fix(session-ui): animate appended inline code chunks

test(session-ui): stream inline code in partial chunks
test(session-ui): demonstrate streamed inline code fade
2026-08-27 20:16:17 +00:00 · 2026-08-25 14:07:29 +08:00 · 2026-08-25 14:01:20 +08:00 · 2026-08-25 13:59:37 +08:00 · 2026-08-25 13:55:47 +08:00
770 changed files with 7364 additions and 48137 deletions
@@ -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.
@@ -1,8 +0,0 @@
---
-"@opencode-ai/core": minor
-"@opencode-ai/schema": patch
---
-
-Open durable sessions with in-process model, tool, instruction, and permission capabilities. Live Sources update at safe boundaries through existing instruction epochs, while capability replacement waits for the next busy period. Capability-owned sessions remain pending after restart until their host reopens and drives them.
-
-Close an open's in-process capabilities after settlement without deleting durable history. Tool executors may yield domain errors, which normalize to tool failures while canonical permission declines retain their interruption behavior.
@@ -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.
@@ -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!

@@ -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
@@ -111,16 +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'
@@ -130,15 +92,8 @@ jobs:
      - 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 }}
+        working-directory: packages/sdk
+        run: bun run verify:package

      - name: Verify compiled service lifecycle
        if: always()
@@ -178,7 +133,7 @@ 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 +144,32 @@ 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 +177,23 @@ 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 }}
@@ -125,7 +125,6 @@
        "@effect/platform-node": "catalog:",
        "@opencode-ai/client": "workspace:*",
        "@opencode-ai/plugin": "workspace:*",
-        "@opencode-ai/pty": "0.1.12",
        "@opencode-ai/schema": "workspace:*",
        "@opencode-ai/server": "workspace:*",
        "@opencode-ai/tui": "workspace:*",
@@ -347,14 +346,18 @@
        "@ai-sdk/amazon-bedrock": "4.0.112",
        "@ai-sdk/anthropic": "3.0.82",
        "@ai-sdk/azure": "3.0.88",
+        "@ai-sdk/cerebras": "2.0.41",
        "@ai-sdk/cohere": "3.0.27",
+        "@ai-sdk/deepinfra": "2.0.41",
        "@ai-sdk/gateway": "3.0.104",
        "@ai-sdk/google-vertex": "4.0.128",
+        "@ai-sdk/groq": "3.0.31",
        "@ai-sdk/mistral": "3.0.51",
        "@ai-sdk/openai-compatible": "2.0.41",
        "@ai-sdk/perplexity": "3.0.26",
        "@ai-sdk/provider": "3.0.8",
        "@ai-sdk/provider-utils": "4.0.23",
+        "@ai-sdk/togetherai": "2.0.41",
        "@ai-sdk/vercel": "2.0.39",
        "@aws-sdk/credential-providers": "3.1057.0",
        "@ff-labs/fff-bun": "0.10.5",
@@ -364,7 +367,6 @@
        "@opencode-ai/ai": "workspace:*",
        "@opencode-ai/codemode": "workspace:*",
        "@opencode-ai/plugin": "workspace:*",
-        "@opencode-ai/pty": "0.1.12",
        "@opencode-ai/schema": "workspace:*",
        "@opencode-ai/util": "workspace:*",
        "@parcel/watcher": "2.5.1",
@@ -428,7 +430,6 @@
      },
      "devDependencies": {
        "@actions/artifact": "4.0.0",
-        "@brendonovich/vite-plugin-opencode": "0.1.1",
        "@lydell/node-pty": "catalog:",
        "@opencode-ai/app": "workspace:*",
        "@opencode-ai/client": "workspace:*",
@@ -553,20 +554,6 @@
        "@typescript/native-preview": "catalog:",
      },
    },
-    "packages/latex": {
-      "name": "@opencode-ai/latex",
-      "version": "0.0.0",
-      "dependencies": {
-        "@opencode-ai/plugin": "workspace:*",
-        "@opentui/core": "catalog:",
-        "string-width": "catalog:",
-      },
-      "devDependencies": {
-        "@tsconfig/bun": "catalog:",
-        "@types/bun": "catalog:",
-        "@typescript/native-preview": "catalog:",
-      },
-    },
    "packages/merman": {
      "name": "@opencode-ai/merman",
      "version": "0.0.0",
@@ -679,7 +666,6 @@
      "dependencies": {
        "@opencode-ai/client": "workspace:*",
        "@opencode-ai/core": "workspace:*",
-        "@opencode-ai/plugin": "workspace:*",
        "@opencode-ai/schema": "workspace:*",
        "@opencode-ai/server": "workspace:*",
        "@opencode-ai/util": "workspace:*",
@@ -699,7 +685,6 @@
      "version": "1.18.4",
      "dependencies": {
        "@effect/platform-node": "catalog:",
-        "@effect/platform-node-shared": "catalog:",
        "@opencode-ai/core": "workspace:*",
        "@opencode-ai/protocol": "workspace:*",
        "@opencode-ai/schema": "workspace:*",
@@ -745,7 +730,6 @@
      },
      "devDependencies": {
        "@happy-dom/global-registrator": "20.0.11",
-        "@playwright/test": "catalog:",
        "@tsconfig/node22": "catalog:",
        "@types/bun": "catalog:",
        "@types/luxon": "catalog:",
@@ -851,7 +835,6 @@
        "@opencode-ai/client": "workspace:*",
        "@opencode-ai/session-ui": "workspace:*",
        "@opencode-ai/ui": "workspace:*",
-        "@playwright/test": "catalog:",
        "@solidjs/meta": "catalog:",
        "@storybook/addon-a11y": "10.4.4",
        "@storybook/addon-docs": "10.4.4",
@@ -893,7 +876,6 @@
      "dependencies": {
        "@opencode-ai/client": "workspace:*",
        "@opencode-ai/core": "workspace:*",
-        "@opencode-ai/latex": "workspace:*",
        "@opencode-ai/merman": "workspace:*",
        "@opencode-ai/plugin": "workspace:*",
        "@opencode-ai/schema": "workspace:*",
@@ -989,7 +971,6 @@
      "dependencies": {
        "@effect/opentelemetry": "catalog:",
        "@effect/platform-node": "catalog:",
-        "@effect/platform-node-shared": "catalog:",
        "@npmcli/arborist": "catalog:",
        "@npmcli/config": "10.8.1",
        "@opentelemetry/api": "1.9.0",
@@ -1195,6 +1176,8 @@

    "@ai-sdk/deepgram": ["@ai-sdk/deepgram@2.0.52", "", { "dependencies": { "@ai-sdk/provider": "3.0.15", "@ai-sdk/provider-utils": "4.0.46" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-8pcrQvEQCbvrrQKnD6hclBbI0hUgSrgyADykRbabxv/g9vPurfMC6n23J7dD+KZ3EcCoW+qz3IUIfySJ58gBOg=="],

+    "@ai-sdk/deepinfra": ["@ai-sdk/deepinfra@2.0.41", "", { "dependencies": { "@ai-sdk/openai-compatible": "2.0.37", "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.21" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-y6RoOP7DGWmDSiSxrUSt5p18sbz+Ixe5lMVPmdE7x+Tr5rlrzvftyHhjWHfqlAtoYERZTGFbP6tPW1OfQcrb4A=="],
+
    "@ai-sdk/deepseek": ["@ai-sdk/deepseek@2.0.47", "", { "dependencies": { "@ai-sdk/provider": "3.0.14", "@ai-sdk/provider-utils": "4.0.38" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-MzcQ321JO8OY+TVLFI81A7cIIuoeLLxrLCDD+8C1E3Ro6UFyfMtRXo9bw9OhTMRSDMo6hgSDOo4Fekz8aJtQYQ=="],

    "@ai-sdk/elevenlabs": ["@ai-sdk/elevenlabs@2.0.52", "", { "dependencies": { "@ai-sdk/provider": "3.0.15", "@ai-sdk/provider-utils": "4.0.46" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-ZgkausouWvO9U4ZtowNJ093bSNYOvH8zqls3uLC3+oxzWvbbTZO8SOdmFk0+gGafsXFJvq2yUX9+rEeJPwOJLw=="],
@@ -1221,6 +1204,8 @@

    "@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.23", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.6" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-z8GlDaCmRSDlqkMF2f4/RFgWxdarvIbyuk+m6WXT1LYgsnGiXRJGTD2Z1+SDl3LqtFuRtGX1aghYvQLoHL/9pg=="],

+    "@ai-sdk/togetherai": ["@ai-sdk/togetherai@2.0.41", "", { "dependencies": { "@ai-sdk/openai-compatible": "2.0.37", "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.21" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-k3p9e3k0/gpDDyTtvafsK4HYR4D/aUQW/kzCwWo1+CzdBU84i4L14gWISC/mv6tgSicMXHcEUd521fPufQwNlg=="],
+
    "@ai-sdk/vercel": ["@ai-sdk/vercel@2.0.39", "", { "dependencies": { "@ai-sdk/openai-compatible": "2.0.37", "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.21" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-8eu3ljJpkCTP4ppcyYB+NcBrkcBoSOFthCSgk5VnjaxnDaOJFaxnPwfddM7wx3RwMk2CiK1O61Px/LlqNc7QkQ=="],

    "@ai-sdk/xai": ["@ai-sdk/xai@3.0.123", "", { "dependencies": { "@ai-sdk/openai-compatible": "2.0.69", "@ai-sdk/provider": "3.0.15", "@ai-sdk/provider-utils": "4.0.46" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-WNASvd1C516oh2qYIj9EvAVPdU+Abads8DQWU6p9lQtvFFeGh8QW+3LDOARZd1GCINUFfw5yadEK845SMQKLsA=="],
@@ -1625,8 +1610,6 @@

    "@braintree/sanitize-url": ["@braintree/sanitize-url@7.1.2", "", {}, "sha512-jigsZK+sMF/cuiB7sERuo9V7N9jx+dhmHHnQyDSVdpZwVutaBu7WvNYqMDLSgFgfB30n452TP3vjDAvFC973mA=="],

-    "@brendonovich/vite-plugin-opencode": ["@brendonovich/vite-plugin-opencode@0.1.1", "", { "dependencies": { "@babel/core": "^7.29.0", "@opencode-ai/client": "0.0.0-beta-18050" }, "peerDependencies": { "vite": "^6.0.0 || ^7.0.0 || ^8.0.0" } }, "sha512-aPG0ct8ctxAqndbNOx7NW0GhU6QY6sOUfi/DaKqH9c5WdxICSsUop6uSkJwPDHP9WpN9eg0dd2D2qwYpG6UdHw=="],
-
    "@bruits/satteri-darwin-arm64": ["@bruits/satteri-darwin-arm64@0.9.5", "", { "os": "darwin", "cpu": "arm64" }, "sha512-iw4nZgx9v30lWo/MTngQqi1pI78KI0DnkSm+lVJGYdmPLgAyDNJigVhpG42/Iq55A6c1Ll8q66ljyyRiQUxwow=="],

    "@bruits/satteri-darwin-x64": ["@bruits/satteri-darwin-x64@0.9.5", "", { "os": "darwin", "cpu": "x64" }, "sha512-6T26Z5Kf3cFW2PSlk9p7zT7yVxvuBSiJvYyz9u8KjYwMTqZyIDOj2wDyNpxKV4+6yUVG7rddq2QwvG/8LJA2+Q=="],
@@ -2165,8 +2148,6 @@

    "@opencode-ai/httpapi-codegen": ["@opencode-ai/httpapi-codegen@workspace:packages/httpapi-codegen"],

-    "@opencode-ai/latex": ["@opencode-ai/latex@workspace:packages/latex"],
-
    "@opencode-ai/merman": ["@opencode-ai/merman@workspace:packages/merman"],

    "@opencode-ai/plugin": ["@opencode-ai/plugin@workspace:packages/plugin"],
@@ -2175,20 +2156,6 @@

