEffect(decodeEvent(route)),
+                protocol.stream.terminal ? Stream.takeUntil(protocol.stream.terminal) : (stream) => stream,
              )
-              const stream = Stream.suspend(() => {
-                let state = protocol.stream.initial(request)
-                const parsed = events.pipe(
-                  Stream.mapEffect(({ event, frame }) =>
-                    protocol.stream.step(state, event).pipe(
-                      Effect.catchCause((cause) =>
-                        Effect.fail(streamError(route, `Failed to parse ${route} event`, cause)),
-                      ),
-                      Effect.map(([next, output]) => {
-                        state = next
-                        return output
-                      }),
-                      Effect.mapError(frameError(frame, event)),
-                    ),
-                  ),
-                  Stream.flatMap(Stream.fromIterable),
-                )
-                const onHalt = protocol.stream.onHalt
-                return onHalt
-                  ? parsed.pipe(
-                      Stream.concat(
-                        Stream.suspend(() => Stream.unwrap(onHalt(state).pipe(Effect.map(Stream.fromIterable)))),
-                      ),
-                    )
-                  : parsed
-              }).pipe(
+              const stream = events.pipe(
+                Stream.mapAccumEffect(
+                  () => protocol.stream.initial(request),
+                  protocol.stream.step,
+                  protocol.stream.onHalt ? { onHalt: protocol.stream.onHalt } : undefined,
+                ),
                Stream.catchCause((cause) => Stream.fail(streamError(route, `Failed to read ${route} stream`, cause))),
                requireTerminalEvent(route),
-                Stream.mapError(
-                  (error) =>
-                    new AIError({
-                      reason: AIErrorReason.make({
-                        ...error.reason,
-                        message: error.reason.message,
-                        cause: error.reason.cause,
-                        http: error.reason.http ?? execution.http,
-                      }),
-                    }),
-                ),
              )
              return execution.complete ? stream.pipe(Stream.onEnd(execution.complete)) : stream
            }),
@@ -445,8 +382,7 @@ export function make<Body, Prepared, Frame, Event, State>(
 }

 const compile = Effect.fn("LLM.compile")(function* (request: LLMRequest, options?: StreamOptions) {
-  const original = applyCachePolicy(resolveRequestOptions(request))
-  const resolved = LLMRequest.update(original, sanitizeSurrogates({ ...LLMRequest.input(original), model: undefined }))
+  const resolved = applyCachePolicy(resolveRequestOptions(request))
  const route = resolved.model.route

  const body = yield* route.body
@@ -7,7 +7,14 @@ import {
  HttpClientRequest,
  HttpClientResponse,
 } from "effect/unstable/http"
-import { HttpContext, HttpRateLimitDetails, AIError, TransportError } from "../schema/index.js"
+import {
+  HttpContext,
+  HttpRateLimitDetails,
+  HttpRequestDetails,
+  HttpResponseDetails,
+  AIError,
+  TransportReason,
+} from "../schema/index.js"
 import { classifyProviderFailure } from "../provider-error.js"

 export interface Interface {
@@ -33,6 +40,17 @@ const headerDetails = (headers: Headers.Headers) =>
 const normalizedHeaders = (headers: Headers.Headers) =>
  Object.fromEntries(Object.entries(headers).map(([key, value]) => [key.toLowerCase(), value]))

+const requestId = (headers: Record<string, string>) => {
+  return (
+    headers["x-request-id"] ??
+    headers["request-id"] ??
+    headers["x-amzn-requestid"] ??
+    headers["x-amz-request-id"] ??
+    headers["x-goog-request-id"] ??
+    headers["cf-ray"]
+  )
+}
+
 const retryAfterMs = (headers: Record<string, string>) => {
  const millis = Number(headers["retry-after-ms"])
  if (Number.isFinite(millis)) return Math.max(0, millis)
@@ -90,13 +108,24 @@ const rateLimitDetails = (headers: Record<string, string>, retryAfter: number |
  })
 }

-export const responseHttp = (response: HttpClientResponse.HttpClientResponse) =>
-  new HttpContext({
-    url: response.request.url,
+const requestDetails = (request: HttpClientRequest.HttpClientRequest) =>
+  new HttpRequestDetails({
+    method: request.method,
+    url: request.url,
+    headers: headerDetails(request.headers),
+  })
+
+const responseDetails = (response: HttpClientResponse.HttpClientResponse) =>
+  new HttpResponseDetails({
    status: response.status,
    headers: headerDetails(response.headers),
  })

+const responseBody = (body: string | void) => {
+  if (body === undefined) return {}
+  return { body }
+}
+
 const decodeProviderBody = Schema.decodeUnknownOption(
  Schema.fromJsonString(
    Schema.Struct({
@@ -114,46 +143,81 @@ const providerMessage = (status: number, body: string | void) => {
  )
 }

-const statusError = (response: HttpClientResponse.HttpClientResponse) =>
-  Effect.gen(function* () {
-    if (response.status < 400) return response
-    const result = yield* response.text.pipe(Effect.result)
-    return yield* httpFailure({
-      message: providerMessage(response.status, result._tag === "Success" ? result.success : undefined),
-      url: response.request.url,
-      status: response.status,
-      responseHeaders: headerDetails(response.headers),
-      responseBody: result._tag === "Success" ? result.success : undefined,
-      cause: result._tag === "Failure" ? (result.failure.cause ?? result.failure) : undefined,
-    })
+const responseHttp = (input: {
+  readonly request: HttpClientRequest.HttpClientRequest
+  readonly response: HttpClientResponse.HttpClientResponse
+  readonly body: ReturnType<typeof responseBody>
+  readonly requestId?: string | undefined
+  readonly rateLimit?: HttpRateLimitDetails | undefined
+}) =>
+  new HttpContext({
+    request: requestDetails(input.request),
+    response: responseDetails(input.response),
+    ...input.body,
+    requestId: input.requestId,
+    rateLimit: input.rateLimit,
  })

-/** Preserve HTTP diagnostics for executor and externally captured failures alike. */
-export const httpFailure = (input: {
+const statusError =
+  (request: HttpClientRequest.HttpClientRequest) => (response: HttpClientResponse.HttpClientResponse) =>
+    Effect.gen(function* () {
+      if (response.status < 400) return response
+      const body = yield* response.text.pipe(Effect.catch(() => Effect.void))
+      const headers = normalizedHeaders(response.headers)
+      const retryAfter = retryAfterMs(headers)
+      const rateLimit = rateLimitDetails(headers, retryAfter)
+      const details = responseBody(body)
+      return yield* new AIError({
+        module: "RequestExecutor",
+        method: "execute",
+        reason: classifyProviderFailure({
+          status: response.status,
+          message: providerMessage(response.status, body),
+          retryAfterMs: retryAfter,
+          rateLimit,
+          http: responseHttp({
+            request,
+            response,
+            body: details,
+            requestId: requestId(headers),
+            rateLimit,
+          }),
+        }),
+      })
+    })
+
+// Classifies an HTTP failure captured outside the executor (for example by the
+// AI SDK's own fetch) onto the same reason types and HttpContext that
+// executor-driven requests produce. The originating request is not available on
+// that path, so the method is assumed (language model calls are always POST),
+// request headers are empty.
+export const classifyHttpFailure = (input: {
  readonly message: string
-  readonly url?: string | undefined
+  readonly url: string
  readonly status?: number | undefined
-  readonly data?: unknown
+  readonly code?: string | undefined
  readonly responseHeaders?: Record<string, string> | undefined
  readonly responseBody?: string | undefined
-  readonly cause?: unknown
 }) => {
  const headers = normalizedHeaders(Headers.fromInput(input.responseHeaders))
  const retryAfter = retryAfterMs(headers)
  const rateLimit = rateLimitDetails(headers, retryAfter)
-  return new AIError({
-    reason: classifyProviderFailure({
-      message: input.message,
-      status: input.status,
-      data: input.data,
-      rawBody: input.responseBody,
-      retryAfterMs: retryAfter,
-      rateLimit,
-      cause: input.cause,
-      http:
-        input.status === undefined || input.url === undefined
+  const details = responseBody(input.responseBody)
+  return classifyProviderFailure({
+    message: input.message,
+    status: input.status,
+    code: input.code,
+    retryAfterMs: retryAfter,
+    rateLimit,
+    http: new HttpContext({
+      request: new HttpRequestDetails({ method: "POST", url: input.url, headers: {} }),
+      response:
+        input.status === undefined
          ? undefined
-          : new HttpContext({ url: input.url, status: input.status, headers }),
+          : new HttpResponseDetails({ status: input.status, headers: headerDetails(Headers.fromInput(headers)) }),
+      ...details,
+      requestId: requestId(headers),
+      rateLimit,
    }),
  })
 }
@@ -180,25 +244,25 @@ const httpError = (input: {
  readonly error: unknown
  readonly request: HttpClientRequest.HttpClientRequest
  readonly operation: HttpOperation
-  readonly http?: HttpContext
 }) => {
  const request = HttpClientError.isHttpClientError(input.error) ? input.error.request : input.request
  const transportError = (failure: { readonly message: string; readonly code?: string | undefined }) =>
    new AIError({
-      reason: new TransportError({
+      module: "RequestExecutor",
+      method: input.operation,
+      reason: new TransportReason({
        message: failure.message,
-        cause: source,
-        http: input.http,
        transport: "http",
        operation: input.operation,
        code: failure.code,
        url: request.url,
+        http: new HttpContext({ request: requestDetails(request) }),
      }),
    })

  const source =
    HttpClientError.isHttpClientError(input.error) && "cause" in input.error.reason
-      ? (input.error.reason.cause ?? input.error)
+      ? input.error.reason.cause
      : input.error
  const native = nativeTransportFailure(source)
  const code = native?.code
@@ -222,13 +286,6 @@ const httpError = (input: {
  })
 }

-export const responseStream = (response: HttpClientResponse.HttpClientResponse): Stream.Stream<Uint8Array, AIError> =>
-  response.stream.pipe(
-    Stream.mapError((error) =>
-      httpError({ error, request: response.request, operation: "read", http: responseHttp(response) }),
-    ),
-  )
-
 export const stream = (
  executor: Interface,
  request: HttpClientRequest.HttpClientRequest,
@@ -237,7 +294,9 @@ export const stream = (
  Stream.unwrap(
    Effect.gen(function* () {
      const response = yield* executor.execute(request, middleware)
-      return responseStream(response)
+      return response.stream.pipe(
+        Stream.mapError((error) => httpError({ error, request: response.request, operation: "read" })),
+      )
    }),
  )

@@ -250,7 +309,7 @@ export const layer: Layer.Layer<Service, never, HttpClient.HttpClient> = Layer.e
        if (!middleware)
          return yield* http.execute(request).pipe(
            Effect.mapError((error) => httpError({ error, request, operation: "request" })),
-            Effect.flatMap(statusError),
+            Effect.flatMap(statusError(request)),
          )

        const response = yield* middleware(request, (input) =>
@@ -258,7 +317,7 @@ export const layer: Layer.Layer<Service, never, HttpClient.HttpClient> = Layer.e
            .execute(input)
            .pipe(Effect.mapError((cause) => (cause instanceof Error ? cause : new Error(String(cause))))),
        ).pipe(Effect.mapError((error) => httpError({ error, request, operation: "request" })))
-        return yield* statusError(response)
+        return yield* statusError(response.request)(response)
      })
    return Service.of({
      execute: executeOnce,
@@ -19,8 +19,6 @@ import type { AIError } from "../schema/index.js"
 export interface Definition<Frame> {
  readonly id: string
  readonly frame: (bytes: Stream.Stream<Uint8Array, AIError>) => Stream.Stream<Frame, AIError>
-  /** Original wire representation when framing transforms the provider payload. */
-  readonly body?: (frame: Frame) => string | undefined
 }

 /** Server-Sent Events framing. Used by every JSON-streaming HTTP provider. */
@@ -59,8 +59,8 @@ export interface ProtocolStream<Frame, Event, State> {
  readonly step: (state: State, event: Event) => Effect.Effect<readonly [State, ReadonlyArray<LLMEvent>], AIError>
  /** Optional request-completion signal for transports that do not end naturally. */
  readonly terminal?: (event: Event) => boolean
-  /** Optional effectful flush emitted when the framed stream ends. */
-  readonly onHalt?: (state: State) => Effect.Effect<ReadonlyArray<LLMEvent>, AIError>
+  /** Optional flush emitted when the framed stream ends. */
+  readonly onHalt?: (state: State) => ReadonlyArray<LLMEvent>
 }