    "@opencode-ai/protocol": ["@opencode-ai/protocol@workspace:packages/protocol"],

-    "@opencode-ai/pty": ["@opencode-ai/pty@0.1.12", "", { "optionalDependencies": { "@opencode-ai/pty-darwin-arm64": "0.1.12", "@opencode-ai/pty-darwin-x64": "0.1.12", "@opencode-ai/pty-linux-arm64-gnu": "0.1.12", "@opencode-ai/pty-linux-arm64-musl": "0.1.12", "@opencode-ai/pty-linux-x64-gnu": "0.1.12", "@opencode-ai/pty-linux-x64-musl": "0.1.12" }, "bin": { "opencode-pty": "bin/opencode-pty.js" } }, "sha512-dl4FyJUhTXThsWYY8txG/8/nwN7dE0M5Sic9r4L9f2pvtJnbR5zrCrPoiPIBIxZle1wVks1dhz4z/CfqLf5sCg=="],
-
-    "@opencode-ai/pty-darwin-arm64": ["@opencode-ai/pty-darwin-arm64@0.1.12", "", { "os": "darwin", "cpu": "arm64" }, "sha512-tMvoriq3VegVlj1uEglc6qE0M7VXy61nyf9Si7tTO7xa8JiyxuFJSXOZ1pGeErDu+pe24hvTyVOR+gkdew8w9g=="],
-
-    "@opencode-ai/pty-darwin-x64": ["@opencode-ai/pty-darwin-x64@0.1.12", "", { "os": "darwin", "cpu": "x64" }, "sha512-Sn5vMLL5giHOhx7J5H6zwDp4YjjXorY+QV0IEYY+SCT4wQfRBliokIyj23pRl6P2RK3u9bDLXJHDNMfDVZ2Rxg=="],
-
-    "@opencode-ai/pty-linux-arm64-gnu": ["@opencode-ai/pty-linux-arm64-gnu@0.1.12", "", { "os": "linux", "cpu": "arm64" }, "sha512-HbnlKZy052l7G527wK0+05EXaUpZ4ykVAmNBEzqWCoi4TeQj2+Nr9kJ9trx9o1KrVcT4Ki58CCvN5QOls6Z0yQ=="],
-
-    "@opencode-ai/pty-linux-arm64-musl": ["@opencode-ai/pty-linux-arm64-musl@0.1.12", "", { "os": "linux", "cpu": "arm64" }, "sha512-2nTN7ggu1h9XgjNcoQMYjP5sirfYnAskpdFCOqjokLqhytX/IMMmkRTQs+foaEaPz0dAIQD3DQplR2jZIgxp1w=="],
-
-    "@opencode-ai/pty-linux-x64-gnu": ["@opencode-ai/pty-linux-x64-gnu@0.1.12", "", { "os": "linux", "cpu": "x64" }, "sha512-FnD5ndnObTQKAoaVvxLKi5W+r3/+dsaMsobz6uK0B9hlmffXxY5CQ6HQyUU/h3aIKLWXxho5XYkA2b9yrp8/gA=="],
-
-    "@opencode-ai/pty-linux-x64-musl": ["@opencode-ai/pty-linux-x64-musl@0.1.12", "", { "os": "linux", "cpu": "x64" }, "sha512-prkrNu6uvjqoffxdGiDHSU5C0Y+kCSfv+lslu7dfRPgPKenVELNpRTAbOduyrWPac2vGt8j5NM61icJyodbJmA=="],
-
    "@opencode-ai/schema": ["@opencode-ai/schema@workspace:packages/schema"],

    "@opencode-ai/script": ["@opencode-ai/script@workspace:packages/script"],
@@ -5941,6 +5908,10 @@

    "@ai-sdk/deepgram/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.46", "", { "dependencies": { "@ai-sdk/provider": "3.0.15", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.8", "undici": "^6.28.0" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-tEtld97plCFiYevsJuOkGkeuhQndeMWFBVrJS4AjnbD5AqrNSXRCe0p+BZ3Cju/sxDeeZ9ym3q9YUV8fASA7aQ=="],

+    "@ai-sdk/deepinfra/@ai-sdk/openai-compatible": ["@ai-sdk/openai-compatible@2.0.37", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.21" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-+POSFVcgiu47BK64dhsI6OpcDC0/VAE2ZSaXdXGNNhpC/ava++uSRJYks0k2bpfY0wwCTgpAWZsXn/dG2Yppiw=="],
+
+    "@ai-sdk/deepinfra/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.21", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.6" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-MtFUYI1/8mgDvRmaBDjbLJPFFrMG777AvSgyIFQtZHIMzm88R/12vYBBpnk7pfiWLFE1DSZzY4WDYzGbKAcmiw=="],
+
    "@ai-sdk/deepseek/@ai-sdk/provider": ["@ai-sdk/provider@3.0.14", "", { "dependencies": { "json-schema": "^0.4.0" } }, "sha512-5X1k57JBJ4H7H1QjX7CnJYAB1I19r/trVZTMcSms7/kLNZ8RaU4Nt2agcwZzv82Hfx6Q7/TOLU7agAKeFfc8cA=="],

    "@ai-sdk/deepseek/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.38", "", { "dependencies": { "@ai-sdk/provider": "3.0.14", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.8" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-/HHGmtKllqjg1OLc023v9w9kK3laW7Z6TzfZukYQWCsGBbzB9p60zTvvpXFVcs44NZBVXL3viOa1HRKUbeee8g=="],
@@ -5979,6 +5950,10 @@

    "@ai-sdk/perplexity/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.21", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.6" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-MtFUYI1/8mgDvRmaBDjbLJPFFrMG777AvSgyIFQtZHIMzm88R/12vYBBpnk7pfiWLFE1DSZzY4WDYzGbKAcmiw=="],

+    "@ai-sdk/togetherai/@ai-sdk/openai-compatible": ["@ai-sdk/openai-compatible@2.0.37", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.21" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-+POSFVcgiu47BK64dhsI6OpcDC0/VAE2ZSaXdXGNNhpC/ava++uSRJYks0k2bpfY0wwCTgpAWZsXn/dG2Yppiw=="],
+
+    "@ai-sdk/togetherai/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.21", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.6" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-MtFUYI1/8mgDvRmaBDjbLJPFFrMG777AvSgyIFQtZHIMzm88R/12vYBBpnk7pfiWLFE1DSZzY4WDYzGbKAcmiw=="],
+
    "@ai-sdk/vercel/@ai-sdk/openai-compatible": ["@ai-sdk/openai-compatible@2.0.37", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.21" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-+POSFVcgiu47BK64dhsI6OpcDC0/VAE2ZSaXdXGNNhpC/ava++uSRJYks0k2bpfY0wwCTgpAWZsXn/dG2Yppiw=="],

    "@ai-sdk/vercel/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.21", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.6" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-MtFUYI1/8mgDvRmaBDjbLJPFFrMG777AvSgyIFQtZHIMzm88R/12vYBBpnk7pfiWLFE1DSZzY4WDYzGbKAcmiw=="],
@@ -6157,8 +6132,6 @@

    "@babel/preset-env/semver": ["semver@6.3.1", "", { "bin": { "semver": "bin/semver.js" } }, "sha512-BR7VvDCVHO+q2xBEWskxS6DJE1qRnb7DxzUrogb71CWoSficBxYsiAGd+Kl0mmq/MprG9yArRkyrQxTO6XjMzA=="],

-    "@brendonovich/vite-plugin-opencode/@opencode-ai/client": ["@opencode-ai/client@0.0.0-beta-18050", "", { "dependencies": { "@opencode-ai/protocol": "0.0.0-beta-18050", "@opencode-ai/schema": "0.0.0-beta-18050" }, "peerDependencies": { "effect": "4.0.0-rc.111", "solid-js": ">=1.9.0" }, "optionalPeers": ["effect", "solid-js"] }, "sha512-zWZv5X23iyx+/mxwiAi18YY/VMjQofTaH7RyKMBt7KL6FmaKAWf9Q05zbQhrLX8DYJD3MOtbjBeQCGLsPUDU8g=="],
-
    "@bruits/satteri-wasm32-wasi/@emnapi/core": ["@emnapi/core@1.11.1", "", { "dependencies": { "@emnapi/wasi-threads": "1.2.2", "tslib": "^2.4.0" } }, "sha512-RSvbQmHzdKzNsLYa/wHrbc3KN4sYLKAdPZxqiM2HATqv/SBk2/ENSHpvXGaLOMcsAyz0poEGqkmmKYG3OWiJEQ=="],

    "@bruits/satteri-wasm32-wasi/@emnapi/runtime": ["@emnapi/runtime@1.11.1", "", { "dependencies": { "tslib": "^2.4.0" } }, "sha512-vgj7R3y3Wgx24IQaGPA/R6YFXLHVMOZ0uVEyIQPaWs+rd1AzfEMXlAC22FYwO1XkKR6NPsq7mUandH8oIRdZFw=="],
@@ -6983,10 +6956,6 @@

    "@babel/helper-compilation-targets/lru-cache/yallist": ["yallist@3.1.1", "", {}, "sha512-a4UGQaWPH59mOXUYnAG2ewncQS4i4F43Tv3JoAM+s2VDAmS9NsK8GpDMLrCHPksFT7h3K6TOoUNn2pb7RoXx4g=="],

-    "@brendonovich/vite-plugin-opencode/@opencode-ai/client/@opencode-ai/protocol": ["@opencode-ai/protocol@0.0.0-beta-18050", "", { "dependencies": { "@opencode-ai/schema": "0.0.0-beta-18050", "effect": "4.0.0-rc.111" } }, "sha512-HDQMnvGp8IU0MdBRbEuydX1WQm09BZ4HJm9iSMQwzweJuQ2HNscgzHJPIH6P02BsbbtfJ8J7sZGPItrz1tWSgw=="],
-
-    "@brendonovich/vite-plugin-opencode/@opencode-ai/client/@opencode-ai/schema": ["@opencode-ai/schema@0.0.0-beta-18050", "", { "dependencies": { "@standard-schema/spec": "1.1.0", "effect": "4.0.0-rc.111" } }, "sha512-/D6VXaWlytTXR3IOiMLIKuPcfp7FQNUzRPm9z3K7UBFd1Bw4q/WZksaf5RVcBGz+0YRxYMc1V4D7MFlceSgtyg=="],
-
    "@bruits/satteri-wasm32-wasi/@emnapi/core/@emnapi/wasi-threads": ["@emnapi/wasi-threads@1.2.2", "", { "dependencies": { "tslib": "^2.4.0" } }, "sha512-c95qOXkHdydNKhscBTebqEC1CVAZpyqOfVfBzQ1qgzyl3gfeldUjIggDbIZgDKsHLgnsM+igH7TJ/eAasaVuMA=="],

    "@electron/asar/minimatch/brace-expansion": ["brace-expansion@1.1.18", "", { "dependencies": { "balanced-match": "^1.0.0", "concat-map": "0.0.1" } }, "sha512-Edep/X9fGqVNmzKBVsDYIOtD+z1tuezV70LBjdCst9Tqu76lsnvRiZ6oTic1n+/BIwX6QDGAO94PN4N2SADvtw=="],
@@ -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-iYdVrLtyKmjlyypisF9SqzgyriWT90kSCh3crxw9AKU=",
-    "aarch64-linux": "sha256-BV2t4w5ujArbtSC/Qfm3gLzevQW9A6hMgOyPVp94g/o=",
-    "aarch64-darwin": "sha256-EwMq7zaxzzcsmH0Pjqu4ftGdcM8Lna8mvHgKzRcVI8g=",
-    "x86_64-darwin": "sha256-PokzxlkQy6JvHADF2ZMIIDI1u9ZjSNNedpmR9gvHS5c="
+    "x86_64-linux": "sha256-2bkzaLe/n63btVRQNhu8LXCtMZJArX1Kedi5U40l1xw=",
+    "aarch64-linux": "sha256-5Cs9M3hvDKAymo71y8oZ7jj3pEm+MI+HHhuSuV7UvtM=",
+    "aarch64-darwin": "sha256-LsJcuxE/NMu+vUFdpBKHc2z0sC0C5bRMlH1Kj+ns9dY=",
+    "x86_64-darwin": "sha256-KDjmKC3JZD8I5A7gi+dYIl0dgHVt20/DwkM9RKBWiJk="
  }
 }
@@ -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),
    })
  }),
 )
@@ -1,5 +1,5 @@
 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"
@@ -69,22 +69,14 @@ export interface OptionsInput {
  // 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 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
-  }
+  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
@@ -267,11 +259,7 @@ const AnthropicToolChoice = Schema.Union([
    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.tag("tool"), name: Schema.String, disable_parallel_tool_use: Schema.optional(Schema.Boolean) }),
 ])

 const AnthropicThinking = Schema.Union([
@@ -373,8 +361,6 @@ 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 +371,13 @@ 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 +390,6 @@ 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 +424,18 @@ 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 => ({
@@ -512,21 +495,14 @@ 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 => {
@@ -574,7 +550,9 @@ const documentContextFromMetadata = (metadata: MediaPart["metadata"]): string |
  return undefined
 }

-const citationsFromMetadata = (metadata: MediaPart["metadata"]): AnthropicDocumentBlock["citations"] | 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)
@@ -724,7 +702,8 @@ const lowerToolResultContent = Effect.fnUntraced(function* (part: ToolResultPart
 })

 const requireThinkingSignature = (request: LLMRequest) => {
-  if (request.model.compatibility?.requireSignature !== undefined) return request.model.compatibility.requireSignature
+  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()
@@ -761,12 +740,9 @@ 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 +784,12 @@ 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
      }
@@ -854,8 +829,8 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
        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)
+          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
@@ -884,7 +859,7 @@ 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(
@@ -918,24 +893,21 @@ 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 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)
+    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 })
+      ? ((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 ??
@@ -986,7 +958,8 @@ const resolveThinking = Effect.fn("AnthropicMessages.resolveThinking")(function*
    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")
+    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 =
@@ -1074,7 +1047,7 @@ 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 +1061,7 @@ 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 +1070,7 @@ 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 +1088,7 @@ 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 +1106,7 @@ 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 +1120,7 @@ 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 }),
  })
 }

@@ -1158,14 +1129,11 @@ 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]

@@ -1211,16 +1179,14 @@ const onContentBlockStart = (
  if (block.type === "thinking" && block.thinking !== undefined) {
    const events: LLMEvent[] = []
    const id = `reasoning-${event.index ?? 0}`
-    const metadata =
-      block.signature === undefined
-        ? undefined
-        : providerMetadata(state.providerMetadataKey, { signature: block.signature })
-    const lifecycle = Lifecycle.reasoningStart(state.lifecycle, events, id, metadata)
+    const providerMetadata =
+      block.signature === undefined ? undefined : anthropicMetadata({ signature: block.signature })
+    const lifecycle = Lifecycle.reasoningStart(state.lifecycle, events, id, providerMetadata)
    return [
      {
        ...state,
        lifecycle: block.thinking
-          ? Lifecycle.reasoningDelta(lifecycle, events, id, block.thinking, metadata)
+          ? Lifecycle.reasoningDelta(lifecycle, events, id, block.thinking, providerMetadata)
          : lifecycle,
        reasoningSignatures:
          event.index === undefined || block.signature === undefined
@@ -1243,14 +1209,14 @@ const onContentBlockStart = (
          state.lifecycle,
          events,
          `reasoning-${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 +1224,11 @@ 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 +1237,6 @@ 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 +1249,6 @@ 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-${index}`)) return [state, NO_EVENTS] satisfies StepResult
    return [
      {
        ...state,
@@ -1330,7 +1293,7 @@ const onContentBlockStop = Effect.fn("AnthropicMessages.onContentBlockStop")(fun
        Lifecycle.textEnd(state.lifecycle, events, `text-${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 +1301,8 @@ 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 +1315,7 @@ 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,
@@ -1396,70 +1356,23 @@ const onError = (event: AnthropicEvent) =>
    }),
  )

-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 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") {
+    const block = event.content_block
+    if (block && (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.fail(
+          ProviderShared.eventError(ADAPTER, `Anthropic tool_use missing id at index ${event.index}`),
+        )
    }
-    return Effect.succeed(onContentBlockStart(state, { ...event, content_block: block }))
-  }
-  if (event.type === "content_block_delta") {
-    if (!ProviderShared.isRecord(event.delta)) return invalidStreamEvent(event)
-    if (typeof event.delta.type === "string" && !isKnownStreamDeltaType(event.delta.type))
-      return Effect.succeed<StepResult>([state, NO_EVENTS])
-    const decoded = decodeAnthropicStreamDelta(event.delta)
-    if (Option.isNone(decoded)) return invalidStreamEvent(event)
-    return onContentBlockDelta(state, { ...event, delta: decoded.value })
+    return Effect.succeed(onContentBlockStart(state, event))
  }
+  if (event.type === "content_block_delta") return onContentBlockDelta(state, event)
  if (event.type === "content_block_stop") return onContentBlockStop(state, event)
-  if (event.type === "message_delta") {
-    const decoded = decodeAnthropicStreamDelta(event.delta)
-    if (Option.isNone(decoded)) return invalidStreamEvent(event)
-    return Effect.succeed(onMessageDelta(state, { ...event, delta: decoded.value }))
-  }
+  if (event.type === "message_delta") return Effect.succeed(onMessageDelta(state, event))
  if (event.type === "message_stop") return onMessageStop(state)
  if (event.type === "error") return onError(event)
  return Effect.succeed<StepResult>([state, NO_EVENTS])
@@ -1481,8 +1394,7 @@ export const protocol = Protocol.make({
  },
  stream: {
    event: Protocol.jsonEvent(AnthropicEvent),
-    initial: (request) => ({
-      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
+    initial: () => ({
      tools: ToolStream.empty<number>(),
      reasoningSignatures: {},
      lifecycle: Lifecycle.initial(),
@@ -1496,9 +1408,10 @@ export const route = Route.make({
  provider: "anthropic",
  providerMetadataKey: "anthropic",
  protocol,
-  endpoint: Endpoint.path((input) => (input.request.model.provider === "anthropic" ? `${PATH}?beta=true` : PATH), {
-    baseURL: DEFAULT_BASE_URL,
-  }),
+  endpoint: Endpoint.path(
+    (input) => (input.request.model.provider === "anthropic" ? `${PATH}?beta=true` : 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,
@@ -644,16 +650,22 @@ const step = (state: ParserState, event: BedrockEvent) =>
      ] as const
    }

-    if (event.exception) {
+    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:
-            event.exception.details.message ??
-            event.exception.details.originalMessage ??
-            "Bedrock Converse stream error",
-          code: event.exception.type,
+          message: exception[1]?.message ?? exception[1]?.originalMessage ?? "Bedrock Converse stream error",
+          code: exception[0],
        }),
      })
    }
@@ -696,8 +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,
@@ -82,9 +82,7 @@ const consumeFrames = (route: string) => (state: FrameBufferState, chunk: Uint8A
        "Failed to parse Bedrock Converse event-stream payload",
      )) as Record<string, unknown>
      delete parsed.p
-      out.push(
-        messageType === "exception" ? { exception: { type: eventType, details: parsed } } : { [eventType]: parsed },
-      )
+      out.push({ [eventType]: parsed })
    }
    return [cursor, out] as const
  })
@@ -229,7 +229,6 @@ type GeminiEvent = Schema.Schema.Type<typeof GeminiEvent>

 interface ParserState {
  readonly route: string
-  readonly providerMetadataKey: string
  readonly finishReason?: string
  readonly hasToolCalls: boolean
  readonly promptFeedback?: GeminiPromptFeedback
@@ -286,23 +285,22 @@ 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) => ({
+const lowerToolCall = (part: ToolCallPart, omitIds: boolean) => ({
  functionCall: { ...(omitIds ? {} : { id: part.id }), name: part.name, args: part.input },
-  thoughtSignature: thoughtSignature(part.providerMetadata, metadataKey),
+  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
@@ -344,19 +342,15 @@ 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, omitCallIds)
          const signature = lowered.thoughtSignature
          parts.push({
            ...lowered,
@@ -504,7 +498,7 @@ 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.
@@ -525,7 +519,7 @@ const mapUsage = (usage: GeminiUsage | undefined, metadataKey: string) => {
    cacheReadInputTokens: cached,
    reasoningTokens: thoughts,
    totalTokens: ProviderShared.totalTokens(promptTokens, outputTokens, usage.totalTokenCount ?? undefined),
-    providerMetadata: providerMetadata(metadataKey, usage),
+    providerMetadata: { google: usage },
  })
 }

@@ -573,15 +567,10 @@ 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:
@@ -590,9 +579,7 @@ 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
 }
@@ -601,9 +588,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
  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)
@@ -647,7 +632,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
          events,
          "reasoning-0",
          part.text,
-          signature ? providerMetadata(state.providerMetadataKey, { thoughtSignature: signature }) : undefined,
+          signature ? googleMetadata({ thoughtSignature: signature }) : undefined,
        )
        continue
      }
@@ -655,16 +640,14 @@ 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
@@ -684,9 +667,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
        lifecycle,
        events,
        "reasoning-0",
-        reasoningSignature
-          ? providerMetadata(state.providerMetadataKey, { thoughtSignature: reasoningSignature })
-          : undefined,
+        reasoningSignature ? googleMetadata({ thoughtSignature: reasoningSignature }) : undefined,
      )
      lifecycle = Lifecycle.stepStart(lifecycle, events)
      events.push(
@@ -694,9 +675,8 @@ const step = (state: ParserState, event: GeminiEvent) => {
          id,
          name: part.functionCall.name,
          input,
-          providerMetadata: part.thoughtSignature
-            ? providerMetadata(state.providerMetadataKey, { thoughtSignature: part.thoughtSignature })
-            : undefined,
+          providerMetadata:
+            part.thoughtSignature ? googleMetadata({ thoughtSignature: part.thoughtSignature }) : undefined,
        }),
      )
      hasToolCalls = true
@@ -734,7 +714,6 @@ export const protocol = Protocol.make({
    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(),
    }),
@@ -154,12 +154,7 @@ 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,
          },
        }
      }),
@@ -79,60 +79,10 @@ 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.
@@ -161,6 +111,7 @@ 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 +124,10 @@ 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 +140,7 @@ 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
 }
@@ -344,17 +288,14 @@ export const Event = Schema.StructWithRest(
    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 +314,9 @@ export const Event = Schema.StructWithRest(
 )
 export type Event = Schema.Schema.Type<typeof Event>

+// Which lowered input item a persisted item id is about to be attached to.
+export type ItemKind = "message" | "reasoning" | "function-call" | "reference"
+
 export interface Extension {
  readonly id: string
  readonly name: string
@@ -381,7 +325,10 @@ export interface Extension {
    readonly media: ProviderShared.NormalizedMedia
    readonly request: LLMRequest
  }) => MediaInput | undefined
-  readonly lowerHostedToolItem?: (item: unknown) => ExtendedHostedToolItem | undefined
+  // Optional grammar check applied before a persisted item id is resent as
+  // part of replayed history. Returning false drops the id; every lowered
+  // item treats a dropped id the same as an absent one.
+  readonly acceptsItemID?: (kind: ItemKind, id: string) => boolean
 }

 const BASE: Extension = { id: ADAPTER, name: NAME }
@@ -393,10 +340,10 @@ 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 messagePhases: Readonly<Record<string, MessagePhase | null>>
  readonly reasoningItems: Readonly<Record<string, ReasoningStreamItem>>
+  readonly store: boolean | undefined
 }

 type ReasoningSummaryStatus = "active" | "can-conclude" | "concluded"
@@ -441,37 +388,53 @@ 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.
+// Servers validate item ids on replayed history, and a malformed or oversized
+// id can fail an otherwise valid request. Only server-issued tokens are worth
+// resending; anything else is treated as absent so the item is resent without
+// an id (or skipped, for items that cannot be expressed without one).
+const ITEM_ID_PATTERN = /^[A-Za-z0-9_-]{1,64}$/
 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" &&
+    ITEM_ID_PATTERN.test(metadata.itemId)
+    ? metadata.itemId
+    : undefined
 }

-const lowerToolCall = (part: ToolCallPart, providerMetadataKey: string): OpenResponsesInputItem => {
+const acceptsItemID = (extension: Extension, kind: ItemKind, id: string | undefined): id is string =>
+  id !== undefined && (extension.acceptsItemID?.(kind, id) ?? true)
+
+const lowerToolCall = (
+  part: ToolCallPart,
+  providerMetadataKey: string,
+  extension: Extension,
+): OpenResponsesInputItem => {
  const id = itemID(part.providerMetadata, providerMetadataKey)
  return {
    type: "function_call",
-    ...(id === undefined ? {} : { id }),
+    ...(acceptsItemID(extension, "function-call", id) ? { id } : {}),
    call_id: part.id,
    name: part.name,
    arguments: ProviderShared.encodeJson(part.input),
  }
 }

-const lowerReasoning = (part: ReasoningPart, providerMetadataKey: string): OpenResponsesReasoningInput | undefined => {
+const lowerReasoning = (
+  part: ReasoningPart,
+  providerMetadataKey: string,
+  extension: Extension,
+): OpenResponsesReasoningInput | undefined => {
  const metadata = part.providerMetadata?.[providerMetadataKey]
-  if (!ProviderShared.isRecord(metadata)) return undefined
  const id = itemID(part.providerMetadata, providerMetadataKey)
+  if (!ProviderShared.isRecord(metadata) || !acceptsItemID(extension, "reasoning", 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,
  }
@@ -562,7 +525,10 @@ 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) {
@@ -585,14 +551,16 @@ 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<
          Array<{ id: string | undefined; phase: MessagePhase | null | undefined; parts: TextPart[] }>
        >((groups, part) => {
          const metadata = part.providerMetadata?.[providerMetadataKey]
-          const id = itemID(part.providerMetadata, providerMetadataKey)
+          const rawID = itemID(part.providerMetadata, providerMetadataKey)
+          const id = acceptsItemID(extension, "message", rawID) ? rawID : undefined
          const phase = ProviderShared.isRecord(metadata) ? messagePhase(metadata.phase) : undefined
          const group = groups.at(-1)
          if (group && group.id === id && group.phase === phase) group.parts.push(part)
@@ -617,51 +585,51 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
        }
        if (part.type === "reasoning") {
          flushText()
-          const reasoning = lowerReasoning(part, providerMetadataKey)
+          const reasoning = lowerReasoning(part, providerMetadataKey, extension)
          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
        }
        if (part.type === "tool-call") {
          flushText()
          if (part.providerExecuted === true) continue
-          input.push(lowerToolCall(part, providerMetadataKey))
+          input.push(lowerToolCall(part, providerMetadataKey, extension))
          continue
        }
        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)
+          const reference = acceptsItemID(extension, "reference", id) ? id : undefined
+          if (store !== false && reference && !hostedToolReferences.has(reference))
+            input.push({ type: "item_reference", id: reference })
+          if (store === false) {
+            // The server is not storing this exchange, so the tool outcome has to
+            // travel in the input. Non-content results degrade to their text form.
+            const content: ReadonlyArray<Content> =
+              part.result.type === "content"
                ? 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
+                : [{ type: "text", text: ProviderShared.toolResultText(part) }]
+            input.push({
+              role: "user",
+              content: yield* Effect.forEach(content, (item) =>
+                lowerHostedToolResultContentItem(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 (reference) hostedToolReferences.add(reference)
          continue
        }
        return yield* ProviderShared.unsupportedContent(extension.name, "assistant", [
@@ -686,16 +654,22 @@ 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 cacheKey = ProviderShared.clampPromptCacheKey(request.promptCacheKey)
  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 } : {}),
@@ -813,7 +787,7 @@ 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>]

@@ -843,9 +817,6 @@ 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]
@@ -892,7 +863,7 @@ 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) {
+  if (item?.type === "message" && item.id) {
    const phase = messagePhase(item.phase)
    return [
      {
@@ -921,28 +892,30 @@ const onOutputItemAdded = (state: ParserState, event: Event): StepResult => {
      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]
+  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]
  if (event.summary_index === 0) return [state, NO_EVENTS]
@@ -989,24 +962,34 @@ 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,
  ]
 }