 /**
@@ -88,13 +88,8 @@ export const httpJson = <Body, Frame>(input: HttpJsonInput<Body, Frame>): HttpJs
      }
    }),
  execute: (prepared, _request, runtime) =>
-    Effect.gen(function* () {
-      const response = yield* runtime.http.execute(prepared.request, prepared.middleware)
-      return {
-        frames: prepared.framing.frame(RequestExecutor.responseStream(response)),
-        http: RequestExecutor.responseHttp(response),
-        body: prepared.framing.body,
-      }
+    Effect.succeed({
+      frames: prepared.framing.frame(RequestExecutor.stream(runtime.http, prepared.request, prepared.middleware)),
    }),
 })

@@ -3,7 +3,7 @@ import { Endpoint } from "../endpoint.js"
 import { Auth } from "../auth.js"
 import type { HttpMiddleware, Interface as RequestExecutorInterface } from "../executor.js"
 import type { WebSocketChannelExecutor } from "./websocket-channel.js"
-import type { AIError, HttpContext, LLMRequest } from "../../schema/index.js"
+import type { AIError, LLMRequest } from "../../schema/index.js"

 export interface TransportRuntime {
  readonly http: RequestExecutorInterface
@@ -11,8 +11,6 @@ export interface TransportRuntime {

 export interface TransportExecution<Frame> {
  readonly frames: Stream.Stream<Frame, AIError>
-  readonly http?: HttpContext
-  body?(frame: Frame): string | undefined
  /** Optional successful-consumption acknowledgement. HTTP leaves this absent. */
  readonly complete?: Effect.Effect<void>
 }
@@ -1,6 +1,6 @@
 import type { Effect, Scope, Stream } from "effect"
 import type { Headers } from "effect/unstable/http"
-import type { AIError, HttpContext } from "../../schema/index.js"
+import type { AIError } from "../../schema/index.js"

 export interface WebSocketChannelExecutor {
  readonly execute: (
@@ -10,7 +10,6 @@ export interface WebSocketChannelExecutor {

 export interface WebSocketChannelExecution {
  readonly frames: Stream.Stream<string, AIError>
-  readonly http?: HttpContext
  /** Commits staged state after the decoded Route stream ends successfully. */
  readonly complete: Effect.Effect<void>
 }
@@ -1,13 +1,7 @@
 import { Cause, Effect, Queue, Stream } from "effect"
 import { Headers } from "effect/unstable/http"
 import { Socket } from "effect/unstable/socket"
-import {
-  AIError,
-  AIErrorReason,
-  TransportError,
-  type HttpContext,
-  type TransportOperation,
-} from "../../schema/index.js"
+import { AIError, TransportReason, type TransportOperation } from "../../schema/index.js"
 import * as HttpTransport from "./http.js"
 import type { Transport } from "./index.js"
 import type {
@@ -23,7 +17,6 @@ export interface WebSocketRequest {
 }

 export interface WebSocketConnection {
-  readonly http?: HttpContext
  readonly sendText: (message: string) => Effect.Effect<void, AIError>
  readonly messages: Stream.Stream<string | Uint8Array, AIError>
  readonly close: Effect.Effect<void, never>
@@ -40,22 +33,21 @@ type WebSocketConstructorWithHeaders = (

 const MAX_FRAME_BYTES = 16 * 1024 * 1024
 const transportError = (
+  method: string,
  message: string,
  input: {
    readonly operation: TransportOperation
    readonly url?: string
    readonly code?: string
-    readonly phase?: TransportError["phase"]
-    readonly delivery?: TransportError["delivery"]
-    readonly body?: string
-    readonly cause?: unknown
+    readonly phase?: TransportReason["phase"]
+    readonly delivery?: TransportReason["delivery"]
  },
 ) =>
  new AIError({
-    reason: new TransportError({
+    module: "WebSocketConnector",
+    method,
+    reason: new TransportReason({
      message,
-      body: input.body,
-      cause: input.cause,
      transport: "websocket",
      operation: input.operation,
      url: input.url,
@@ -67,16 +59,22 @@ const transportError = (

 const annotateTransportError = (
  error: AIError,
-  input: { readonly phase: TransportError["phase"]; readonly delivery: TransportError["delivery"] },
+  input: { readonly phase: TransportReason["phase"]; readonly delivery: TransportReason["delivery"] },
 ) =>
  error.reason._tag === "Transport"
    ? new AIError({
-        reason: new TransportError({
-          ...error.reason,
+        module: error.module,
+        method: error.method,
+        reason: new TransportReason({
          message: error.reason.message,
-          cause: error.reason.cause,
+          transport: error.reason.transport,
+          operation: error.reason.operation,
+          code: error.reason.code,
+          url: error.reason.url,
+          http: error.reason.http,
          phase: input.phase,
          delivery: input.delivery,
+          recovery: error.reason.recovery,
        }),
      })
    : error
@@ -97,7 +95,7 @@ const waitOpen = (ws: globalThis.WebSocket, input: WebSocketRequest) => {
  if (ws.readyState === globalThis.WebSocket.OPEN) return Effect.void
  if (ws.readyState === globalThis.WebSocket.CLOSING || ws.readyState === globalThis.WebSocket.CLOSED) {
    return Effect.fail(
-      transportError(`WebSocket closed before opening (state ${ws.readyState})`, {
+      transportError("open", `WebSocket closed before opening (state ${ws.readyState})`, {
        url: input.url,
        operation: "request",
        code: "closed",
@@ -126,8 +124,7 @@ const waitOpen = (ws: globalThis.WebSocket, input: WebSocketRequest) => {
      cleanup()
      resume(
        Effect.fail(
-          transportError(`Failed to open WebSocket: ${eventMessage(event)}`, {
-            cause: "error" in event ? (event.error ?? event) : event,
+          transportError("open", `Failed to open WebSocket: ${eventMessage(event)}`, {
            url: input.url,
            operation: "request",
            phase: "connect",
@@ -140,9 +137,7 @@ const waitOpen = (ws: globalThis.WebSocket, input: WebSocketRequest) => {
      cleanup()
      resume(
        Effect.fail(
-          transportError(`WebSocket closed before opening with code ${event.code}`, {
-            body: event.reason,
-            cause: event,
+          transportError("open", `WebSocket closed before opening with code ${event.code}`, {
            url: input.url,
            operation: "request",
            code: String(event.code),
@@ -174,8 +169,7 @@ export const toWebSocketUrl = (value: string) =>
      throw new Error(`Unsupported WebSocket URL protocol ${url.protocol}`)
    },
    catch: (error) =>
-      transportError(error instanceof Error ? error.message : "Invalid WebSocket URL", {
-        cause: error,
+      transportError("prepare", error instanceof Error ? error.message : "Invalid WebSocket URL", {
        url: value,
        operation: "request",
        code: "invalid-url",
@@ -195,8 +189,7 @@ export const open = (input: WebSocketRequest) =>
          headers: input.headers,
        }),
      catch: (error) =>
-        transportError(error instanceof Error ? error.message : "Failed to construct WebSocket", {
-          cause: error,
+        transportError("open", error instanceof Error ? error.message : "Failed to construct WebSocket", {
          url: input.url,
          operation: "request",
          phase: "connect",
@@ -221,8 +214,7 @@ export const fromWebSocket = (
      Queue.failCauseUnsafe(
        messages,
        Cause.fail(
-          transportError("WebSocket message exceeds the 16 MiB limit", {
-            body: typeof message === "string" ? message : new TextDecoder().decode(message),
+          transportError("message", "WebSocket message exceeds the 16 MiB limit", {
            url: input.url,
            operation: "read",
            code: "message-too-large",
@@ -239,8 +231,7 @@ export const fromWebSocket = (
      Queue.failCauseUnsafe(
        messages,
        Cause.fail(
-          transportError("WebSocket inbound queue overflow", {
-            body: typeof message === "string" ? message : new TextDecoder().decode(message),
+          transportError("message", "WebSocket inbound queue overflow", {
            url: input.url,
            operation: "read",
            code: "queue-overflow",
@@ -257,8 +248,7 @@ export const fromWebSocket = (
      Queue.failCauseUnsafe(
        messages,
        Cause.fail(
-          transportError("Unsupported WebSocket message payload", {
-            cause: event,
+          transportError("message", "Unsupported WebSocket message payload", {
            url: input.url,
            operation: "read",
            code: "message",
@@ -271,8 +261,7 @@ export const fromWebSocket = (
      Queue.failCauseUnsafe(
        messages,
        Cause.fail(
-          transportError(`WebSocket error: ${eventMessage(event)}`, {
-            cause: "error" in event ? (event.error ?? event) : event,
+          transportError("message", `WebSocket error: ${eventMessage(event)}`, {
            url: input.url,
            operation: "read",
            code: "message",
@@ -285,9 +274,7 @@ export const fromWebSocket = (
      Queue.failCauseUnsafe(
        messages,
        Cause.fail(
-          transportError(`WebSocket closed with code ${event.code}`, {
-            body: event.reason,
-            cause: event,
+          transportError("message", `WebSocket closed with code ${event.code}`, {
            url: input.url,
            operation: "read",
            code: String(event.code),
@@ -311,7 +298,7 @@ export const fromWebSocket = (
        Effect.suspend(() => {
          if (ws.readyState !== globalThis.WebSocket.OPEN)
            return Effect.fail(
-              transportError(`WebSocket is not open (state ${ws.readyState})`, {
+              transportError("sendText", `WebSocket is not open (state ${ws.readyState})`, {
                url: input.url,
                operation: "write",
                phase: "send",
@@ -321,8 +308,7 @@ export const fromWebSocket = (
          return Effect.try({
            try: () => ws.send(message),
            catch: (error) =>
-              transportError(error instanceof Error ? error.message : "Failed to send WebSocket message", {
-                cause: error,
+              transportError("sendText", error instanceof Error ? error.message : "Failed to send WebSocket message", {
                url: input.url,
                operation: "write",
                phase: "send",
@@ -363,23 +349,10 @@ export const makeDirect = (connector: WebSocketConnector): WebSocketChannelExecu
        (connection) => connection.close,
      )
      const create = yield* exchange.driver.create(undefined)
-      yield* connection.sendText(create.message).pipe(
-        Effect.mapError(
-          (error) =>
-            new AIError({
-              reason: AIErrorReason.make({
-                ...error.reason,
-                message: error.reason.message,
-                cause: error.reason.cause,
-                http: error.reason.http ?? connection.http,
-              }),
-            }),
-        ),
-      )
+      yield* connection.sendText(create.message)
      const decoder = new TextDecoder()
      let observed = false
      return {
-        http: connection.http,
        frames: connection.messages.pipe(
          Stream.map((message) => {
            observed = true
@@ -391,49 +364,9 @@ export const makeDirect = (connector: WebSocketConnector): WebSocketChannelExecu
              delivery: observed ? "accepted" : "ambiguous",
            }),
          ),
-          Stream.mapEffect((frame) =>
-            exchange.driver.observe(create, frame).pipe(
-              Effect.mapError(
-                (error) =>
-                  new AIError({
-                    reason: AIErrorReason.make({
-                      ...error.reason,
-                      message: error.reason.message,
-                      cause: error.reason.cause,
-                      body: frame,
-                    }),
-                  }),
-              ),
-              Effect.map((observation) =>
-                "error" in observation
-                  ? {
-                      ...observation,
-                      error: new AIError({
-                        reason: AIErrorReason.make({
-                          ...observation.error.reason,
-                          message: observation.error.reason.message,
-                          cause: observation.error.reason.cause,
-                          body: frame,
-                        }),
-                      }),
-                    }
-                  : observation,
-              ),
-            ),
-          ),
+          Stream.mapEffect((frame) => exchange.driver.observe(create, frame)),
          Stream.takeUntil(observationTerminal),
          Stream.mapEffect(observationFrame),
-          Stream.mapError(
-            (error) =>
-              new AIError({
-                reason: AIErrorReason.make({
-                  ...error.reason,
-                  message: error.reason.message,
-                  cause: error.reason.cause,
-                  http: error.reason.http ?? connection.http,
-                }),
-              }),
-          ),
        ),
        complete: Effect.void,
      }
@@ -484,7 +417,7 @@ export const json = <Body, Message>(input: JsonInput<Body, Message>): JsonTransp
    const webSocket = options?.webSocket
    if (!webSocket) {
      return Effect.fail(
-        transportError("WebSocket JSON transport requires StreamOptions.webSocket", {
+        transportError("json", "WebSocket JSON transport requires StreamOptions.webSocket", {
          url: prepared.url,
          operation: "request",
          code: "unavailable",
@@ -502,7 +435,7 @@ export const json = <Body, Message>(input: JsonInput<Body, Message>): JsonTransp
      connect: { url: prepared.url, headers: prepared.headers },
      fallback: () =>
        Stream.fail(
-          transportError("WebSocket JSON transport does not provide HTTP fallback", {
+          transportError("fallback", "WebSocket JSON transport does not provide HTTP fallback", {
            url: prepared.url,
            operation: "request",
            code: "websocket",
@@ -1,13 +1,19 @@
 import { Schema } from "effect"
 import { Tool } from "@opencode-ai/schema/tool"
 import { ModelID, ProviderID, RouteID } from "./ids.js"
+import { ProviderMetadata } from "./messages.js"

 export const ProviderFailureClassification = Schema.Literals(["context-overflow", "payload-too-large"])
 export type ProviderFailureClassification = typeof ProviderFailureClassification.Type

-export class HttpContext extends Schema.Class<HttpContext>("AI.HttpContext")({
+export class HttpRequestDetails extends Schema.Class<HttpRequestDetails>("AI.HttpRequestDetails")({
+  method: Schema.String,
  url: Schema.String,
-  status: Schema.Int.check(Schema.isBetween({ minimum: 100, maximum: 599 })),
+  headers: Schema.Record(Schema.String, Schema.String),
+}) {}
+
+export class HttpResponseDetails extends Schema.Class<HttpResponseDetails>("AI.HttpResponseDetails")({
+  status: Schema.Number,
  headers: Schema.Record(Schema.String, Schema.String),
 }) {}

@@ -18,61 +24,74 @@ export class HttpRateLimitDetails extends Schema.Class<HttpRateLimitDetails>("AI
  reset: Schema.optional(Schema.Record(Schema.String, Schema.String)),
 }) {}

-const ReasonFields = {
-  message: Schema.String,
-  // Preserve the complete original response or triggering event before decoding narrows it.
+export class HttpContext extends Schema.Class<HttpContext>("AI.HttpContext")({
+  request: HttpRequestDetails,
+  response: Schema.optional(HttpResponseDetails),
  body: Schema.optional(Schema.String),
-  http: Schema.optional(HttpContext),
-  cause: Schema.optional(Schema.Defect({ includeStack: true })),
-}
-
-export class InvalidRequestError extends Schema.TaggedError<InvalidRequestError>("AI.Error.InvalidRequest")(
-  "InvalidRequest",
-  {
-    ...ReasonFields,
-    parameter: Schema.optional(Schema.String),
-    classification: Schema.optional(ProviderFailureClassification),
-  },
-) {}
-
-export class NoRouteError extends Schema.TaggedError<NoRouteError>("AI.Error.NoRoute")("NoRoute", {
-  ...ReasonFields,
-  route: RouteID,
-  provider: ProviderID,
-  model: ModelID,
-}) {}
-
-export class AuthenticationError extends Schema.TaggedError<AuthenticationError>("AI.Error.Authentication")(
-  "Authentication",
-  {
-    ...ReasonFields,
-    kind: Schema.Literals(["missing", "invalid", "expired", "insufficient-permissions", "unknown"]),
-  },
-) {}
-
-export class RateLimitError extends Schema.TaggedError<RateLimitError>("AI.Error.RateLimit")("RateLimit", {
-  ...ReasonFields,
-  retryAfterMs: Schema.optional(Schema.Number),
+  bodyTruncated: Schema.optional(Schema.Boolean),
+  requestId: Schema.optional(Schema.String),
  rateLimit: Schema.optional(HttpRateLimitDetails),
 }) {}

-export class QuotaExceededError extends Schema.TaggedError<QuotaExceededError>("AI.Error.QuotaExceeded")(
-  "QuotaExceeded",
-  ReasonFields,
-) {}
+export class InvalidRequestReason extends Schema.Class<InvalidRequestReason>("AI.Error.InvalidRequest")({
+  _tag: Schema.tag("InvalidRequest"),
+  message: Schema.String,
+  parameter: Schema.optional(Schema.String),
+  classification: Schema.optional(ProviderFailureClassification),
+  providerMetadata: Schema.optional(ProviderMetadata),
+  http: Schema.optional(HttpContext),
+}) {}

-export class ContentPolicyError extends Schema.TaggedError<ContentPolicyError>("AI.Error.ContentPolicy")(
-  "ContentPolicy",
-  ReasonFields,
-) {}
+export class NoRouteReason extends Schema.Class<NoRouteReason>("AI.Error.NoRoute")({
+  _tag: Schema.tag("NoRoute"),
+  route: RouteID,
+  provider: ProviderID,
+  model: ModelID,
+}) {
+  get message() {
+    return `No AI route for ${this.provider}/${this.model} using ${this.route}`
+  }
+}

-export class ProviderInternalError extends Schema.TaggedError<ProviderInternalError>("AI.Error.ProviderInternal")(
-  "ProviderInternal",
-  {
-    ...ReasonFields,
-    retryAfterMs: Schema.optional(Schema.Number),
-  },
-) {}
+export class AuthenticationReason extends Schema.Class<AuthenticationReason>("AI.Error.Authentication")({
+  _tag: Schema.tag("Authentication"),
+  message: Schema.String,
+  kind: Schema.Literals(["missing", "invalid", "expired", "insufficient-permissions", "unknown"]),
+  providerMetadata: Schema.optional(ProviderMetadata),
+  http: Schema.optional(HttpContext),
+}) {}
+
+export class RateLimitReason extends Schema.Class<RateLimitReason>("AI.Error.RateLimit")({
+  _tag: Schema.tag("RateLimit"),
+  message: Schema.String,
+  retryAfterMs: Schema.optional(Schema.Number),
+  rateLimit: Schema.optional(HttpRateLimitDetails),
+  providerMetadata: Schema.optional(ProviderMetadata),
+  http: Schema.optional(HttpContext),
+}) {}
+
+export class QuotaExceededReason extends Schema.Class<QuotaExceededReason>("AI.Error.QuotaExceeded")({
+  _tag: Schema.tag("QuotaExceeded"),
+  message: Schema.String,
+  providerMetadata: Schema.optional(ProviderMetadata),
+  http: Schema.optional(HttpContext),
+}) {}
+
+export class ContentPolicyReason extends Schema.Class<ContentPolicyReason>("AI.Error.ContentPolicy")({
+  _tag: Schema.tag("ContentPolicy"),
+  message: Schema.String,
+  providerMetadata: Schema.optional(ProviderMetadata),
+  http: Schema.optional(HttpContext),
+}) {}
+
+export class ProviderInternalReason extends Schema.Class<ProviderInternalReason>("AI.Error.ProviderInternal")({
+  _tag: Schema.tag("ProviderInternal"),
+  message: Schema.String,
+  status: Schema.optional(Schema.Number),
+  retryAfterMs: Schema.optional(Schema.Number),
+  providerMetadata: Schema.optional(ProviderMetadata),
+  http: Schema.optional(HttpContext),
+}) {}

 export const TransportType = Schema.Literals(["http", "websocket"])
 export type TransportType = typeof TransportType.Type
@@ -80,12 +99,14 @@ export type TransportType = typeof TransportType.Type
 export const TransportOperation = Schema.Literals(["request", "read", "write"])
 export type TransportOperation = typeof TransportOperation.Type

-export class TransportError extends Schema.TaggedError<TransportError>("AI.Error.Transport")("Transport", {
-  ...ReasonFields,
+export class TransportReason extends Schema.Class<TransportReason>("AI.Error.Transport")({
+  _tag: Schema.tag("Transport"),
+  message: Schema.String,
  transport: TransportType,
  operation: TransportOperation,
  code: Schema.optional(Schema.String),
  url: Schema.optional(Schema.String),
+  http: Schema.optional(HttpContext),
  phase: Schema.optional(
    Schema.Literals(["prepare", "queue", "connect", "send", "receive", "decode", "complete", "fallback", "close"]),
  ),
@@ -95,40 +116,48 @@ export class TransportError extends Schema.TaggedError<TransportError>("AI.Error
  ),
 }) {}

-export class InvalidProviderOutputError extends Schema.TaggedError<InvalidProviderOutputError>(
+export class InvalidProviderOutputReason extends Schema.Class<InvalidProviderOutputReason>(
  "AI.Error.InvalidProviderOutput",
-)("InvalidProviderOutput", {
-  ...ReasonFields,
+)({
+  _tag: Schema.tag("InvalidProviderOutput"),
+  message: Schema.String,
  classification: Schema.optional(Schema.Literals(["incomplete-stream"])),
  route: Schema.optional(Schema.String),
+  raw: Schema.optional(Schema.String),
+  providerMetadata: Schema.optional(ProviderMetadata),
 }) {}

-export class UnknownProviderError extends Schema.TaggedError<UnknownProviderError>("AI.Error.UnknownProvider")(
-  "UnknownProvider",
-  ReasonFields,
-) {}
+export class UnknownProviderReason extends Schema.Class<UnknownProviderReason>("AI.Error.UnknownProvider")({
+  _tag: Schema.tag("UnknownProvider"),
+  message: Schema.String,
+  status: Schema.optional(Schema.Number),
+  providerMetadata: Schema.optional(ProviderMetadata),
+  http: Schema.optional(HttpContext),
+}) {}

 export const AIErrorReason = Schema.Union([
-  InvalidRequestError,
-  NoRouteError,
-  AuthenticationError,
-  RateLimitError,
-  QuotaExceededError,
-  ContentPolicyError,
-  ProviderInternalError,
-  TransportError,
-  InvalidProviderOutputError,
-  UnknownProviderError,
+  InvalidRequestReason,
+  NoRouteReason,
+  AuthenticationReason,
+  RateLimitReason,
+  QuotaExceededReason,
+  ContentPolicyReason,
+  ProviderInternalReason,
+  TransportReason,
+  InvalidProviderOutputReason,
+  UnknownProviderReason,
 ]).pipe(Schema.toTaggedUnion("_tag"))
 export type AIErrorReason = Schema.Schema.Type<typeof AIErrorReason>

 export class AIError extends Schema.TaggedError<AIError>()("AI.Error", {
+  module: Schema.String,
+  method: Schema.String,
  reason: AIErrorReason,
 }) {
  override readonly cause = this.reason

-  override get message(): string {
-    return this.reason.message
+  override get message() {
+    return `${this.module}.${this.method}: ${this.reason.message}`
  }
 }

@@ -152,8 +152,6 @@ export const ToolInputDelta = Schema.Struct({
  id: ToolCallID,
  name: Schema.String,
  text: Schema.String,
-  /** Best-effort parse of all input fragments received through this delta. */
-  input: Schema.optional(Schema.Unknown),
 }).annotate({ identifier: "LLM.Event.ToolInputDelta" })
 export type ToolInputDelta = Schema.Schema.Type<typeof ToolInputDelta>