@@ -1014,7 +997,7 @@ const onFunctionCallArgumentsDelta = Effect.fn("OpenResponses.onFunctionCallArgu
  state: ParserState,
  event: Event,
 ) {
-  if (event.item_id === undefined) return [state, NO_EVENTS] satisfies StepResult
+  if (!event.item_id) 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
@@ -1045,7 +1028,7 @@ 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) {
+  if (item.type === "message" && item.id) {
    const itemPhase = messagePhase(item.phase)
    const phase = itemPhase === undefined ? state.messagePhases[item.id] : itemPhase
    const events: LLMEvent[] = []
@@ -1069,19 +1052,18 @@ 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
@@ -1129,50 +1111,30 @@ 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.
@@ -1219,14 +1181,9 @@ export const providerFailure = (id: string, event: Event, fallback: string) => {

 const providerError = (state: ParserState, event: Event, fallback: string) => providerFailure(state.id, event, fallback)

-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
+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)
@@ -1235,7 +1192,7 @@ 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")
+    if (!event.item_id || typeof value !== "string")
      return ProviderShared.eventError(state.id, `${event.type} is malformed`)
    return Effect.succeed(
      event.type === "response.refusal.delta"
@@ -1244,7 +1201,7 @@ export const step = (state: ParserState, input: Event) => {
    )
  }
  if (event.type === "response.reasoning.delta" || event.type === "response.reasoning_summary_text.delta") {
-    if (event.item_id === undefined) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
+    if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
    return Effect.succeed(onReasoningDelta(state, event, event.item_id))
  }
  if (
@@ -1252,36 +1209,28 @@ export const step = (state: ParserState, input: Event) => {
    event.type === "response.reasoning_summary_text.done" ||
    event.type === "response.reasoning_text.done"
  ) {
-    if (event.item_id === undefined) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
+    if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
    return Effect.succeed(onReasoningDone(state, event, event.item_id))
  }
  if (event.type === "response.reasoning_summary_part.added")
-    return event.item_id !== undefined
+    return event.item_id
      ? Effect.succeed(onReasoningSummaryPartAdded(state, event))
      : ProviderShared.eventError(state.id, `${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
+    return event.item_id
      ? onFunctionCallArgumentsDelta(state, event)
      : ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
  if (event.type === "response.output_item.done") {
-    if (event.item?.type === "message" && event.item.id === undefined)
+    if (event.item?.type === "message" && !event.item.id)
      return ProviderShared.eventError(state.id, `${event.type} message is missing id`)
    return onOutputItemDone(state, event)
  }
@@ -1309,10 +1258,10 @@ export const initial = (request: LLMRequest, extension: Extension = BASE): Parse
  hasFunctionCall: false,
  tools: ToolStream.empty<string>(),
  lifecycle: Lifecycle.initial(),
-  outputItems: {},
  messageItems: new Set<string>(),
  messagePhases: {},
  reasoningItems: {},
+  store: OpenResponsesOptions.resolve(request).store,
 })

 export const protocol = Protocol.make({
@@ -253,7 +253,6 @@ interface PendingToolDelta {
 }

 export interface ParserState {
-  readonly providerMetadataKey: string
  readonly tools: ToolStream.State<number>
  readonly pendingTools: Partial<Record<number, PendingToolDelta>>
  readonly toolCallEvents: ReadonlyArray<LLMEvent>
@@ -279,7 +278,6 @@ interface LoweringOptions {
  readonly cacheControl?: (
    cache: CacheHint | undefined,
  ) => Schema.Schema.Type<typeof OpenAIChatCacheControl> | undefined
-  readonly toolCallID?: (id: string) => string
 }

 const lowerTool = (
@@ -306,8 +304,8 @@ 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,
@@ -325,18 +323,17 @@ 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
 }

@@ -366,9 +363,8 @@ 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[] = []
@@ -385,31 +381,25 @@ 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)
@@ -437,7 +427,7 @@ 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),
      })
@@ -447,7 +437,7 @@ 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),
    })
@@ -463,13 +453,11 @@ 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
 })

@@ -490,43 +478,12 @@ 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) {
@@ -569,19 +526,14 @@ 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
@@ -603,10 +555,7 @@ const hasToolHistory = (messages: ReadonlyArray<LLMRequest["messages"][number]>)
 // 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 detectMaxTokensField = (provider: string, baseURL: string | undefined): "max_tokens" | "max_completion_tokens" => {
  const p = provider.toLowerCase()
  const url = (baseURL ?? "").toLowerCase()
  if (
@@ -656,8 +605,7 @@ const detectSupportsStore = (provider: string, baseURL: string | undefined): boo
  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 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 =
@@ -689,7 +637,11 @@ const detectSupportsStrictMode = (provider: string, baseURL: string | undefined)
  return !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia
 }

-const detectZaiToolStream = (provider: string, baseURL: string | undefined, modelID: string): boolean => {
+const detectZaiToolStream = (
+  provider: string,
+  baseURL: string | undefined,
+  modelID: string,
+): boolean => {
  const p = provider.toLowerCase()
  const url = (baseURL ?? "").toLowerCase()
  const isZai =
@@ -707,7 +659,7 @@ const detectZaiToolStream = (provider: string, baseURL: string | undefined, mode

 const lowerOptions = (request: LLMRequest, supportsStore: boolean) => {
  const options = OpenAIOptions.resolve(request)
-  const cacheKey = ProviderShared.promptCacheKey(request)
+  const cacheKey = ProviderShared.clampPromptCacheKey(request.promptCacheKey)
  return {
    ...(supportsStore && options.store !== undefined ? { store: options.store } : {}),
    // For providers that support `store`, ensure stateless `store:false` is sent
@@ -739,10 +691,10 @@ export const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (
  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 supportsStrictMode = request.model.compatibility?.supportsStrictMode ?? detectSupportsStrictMode(provider, baseURL)
  const zaiToolStream =
-    request.model.compatibility?.zaiToolStream ?? detectZaiToolStream(provider, baseURL, request.model.id)
+    request.model.compatibility?.zaiToolStream ??
+    detectZaiToolStream(provider, baseURL, request.model.id)
  const hasHistory = hasToolHistory(request.messages)
  const hasActiveTools = request.tools.length > 0
  return {
@@ -827,14 +779,15 @@ const mapFinishReason = Effect.fn("OpenAIChat.mapFinishReason")(function* (event
 // 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 ??
+      (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 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))
@@ -846,7 +799,7 @@ 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 },
  })
 }

@@ -920,12 +873,8 @@ 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 } : {}),
  },
@@ -952,17 +901,15 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
    // 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 usage = mapUsage(event.usage) ?? (choiceUsage ? mapUsage(choiceUsage) : 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
+        ? {
+            normalized: yield* mapFinishReason(event, rawFinishReason),
+            raw: choice?.native_finish_reason ?? rawFinishReason,
+          }
+        : state.finishReason
    const delta = choice?.delta
    const toolDeltas = delta?.tool_calls ?? []
    let tools = state.tools
@@ -993,7 +940,7 @@ 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 (
@@ -1009,11 +956,7 @@ 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)
    }
@@ -1023,11 +966,7 @@ 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)
    }
@@ -1088,7 +1027,6 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>

    return [
      {
-        providerMetadataKey: state.providerMetadataKey,
        tools: finished?.tools ?? tools,
        pendingTools,
        toolCallEvents: finished?.events ?? state.toolCallEvents,
@@ -1132,18 +1070,12 @@ const finishEvents = Effect.fn("OpenAIChat.finishEvents")(function* (state: Pars
      }
    : { normalized: hasToolCalls ? ("tool-calls" as const) : ("stop" 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
@@ -1170,7 +1102,6 @@ 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: [],
@@ -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"
@@ -7,7 +7,7 @@ import { Protocol } from "../route/protocol.js"
 import { HttpTransport } from "../route/transport/index.js"
 import { 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 { OpenAIImage } from "./utils/openai-image.js"
 import { ResponsesHostedTools } from "./utils/responses-hosted-tools.js"
 import { ToolSchemaProjection } from "./utils/tool-schema.js"
@@ -32,40 +32,6 @@ 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([
@@ -75,7 +41,6 @@ const OpenAIResponsesToolChoice = Schema.Union([

 const OpenAIResponsesCoreFields = {
  ...OpenResponses.coreFields,
-  input: Schema.Array(Schema.Union([OpenResponses.InputItem, OpenAIResponsesHostedToolItem])),
  tools: optionalArray(OpenAIResponsesTools),
  tool_choice: Schema.optional(OpenAIResponsesToolChoice),
 }
@@ -86,10 +51,28 @@ const OpenAIResponsesBody = Schema.Struct({
 })
 export type OpenAIResponsesBody = Schema.Schema.Type<typeof OpenAIResponsesBody>

+// Replayed items are paired with stored server state by id, so a foreign or
+// synthetic token can fail request validation even when `call_id` pairing is
+// intact. Only resend ids in each item kind's own grammar; hosted tool
+// references keep generic validation because every hosted tool mints its own
+// prefix. The same allowlist approach codex uses before resending history
+// (codex-rs core/src/client.rs, `prepare_response_items_for_request`).
+const ITEM_ID_PREFIXES: Record<OpenResponses.ItemKind, ReadonlyArray<string>> = {
+  message: ["msg_"],
+  reasoning: ["rs_"],
+  "function-call": ["fc_"],
+  // Every hosted tool mints its own id prefix, so references keep generic
+  // validation only.
+  reference: [],
+}
+
 const extension = {
  id: ADAPTER,
  name: NAME,
-  lowerHostedToolItem: (item: unknown) => (Schema.is(OpenAIResponsesHostedToolItem)(item) ? item : undefined),
+  acceptsItemID: (kind: OpenResponses.ItemKind, id: string) => {
+    const prefixes = ITEM_ID_PREFIXES[kind]
+    return prefixes.length === 0 || prefixes.some((prefix) => id.startsWith(prefix))
+  },
 } satisfies OpenResponses.Extension

 const nativeImageToolInput = (tool: ToolDefinition) => {
@@ -122,8 +105,6 @@ 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 }),
@@ -131,7 +112,7 @@ const fromRequest = Effect.fn("OpenAIResponses.fromRequest")(function* (request:
  )
  const toolSchemaCompatibility = request.model.compatibility?.toolSchema
  const parallelToolCalls = OpenResponses.resolveParallelToolCalls(request)
-  return yield* decodeBody({
+  return {
    ...body,
    ...(parallelToolCalls === undefined ? {} : { parallel_tool_calls: parallelToolCalls }),
    tools:
@@ -142,7 +123,7 @@ 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) {
@@ -184,10 +165,8 @@ const HOSTED_TOOLS = {

 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),
-        )
+    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)
@@ -228,7 +207,7 @@ 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"
@@ -28,10 +28,10 @@ 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
+export const clampPromptCacheKey = (key: string | undefined): string | undefined => {
+  if (key === undefined) return undefined
+  const chars = Array.from(key)
+  if (chars.length <= OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH) return key
  return chars.slice(0, OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH).join("")
 }

@@ -210,9 +210,10 @@ 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>,
@@ -220,23 +221,9 @@ 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)
-          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)) &&
@@ -29,9 +29,10 @@ 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 const ServiceTier = Schema.declare<ServiceTier>(
+  (value): value is ServiceTier => typeof value === "string",
+  { title: "ServiceTier" },
+)

 export const Truncations = ["auto", "disabled"] as const
 export type Truncation = (typeof Truncations)[number]
@@ -55,6 +56,7 @@ 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),
@@ -42,61 +42,7 @@ export function parseJSON(jsonString: string, allowPartial = Allow.ALL): unknown
  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
+  return _parseJSON(input, allowPartial)
 }

 const _parseJSON = (jsonString: string, allow: number) => {
@@ -202,12 +148,7 @@ const _parseJSON = (jsonString: string, allow: number) => {
        skipBlank()
        index++
        try {
-          Object.defineProperty(object, key, {
-            value: parseAny(),
-            enumerable: true,
-            configurable: true,
-            writable: true,
-          })
+          object[key] = parseAny()
        } catch (error) {
          if (Allow.OBJ & allow) return object
          throw error
@@ -34,35 +34,37 @@ 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
-  },
-)
+) => 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"
@@ -1,5 +1,5 @@
 import { Effect, Option } from "effect"
-import { AIError, LLMEvent, type ProviderMetadata, type ToolCall } from "../../schema/index.js"
+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"

@@ -59,44 +59,46 @@ const inputStart = (tool: PendingTool) =>
    providerMetadata: tool.providerMetadata,
  })

-const inputDelta = (tool: PendingTool, text: string) =>
-  LLMEvent.toolInputDelta({
+const inputDelta = (tool: PendingTool, text: string) => {
+  const input = parsePartialInput(tool.input)
+  return LLMEvent.toolInputDelta({
    id: tool.id,
    name: tool.name,
    text,
-    input: Option.getOrElse(parsePartialInput(tool.input), () => ({})),
+    ...(Option.isSome(input) ? { input: input.value } : {}),
  })
+}

 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 +181,7 @@ 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.
 */
@@ -1,52 +1,15 @@
-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 ?? {} },
@@ -72,10 +35,7 @@ const step = (state: OpenResponses.ParserState, event: OpenResponses.Event) => {

 export const protocol = Protocol.make({
  id: ADAPTER,
-  body: {
-    schema: XAIResponsesBody,
-    from: fromRequest,
-  },
+  body: OpenResponses.protocol.body,
  stream: {
    event: OpenResponses.protocol.stream.event,
    initial: (request) => OpenResponses.initial(request, extension),
@@ -40,16 +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))

@@ -107,7 +106,6 @@ export function classifyProviderFailure(input: ProviderFailure): AIError["reason
    clientScoped &&
    (codes.includes("context_length_exceeded") ||
      codes.includes("model_context_window_exceeded") ||
-      codes.includes("request_too_large") ||
      isContextOverflow(text))
  )
    return new InvalidRequestReason({ ...common, classification: "context-overflow" })
@@ -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 },
 })