@@ -51,7 +51,6 @@ export const MediaPart = Schema.Struct({
  mediaType: Schema.String,
  data: Schema.Union([Schema.String, Schema.Uint8Array]),
  filename: Schema.optional(Schema.String),
-  cache: Schema.optional(CacheHint),
  metadata: Schema.optional(Schema.Record(Schema.String, Schema.Unknown)),
 }).annotate({ identifier: "LLM.Content.Media" })
 export type MediaPart = Schema.Schema.Type<typeof MediaPart>
@@ -256,7 +255,6 @@ export namespace ToolDefinition {
 export class ToolChoice extends Schema.Class<ToolChoice>("LLM.ToolChoice")({
  type: Schema.Literals(["auto", "none", "required", "tool"]),
  name: Schema.optional(Schema.String),
-  disableParallelToolUse: Schema.optional(Schema.Boolean),
 }) {}

 export namespace ToolChoice {
@@ -153,16 +153,8 @@ export class LanguageModelCompatibility extends Schema.Class<LanguageModelCompat
 )({
  toolSchema: Schema.optional(LanguageModelToolSchemaCompatibility),
  reasoningField: Schema.optional(Schema.String),
-  /** Require every assistant message to include its reasoning field, even when empty. */
-  requireReasoning: Schema.optional(Schema.Boolean),
  maxTokensField: Schema.optional(LanguageModelMaxTokensFieldCompatibility),
  requireFinishReason: Schema.optional(Schema.Boolean),
-  requireAssistantAfterTool: Schema.optional(Schema.Boolean),
-  supportsStore: Schema.optional(Schema.Boolean),
-  supportsUsageInStreaming: Schema.optional(Schema.Boolean),
-  supportsStrictMode: Schema.optional(Schema.Boolean),
-  zaiToolStream: Schema.optional(Schema.Boolean),
-  requireSignature: Schema.optional(Schema.Boolean),
 }) {}

 export namespace LanguageModelCompatibility {
@@ -1,12 +0,0 @@
-import { isRecord } from "./record.js"
-
-export const sanitizeSurrogates = <T>(value: T): T => {
-  if (typeof value === "string") return value.toWellFormed() as T
-  if (Array.isArray(value)) return value.map(sanitizeSurrogates) as T
-  if (value instanceof Uint8Array || value instanceof Error) return value
-  if (isRecord(value))
-    return Object.fromEntries(
-      Object.entries(value).map(([key, entry]) => [key.toWellFormed(), sanitizeSurrogates(entry)]),
-    ) as T
-  return value
-}
@@ -3,7 +3,7 @@ import { Effect } from "effect"
 import { CacheHint, LLM, Message } from "../src/index.js"
 import { Auth } from "../src/route.js"
 import { compileRequest } from "../src/route/client.js"
-import { AmazonBedrock, GoogleVertexMessages } from "../src/providers.js"
+import { AmazonBedrock } from "../src/providers.js"
 import * as AnthropicMessages from "../src/protocols/anthropic-messages.js"
 import * as Gemini from "../src/protocols/gemini.js"
 import * as OpenAIChat from "../src/protocols/openai-chat.js"
@@ -86,27 +86,6 @@ describe("applyCachePolicy", () => {
    }),
  )

-  it.effect("'auto' emits Anthropic cache markers on Vertex", () =>
-    Effect.gen(function* () {
-      const prepared = yield* compileRequest(
-        LLM.request({
-          model: GoogleVertexMessages.configure({ accessToken: "test", location: "global", project: "test" }).model(
-            "claude-opus-4-8",
-          ),
-          system: "You are concise.",
-          tools: [{ name: "lookup", description: "Look up a value", inputSchema: { type: "object", properties: {} } }],
-          prompt: "hi",
-        }),
-      )
-
-      expect(prepared.body).toMatchObject({
-        tools: [{ name: "lookup", cache_control: { type: "ephemeral" } }],
-        system: [{ type: "text", text: "You are concise.", cache_control: { type: "ephemeral" } }],
-        messages: [{ role: "user", content: [{ type: "text", text: "hi", cache_control: { type: "ephemeral" } }] }],
-      })
-    }),
-  )
-
  it.effect("'auto' is a no-op on OpenAI (implicit caching protocol)", () =>
    Effect.gen(function* () {
      const prepared = yield* compileRequest(
@@ -1,7 +1,7 @@
 import { describe, expect, test } from "bun:test"
 import { Effect, Ref, Schema } from "effect"
 import { HttpClientRequest, HttpClientResponse } from "effect/unstable/http"
-import { LLM, Message, ToolCallPart, mergeProviderOptions } from "../src/index.js"
+import { LLM, mergeProviderOptions } from "../src/index.js"
 import { AnthropicMessages, OpenAIChat } from "../src/protocols.js"
 import { Auth, LLMClient } from "../src/route.js"
 import { compileRequest } from "../src/route/client.js"
@@ -66,7 +66,7 @@ describe("request option precedence", () => {
      expect(prepared.body).toMatchObject({
        model: "gpt-4o-mini",
        stream: true,
-        max_completion_tokens: 30,
+        max_tokens: 30,
        temperature: 0.5,
        top_p: 0.9,
        frequency_penalty: 0.25,
@@ -247,73 +247,6 @@ describe("request option precedence", () => {
    }),
  )

-  it.effect("sanitizes outbound JSON without an HTTP overlay", () =>
-    LLMClient.generate(
-      LLM.request({
-        model: OpenAIChat.route
-          .with({ endpoint: { baseURL: "https://api.openai.test/v1/" }, auth: Auth.bearer("test") })
-          .model({ id: "gpt-4o-mini" }),
-        prompt: "hello \uD800 \u{1F600}",
-      }),
-    ).pipe(
-      Effect.provide(
-        dynamicResponse((input) =>
-          Effect.gen(function* () {
-            expect(decodeJson(input.text)).toMatchObject({
-              messages: [{ role: "user", content: "hello \uFFFD \u{1F600}" }],
-            })
-            return input.respond(sseEvents(deltaChunk({}, "stop")), {
-              headers: { "content-type": "text/event-stream" },
-            })
-          }),
-        ),
-      ),
-    ),
-  )
-
-  it.effect("sanitizes unpaired surrogates throughout outbound JSON", () =>
-    LLMClient.generate(
-      LLM.request({
-        model: OpenAIChat.route
-          .with({ endpoint: { baseURL: "https://api.openai.test/v1/" }, auth: Auth.bearer("test") })
-          .model({ id: "gpt-4o-mini" }),
-        system: "system \uD800 \u{1F600}",
-        messages: [
-          Message.user("user \uDC00"),
-          Message.assistant([
-            Message.text("assistant \uD800"),
-            ToolCallPart.make({ id: "call_1", name: "lookup", input: { query: "input \uDC00" } }),
-          ]),
-          Message.tool({ id: "call_1", name: "lookup", result: { output: "result \uD800" } }),
-        ],
-        http: { body: { metadata: { "key\uD800": ["overlay \uDC00", "valid \u{1F600}"] } } },
-      }),
-    ).pipe(
-      Effect.provide(
-        dynamicResponse((input) =>
-          Effect.gen(function* () {
-            expect(decodeJson(input.text)).toMatchObject({
-              messages: [
-                { role: "system", content: "system \uFFFD \u{1F600}" },
-                { role: "user", content: "user \uFFFD" },
-                {
-                  role: "assistant",
-                  content: "assistant \uFFFD",
-                  tool_calls: [{ function: { arguments: '{"query":"input \uFFFD"}' } }],
-                },
-                { role: "tool", content: '{"output":"result \uFFFD"}' },
-              ],
-              metadata: { "key\uFFFD": ["overlay \uFFFD", "valid \u{1F600}"] },
-            })
-            return input.respond(sseEvents(deltaChunk({}, "stop")), {
-              headers: { "content-type": "text/event-stream" },
-            })
-          }),
-        ),
-      ),
-    ),
-  )
-
  it.effect("applies raw body overlays after protocol lowering", () =>
    LLMClient.generate(
      LLM.request({
@@ -1,7 +1,7 @@
 import { describe, expect } from "bun:test"
 import { Deferred, Effect, Fiber, Layer, Ref, Stream } from "effect"
 import { Headers, HttpClient, HttpClientError, HttpClientRequest, HttpClientResponse } from "effect/unstable/http"
-import { LLM, AIError, HttpContext, InvalidProviderOutputError, TransportError } from "../src/index.js"
+import { LLM, AIError } from "../src/index.js"
 import { LLMClient, RequestExecutor, WebSocketTransport, type WebSocketChannelExecutor } from "../src/route.js"
 import * as OpenAIChat from "../src/protocols/openai-chat.js"
 import * as OpenAI from "../src/providers/openai.js"
@@ -62,85 +62,22 @@ const countedResponsesLayer = (attempts: Ref.Ref<number>, responses: ReadonlyArr
 const expectAIError = (error: unknown) => {
  expect(error).toBeInstanceOf(AIError)
  if (!(error instanceof AIError)) throw new Error("expected AIError")
-  expect(error.reason).toBeInstanceOf(Error)
-  expect(error.cause).toBe(error.reason)
  return error
 }

+const errorHttp = (error: AIError) => ("http" in error.reason ? error.reason.http : undefined)
 const largeProviderMessage = `Upstream request failed: ${"validation failed; ".repeat(1_000)}`

 describe("RequestExecutor", () => {
-  it.effect("preserves externally captured HTTP errors without inventing response context", () =>
-    Effect.sync(() => {
-      const cause = new Error("upstream request failed")
-      const body = '{"error":{"message":"Rate limited","trace":"original"}}'
-      const error = RequestExecutor.httpFailure({
-        message: "Rate limited",
-        url: request.url,
-        status: 429,
-        responseHeaders: { "Retry-After": "2", "X-Request-ID": "req_external" },
-        responseBody: body,
-        cause,
-      })
-
-      expect(error.message).toBe("Rate limited")
-      expect(error.reason).toMatchObject({ _tag: "RateLimit", retryAfterMs: 2000 })
-      expect(error.reason.body).toBe(body)
-      expect(error.reason.cause).toBe(cause)
-      expect(error.reason.http).toEqual(
-        new HttpContext({
-          url: request.url,
-          status: 429,
-          headers: { "retry-after": "2", "x-request-id": "req_external" },
-        }),
-      )
-      expect(RequestExecutor.httpFailure({ message: "No response", url: request.url }).reason.http).toBeUndefined()
-      expect(RequestExecutor.httpFailure({ message: "No URL", status: 500 }).reason.http).toBeUndefined()
-    }),
-  )
-
-  it.effect("retains the original body-read failure on an HTTP status error", () =>
-    Effect.gen(function* () {
-      const cause = new Error("response body disconnected")
-      const error = yield* Effect.gen(function* () {
-        const executor = yield* RequestExecutor.Service
-        return yield* executor.execute(request).pipe(Effect.flip)
-      }).pipe(
-        Effect.provide(
-          responsesLayer([
-            new Response(
-              new ReadableStream({
-                start(controller) {
-                  controller.error(cause)
-                },
-              }),
-              {
-                status: 503,
-                headers: { "x-request-id": "req_failed_body" },
-              },
-            ),
-          ]),
-        ),
-      )
-
-      expect(error.reason._tag).toBe("ProviderInternal")
-      expect(error.reason.cause).toBe(cause)
-      expect(error.reason.body).toBeUndefined()
-      expect(error.reason.http).toMatchObject({ status: 503, headers: { "x-request-id": "req_failed_body" } })
-    }),
-  )
-
  it.effect("parses response body failures at the executor seam", () =>
    Effect.gen(function* () {
      const executor = yield* RequestExecutor.Service
      const error = yield* RequestExecutor.stream(executor, secretRequest).pipe(Stream.runDrain, Effect.flip)

      expectAIError(error)
-      expect(error.message).toBe("ECONNRESET: disconnected query-secret-123 header-secret-456")
-      expect(error.reason.http).toMatchObject({ status: 200, url: secretRequest.url })
-      expect(error.reason.cause).toMatchObject({ code: "ECONNRESET" })
      expect(error.reason).toMatchObject({
        _tag: "Transport",
+        message: "ECONNRESET: disconnected query-secret-123 header-secret-456",
        transport: "http",
        operation: "read",
        code: "ECONNRESET",
@@ -167,10 +104,9 @@ describe("RequestExecutor", () => {
      const error = yield* RequestExecutor.stream(executor, secretRequest).pipe(Stream.runDrain, Effect.flip)

      expectAIError(error)
-      expect(error.message).toBe("ECONNRESET: socket closed")
-      expect(error.reason.cause).toBeInstanceOf(TypeError)
      expect(error.reason).toMatchObject({
        _tag: "Transport",
+        message: "ECONNRESET: socket closed",
        operation: "read",
        code: "ECONNRESET",
      })
@@ -197,9 +133,7 @@ describe("RequestExecutor", () => {
        .pipe(Effect.flip)

      expectAIError(error)
-      expect(error.message).toBe("plugin rejected request")
-      expect(error.reason.cause).toBeInstanceOf(Error)
-      expect(error.reason.http).toBeUndefined()
+      expect(error.reason.message).toBe("plugin rejected request")
    }).pipe(Effect.provide(responsesLayer([]))),
  )

@@ -218,11 +152,16 @@ describe("RequestExecutor", () => {
        .pipe(Effect.flip)

      expectAIError(error)
-      expect(error.message).toBe("ECONNRESET: proxy disconnected proxy-secret")
-      expect(error.reason.http).toBeUndefined()
      expect(error.reason).toMatchObject({
        _tag: "Transport",
+        message: "ECONNRESET: proxy disconnected proxy-secret",
        url: "https://proxy.test/v1/chat?api_key=proxy-secret",
+        http: {
+          request: {
+            url: "https://proxy.test/v1/chat?api_key=proxy-secret",
+            headers: { authorization: "Bearer proxy-secret" },
+          },
+        },
      })
    }).pipe(
      Effect.provide(
@@ -267,33 +206,11 @@ describe("RequestExecutor", () => {
      expect(error.reason).toMatchObject({
        _tag: "InvalidRequest",
        classification: "payload-too-large",
+        http: { response: { status: 413 } },
      })
-      expect(error.reason.http?.status).toBe(413)
    }).pipe(Effect.provide(responsesLayer([new Response("request too large", { status: 413 })]))),
  )

-  it.effect("classifies Anthropic request_too_large as context overflow", () =>
-    Effect.gen(function* () {
-      const executor = yield* RequestExecutor.Service
-      const error = yield* executor.execute(request).pipe(Effect.flip)
-
-      expectAIError(error)
-      expect(error.reason).toMatchObject({
-        _tag: "InvalidRequest",
-        classification: "context-overflow",
-      })
-      expect(error.reason.http?.status).toBe(413)
-    }).pipe(
-      Effect.provide(
-        responsesLayer([
-          new Response('{"error":{"type":"request_too_large","message":"Request exceeds the maximum size"}}', {
-            status: 413,
-          }),
-        ]),
-      ),
-    ),
-  )
-
  it.effect("does not classify ordinary invalid requests as context overflow", () =>
    Effect.gen(function* () {
      const executor = yield* RequestExecutor.Service
@@ -302,7 +219,7 @@ describe("RequestExecutor", () => {
      expectAIError(error)
      expect(error.reason).toMatchObject({ _tag: "InvalidRequest" })
      expect("classification" in error.reason ? error.reason.classification : undefined).toBeUndefined()
-      expect(error.message).toBe("Provider request failed with HTTP 400")
+      expect(error.reason.message).toBe("Provider request failed with HTTP 400")
    }).pipe(Effect.provide(responsesLayer([new Response("invalid parameter", { status: 400 })]))),
  )

@@ -312,9 +229,9 @@ describe("RequestExecutor", () => {
      const error = yield* executor.execute(request).pipe(Effect.flip)

      expectAIError(error)
-      expect(error.reason).toMatchObject({ _tag: "InvalidRequest" })
-      expect(error.message).toBe(largeProviderMessage)
-      expect(error.reason.body).toContain(largeProviderMessage)
+      expect(error.reason).toMatchObject({ _tag: "InvalidRequest", message: largeProviderMessage })
+      expect(errorHttp(error)?.body).toContain(largeProviderMessage)
+      expect(errorHttp(error)?.bodyTruncated).toBeUndefined()
    }).pipe(
      Effect.provide(
        responsesLayer([
@@ -338,8 +255,8 @@ describe("RequestExecutor", () => {
      expectAIError(error)
      expect(error.reason).toMatchObject({
        _tag: "InvalidRequest",
+        message: "Provider request failed with HTTP 400",
      })
-      expect(error.message).toBe("Provider request failed with HTTP 400")
    }).pipe(Effect.provide(responsesLayer([new Response('{"error":{"message":"  "}}', { status: 400 })]))),
  )

@@ -382,21 +299,30 @@ describe("RequestExecutor", () => {
      const error = yield* executor.execute(request).pipe(Effect.flip)

      expectAIError(error)
-      expect(error.reason).toMatchObject({
-        _tag: "RateLimit",
-        retryAfterMs: 0,
-        rateLimit: { retryAfterMs: 0 },
-        http: {
-          url: "https://provider.test/v1/chat?api_key=secret&key=secret&debug=1",
-          status: 429,
-          headers: {
-            "retry-after-ms": "0",
-            "x-request-id": "req_123",
-            "x-api-key": "secret",
+      expect(error).toMatchObject({
+        reason: {
+          _tag: "RateLimit",
+          retryAfterMs: 0,
+          rateLimit: { retryAfterMs: 0 },
+          http: {
+            requestId: "req_123",
+            request: {
+              method: "POST",
+              url: "https://provider.test/v1/chat?api_key=secret&key=secret&debug=1",
+              headers: { authorization: "Bearer secret", "x-safe": "visible" },
+            },
+            response: {
+              status: 429,
+              headers: {
+                "retry-after-ms": "0",
+                "x-request-id": "req_123",
+                "x-api-key": "secret",
+              },
+            },
          },
        },
      })
-      expect(error.reason.body).toBe("rate limited")
+      expect(errorHttp(error)?.body).toBe("rate limited")
    }).pipe(
      Effect.provide(
        responsesLayer([
@@ -415,7 +341,8 @@ describe("RequestExecutor", () => {
      const error = yield* executor.execute(request).pipe(Effect.flip)

      expectAIError(error)
-      expect(error.reason.http?.headers["x-safe"]).toBe("response-secret")
+      expect(errorHttp(error)?.request.headers["x-safe"]).toBe("visible")
+      expect(errorHttp(error)?.response?.headers["x-safe"]).toBe("response-secret")
    }).pipe(
      Effect.provide(responsesLayer([new Response("bad", { status: 400, headers: { "x-safe": "response-secret" } })])),
      Effect.provideService(Headers.CurrentRedactedNames, ["x-safe"]),
@@ -461,8 +388,8 @@ describe("RequestExecutor", () => {
      const error = yield* executor.execute(request).pipe(Effect.flip)

      expectAIError(error)
-      expect(error.reason).toMatchObject({ _tag: "RateLimit" })
-      expect(error.reason._tag === "RateLimit" ? error.reason.rateLimit : undefined).toEqual({
+      expect(error.reason).toMatchObject({ _tag: "ProviderInternal" })
+      expect(errorHttp(error)?.rateLimit).toEqual({
        retryAfterMs: 0,
        limit: { requests: "100", "input-tokens": "10000" },
        remaining: { requests: "12", "input-tokens": "9000" },
@@ -471,8 +398,8 @@ describe("RequestExecutor", () => {
    }).pipe(
      Effect.provide(
        responsesLayer([
-          new Response("rate limited", {
-            status: 429,
+          new Response("overloaded", {
+            status: 529,
            headers: {
              "retry-after-ms": "0",
              "anthropic-ratelimit-requests-limit": "100",
@@ -504,8 +431,7 @@ describe("RequestExecutor", () => {
      )

      expectAIError(error)
-      expect(error.reason).toMatchObject({ _tag: "ProviderInternal" })
-      expect(error.reason.http?.status).toBe(503)
+      expect(error.reason).toMatchObject({ _tag: "ProviderInternal", status: 503 })
      expect(yield* Ref.get(attempts)).toBe(1)
    }),
  )
@@ -518,8 +444,7 @@ describe("RequestExecutor", () => {
          const error = yield* executor.execute(request).pipe(Effect.flip)

          expectAIError(error)
-          expect(error.reason).toMatchObject({ _tag: "ProviderInternal" })
-          expect(error.reason.http?.status).toBe(status)
+          expect(error.reason).toMatchObject({ _tag: "ProviderInternal", status })
        }).pipe(
          Effect.provide(
            responsesLayer([
@@ -543,7 +468,8 @@ describe("RequestExecutor", () => {

      expectAIError(error)
      expect(error.reason).toMatchObject({ _tag: "Authentication" })
-      expect(error.reason.body).toHaveLength(20_000)
+      expect(errorHttp(error)?.bodyTruncated).toBeUndefined()
+      expect(errorHttp(error)?.body).toHaveLength(20_000)
    }).pipe(
      Effect.provide(
        responsesLayer([
@@ -560,7 +486,9 @@ describe("RequestExecutor", () => {
      const error = yield* executor.execute(request).pipe(Effect.flip)

      expectAIError(error)
-      expect(error.reason.body).toBe('{"error":{"message":"bad","key":"body-secret","detail":"api_key=query-secret"}}')
+      expect(errorHttp(error)?.body).toBe(
+        '{"error":{"message":"bad","key":"body-secret","detail":"api_key=query-secret"}}',
+      )
    }).pipe(
      Effect.provide(
        responsesLayer([
@@ -578,7 +506,7 @@ describe("RequestExecutor", () => {
      const error = yield* executor.execute(secretRequest).pipe(Effect.flip)

      expectAIError(error)
-      expect(error.reason.body).toBe("provider echoed query-secret-123 and authorization header-secret-456")
+      expect(errorHttp(error)?.body).toBe("provider echoed query-secret-123 and authorization header-secret-456")
    }).pipe(
      Effect.provide(
        responsesLayer([
@@ -615,9 +543,6 @@ describe("RequestExecutor", () => {

      expectAIError(error)
      expect(error.reason).toMatchObject({ _tag: "InvalidProviderOutput" })
-      expect(error.reason.body).toBe("not-json")
-      expect(error.reason.cause).toBeDefined()
-      expect(error.reason.http).toMatchObject({ status: 200, headers: { "content-type": "text/event-stream" } })
      expect(yield* Ref.get(attempts)).toBe(1)
    }),
  )
@@ -631,154 +556,6 @@ describe("WebSocket channel execution", () => {
    JSON.stringify({ type: "response.completed", response: { id: "resp_1" } }),
  ]