@@ -68,7 +68,7 @@ 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"
@@ -9,7 +9,7 @@ import type { ProviderPackage } from "../provider-package.js"
 import * as OpenAICompatibleProfiles from "./openai-compatible-profile.js"
 import * as OpenAIChat from "../protocols/openai-chat.js"
 import { newBreakpoints, ttlBucket } from "../protocols/utils/cache.js"
-import { isRecord } from "../protocols/shared.js"
+import { isRecord, ProviderShared } from "../protocols/shared.js"

 export const profile = OpenAICompatibleProfiles.profiles.openrouter
 export const id = ProviderID.make(profile.provider)
@@ -115,10 +115,12 @@ export const protocol = Protocol.make({
              reasoning_details: reasoningDetails,
            }
          })
+          const cacheKey = ProviderShared.clampPromptCacheKey(request.promptCacheKey)
          return {
            ...body,
            messages,
            ...bodyOptions(request.providerOptions),
+            ...(cacheKey ? { prompt_cache_key: cacheKey } : {}),
          } as OpenRouterBody
        }),
      ),
@@ -164,7 +166,6 @@ 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)
@@ -42,7 +42,7 @@ const responsesRoute = Route.make({
    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({
@@ -7,7 +7,6 @@ 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 {
@@ -89,7 +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>
@@ -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,
@@ -345,7 +338,9 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
                return onHalt
                  ? parsed.pipe(
                      Stream.concat(
-                        Stream.suspend(() => Stream.unwrap(onHalt(state).pipe(Effect.map(Stream.fromIterable)))),
+                        Stream.suspend(() =>
+                          Stream.unwrap(onHalt(state).pipe(Effect.map(Stream.fromIterable))),
+                        ),
                      ),
                    )
                  : parsed
@@ -405,8 +400,7 @@ 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
@@ -40,6 +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)
@@ -136,12 +147,14 @@ 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,
  })

@@ -166,6 +179,7 @@ const statusError =
            request,
            response,
            body: details,
+            requestId: requestId(headers),
            rateLimit,
          }),
        }),
@@ -202,6 +216,7 @@ export const classifyHttpFailure = (input: {
          ? undefined
          : new HttpResponseDetails({ status: input.status, headers: headerDetails(Headers.fromInput(headers)) }),
      ...details,
+      requestId: requestId(headers),
      rateLimit,
    }),
  })
@@ -29,6 +29,7 @@ export class HttpContext extends Schema.Class<HttpContext>("AI.HttpContext")({
  response: Schema.optional(HttpResponseDetails),
  body: Schema.optional(Schema.String),
  bodyTruncated: Schema.optional(Schema.Boolean),
+  requestId: Schema.optional(Schema.String),
  rateLimit: Schema.optional(HttpRateLimitDetails),
 }) {}

@@ -153,11 +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),
@@ -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"
@@ -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({
@@ -211,28 +211,6 @@ describe("RequestExecutor", () => {
    }).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",
-        http: { response: { status: 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
@@ -327,6 +305,7 @@ describe("RequestExecutor", () => {
          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",
@@ -1,7 +1,10 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:azure", "provider:azure"],
+    "tags": [
+      "prefix:azure",
+      "provider:azure"
+    ],
    "name": "azure/chat-streams-text",
    "recordedAt": "2026-08-23T17:21:53.198Z"
  },
@@ -1,7 +1,10 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:azure", "provider:azure"],
+    "tags": [
+      "prefix:azure",
+      "provider:azure"
+    ],
    "name": "azure/responses-calls-a-tool",
    "recordedAt": "2026-08-23T17:21:55.170Z"
  },
@@ -1,7 +1,10 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:azure", "provider:azure"],
+    "tags": [
+      "prefix:azure",
+      "provider:azure"
+    ],
    "name": "azure/responses-continues-after-a-tool-result",
    "recordedAt": "2026-08-23T17:21:56.397Z"
  },
@@ -1,7 +1,10 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:azure", "provider:azure"],
+    "tags": [
+      "prefix:azure",
+      "provider:azure"
+    ],
    "name": "azure/responses-streams-text",
    "recordedAt": "2026-08-23T17:21:54.158Z"
  },
@@ -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"
-      }
-    }
-  ]
-}
@@ -7,7 +7,13 @@
    "route": "cloudflare-workers-ai",
    "transport": "http",
    "model": "@cf/openai/gpt-oss-20b",
-    "tags": ["prefix:cloudflare-workers-ai", "provider:cloudflare-workers-ai", "tool", "tool-call", "golden"]
+    "tags": [
+      "prefix:cloudflare-workers-ai",
+      "provider:cloudflare-workers-ai",
+      "tool",
+      "tool-call",
+      "golden"
+    ]
  },
  "interactions": [
    {
@@ -29,4 +35,4 @@
      }
    }
  ]
-}
+}
@@ -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"
-      }
-    }
-  ]
-}
@@ -1,7 +1,11 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:google-vertex", "provider:google-vertex", "protocol:gemini"],
+    "tags": [
+      "prefix:google-vertex",
+      "provider:google-vertex",
+      "protocol:gemini"
+    ],
    "name": "google-vertex/calls-a-tool",
    "recordedAt": "2026-08-23T17:21:51.036Z"
  },
@@ -1,7 +1,11 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:google-vertex", "provider:google-vertex", "protocol:gemini"],
+    "tags": [
+      "prefix:google-vertex",
+      "provider:google-vertex",
+      "protocol:gemini"
+    ],
    "name": "google-vertex/continues-after-a-tool-result",
    "recordedAt": "2026-08-23T17:21:51.853Z"
  },
@@ -1,7 +1,11 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:google-vertex", "provider:google-vertex", "protocol:gemini"],
+    "tags": [
+      "prefix:google-vertex",
+      "provider:google-vertex",
+      "protocol:gemini"
+    ],
    "name": "google-vertex/streams-text",
    "recordedAt": "2026-08-23T17:21:50.112Z"
  },
@@ -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",
@@ -32,7 +32,7 @@
        {
          "direction": "client",
          "kind": "text",
-          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Alpha.\"}]}],\"store\":false,\"max_output_tokens\":30,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"instructions\":\"Follow the user's exact reply instruction.\"}"
+          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"system\",\"content\":\"Follow the user's exact reply instruction.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Alpha.\"}]}],\"store\":false,\"max_output_tokens\":30,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"}}"
        },
        {
          "direction": "server",
@@ -81,7 +81,7 @@
        {
          "direction": "client",
          "kind": "text",
-          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Alpha.\"}]},{\"type\":\"message\",\"id\":\"msg_ws_reconnect_1\",\"role\":\"assistant\",\"content\":[{\"type\":\"output_text\",\"text\":\"Alpha.\"}]},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Beta.\"}]}],\"store\":false,\"max_output_tokens\":30,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"instructions\":\"Follow the user's exact reply instruction.\"}"
+          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"system\",\"content\":\"Follow the user's exact reply instruction.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Alpha.\"}]},{\"type\":\"message\",\"id\":\"msg_ws_reconnect_1\",\"role\":\"assistant\",\"content\":[{\"type\":\"output_text\",\"text\":\"Alpha.\"}]},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Beta.\"}]}],\"store\":false,\"max_output_tokens\":30,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"}}"
        },
        {
          "direction": "server",
@@ -32,7 +32,7 @@
        {
          "direction": "client",
          "kind": "text",
-          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Ready.\"}]}],\"store\":false,\"max_output_tokens\":30,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"instructions\":\"Follow the user's exact reply instruction.\"}"
+          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"system\",\"content\":\"Follow the user's exact reply instruction.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Ready.\"}]}],\"store\":false,\"max_output_tokens\":30,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"}}"
        },
        {
          "direction": "server",
@@ -81,7 +81,7 @@
        {
          "direction": "client",
          "kind": "text",
-          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Recovered.\"}]}],\"store\":false,\"max_output_tokens\":30,\"previous_response_id\":\"resp_ws_rejection_1\",\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"instructions\":\"Follow the user's exact reply instruction.\"}"
+          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Recovered.\"}]}],\"store\":false,\"max_output_tokens\":30,\"previous_response_id\":\"resp_ws_rejection_1\",\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"}}"
        },
        {
          "direction": "server",
@@ -91,7 +91,7 @@
        {
          "direction": "client",
          "kind": "text",
-          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Ready.\"}]},{\"type\":\"message\",\"id\":\"msg_ws_rejection_1\",\"role\":\"assistant\",\"content\":[{\"type\":\"output_text\",\"text\":\"Ready.\"}]},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Recovered.\"}]}],\"store\":false,\"max_output_tokens\":30,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"instructions\":\"Follow the user's exact reply instruction.\"}"
+          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"system\",\"content\":\"Follow the user's exact reply instruction.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Ready.\"}]},{\"type\":\"message\",\"id\":\"msg_ws_rejection_1\",\"role\":\"assistant\",\"content\":[{\"type\":\"output_text\",\"text\":\"Ready.\"}]},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Recovered.\"}]}],\"store\":false,\"max_output_tokens\":30,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"}}"
        },
        {
          "direction": "server",
@@ -26,7 +26,7 @@
        "headers": {
          "content-type": "application/json"
        },
-        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Think briefly, then reply exactly with: Hello!\"}]}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"low\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":120,\"stream\":true,\"instructions\":\"Show concise reasoning when the provider supports visible reasoning summaries.\"}"
+        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"system\",\"content\":\"Show concise reasoning when the provider supports visible reasoning summaries.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Think briefly, then reply exactly with: Hello!\"}]}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"low\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":120,\"stream\":true}"
      },
      "response": {
        "status": 200,
@@ -18,7 +18,7 @@
        "headers": {
          "content-type": "application/json"
        },
-        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Think briefly, then reply exactly with: Hello!\"}]}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"low\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":120,\"stream\":true,\"instructions\":\"Show concise reasoning when the provider supports visible reasoning summaries.\"}"
+        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"system\",\"content\":\"Show concise reasoning when the provider supports visible reasoning summaries.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Think briefly, then reply exactly with: Hello!\"}]}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"low\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":120,\"stream\":true}"
      },
      "response": {
        "status": 200,
@@ -5,7 +5,14 @@
    "route": "openai-responses",
    "transport": "http",
    "model": "gpt-5.5",
-    "tags": ["prefix:openai-responses", "provider:openai", "flagship", "tool", "tool-loop", "golden"],
+    "tags": [
+      "prefix:openai-responses",
+      "provider:openai",
+      "flagship",
+      "tool",
+      "tool-loop",
+      "golden"
+    ],
    "name": "openai-responses/openai-responses-gpt-5-5-tool-loop",
    "recordedAt": "2026-08-20T06:30:22.262Z"
  },
@@ -18,7 +25,7 @@
        "headers": {
          "content-type": "application/json"
        },
-        "body": "{\"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,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":80,\"stream\":true,\"instructions\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"}"
+        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"system\",\"content\":\"Use the get_weather tool exactly once. After the tool result, 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,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":80,\"stream\":true}"
      },
      "response": {
        "status": 200,
@@ -36,7 +43,7 @@
        "headers": {
          "content-type": "application/json"
        },
-        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"What is the weather in Paris?\"}]},{\"type\":\"reasoning\",\"id\":\"rs_0ad67c31d9ddad95016a869efbd02487d1a51eed850e6f87f5\",\"summary\":[],\"encrypted_content\":\"gAAAAABqhp79U9UPKmTdmo9tmdil0C2KXpkFqUc4MNkYHT53Lzos9omncFPg76QzUmmSOdBcajisWBEo-xiTCvhp135uACUq8TJcdw4DluieYq6dWszijy28PFFfeO-6MmHwi7zeln1Z202zErJUEyuf1bML68VAeam5PqlMLG-a4-pmnWiH2ExWKibTUX37QoMQoArrkccJOCmxwDflV_kWDPMFxQVDfeMg9fd1gVv2u-x1Mjk0b9mJDOq0Fe5Gh-IkpWzfXgZTdptFmCM75cksvs61Rqsx6P33czal-LSixEF0WMizCvbMQmqKGs7MKGMeoa6j6vWOnB3ICIbv6FShnSaTpZWJFwejvOurkfuxa-2q6xVDZsBoQCgMWPHsqLxwAo1JKdfBk0pMvSuvpw2BRxykUZ1ULCYJ-BypST65292-EuSZFIuXPMPir-_raSCTsgsZNMscDG6ll3qksDTDS6_o5NutD7Ra-WZzaUe_HQlSLKLACTc4qv2EK1QoC4aYv4goxkTSx17WhS2D86lILgkUd-TIHjJ6iR3uxSNx7YeBNxiJgddIAEjAaSrdF-WDouSNT9k3efd5HhT3zahIOMKgb3XIQzFOYWfWgea5-SbaIdKwne9hU0QyhcBQs6yoifSg-fJZtahbPb-GCDYnOLlH-bV94vldoccb-2P1JdB3jaLj5tJUecfr2H4qiu8MgkPj0TkwYNbJynYmJo9H5Lm-XJ9gfzIXzJh0arKwsS4gwDLf4J3LOEF3WEW3mknOjjb9PrLmHRYXQQh9tTiX9ILPZpbufkyCurTUMQgWiSCitXBC6FoLXRHilSmb-6_avBnlUMziMfey-FkKvRfiPox6BaJrnOq6SGlOv11y7EKvzrn29la7HKPygYenDAkyq2mq0Zk2nLWNmJcv9sQTBrkBdFMmJYPi2J2im8XD5MmAjEL8R4FCBHoPIIZ6pENQykvH8PhpWKuzF5gJlY3Vwz4iJ0Qb9TrNI0hzBoI1U0LeB5FJ2HgjZQwCFF5x3ubh72xrUsFpuyYyYPa8GDT0Bo-LW_IlJ_mN4EwI5Nk9n-8Bt015yxsfpa5YaDeCeQFcdj8SD0UAd7QWtGACpzKcIj1-vJJU7OiwscV_v1dLvoiEe1ehI9jcvPn28TgHlo_dippe0iMN4FAm1Bf8vtWVMFDvfV1rPv1pAFFnSa9XqFszD5Exo_xzcQEoKXvQv3OnUtoiM4Db4uadClazLjoep2TQgHcJBVbTbLySTVPmok4ROFQZsU_mq4vu2M__d8HOjADfIIYz5VLVQKNpo0Hv_QkT2bn56Q==\"},{\"type\":\"function_call\",\"id\":\"fc_0ad67c31d9ddad95016a869efd126887d1b7e2f17f155cb5dc\",\"call_id\":\"call_qrzOfKDfzaq8fqbSNNVHlNsV\",\"name\":\"get_weather\",\"arguments\":\"{\\\"city\\\":\\\"Paris\\\"}\"},{\"type\":\"function_call_output\",\"call_id\":\"call_qrzOfKDfzaq8fqbSNNVHlNsV\",\"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,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":80,\"stream\":true,\"instructions\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"}"
+        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"system\",\"content\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"What is the weather in Paris?\"}]},{\"type\":\"reasoning\",\"id\":\"rs_0ad67c31d9ddad95016a869efbd02487d1a51eed850e6f87f5\",\"summary\":[],\"encrypted_content\":\"gAAAAABqhp79U9UPKmTdmo9tmdil0C2KXpkFqUc4MNkYHT53Lzos9omncFPg76QzUmmSOdBcajisWBEo-xiTCvhp135uACUq8TJcdw4DluieYq6dWszijy28PFFfeO-6MmHwi7zeln1Z202zErJUEyuf1bML68VAeam5PqlMLG-a4-pmnWiH2ExWKibTUX37QoMQoArrkccJOCmxwDflV_kWDPMFxQVDfeMg9fd1gVv2u-x1Mjk0b9mJDOq0Fe5Gh-IkpWzfXgZTdptFmCM75cksvs61Rqsx6P33czal-LSixEF0WMizCvbMQmqKGs7MKGMeoa6j6vWOnB3ICIbv6FShnSaTpZWJFwejvOurkfuxa-2q6xVDZsBoQCgMWPHsqLxwAo1JKdfBk0pMvSuvpw2BRxykUZ1ULCYJ-BypST65292-EuSZFIuXPMPir-_raSCTsgsZNMscDG6ll3qksDTDS6_o5NutD7Ra-WZzaUe_HQlSLKLACTc4qv2EK1QoC4aYv4goxkTSx17WhS2D86lILgkUd-TIHjJ6iR3uxSNx7YeBNxiJgddIAEjAaSrdF-WDouSNT9k3efd5HhT3zahIOMKgb3XIQzFOYWfWgea5-SbaIdKwne9hU0QyhcBQs6yoifSg-fJZtahbPb-GCDYnOLlH-bV94vldoccb-2P1JdB3jaLj5tJUecfr2H4qiu8MgkPj0TkwYNbJynYmJo9H5Lm-XJ9gfzIXzJh0arKwsS4gwDLf4J3LOEF3WEW3mknOjjb9PrLmHRYXQQh9tTiX9ILPZpbufkyCurTUMQgWiSCitXBC6FoLXRHilSmb-6_avBnlUMziMfey-FkKvRfiPox6BaJrnOq6SGlOv11y7EKvzrn29la7HKPygYenDAkyq2mq0Zk2nLWNmJcv9sQTBrkBdFMmJYPi2J2im8XD5MmAjEL8R4FCBHoPIIZ6pENQykvH8PhpWKuzF5gJlY3Vwz4iJ0Qb9TrNI0hzBoI1U0LeB5FJ2HgjZQwCFF5x3ubh72xrUsFpuyYyYPa8GDT0Bo-LW_IlJ_mN4EwI5Nk9n-8Bt015yxsfpa5YaDeCeQFcdj8SD0UAd7QWtGACpzKcIj1-vJJU7OiwscV_v1dLvoiEe1ehI9jcvPn28TgHlo_dippe0iMN4FAm1Bf8vtWVMFDvfV1rPv1pAFFnSa9XqFszD5Exo_xzcQEoKXvQv3OnUtoiM4Db4uadClazLjoep2TQgHcJBVbTbLySTVPmok4ROFQZsU_mq4vu2M__d8HOjADfIIYz5VLVQKNpo0Hv_QkT2bn56Q==\"},{\"type\":\"function_call\",\"id\":\"fc_0ad67c31d9ddad95016a869efd126887d1b7e2f17f155cb5dc\",\"call_id\":\"call_qrzOfKDfzaq8fqbSNNVHlNsV\",\"name\":\"get_weather\",\"arguments\":\"{\\\"city\\\":\\\"Paris\\\"}\"},{\"type\":\"function_call_output\",\"call_id\":\"call_qrzOfKDfzaq8fqbSNNVHlNsV\",\"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,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":80,\"stream\":true}"
      },
      "response": {
        "status": 200,
@@ -52,4 +52,4 @@
      }
    }
  ]
-}
+}
@@ -2,7 +2,12 @@
  "version": 1,
  "metadata": {
    "model": "anthropic/claude-sonnet-4.6",
-    "tags": ["prefix:openai-compatible-chat", "provider:vercel-ai-gateway", "protocol:openai-chat", "reasoning"],
+    "tags": [
+      "prefix:openai-compatible-chat",
+      "provider:vercel-ai-gateway",
+      "protocol:openai-chat",
+      "reasoning"
+    ],
    "name": "vercel-ai-gateway-reasoning",
    "recordedAt": "2026-07-18T11:28:42.077Z"
  },
@@ -26,4 +31,4 @@
      }
    }
  ]
-}
+}
@@ -18,21 +18,6 @@ describe("partial JSON", () => {
    expect(() => parse('"hello', ~Allow.STR)).toThrow(PartialJSON)
  })