-  it.effect("preserves close reasons and native event causes without fabricated HTTP metadata", () =>
-    Effect.gen(function* () {
-      class TestSocket extends EventTarget {
-        readyState = globalThis.WebSocket.OPEN
-        send() {}
-        close() {}
-      }
-      const socket = new TestSocket()
-      const connection = yield* WebSocketTransport.fromWebSocket(
-        // oxlint-disable-next-line typescript-eslint/no-unsafe-type-assertion
-        socket as unknown as globalThis.WebSocket,
-        { url: "wss://provider.test/responses", headers: Headers.empty },
-      )
-      const event = new CloseEvent("close", { code: 1011, reason: "upstream trace: req_close" })
-      socket.dispatchEvent(event)
-      const error = yield* connection.messages.pipe(Stream.runDrain, Effect.flip)
-
-      expect(error.reason).toMatchObject({ _tag: "Transport", code: "1011", phase: "close" })
-      expect(error.message).toBe("WebSocket closed with code 1011")
-      expect(error.reason.body).toBe(event.reason)
-      expect(error.reason.cause).toBe(event)
-      expect(error.reason.http).toBeUndefined()
-      yield* connection.close
-    }),
-  )
-
-  it.effect("preserves opening event errors and native send exceptions", () =>
-    Effect.gen(function* () {
-      const cause = new Error("native send failed")
-      class TestSocket extends EventTarget {
-        readyState = globalThis.WebSocket.CONNECTING
-        send() {
-          throw cause
-        }
-        close() {}
-      }
-      const socket = new TestSocket()
-      const open = WebSocketTransport.fromWebSocket(
-        // oxlint-disable-next-line typescript-eslint/no-unsafe-type-assertion
-        socket as unknown as globalThis.WebSocket,
-        { url: "wss://provider.test/responses", headers: Headers.empty },
-      )
-      const fiber = yield* open.pipe(Effect.flip, Effect.forkChild({ startImmediately: true }))
-      const event = new ErrorEvent("error", { message: "handshake rejected", error: cause })
-      socket.dispatchEvent(event)
-      const error = yield* Fiber.join(fiber)
-      expect(error.reason.cause).toBe(cause)
-      expect(error.message).toContain("handshake rejected")
-      expect(error.reason.http).toBeUndefined()
-
-      socket.readyState = globalThis.WebSocket.OPEN
-      const connection = yield* open
-      const sent = yield* connection.sendText("create").pipe(Effect.flip)
-      expect(sent.reason.cause).toBe(cause)
-      expect(sent.message).toBe(cause.message)
-      yield* connection.close
-    }),
-  )
-
-  it.effect("preserves raw driver failures and known upgrade metadata", () =>
-    Effect.gen(function* () {
-      const cause = new Error("driver validation failed")
-      const frame = '{ "error": "failed", "trace": "original" }'
-      const http = new HttpContext({
-        url: "https://provider.test/responses",
-        status: 101,
-        headers: { upgrade: "websocket" },
-      })
-      const executor = WebSocketTransport.makeDirect({
-        open: () =>
-          Effect.succeed({
-            http,
-            sendText: () => Effect.void,
-            messages: Stream.make(frame),
-            close: Effect.void,
-          }),
-      })
-      const execution = yield* executor.execute({
-        id: "exchange_error",
-        connect: { url: "wss://provider.test/responses", headers: Headers.empty },
-        fallback: () => Stream.empty,
-        driver: {
-          create: () => Effect.succeed({ message: "create", mode: "full" }),
-          observe: () =>
-            Effect.succeed({
-              type: "provider-failure",
-              error: new AIError({
-                reason: new InvalidProviderOutputError({
-                  message: "Driver failed",
-                  cause,
-                  body: "narrowed",
-                }),
-              }),
-            }),
-        },
-      })
-      const error = yield* execution.frames.pipe(Stream.runDrain, Effect.flip)
-
-      expect(error.message).toBe("Driver failed")
-      expect(error.reason.body).toBe(frame)
-      expect(error.reason.cause).toBe(cause)
-      expect(error.reason.http).toBe(http)
-      expect(execution.http).toBe(http)
-    }),
-  )
-
-  it.effect("retains diagnostic fields when annotating transport delivery", () =>
-    Effect.gen(function* () {
-      const cause = new Error("connection closed")
-      const executor = WebSocketTransport.makeDirect({
-        open: () =>
-          Effect.succeed({
-            sendText: () => Effect.void,
-            messages: Stream.fail(
-              new AIError({
-                reason: new TransportError({
-                  message: "Socket closed",
-                  transport: "websocket",
-                  operation: "read",
-                  phase: "close",
-                  recovery: "retry-full",
-                  body: "server close detail",
-                  cause,
-                }),
-              }),
-            ),
-            close: Effect.void,
-          }),
-      })
-      const execution = yield* executor.execute({
-        id: "exchange_closed",
-        connect: { url: "wss://provider.test/responses", headers: Headers.empty },
-        fallback: () => Stream.empty,
-        driver: {
-          create: () => Effect.succeed({ message: "create", mode: "full" }),
-          observe: (_create, frame) => Effect.succeed({ type: "frame", frame }),
-        },
-      })
-      const error = yield* execution.frames.pipe(Stream.runDrain, Effect.flip)
-
-      expect(error.message).toBe("Socket closed")
-      expect(error.reason.body).toBe("server close detail")
-      expect(error.reason.cause).toBe(cause)
-      expect(error.reason).toMatchObject({ phase: "close", delivery: "ambiguous", recovery: "retry-full" })
-      expect(error.reason.http).toBeUndefined()
-    }),
-  )
-
  it.effect("runs a channel driver through the direct executor", () =>
    Effect.gen(function* () {
      const sent = yield* Ref.make("")
@@ -1,28 +0,0 @@
-{
-  "version": 1,
-  "metadata": {
-    "tags": ["prefix:azure", "provider:azure"],
-    "name": "azure/chat-streams-text",
-    "recordedAt": "2026-08-23T17:21:53.198Z"
-  },
-  "interactions": [
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://aiden-azury-group.openai.azure.com/openai/v1/chat/completions?api-version=v1",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"model\":\"gpt-5.6-luna\",\"messages\":[{\"role\":\"user\",\"content\":\"Reply with exactly one word: hello\"}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"store\":false,\"reasoning_effort\":\"medium\"}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream; charset=utf-8"
-        },
-        "body": "data: {\"choices\":[],\"created\":0,\"id\":\"\",\"model\":\"\",\"object\":\"\",\"prompt_filter_results\":[{\"prompt_index\":0,\"content_filter_results\":{}}]}\n\ndata: {\"choices\":[{\"content_filter_results\":{},\"delta\":{\"content\":\"\",\"refusal\":null,\"role\":\"assistant\"},\"finish_reason\":null,\"index\":0,\"logprobs\":null}],\"created\":1787505712,\"id\":\"chatcmpl-EG6BEiYSfrcTSI2WX8PqNzERZDcPc\",\"model\":\"gpt-5.6-luna-2026-07-09\",\"obfuscation\":\"Mxr\",\"object\":\"chat.completion.chunk\",\"service_tier\":\"default\",\"system_fingerprint\":null,\"usage\":null}\n\ndata: {\"choices\":[{\"content_filter_results\":{},\"delta\":{\"content\":\"hello\"},\"finish_reason\":null,\"index\":0,\"logprobs\":null}],\"created\":1787505712,\"id\":\"chatcmpl-EG6BEiYSfrcTSI2WX8PqNzERZDcPc\",\"model\":\"gpt-5.6-luna-2026-07-09\",\"obfuscation\":\"\",\"object\":\"chat.completion.chunk\",\"service_tier\":\"default\",\"system_fingerprint\":null,\"usage\":null}\n\ndata: {\"choices\":[{\"content_filter_results\":{},\"delta\":{},\"finish_reason\":\"stop\",\"index\":0,\"logprobs\":null}],\"created\":1787505712,\"id\":\"chatcmpl-EG6BEiYSfrcTSI2WX8PqNzERZDcPc\",\"model\":\"gpt-5.6-luna-2026-07-09\",\"obfuscation\":\"WyZa5AY1CaCeFdS\",\"object\":\"chat.completion.chunk\",\"service_tier\":\"default\",\"system_fingerprint\":null,\"usage\":null}\n\ndata: {\"choices\":[],\"created\":1787505712,\"id\":\"chatcmpl-EG6BEiYSfrcTSI2WX8PqNzERZDcPc\",\"latency_checkpoint\":{\"engine_tbt_ms\":20,\"engine_ttft_ms\":106,\"engine_ttlt_ms\":206,\"pre_inference_ms\":89,\"service_tbt_ms\":20,\"service_ttft_ms\":480,\"service_ttlt_ms\":576,\"total_duration_ms\":491,\"user_visible_ttft_ms\":391},\"model\":\"gpt-5.6-luna-2026-07-09\",\"obfuscation\":\"6\",\"object\":\"chat.completion.chunk\",\"service_tier\":\"default\",\"system_fingerprint\":null,\"usage\":{\"completion_tokens\":5,\"completion_tokens_details\":{\"accepted_prediction_tokens\":0,\"audio_tokens\":0,\"reasoning_tokens\":0,\"rejected_prediction_tokens\":0},\"prompt_tokens\":13,\"prompt_tokens_details\":{\"audio_tokens\":0,\"cache_write_tokens\":0,\"cached_tokens\":0},\"total_tokens\":18}}\n\ndata: [DONE]\n\n"
-      }
-    }
-  ]
-}
@@ -1,32 +0,0 @@
-{
-  "version": 1,
-  "metadata": {
-    "provider": "cerebras",
-    "route": "cerebras-chat",
-    "transport": "http",
-    "model": "gpt-oss-120b",
-    "tags": ["prefix:cerebras-chat", "provider:cerebras", "text", "golden"],
-    "name": "cerebras-chat/cerebras-gpt-oss-120b-text",
-    "recordedAt": "2026-08-25T23:55:27.619Z"
-  },
-  "interactions": [
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://api.cerebras.ai/v1/chat/completions",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"model\":\"gpt-oss-120b\",\"messages\":[{\"role\":\"system\",\"content\":\"You are concise.\"},{\"role\":\"user\",\"content\":\"Reply exactly with: Hello!\"}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":256}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream; charset=utf-8"
-        },
-        "body": "data: {\"id\":\"chatcmpl-cd45cd6e-886a-433e-8ba9-caca78c1f13a\",\"choices\":[{\"delta\":{\"role\":\"assistant\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_e2cabf4999eb0aead3d1\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-cd45cd6e-886a-433e-8ba9-caca78c1f13a\",\"choices\":[{\"delta\":{\"reasoning\":\"The\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_e2cabf4999eb0aead3d1\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-cd45cd6e-886a-433e-8ba9-caca78c1f13a\",\"choices\":[{\"delta\":{\"reasoning\":\" user says: \\\"Reply exactly with: Hello!\\\" So we must output exactly\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_e2cabf4999eb0aead3d1\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-cd45cd6e-886a-433e-8ba9-caca78c1f13a\",\"choices\":[{\"delta\":{\"reasoning\":\" \\\"Hello!\\\" with no extra characters, no formatting\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_e2cabf4999eb0aead3d1\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-cd45cd6e-886a-433e-8ba9-caca78c1f13a\",\"choices\":[{\"delta\":{\"reasoning\":\". Ensure no extra spaces or new\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_e2cabf4999eb0aead3d1\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-cd45cd6e-886a-433e-8ba9-caca78c1f13a\",\"choices\":[{\"delta\":{\"reasoning\":\"lines? Probably just \\\"Hello!\\\".\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_e2cabf4999eb0aead3d1\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-cd45cd6e-886a-433e-8ba9-caca78c1f13a\",\"choices\":[{\"delta\":{\"reasoning\":\" Usually we output exactly that.\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_e2cabf4999eb0aead3d1\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-cd45cd6e-886a-433e-8ba9-caca78c1f13a\",\"choices\":[{\"delta\":{\"content\":\"Hello!\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_e2cabf4999eb0aead3d1\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-cd45cd6e-886a-433e-8ba9-caca78c1f13a\",\"choices\":[{\"delta\":{},\"finish_reason\":\"stop\",\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_e2cabf4999eb0aead3d1\",\"object\":\"chat.completion.chunk\",\"usage\":{\"total_tokens\":142,\"completion_tokens\":58,\"completion_tokens_details\":{\"accepted_prediction_tokens\":0,\"rejected_prediction_tokens\":0,\"reasoning_tokens\":46},\"prompt_tokens\":84,\"prompt_tokens_details\":{\"cached_tokens\":0}},\"time_info\":{\"created\":1787702127.645281,\"queue_time\":0.003817115,\"prompt_time\":0.001587193,\"completion_time\":0.029805929,\"total_time\":0.036823272705078125}}\n\ndata: [DONE]\n\n"
-      }
-    }
-  ]
-}
@@ -1,32 +0,0 @@
-{
-  "version": 1,
-  "metadata": {
-    "provider": "cerebras",
-    "route": "cerebras-chat",
-    "transport": "http",
-    "model": "gpt-oss-120b",
-    "tags": ["prefix:cerebras-chat", "provider:cerebras", "tool", "tool-call", "golden"],
-    "name": "cerebras-chat/cerebras-gpt-oss-120b-tool-call",
-    "recordedAt": "2026-08-25T23:55:28.454Z"
-  },
-  "interactions": [
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://api.cerebras.ai/v1/chat/completions",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"model\":\"gpt-oss-120b\",\"messages\":[{\"role\":\"system\",\"content\":\"Call tools exactly as requested.\"},{\"role\":\"user\",\"content\":\"Call get_weather with city exactly Paris.\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}}],\"tool_choice\":{\"type\":\"function\",\"function\":{\"name\":\"get_weather\"}},\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":512}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream; charset=utf-8"
-        },
-        "body": "data: {\"id\":\"chatcmpl-402a09cc-8668-446d-8f9c-3e4121f5fa51\",\"choices\":[{\"delta\":{\"role\":\"assistant\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_4cfabdd6620dc0120785\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-402a09cc-8668-446d-8f9c-3e4121f5fa51\",\"choices\":[{\"delta\":{\"reasoning\":\"We\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_4cfabdd6620dc0120785\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-402a09cc-8668-446d-8f9c-3e4121f5fa51\",\"choices\":[{\"delta\":{\"reasoning\":\" need to call the function get_weather with city \\\"\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_4cfabdd6620dc0120785\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-402a09cc-8668-446d-8f9c-3e4121f5fa51\",\"choices\":[{\"delta\":{\"reasoning\":\"Paris\\\".\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_4cfabdd6620dc0120785\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-402a09cc-8668-446d-8f9c-3e4121f5fa51\",\"choices\":[{\"delta\":{\"tool_calls\":[{\"function\":{\"name\":\"get_weather\",\"arguments\":\"\"},\"type\":\"function\",\"id\":\"3d860cefe\",\"index\":0}]},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_4cfabdd6620dc0120785\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-402a09cc-8668-446d-8f9c-3e4121f5fa51\",\"choices\":[{\"delta\":{\"tool_calls\":[{\"function\":{\"arguments\":\"{\\\"city\\\":\\\"Paris\\\"}\"},\"type\":\"function\",\"index\":0}]},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_4cfabdd6620dc0120785\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-402a09cc-8668-446d-8f9c-3e4121f5fa51\",\"choices\":[{\"delta\":{},\"finish_reason\":\"tool_calls\",\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_4cfabdd6620dc0120785\",\"object\":\"chat.completion.chunk\",\"usage\":{\"total_tokens\":174,\"completion_tokens\":37,\"completion_tokens_details\":{\"accepted_prediction_tokens\":0,\"rejected_prediction_tokens\":0,\"reasoning_tokens\":13},\"prompt_tokens\":137,\"prompt_tokens_details\":{\"cached_tokens\":0}},\"time_info\":{\"created\":1787702127.8019717,\"queue_time\":0.31196235,\"prompt_time\":0.005234764,\"completion_time\":0.020198402,\"total_time\":0.702225923538208}}\n\ndata: [DONE]\n\n"
-      }
-    }
-  ]
-}
@@ -18,7 +18,7 @@
        "headers": {
          "content-type": "application/json"
        },
-        "body": "{\"model\": \"@cf/openai/gpt-oss-20b\", \"messages\": [{\"role\": \"system\", \"content\": \"Call tools exactly as requested.\"}, {\"role\": \"user\", \"content\": \"Call get_weather with city exactly Paris.\"}], \"tools\": [{\"type\": \"function\", \"function\": {\"name\": \"get_weather\", \"description\": \"Get current weather for a city.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"city\": {\"type\": \"string\"}}, \"required\": [\"city\"], \"additionalProperties\": false}, \"strict\": false}}], \"tool_choice\": {\"type\": \"function\", \"function\": {\"name\": \"get_weather\"}}, \"stream\": true, \"stream_options\": {\"include_usage\": true}, \"max_tokens\": 120, \"temperature\": 0}"
+        "body": "{\"model\":\"@cf/openai/gpt-oss-20b\",\"messages\":[{\"role\":\"system\",\"content\":\"Call tools exactly as requested.\"},{\"role\":\"user\",\"content\":\"Call get_weather with city exactly Paris.\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false}}}],\"tool_choice\":{\"type\":\"function\",\"function\":{\"name\":\"get_weather\"}},\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":120,\"temperature\":0}"
      },
      "response": {
        "status": 200,
@@ -1,32 +0,0 @@
-{
-  "version": 1,
-  "metadata": {
-    "provider": "deepinfra",
-    "route": "deepinfra-chat",
-    "transport": "http",
-    "model": "meta-llama/Llama-3.3-70B-Instruct-Turbo",
-    "tags": ["prefix:deepinfra-chat", "provider:deepinfra", "text", "golden"],
-    "name": "deepinfra-chat/deepinfra-llama-3-3-70b-text",
-    "recordedAt": "2026-08-26T00:34:03.019Z"
-  },
-  "interactions": [
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://api.deepinfra.com/v1/openai/chat/completions",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"messages\":[{\"role\":\"system\",\"content\":\"You are concise.\"},{\"role\":\"user\",\"content\":\"Reply exactly with: Hello!\"}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":40,\"temperature\":0}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream; charset=utf-8"
-        },
-        "body": "data: {\"service_tier\":null,\"id\":\"chatcmpl-RbZ8MyY5pos2MShihSmoXoRe\",\"object\":\"chat.completion.chunk\",\"created\":1787704442,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RbZ8MyY5pos2MShihSmoXoRe\",\"object\":\"chat.completion.chunk\",\"created\":1787704442,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"Hello\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RbZ8MyY5pos2MShihSmoXoRe\",\"object\":\"chat.completion.chunk\",\"created\":1787704442,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"!\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RbZ8MyY5pos2MShihSmoXoRe\",\"object\":\"chat.completion.chunk\",\"created\":1787704442,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":25,\"total_tokens\":28,\"completion_tokens\":3,\"estimated_cost\":null,\"prompt_tokens_details\":null}}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RbZ8MyY5pos2MShihSmoXoRe\",\"object\":\"chat.completion.chunk\",\"created\":1787704442,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[],\"usage\":{\"prompt_tokens\":25,\"total_tokens\":28,\"completion_tokens\":3,\"estimated_cost\":3.46e-6,\"prompt_tokens_details\":null}}\n\ndata: [DONE]\n\n"
-      }
-    }
-  ]
-}
@@ -1,32 +0,0 @@
-{
-  "version": 1,
-  "metadata": {
-    "provider": "deepinfra",
-    "route": "deepinfra-chat",
-    "transport": "http",
-    "model": "meta-llama/Llama-3.3-70B-Instruct-Turbo",
-    "tags": ["prefix:deepinfra-chat", "provider:deepinfra", "tool", "tool-call", "golden"],
-    "name": "deepinfra-chat/deepinfra-llama-3-3-70b-tool-call",
-    "recordedAt": "2026-08-26T00:34:04.173Z"
-  },
-  "interactions": [
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://api.deepinfra.com/v1/openai/chat/completions",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"messages\":[{\"role\":\"system\",\"content\":\"Call tools exactly as requested.\"},{\"role\":\"user\",\"content\":\"Call get_weather with city exactly Paris.\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}}],\"tool_choice\":{\"type\":\"function\",\"function\":{\"name\":\"get_weather\"}},\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":80,\"temperature\":0}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream; charset=utf-8"
-        },
-        "body": "data: {\"service_tier\":null,\"id\":\"chatcmpl-RFOwhlB2PhZrgMLviGrx5BQf\",\"object\":\"chat.completion.chunk\",\"created\":1787704443,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RFOwhlB2PhZrgMLviGrx5BQf\",\"object\":\"chat.completion.chunk\",\"created\":1787704443,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":[{\"index\":0,\"id\":\"call_SMfBjXa8eCmHLyjfeyARxe3a\",\"function\":{\"arguments\":\"{\\\"city\\\": \\\"Paris\\\"}\",\"name\":\"get_weather\"},\"type\":\"function\"}]},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RFOwhlB2PhZrgMLviGrx5BQf\",\"object\":\"chat.completion.chunk\",\"created\":1787704443,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":230,\"total_tokens\":244,\"completion_tokens\":14,\"estimated_cost\":null,\"prompt_tokens_details\":null}}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RFOwhlB2PhZrgMLviGrx5BQf\",\"object\":\"chat.completion.chunk\",\"created\":1787704443,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[],\"usage\":{\"prompt_tokens\":230,\"total_tokens\":244,\"completion_tokens\":14,\"estimated_cost\":0.000027480000000000005,\"prompt_tokens_details\":null}}\n\ndata: [DONE]\n\n"
-      }
-    }
-  ]
-}
@@ -1,50 +0,0 @@
-{
-  "version": 1,
-  "metadata": {
-    "provider": "deepinfra",
-    "route": "deepinfra-chat",
-    "transport": "http",
-    "model": "meta-llama/Llama-3.3-70B-Instruct-Turbo",
-    "tags": ["prefix:deepinfra-chat", "provider:deepinfra", "tool", "tool-loop", "golden"],
-    "name": "deepinfra-chat/deepinfra-llama-3-3-70b-tool-loop",
-    "recordedAt": "2026-08-26T00:34:05.656Z"
-  },
-  "interactions": [
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://api.deepinfra.com/v1/openai/chat/completions",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"messages\":[{\"role\":\"system\",\"content\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"},{\"role\":\"user\",\"content\":\"What is the weather in Paris?\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":80,\"temperature\":0}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream; charset=utf-8"
-        },
-        "body": "data: {\"service_tier\":null,\"id\":\"chatcmpl-RxlHFSnlLbUUz6XSxqBQj7TC\",\"object\":\"chat.completion.chunk\",\"created\":1787704444,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RxlHFSnlLbUUz6XSxqBQj7TC\",\"object\":\"chat.completion.chunk\",\"created\":1787704444,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":[{\"index\":0,\"id\":\"call_W3stxe7FNHozlB4tDxlTVZou\",\"function\":{\"arguments\":\"{\\\"city\\\": \\\"Paris\\\"}\",\"name\":\"get_weather\"},\"type\":\"function\"}]},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RxlHFSnlLbUUz6XSxqBQj7TC\",\"object\":\"chat.completion.chunk\",\"created\":1787704444,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":243,\"total_tokens\":257,\"completion_tokens\":14,\"estimated_cost\":null,\"prompt_tokens_details\":null}}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RxlHFSnlLbUUz6XSxqBQj7TC\",\"object\":\"chat.completion.chunk\",\"created\":1787704444,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[],\"usage\":{\"prompt_tokens\":243,\"total_tokens\":257,\"completion_tokens\":14,\"estimated_cost\":0.000028780000000000002,\"prompt_tokens_details\":null}}\n\ndata: [DONE]\n\n"
-      }
-    },
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://api.deepinfra.com/v1/openai/chat/completions",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"messages\":[{\"role\":\"system\",\"content\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"},{\"role\":\"user\",\"content\":\"What is the weather in Paris?\"},{\"role\":\"assistant\",\"content\":null,\"tool_calls\":[{\"id\":\"call_W3stxe7FNHozlB4tDxlTVZou\",\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"arguments\":\"{\\\"city\\\":\\\"Paris\\\"}\"}}],\"reasoning_content\":\"\"},{\"role\":\"tool\",\"tool_call_id\":\"call_W3stxe7FNHozlB4tDxlTVZou\",\"content\":\"{\\\"temperature\\\":22,\\\"condition\\\":\\\"sunny\\\"}\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":80,\"temperature\":0}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream; charset=utf-8"
-        },
-        "body": "data: {\"service_tier\":null,\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"Paris\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\" is\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\" sunny\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\".\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":276,\"total_tokens\":281,\"completion_tokens\":5,\"estimated_cost\":null,\"prompt_tokens_details\":null}}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[],\"usage\":{\"prompt_tokens\":276,\"total_tokens\":281,\"completion_tokens\":5,\"estimated_cost\":0.0000292,\"prompt_tokens_details\":null}}\n\ndata: [DONE]\n\n"