-  test("repairs invalid escapes and raw control characters", () => {
-    expect(parse('{"path":"A\\H","text":"first\tsecond"}')).toEqual({
-      path: "A\\H",
-      text: "first\tsecond",
-    })
-  })
-
-  test("preserves prototype keys in partial objects", () => {
-    const object = parse('{"__proto__":{"safe":true}') as Record<string, unknown>
-
-    expect(Object.hasOwn(object, "__proto__")).toBe(true)
-    expect(Object.getPrototypeOf(object)).toBe(Object.prototype)
-    expect(object.__proto__).toEqual({ safe: true })
-  })
-
  test("controls partial collection values independently", () => {
    expect(parse('["', Allow.ARR)).toEqual([])
    expect(parse('["', Allow.ARR | Allow.STR)).toEqual([""])
@@ -18,19 +18,13 @@ describe("provider error classification", () => {
    expect(messages.every(isContextOverflow)).toBe(true)
  })

-  test("classifies Anthropic request_too_large as recoverable overflow", () => {
-    expect(
+  test("classifies request size failures separately from context overflow", () => {
+    const failures = [
+      classifyProviderFailure({ message: "request too large", status: 413 }),
      classifyProviderFailure({
        message: '{"error":{"type":"request_too_large","message":"Request exceeds the maximum size"}}',
        status: 400,
      }),
-    ).toMatchObject({ _tag: "InvalidRequest", classification: "context-overflow" })
-    expect(isContextOverflow("413 status code (no body)")).toBe(true)
-  })
-
-  test("classifies generic request size failures separately from context overflow", () => {
-    const failures = [
-      classifyProviderFailure({ message: "request too large", status: 413 }),
      classifyProviderFailure({ message: "upstream request entity too large", status: 502 }),
    ]

@@ -39,6 +33,7 @@ describe("provider error classification", () => {
        expect.objectContaining({ _tag: "InvalidRequest", classification: "payload-too-large" }),
      ),
    )
+    expect(isContextOverflow("413 status code (no body)")).toBe(false)
  })

  test("does not classify rate limits as context overflow", () => {
@@ -89,7 +84,9 @@ describe("provider error classification", () => {

  test("classifies network error text as provider internal", () => {
    expect(
-      ["network error", "network-error", "network_error"].map((message) => classifyProviderFailure({ message })._tag),
+      ["network error", "network-error", "network_error"].map(
+        (message) => classifyProviderFailure({ message })._tag,
+      ),
    ).toEqual(["ProviderInternal", "ProviderInternal", "ProviderInternal"])
  })

@@ -26,10 +26,6 @@ describe("provider package entrypoints", () => {
      import("@opencode-ai/ai/providers/amazon-bedrock/mantle"),
      import("@opencode-ai/ai/providers/amazon-bedrock/mantle/chat"),
      import("@opencode-ai/ai/providers/amazon-bedrock/mantle/responses"),
-      import("@opencode-ai/ai/providers/togetherai"),
-      import("@opencode-ai/ai/providers/cerebras"),
-      import("@opencode-ai/ai/providers/deepinfra"),
-      import("@opencode-ai/ai/providers/groq"),
    ])

    for (const module of modules) expect(module.model).toBeFunction()
@@ -39,24 +35,6 @@ describe("provider package entrypoints", () => {
    expect(modules[19].model).toBe(modules[20].model)
  })

-  test("maps DeepInfra package settings onto its native executable model", async () => {
-    const DeepInfra = await import("@opencode-ai/ai/providers/deepinfra")
-    const settings = {
-      apiKey: "fixture",
-      baseURL: "https://provider.example.test/v1/",
-      headers: { "x-application": "opencode" },
-      body: { service_tier: "priority" },
-      providerOptions: { reasoningEffort: "high" as const },
-    }
-    const deepinfra = DeepInfra.model("google/gemma-3-27b-it", settings)
-
-    expect(deepinfra.route.id).toBe("deepinfra-chat")
-    expect(deepinfra.route.endpoint.baseURL).toBe("https://provider.example.test/v1/openai")
-    expect(deepinfra.route.defaults.providerOptions).toEqual(settings.providerOptions)
-    expect(deepinfra.route.defaults.headers).toEqual(settings.headers)
-    expect(deepinfra.route.defaults.http?.body).toEqual(settings.body)
-  })
-
  test("maps OpenRouter and xAI package settings onto executable models", async () => {
    const OpenRouter = await import("@opencode-ai/ai/providers/openrouter")
    const XAI = await import("@opencode-ai/ai/providers/xai")
@@ -117,11 +95,7 @@ describe("provider package entrypoints", () => {
    })
    expect(selected.route.defaults.headers).toEqual({ "x-application": "opencode" })
    expect(selected.route.defaults.http?.body).toEqual({ service_tier: "priority" })
-    expect(selected.route.defaults.providerOptions).toEqual({
-      reasoningEffort: "low",
-      store: true,
-      include: ["reasoning.encrypted_content"],
-    })
+    expect(selected.route.defaults.providerOptions).toEqual({ reasoningEffort: "low", store: true })
  })

  test("maps Anthropic-compatible settings onto the executable model", async () => {
@@ -311,10 +285,7 @@ describe("provider package entrypoints", () => {
      baseURL: "https://aiplatform.googleapis.com/v1/projects/vertex-project/locations/global/endpoints/openapi",
      path: "/responses",
    })
-    expect(responses.route.defaults.providerOptions).toEqual({
-      store: false,
-      include: ["reasoning.encrypted_content"],
-    })
+    expect(responses.route.defaults.providerOptions).toEqual({ store: false })
  })

  test("rejects conflicting Vertex auth settings at runtime", async () => {
@@ -2,10 +2,9 @@ import { describe, expect } from "bun:test"
 import { Effect } from "effect"
 import { HttpClientRequest } from "effect/unstable/http"
 import { CacheHint, LLM, AIError, LLMRequest, Message, ToolCallPart, ToolDefinition, Usage } from "../../src/index.js"
-import { Auth, Endpoint, LLMClient, Route } from "../../src/route.js"
+import { Auth, LLMClient } from "../../src/route.js"
 import { compileRequest } from "../../src/route/client.js"
 import * as AnthropicMessages from "../../src/protocols/anthropic-messages.js"
-import { GoogleVertexMessages } from "../../src/providers.js"
 import { continuationRequest, nativeAnthropicMessagesContinuation } from "../continuation-scenarios.js"
 import { it } from "../lib/effect.js"
 import { dynamicResponse, fixedResponse } from "../lib/http.js"
@@ -28,12 +27,6 @@ const compileUnsignedReasoning = (model: LLMRequest["model"]) =>
    }),
  )

-const vertexOpus48 = GoogleVertexMessages.configure({
-  accessToken: "test",
-  location: "global",
-  project: "test",
-}).model("claude-opus-4-8")
-
 const request = LLM.request({
  id: "req_1",
  model,
@@ -293,149 +286,6 @@ describe("Anthropic Messages route", () => {
    }),
  )