-      }
-    }
-  ]
-}
@@ -1,28 +0,0 @@
-{
-  "version": 1,
-  "metadata": {
-    "tags": ["prefix:google-vertex", "provider:google-vertex", "protocol:gemini"],
-    "name": "google-vertex/calls-a-tool",
-    "recordedAt": "2026-08-23T17:21:51.036Z"
-  },
-  "interactions": [
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://aiplatform.googleapis.com/v1/publishers/google/models/gemini-3.5-flash:streamGenerateContent?alt=sse",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"contents\":[{\"role\":\"user\",\"parts\":[{\"text\":\"What is the weather in Paris? Use the lookup_weather tool.\"}]}],\"tools\":[{\"functionDeclarations\":[{\"name\":\"lookup_weather\",\"description\":\"Look up the current weather for a city\",\"parameters\":{\"required\":[\"city\"],\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}}}}]}]}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream"
-        },
-        "body": "data: {\"candidates\": [{\"content\": {\"role\": \"model\",\"parts\": [{\"functionCall\": {\"name\": \"lookup_weather\",\"args\": {\"city\": \"Paris\"},\"id\": \"call_425130\"},\"thoughtSignature\": \"AY89a1+1fXnLgYhHMuN3Ak6LBhT6PcrYOW7iPav4LfsacvG/Z6l1yJ+AsU7vWhFj/JyPIbsJJQ+GjohM9sCIZ6nqUOIg3reo/7osmrCvFrVHedTHQcwiPzoz2Kp3gb+uWjFAXxk1EX4IRAKcu0ox1W/Z9PpuZvHkTerGO2a82e02N6MAF1YhhtbXFvSdqLRih2Os68rdOk5/Bcld7ol8qUgeyIZ3CtI3OJ5jwRcD8LjvK33A7ZFzH5Bxp/peUmXvqnu5iNhnGBxZaJy/vupCtxRZxjaS+ojG0/UhyrnRiKIpbzQ0FBkxePPn8GCX/LOe2y3GUc98co8lN8OOuCd9ZmEdx5AjHmQkPO9fAV9SxG6Bda6SDWVL8o/Uz3WSQYoUEfAdoajEWIBvcisoeCJjb7zgmRRZ9VQSPl3RXj5LFRvX8jn0YKV1CahYbc24jA==\"}]}}],\"usageMetadata\": {\"trafficType\": \"ON_DEMAND\"},\"modelVersion\": \"gemini-3.5-flash\",\"createTime\": \"2026-08-23T17:21:50.308576Z\",\"responseId\": \"LiyLauDqErCErb8Pj8aWkAs\"}\r\n\r\ndata: {\"candidates\": [{\"content\": {\"role\": \"model\",\"parts\": [{\"text\": \"\"}]},\"finishReason\": \"STOP\"}],\"usageMetadata\": {\"promptTokenCount\": 39,\"candidatesTokenCount\": 16,\"totalTokenCount\": 102,\"trafficType\": \"ON_DEMAND\",\"promptTokensDetails\": [{\"modality\": \"TEXT\",\"tokenCount\": 39}],\"candidatesTokensDetails\": [{\"modality\": \"TEXT\",\"tokenCount\": 16}],\"thoughtsTokenCount\": 47},\"modelVersion\": \"gemini-3.5-flash\",\"createTime\": \"2026-08-23T17:21:50.308576Z\",\"responseId\": \"LiyLauDqErCErb8Pj8aWkAs\"}\r\n\r\n"
-      }
-    }
-  ]
-}
@@ -1,28 +0,0 @@
-{
-  "version": 1,
-  "metadata": {
-    "tags": ["prefix:google-vertex", "provider:google-vertex", "protocol:gemini"],
-    "name": "google-vertex/continues-after-a-tool-result",
-    "recordedAt": "2026-08-23T17:21:51.853Z"
-  },
-  "interactions": [
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://aiplatform.googleapis.com/v1/publishers/google/models/gemini-3.5-flash:streamGenerateContent?alt=sse",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"contents\":[{\"role\":\"user\",\"parts\":[{\"text\":\"What is the weather in Paris?\"}]},{\"role\":\"model\",\"parts\":[{\"functionCall\":{\"name\":\"lookup_weather\",\"args\":{\"city\":\"Paris\"}},\"thoughtSignature\":\"skip_thought_signature_validator\"}]},{\"role\":\"user\",\"parts\":[{\"functionResponse\":{\"name\":\"lookup_weather\",\"response\":{\"name\":\"lookup_weather\",\"content\":\"18C, light rain\"}}}]}],\"tools\":[{\"functionDeclarations\":[{\"name\":\"lookup_weather\",\"description\":\"Look up the current weather for a city\",\"parameters\":{\"required\":[\"city\"],\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}}}}]}]}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream"
-        },
-        "body": "data: {\"candidates\": [{\"content\": {\"role\": \"model\",\"parts\": [{\"text\": \"The weather in Paris is currently 18°C with light rain.\"}]}}],\"usageMetadata\": {\"trafficType\": \"ON_DEMAND\"},\"modelVersion\": \"gemini-3.5-flash\",\"createTime\": \"2026-08-23T17:21:51.220919Z\",\"responseId\": \"LyyLave9DbWnrb8P1IjLmQQ\"}\r\n\r\ndata: {\"candidates\": [{\"content\": {\"role\": \"model\",\"parts\": [{\"text\": \"\",\"thoughtSignature\": \"AY89a197c+fpHJftPtcufnqMAyoRQKVEQK+KeG+RVHVx2wKil3L4jP4YWvfVbcuOFr2jio4Kre/hCrDANAoMFSvaZrdaPeo1b5bXQSmJKMH03yM5M6q6ME6JiBvXym143U4exIde4UbOh2tMeyXMvB3aWxcavIHd78g5G5QPLreo6A3LO5871cYYVeRwteY+/zbEdqfaAq1hlk6WYpWkNljYpjMyKwr15YC8rFLh3HYayS9tTN++GGrk/reZn6C3OEPlzPou/pXRATzcEAGVl/TW\"}]},\"finishReason\": \"STOP\"}],\"usageMetadata\": {\"promptTokenCount\": 59,\"candidatesTokenCount\": 15,\"totalTokenCount\": 98,\"trafficType\": \"ON_DEMAND\",\"promptTokensDetails\": [{\"modality\": \"TEXT\",\"tokenCount\": 59}],\"candidatesTokensDetails\": [{\"modality\": \"TEXT\",\"tokenCount\": 15}],\"thoughtsTokenCount\": 24},\"modelVersion\": \"gemini-3.5-flash\",\"createTime\": \"2026-08-23T17:21:51.220919Z\",\"responseId\": \"LyyLave9DbWnrb8P1IjLmQQ\"}\r\n\r\n"
-      }
-    }
-  ]
-}
@@ -1,28 +0,0 @@
-{
-  "version": 1,
-  "metadata": {
-    "tags": ["prefix:google-vertex", "provider:google-vertex", "protocol:gemini"],
-    "name": "google-vertex/streams-text",
-    "recordedAt": "2026-08-23T17:21:50.112Z"
-  },
-  "interactions": [
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://aiplatform.googleapis.com/v1/publishers/google/models/gemini-3.5-flash:streamGenerateContent?alt=sse",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"contents\":[{\"role\":\"user\",\"parts\":[{\"text\":\"Reply with exactly one word: hello\"}]}]}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream"
-        },
-        "body": "data: {\"candidates\": [{\"content\": {\"role\": \"model\",\"parts\": [{\"text\": \"Hello\"}]}}],\"usageMetadata\": {\"trafficType\": \"ON_DEMAND\"},\"modelVersion\": \"gemini-3.5-flash\",\"createTime\": \"2026-08-23T17:21:48.528714Z\",\"responseId\": \"LCyLasqiIO6crb8P1sDboQc\"}\r\n\r\ndata: {\"candidates\": [{\"content\": {\"role\": \"model\",\"parts\": [{\"text\": \"\",\"thoughtSignature\": \"AY89a1+BGsRqlGpfT0psLB4jeTkT5rDV2HFOlrRuF7aVxDOjqNVUku6t4azeSnxpd+msHWuwXj4RS+7gmVlzVs+JNi8uj+iZWTBCi71vSh9kdK9ed/sHv9J7uL9ZWSOcgbhX/hxdXaUp5yVbQzHFXPjR9A/IkEkHV8VKarDZVFE1T1uASia74lkmyBeZZz+DQmRsLwbUHzFUKlF3qnk/SliLo21ZgASd7itlALQ0PBLJZwgeI3g7tDscDSE18hnB11Fky8q7MLd3HY16zbDvHBEMb18pmmPelPI01KdrCIwMSou/01/u5jiSUCc3pFksZawUj3tAHocHSC3ZKAQQQuUXGe5tm61C2E40/NANBeePc1S4HYE6Yo/vtX6tE02LDky5IQWX09H6+DZ7fpopP5nCUfcKPHa3hVjYquWYYMtZgXO4ZpxfVd3lt1VUDuJNN3BMMCZapjBoJZFPXPJ5t/yg9Rnd791+msGH77b4wztz1vtsPrT9oV9g6SDo9ZUH6BaOcbK7fw8FaXcGw+55malEwQy6zpRLGecooBu70p6RwhaAUyKIMX49y+F2hkNxQxDeBUNckJnu6n4w+KLyjP+bR0gqPJbGjVfteHm+QujqjJdBBT/m1u9kPo1nIbzdEs/PIADBdbuV7TkD/HoRFKpLnNmM2no8ioTtFEjKBDz4ippGi15r8pGgA6wIb/1HAvOGh+PVERdGcbelVTgfONwBqjQ7B1wmEizCfyYuMIskfwjxDGayfKlpDxrnNeogtEct9u5/DjEKlURlg9MtmW1B9P8BXYJ+7SCiRJWwW6bzB+5C+MLCnETl/mljDizoJMHK8DKIhI4oxBsrWXEuoHFwEwGIeOZq0BofH2Jz/l6+KIboV/zd581Kk0zPg/rlI6acfjUEtXtbF+t0+jzoJN7006x4i2tqXeJZ+4e5yisSArEsfJ0YzNWoJtBHG9V9/euDcEP3+jsr98efaQaQbLMPvT/Hb7CYQ7ChhGfcGxQ=\"}]},\"finishReason\": \"STOP\"}],\"usageMetadata\": {\"promptTokenCount\": 7,\"candidatesTokenCount\": 1,\"totalTokenCount\": 150,\"trafficType\": \"ON_DEMAND\",\"promptTokensDetails\": [{\"modality\": \"TEXT\",\"tokenCount\": 7}],\"candidatesTokensDetails\": [{\"modality\": \"TEXT\",\"tokenCount\": 1}],\"thoughtsTokenCount\": 142},\"modelVersion\": \"gemini-3.5-flash\",\"createTime\": \"2026-08-23T17:21:48.528714Z\",\"responseId\": \"LCyLasqiIO6crb8P1sDboQc\"}\r\n\r\n"
-      }
-    }
-  ]
-}
@@ -1,29 +0,0 @@
-{
-  "version": 1,
-  "metadata": {
-    "model": "openai/gpt-oss-20b",
-    "tags": ["prefix:groq-chat", "provider:groq", "protocol:groq-chat", "text", "usage"],
-    "name": "groq-chat/streams-text-with-usage",
-    "recordedAt": "2026-08-26T14:40:09.833Z"
-  },
-  "interactions": [
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://api.groq.com/openai/v1/chat/completions",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"model\":\"openai/gpt-oss-20b\",\"messages\":[{\"role\":\"user\",\"content\":\"Reply with exactly one word: hello\"}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"reasoning_effort\":\"low\",\"max_completion_tokens\":512,\"include_reasoning\":false,\"service_tier\":\"on_demand\",\"user\":\"recorded-test\"}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream"
-        },
-        "body": "data: {\"id\":\"chatcmpl-660048b0-3c99-4215-a582-e116b77eb881\",\"object\":\"chat.completion.chunk\",\"created\":1787755209,\"model\":\"openai/gpt-oss-20b\",\"system_fingerprint\":\"fp_66891002f6\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\"},\"logprobs\":null,\"finish_reason\":null}],\"x_groq\":{\"id\":\"req_01m0z87915eep9bpf10gg7331e\",\"seed\":94036161}}\n\ndata: {\"id\":\"chatcmpl-660048b0-3c99-4215-a582-e116b77eb881\",\"object\":\"chat.completion.chunk\",\"created\":1787755209,\"model\":\"openai/gpt-oss-20b\",\"system_fingerprint\":\"fp_66891002f6\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\"hello\"},\"logprobs\":null,\"finish_reason\":null}]}\n\ndata: {\"id\":\"chatcmpl-660048b0-3c99-4215-a582-e116b77eb881\",\"object\":\"chat.completion.chunk\",\"created\":1787755209,\"model\":\"openai/gpt-oss-20b\",\"system_fingerprint\":\"fp_66891002f6\",\"choices\":[{\"index\":0,\"delta\":{},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"x_groq\":{\"id\":\"req_01m0z87915eep9bpf10gg7331e\",\"usage\":{\"queue_time\":0.10886435,\"prompt_tokens\":78,\"prompt_time\":0.003693734,\"completion_tokens\":20,\"completion_time\":0.020459983,\"total_tokens\":98,\"total_time\":0.024153717,\"completion_tokens_details\":{\"reasoning_tokens\":10}}},\"usage\":{\"queue_time\":0.10886435,\"prompt_tokens\":78,\"prompt_time\":0.003693734,\"completion_tokens\":20,\"completion_time\":0.020459983,\"total_tokens\":98,\"total_time\":0.024153717,\"completion_tokens_details\":{\"reasoning_tokens\":10}}}\n\ndata: {\"id\":\"chatcmpl-660048b0-3c99-4215-a582-e116b77eb881\",\"object\":\"chat.completion.chunk\",\"created\":1787755209,\"model\":\"openai/gpt-oss-20b\",\"system_fingerprint\":\"fp_66891002f6\",\"choices\":[],\"usage\":{\"queue_time\":0.10886435,\"prompt_tokens\":78,\"prompt_time\":0.003693734,\"completion_tokens\":20,\"completion_time\":0.020459983,\"total_tokens\":98,\"total_time\":0.024153717,\"completion_tokens_details\":{\"reasoning_tokens\":10}},\"service_tier\":\"on_demand\"}\n\ndata: [DONE]\n\n"
-      }
-    }
-  ]
-}
@@ -32,7 +32,7 @@
        {
          "direction": "client",
          "kind": "text",
-          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"What is the weather in Paris?\"}]}],\"tools\":[{\"type\":\"function\",\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"max_output_tokens\":50,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"instructions\":\"Call get_weather once, then reply exactly: Paris is sunny.\"}"
+          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"system\",\"content\":\"Call get_weather once, then reply exactly: Paris is sunny.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"What is the weather in Paris?\"}]}],\"tools\":[{\"type\":\"function\",\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"max_output_tokens\":50,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"}}"
        },
        {
          "direction": "server",
@@ -62,7 +62,7 @@
        {
          "direction": "client",
          "kind": "text",
-          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"type\":\"function_call_output\",\"call_id\":\"call_ws_weather\",\"output\":\"{\\\"temperature\\\":22,\\\"condition\\\":\\\"sunny\\\"}\"}],\"tools\":[{\"type\":\"function\",\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"max_output_tokens\":50,\"previous_response_id\":\"resp_ws_tool_1\",\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"instructions\":\"Call get_weather once, then reply exactly: Paris is sunny.\"}"
+          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"type\":\"function_call_output\",\"call_id\":\"call_ws_weather\",\"output\":\"{\\\"temperature\\\":22,\\\"condition\\\":\\\"sunny\\\"}\"}],\"tools\":[{\"type\":\"function\",\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"max_output_tokens\":50,\"previous_response_id\":\"resp_ws_tool_1\",\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"}}"
        },
        {
          "direction": "server",
--- a/Show More
+++ b/Show More