-  it.effect("keeps a terminal Vertex system update in the tool-result turn", () =>
-    Effect.gen(function* () {
-      const prepared = yield* compileRequest(
-        LLM.request({
-          model: vertexOpus48,
-          messages: [
-            Message.assistant([ToolCallPart.make({ id: "call_1", name: "lookup", input: {} })]),
-            Message.tool({ id: "call_1", name: "lookup", result: "Done." }),
-            Message.system("Operator update."),
-          ],
-          cache: "none",
-        }),
-      )
-
-      expect(prepared.body.messages).toEqual([
-        {
-          role: "assistant",
-          content: [{ type: "tool_use", id: "call_1", name: "lookup", input: {} }],
-        },
-        {
-          role: "user",
-          content: [
-            {
-              type: "tool_result",
-              tool_use_id: "call_1",
-              content: '"Done."',
-              is_error: undefined,
-              cache_control: undefined,
-            },
-            {
-              type: "text",
-              text: "<system-update>\nOperator update.\n</system-update>",
-              cache_control: undefined,
-            },
-          ],
-        },
-      ])
-    }),
-  )
-
-  it.effect("preserves folded tool-result system updates across multi-turn Vertex history", () =>
-    Effect.gen(function* () {
-      const prepared = yield* compileRequest(
-        LLM.request({
-          model: vertexOpus48,
-          messages: [
-            Message.assistant([ToolCallPart.make({ id: "call_1", name: "lookup", input: {} })]),
-            Message.tool({ id: "call_1", name: "lookup", result: "Done." }),
-            Message.system("Operator update."),
-            Message.assistant("Acknowledged."),
-            Message.user("Next step."),
-          ],
-          cache: "none",
-        }),
-      )
-
-      expect(prepared.body.messages).toEqual([
-        {
-          role: "assistant",
-          content: [{ type: "tool_use", id: "call_1", name: "lookup", input: {} }],
-        },
-        {
-          role: "user",
-          content: [
-            {
-              type: "tool_result",
-              tool_use_id: "call_1",
-              content: '"Done."',
-              is_error: undefined,
-              cache_control: undefined,
-            },
-            {
-              type: "text",
-              text: "<system-update>\nOperator update.\n</system-update>",
-              cache_control: undefined,
-            },
-          ],
-        },
-        { role: "assistant", content: [{ type: "text", text: "Acknowledged." }] },
-        { role: "user", content: [{ type: "text", text: "Next step." }] },
-      ])
-    }),
-  )
-
-  it.effect("keeps a terminal direct Anthropic system update native", () =>
-    Effect.gen(function* () {
-      const prepared = yield* compileRequest(
-        LLM.request({
-          model: opus48,
-          messages: [
-            Message.assistant([ToolCallPart.make({ id: "call_1", name: "lookup", input: {} })]),
-            Message.tool({ id: "call_1", name: "lookup", result: "Done." }),
-            Message.system("Operator update."),
-          ],
-          cache: "none",
-        }),
-      )
-
-      expect(prepared.body.messages).toEqual([
-        {
-          role: "assistant",
-          content: [{ type: "tool_use", id: "call_1", name: "lookup", input: {} }],
-        },
-        {
-          role: "user",
-          content: [
-            {
-              type: "tool_result",
-              tool_use_id: "call_1",
-              content: '"Done."',
-              is_error: undefined,
-              cache_control: undefined,
-            },
-          ],
-        },
-        {
-          role: "system",
-          content: [{ type: "text", text: "Operator update.", cache_control: undefined }],
-        },
-      ])
-    }),
-  )
-
-  it.effect("keeps an ordinary terminal Vertex system update native", () =>
-    Effect.gen(function* () {
-      const prepared = yield* compileRequest(
-        LLM.request({
-          model: vertexOpus48,
-          messages: [Message.user("Before."), Message.system("Operator update.")],
-          cache: "none",
-        }),
-      )
-
-      expect(prepared.body.messages).toEqual([
-        { role: "user", content: [{ type: "text", text: "Before." }] },
-        {
-          role: "system",
-          content: [{ type: "text", text: "Operator update.", cache_control: undefined }],
-        },
-      ])
-    }),
-  )
-
  it.effect("rejects a system update between a local tool call and its result", () =>
    Effect.gen(function* () {
      const error = yield* compileRequest(
@@ -810,99 +660,6 @@ describe("Anthropic Messages route", () => {
    }),
  )

-  it.effect("round-trips compatible provider metadata in its own namespace", () =>
-    Effect.gen(function* () {
-      const compatible = Route.make({
-        id: "custom-anthropic-messages",
-        provider: "custom-anthropic",
-        protocol: AnthropicMessages.protocol,
-        endpoint: Endpoint.path("/messages", { baseURL: "https://compatible.test/v1" }),
-        auth: Auth.header("x-api-key", "test"),
-        framing: AnthropicMessages.framing,
-      }).model({ id: "custom-model" })
-      const result = [
-        {
-          type: "web_search_result",
-          url: "https://example.com",
-          citations: [{ type: "web_search_result_location", cited_text: "Example" }],
-        },
-      ]
-      const response = yield* LLMClient.generate(LLM.request({ model: compatible, prompt: "Search." })).pipe(
-        Effect.provide(
-          fixedResponse(
-            sseEvents(
-              { type: "message_start", message: { usage: { input_tokens: 5, custom_start: true } } },
-              { type: "content_block_start", index: 0, content_block: { type: "thinking", thinking: "Thinking." } },
-              { type: "content_block_delta", index: 0, delta: { type: "signature_delta", signature: "custom_sig" } },
-              { type: "content_block_stop", index: 0 },
-              {
-                type: "content_block_start",
-                index: 1,
-                content_block: { type: "redacted_thinking", data: "custom_redacted" },
-              },
-              { type: "content_block_stop", index: 1 },
-              {
-                type: "content_block_start",
-                index: 2,
-                content_block: {
-                  type: "server_tool_use",
-                  id: "custom_tool",
-                  name: "web_search",
-                  input: { query: "example" },
-                },
-              },
-              { type: "content_block_stop", index: 2 },
-              {
-                type: "content_block_start",
-                index: 3,
-                content_block: { type: "web_search_tool_result", tool_use_id: "custom_tool", content: result },
-              },
-              { type: "content_block_stop", index: 3 },
-              {
-                type: "message_delta",
-                delta: { stop_reason: "end_turn", stop_sequence: "custom_stop" },
-                usage: { output_tokens: 2, custom_terminal: true },
-              },
-              { type: "message_stop" },
-            ),
-          ),
-        ),
-      )
-
-      expect(response.message.content).toMatchObject([
-        { type: "reasoning", text: "Thinking.", providerMetadata: { "custom-anthropic": { signature: "custom_sig" } } },
-        { type: "reasoning", text: "", providerMetadata: { "custom-anthropic": { redactedData: "custom_redacted" } } },
-        { type: "tool-call", id: "custom_tool", providerExecuted: true },
-        {
-          type: "tool-result",
-          providerExecuted: true,
-          providerMetadata: { "custom-anthropic": { blockType: "web_search_tool_result", result } },
-        },
-      ])
-      expect(response.usage?.providerMetadata).toEqual({
-        "custom-anthropic": { input_tokens: 5, custom_start: true, output_tokens: 2, custom_terminal: true },
-      })
-      expect(response.events.at(-1)).toMatchObject({
-        providerMetadata: { "custom-anthropic": { stopSequence: "custom_stop" } },
-      })
-
-      const prepared = yield* compileRequest(
-        LLM.request({ model: compatible, messages: [response.message], cache: "none" }),
-      )
-      expect(prepared.body.messages).toEqual([
-        {
-          role: "assistant",
-          content: [
-            { type: "thinking", thinking: "Thinking.", signature: "custom_sig" },
-            { type: "redacted_thinking", data: "custom_redacted" },
-            { type: "server_tool_use", id: "custom_tool", name: "web_search", input: { query: "example" } },
-            { type: "web_search_tool_result", tool_use_id: "custom_tool", content: result },
-          ],
-        },
-      ])
-    }),
-  )
-
  it.effect("parses text, reasoning, and usage stream fixtures", () =>
    Effect.gen(function* () {
      const body = sseEvents(
@@ -1013,108 +770,6 @@ describe("Anthropic Messages route", () => {
    }),
  )

-  it.effect("ignores unknown content block and delta variants", () =>
-    Effect.gen(function* () {
-      const response = yield* LLMClient.generate(request).pipe(
-        Effect.provide(
-          fixedResponse(
-            sseEvents(
-              { type: "message_start", message: { usage: { input_tokens: 5 } } },
-              { type: "future_event", content_block: 42, delta: 42 },
-              { type: "content_block_start", index: 0, content_block: { type: "future_block", text: 42 } },
-              { type: "content_block_delta", index: 0, delta: { text: "ignored" } },
-              { type: "content_block_delta", index: 0, delta: { type: "future_delta", text: 42 } },
-              { type: "content_block_delta", index: 0, delta: { type: "text_delta", text: "hidden" } },
-              { type: "content_block_delta", index: 0, delta: { type: "thinking_delta", thinking: "hidden" } },
-              { type: "content_block_delta", index: 0, delta: { type: "signature_delta", signature: "hidden" } },
-              { type: "content_block_stop", index: 0 },
-              { type: "content_block_start", index: 1, content_block: { type: "text", text: "" } },
-              { type: "content_block_delta", index: 1, delta: { type: "text_delta", text: "Hello" } },
-              { type: "content_block_stop", index: 1 },
-              { type: "message_delta", delta: { stop_reason: "end_turn" }, usage: { output_tokens: 1 } },
-              { type: "message_stop" },
-            ),
-          ),
-        ),
-      )
-
-      expect(response.message.content).toEqual([{ type: "text", text: "Hello" }])
-      expect(response.finishReason).toEqual({ normalized: "stop", raw: "end_turn" })
-    }),
-  )
-
-  it.effect("rejects malformed recognized content block variants", () =>
-    Effect.gen(function* () {
-      const error = yield* LLMClient.generate(request).pipe(
-        Effect.provide(
-          fixedResponse(
-            sseEvents(
-              { type: "message_start", message: { usage: { input_tokens: 5 } } },
-              { type: "content_block_start", index: 0, content_block: { type: "text", text: 42 } },
-            ),
-          ),
-        ),
-        Effect.flip,
-      )
-
-      expect(error.reason).toMatchObject({
-        _tag: "InvalidProviderOutput",
-        message: "Invalid anthropic/anthropic-messages stream event",
-      })
-    }),
-  )
-
-  it.effect("rejects malformed recognized content delta variants", () =>
-    Effect.gen(function* () {
-      const error = yield* LLMClient.generate(request).pipe(
-        Effect.provide(
-          fixedResponse(
-            sseEvents(
-              { type: "message_start", message: { usage: { input_tokens: 5 } } },
-              { type: "content_block_start", index: 0, content_block: { type: "text", text: "" } },
-              { type: "content_block_delta", index: 0, delta: { type: "text_delta", text: 42 } },
-            ),
-          ),
-        ),
-        Effect.flip,
-      )
-
-      expect(error.reason).toMatchObject({
-        _tag: "InvalidProviderOutput",
-        message: "Invalid anthropic/anthropic-messages stream event",
-      })
-    }),
-  )
-
-  it.effect("rejects malformed payloads on unrelated stream events", () =>
-    Effect.gen(function* () {
-      const events = [
-        { type: "message_start", message: { usage: { input_tokens: 1 } }, delta: 42 },
-        { type: "content_block_start", index: 0 },
-        { type: "content_block_delta", index: 0 },
-        { type: "content_block_stop", index: 0, content_block: { type: "text", text: 42 } },
-        { type: "message_delta" },
-        { type: "message_delta", delta: { stop_reason: 42 } },
-        { type: "message_stop", delta: { text: 42 } },
-        { type: "error", error: { type: "overloaded_error", message: "busy" }, content_block: 42 },
-      ]
-
-      yield* Effect.forEach(events, (event) =>
-        Effect.gen(function* () {
-          const error = yield* LLMClient.generate(request).pipe(
-            Effect.provide(fixedResponse(sseEvents(event))),
-            Effect.flip,
-          )
-
-          expect(error.reason).toMatchObject({
-            _tag: "InvalidProviderOutput",
-            message: "Invalid anthropic/anthropic-messages stream event",
-          })
-        }),
-      )
-    }),
-  )
-
  it.effect("rejects malformed recognized SSE events", () =>
    Effect.gen(function* () {
      const error = yield* LLMClient.generate(request).pipe(
@@ -491,7 +491,7 @@ describe("Bedrock Converse route", () => {
    }),
  )

-  it.effect("recovers incomplete tool input at finalization", () =>
+  it.effect("emits malformed tool input as an unexecuted tool error", () =>
    Effect.gen(function* () {
      const body = eventStreamBody(
        ["messageStart", { role: "assistant" }],
@@ -508,10 +508,10 @@ describe("Bedrock Converse route", () => {
      )
      const response = yield* LLMClient.generate(baseRequest).pipe(Effect.provide(fixedBytes(body)))

-      expect(response.events.find((event) => event.type === "tool-call")).toMatchObject({
+      expect(response.events.find((event) => event.type === "tool-input-error")).toMatchObject({
        id: "tool_1",
        name: "lookup",
-        input: { query: "partial" },
+        raw: '{"query":"partial',
      })
      expect(response.finishReason).toEqual({ normalized: "tool-calls", raw: "end_turn" })
    }),
@@ -569,57 +569,6 @@ describe("Bedrock Converse route", () => {
    }),
  )

-  it.effect("round-trips reassigned provider reasoning and usage metadata in its own namespace", () =>
-    Effect.gen(function* () {
-      const compatible = model.route.with({ provider: "custom-bedrock" }).model({ id: model.id })
-      const redactedData = "cmVkYWN0ZWQtdGhpbmtpbmc="
-      const response = yield* LLMClient.generate(LLMRequest.update(baseRequest, { model: compatible })).pipe(
-        Effect.provide(
-          fixedBytes(
-            eventStreamBody(
-              ["messageStart", { role: "assistant" }],
-              ["contentBlockDelta", { contentBlockIndex: 0, delta: { reasoningContent: { text: "Let me think." } } }],
-              ["contentBlockDelta", { contentBlockIndex: 0, delta: { reasoningContent: { signature: "custom_sig" } } }],
-              ["contentBlockStop", { contentBlockIndex: 0 }],
-              [
-                "contentBlockDelta",
-                { contentBlockIndex: 1, delta: { reasoningContent: { redactedContent: redactedData } } },
-              ],
-              ["contentBlockStop", { contentBlockIndex: 1 }],
-              ["messageStop", { stopReason: "end_turn" }],
-              ["metadata", { usage: { inputTokens: 5, outputTokens: 2, totalTokens: 7 } }],
-            ),
-          ),
-        ),
-      )
-
-      expect(response.message.content).toEqual([
-        {
-          type: "reasoning",
-          text: "Let me think.",
-          providerMetadata: { "custom-bedrock": { signature: "custom_sig" } },
-        },
-        { type: "reasoning", text: "", providerMetadata: { "custom-bedrock": { redactedData } } },
-      ])
-      expect(response.usage?.providerMetadata).toEqual({
-        "custom-bedrock": { inputTokens: 5, outputTokens: 2, totalTokens: 7 },
-      })
-
-      const prepared = yield* compileRequest(
-        LLM.request({ model: compatible, messages: [response.message], cache: "none" }),
-      )
-      expect(prepared.body.messages).toEqual([
-        {
-          role: "assistant",
-          content: [
-            { reasoningContent: { reasoningText: { text: "Let me think.", signature: "custom_sig" } } },
-            { reasoningContent: { redactedContent: redactedData } },
-          ],
-        },
-      ])
-    }),
-  )
-
  it.effect("preserves reasoning signatures when contentBlockStop is missing", () =>
    Effect.gen(function* () {
      const response = yield* LLMClient.generate(baseRequest).pipe(
@@ -767,32 +716,6 @@ describe("Bedrock Converse route", () => {
    }),
  )

-  it.effect("ignores unknown normal stream events", () =>
-    Effect.gen(function* () {
-      const body = concat([
-        eventFrame("messageStart", { role: "assistant" }),
-        eventFrame("futureEvent", { message: "Ignore this" }),
-        eventFrame("messageStop", { stopReason: "end_turn" }),
-      ])
-      const response = yield* LLMClient.generate(baseRequest).pipe(Effect.provide(fixedBytes(body)))
-
-      expect(response.finishReason).toEqual({ normalized: "stop", raw: "end_turn" })
-    }),
-  )
-
-  it.effect("fails unknown stream exceptions after message stop", () =>
-    Effect.gen(function* () {
-      const body = concat([
-        eventFrame("messageStart", { role: "assistant" }),
-        eventFrame("messageStop", { stopReason: "end_turn" }),
-        exceptionFrame("futureException", { message: "A future provider failure" }),
-      ])
-      const error = yield* LLMClient.generate(baseRequest).pipe(Effect.provide(fixedBytes(body)), Effect.flip)
-
-      expect(error.reason).toMatchObject({ _tag: "UnknownProvider", message: "A future provider failure" })
-    }),
-  )
-
  it.effect("classifies throttlingException as a rate limit", () =>
    Effect.gen(function* () {
      const body = concat([
@@ -1,13 +1,11 @@
 import { describe, expect } from "bun:test"
 import { Effect } from "effect"
 import { HttpClientRequest } from "effect/unstable/http"
-import { LLM, Message } from "../../src/index.js"
+import { LLM } from "../../src/index.js"
 import { AmazonBedrockMantle } from "../../src/providers.js"
-import { model } from "../../src/providers/amazon-bedrock/mantle.js"
-import { OpenAIResponses } from "../../src/protocols/openai-responses.js"
 import { compileRequest, LLMClient } from "../../src/route/client.js"
 import { it } from "../lib/effect.js"
-import { dynamicResponse, fixedResponse } from "../lib/http.js"
+import { dynamicResponse } from "../lib/http.js"
 import { sseEvents } from "../lib/sse.js"
 import { recordedTests } from "../recorded-test.js"

@@ -18,16 +16,12 @@ const credentials = {
 }

 describe("Amazon Bedrock Mantle provider", () => {
-  it.effect("uses Responses by default and exposes Chat explicitly", () =>
+  it.effect("uses Chat by default and exposes Responses", () =>
    Effect.gen(function* () {
      const provider = AmazonBedrockMantle.configure({ credentials })
-      expect(provider.model).toBe(provider.responses)
-      expect(AmazonBedrockMantle.model).toBe(AmazonBedrockMantle.responsesModel)
-      expect(model).toBe(AmazonBedrockMantle.responsesModel)
-      expect(provider.model("openai.gpt-oss-120b").route.transport).toBe(OpenAIResponses.httpTransport)
-      const chat = yield* compileRequest(LLM.request({ model: provider.chat("openai.gpt-oss-120b"), prompt: "Hi" }))
+      const chat = yield* compileRequest(LLM.request({ model: provider.model("openai.gpt-oss-120b"), prompt: "Hi" }))
      const responses = yield* compileRequest(
-        LLM.request({ model: provider.model("openai.gpt-oss-120b"), prompt: "Hi" }),
+        LLM.request({ model: provider.responses("openai.gpt-oss-120b"), prompt: "Hi" }),
      )

      expect(chat).toMatchObject({
@@ -40,23 +34,6 @@ describe("Amazon Bedrock Mantle provider", () => {
        protocol: "openai-responses",
        body: { model: "openai.gpt-oss-120b", store: false },
      })
-      expect(provider.model("openai.gpt-oss-120b").route.providerMetadataKey).toBe("mantle")
-      expect(provider.chat("openai.gpt-oss-120b").route.providerMetadataKey).toBe("mantle")
-    }),
-  )
-
-  it.effect("preserves configured top-p generation defaults for Chat and Responses", () =>
-    Effect.gen(function* () {
-      const settings = { apiKey: "test-key", topP: 0.8 }
-      const chat = yield* compileRequest(
-        LLM.request({ model: AmazonBedrockMantle.chatModel("openai.gpt-oss-safeguard-20b", settings), prompt: "Hi" }),
-      )
-      const responses = yield* compileRequest(
-        LLM.request({ model: AmazonBedrockMantle.responsesModel("openai.gpt-oss-120b", settings), prompt: "Hi" }),
-      )
-
-      expect(chat.body.top_p).toBe(0.8)
-      expect(responses.body.top_p).toBe(0.8)
    }),
  )

@@ -106,42 +83,6 @@ describe("Amazon Bedrock Mantle provider", () => {
      expect(seen).toEqual([{ url: "https://mantle.test/v1/chat/completions", authorization: "Bearer test-key" }])
    }),
  )
-
-  it.effect("replays reasoning with Mantle's message-prefixed item ids", () =>
-    Effect.gen(function* () {
-      const model = AmazonBedrockMantle.configure({ apiKey: "test-key" }).responses("openai.gpt-oss-120b")
-      const item = { type: "reasoning", id: "msg_95d4d0af4350432a", encrypted_content: "mantle-state" }
-      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Think." })).pipe(
-        Effect.provide(
-          fixedResponse(
-            sseEvents(
-              { type: "response.output_item.added", item },
-              { type: "response.reasoning_summary_text.delta", item_id: item.id, delta: "Considering." },
-              { type: "response.output_item.done", item },
-              { type: "response.completed", response: { id: "resp_1" } },
-            ),
-          ),
-        ),
-      )
-
-      const prepared = yield* compileRequest(
-        LLM.request({ model, messages: [response.message, Message.user("Continue.")] }),
-      )
-
-      expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
-        mantle: { itemId: "msg_95d4d0af4350432a", reasoningEncryptedContent: "mantle-state" },
-      })
-      expect(prepared.body.input).toEqual([
-        {
-          type: "reasoning",
-          id: "msg_95d4d0af4350432a",
-          summary: [{ type: "summary_text", text: "Considering." }],
-          encrypted_content: "mantle-state",
-        },
-        { role: "user", content: [{ type: "input_text", text: "Continue." }] },
-      ])
-    }),
-  )
 })

 const recorded = recordedTests({
@@ -126,7 +126,7 @@ describe("Cloudflare", () => {
      expect(response.reasoning).toBe("Thinking")
      expect(response.events.filter(LLMEvent.is.reasoningDelta)).toHaveLength(2)
      expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
-        "cloudflare-ai-gateway": { reasoningField: "reasoning", reasoningDetails: merged },
+        openai: { reasoningField: "reasoning", reasoningDetails: merged },
      })

      const replay = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
@@ -515,10 +515,7 @@ describe("Gemini route", () => {
        {
          role: "model",
          parts: [
-            {
-              functionCall: { id: "call_image", name: "read", args: { path: "pixel.png" } },
-              thoughtSignature: "sig_1",
-            },
+            { functionCall: { id: "call_image", name: "read", args: { path: "pixel.png" } }, thoughtSignature: "sig_1" },
          ],
        },
        {
@@ -609,7 +606,10 @@ describe("Gemini route", () => {
      expect(prepared.body.contents).toEqual([
        {
          role: "model",
-          parts: [{ functionCall: { name: "shot", args: {} } }, { functionCall: { name: "shot", args: {} } }],
+          parts: [
+            { functionCall: { name: "shot", args: {} } },
+            { functionCall: { name: "shot", args: {} } },
+          ],
        },
        {
          role: "user",
@@ -1071,9 +1071,7 @@ describe("Gemini route", () => {
      const prepared = yield* compileRequest(
        LLM.request({
          model,
-          messages: [
-            Message.assistant([{ type: "text", text: "All done.", providerMetadata: delta?.providerMetadata }]),
-          ],
+          messages: [Message.assistant([{ type: "text", text: "All done.", providerMetadata: delta?.providerMetadata }])],
        }),
      )
      expect(prepared.body.contents).toEqual([
@@ -1574,7 +1572,9 @@ describe("Gemini route", () => {
              { candidates: [{ content: { role: "model", parts: null } }] },
              { candidates: [{ content: null, finishReason: null }] },
              {
-                candidates: [{ content: { role: "model", parts: [{ text: "Hello" }] }, finishReason: "STOP" as const }],
+                candidates: [
+                  { content: { role: "model", parts: [{ text: "Hello" }] }, finishReason: "STOP" as const },
+                ],
              },
            ),
          ),
@@ -1,6 +1,5 @@
 import * as Anthropic from "../../src/providers/anthropic.js"
 import * as AnthropicCompatible from "../../src/providers/anthropic-compatible.js"
-import { Cerebras, DeepInfra, TogetherAI } from "../../src/providers/index.js"
 import { CloudflareAIGateway, CloudflareWorkersAI } from "../../src/providers/cloudflare.js"
 import * as Google from "../../src/providers/google.js"
 import * as OpenAI from "../../src/providers/openai.js"
@@ -48,16 +47,14 @@ const cloudflareWorkersAITools = cloudflareWorkers.model("@cf/openai/gpt-oss-20b
 const deepseek = OpenAICompatible.deepseek
  .configure({ apiKey: process.env.DEEPSEEK_API_KEY ?? "fixture" })
  .model("deepseek-chat")
-const together = TogetherAI.configure({
-  apiKey: process.env.TOGETHER_API_KEY ?? process.env.TOGETHER_AI_API_KEY ?? "fixture",
-}).model("meta-llama/Llama-3.3-70B-Instruct-Turbo")
-const cerebras = Cerebras.configure({ apiKey: process.env.CEREBRAS_API_KEY ?? "fixture" }).model("gpt-oss-120b")
+const together = OpenAICompatible.togetherai
+  .configure({
+    apiKey: process.env.TOGETHER_AI_API_KEY ?? "fixture",
+  })
+  .model("meta-llama/Llama-3.3-70B-Instruct-Turbo")
 const groq = OpenAICompatible.groq
  .configure({ apiKey: process.env.GROQ_API_KEY ?? "fixture" })
  .model("llama-3.3-70b-versatile")
-const deepInfra = DeepInfra.configure({ apiKey: process.env.DEEPINFRA_API_KEY ?? "fixture" }).model(
-  "meta-llama/Llama-3.3-70B-Instruct-Turbo",
-)
 const openRouter = OpenRouter.configure({ apiKey: process.env.OPENROUTER_API_KEY ?? "fixture" })
 const openrouter = openRouter.model("openai/gpt-4o-mini")
 const openrouterGpt55 = openRouter.model("openai/gpt-5.5")
@@ -196,27 +193,8 @@ describeRecordedGoldenScenarios([
    name: "TogetherAI Llama 3.3 70B",
    prefix: "openai-compatible-chat",
    model: together,
-    requires: ["TOGETHER_API_KEY"],
-    scenarios: [
-      {
-        id: "text",
-        cassette: "openai-compatible-chat/togetherai-streams-text",
-        prompt: "Reply with exactly: Hello!",
-        maxTokens: 20,
-      },
-      { id: "tool-call", cassette: "openai-compatible-chat/togetherai-streams-tool-call" },
-    ],
-  },
-  {
-    name: "Cerebras GPT OSS 120B",
-    prefix: "cerebras-chat",
-    model: cerebras,
-    requires: ["CEREBRAS_API_KEY"],
-    scenarios: [
-      { id: "text", maxTokens: 256, temperature: false },
-      { id: "tool-call", maxTokens: 512, temperature: false },
-      { id: "tool-loop", maxTokens: 512, temperature: false, timeout: 30_000 },
-    ],
+    requires: ["TOGETHER_AI_API_KEY"],
+    scenarios: ["text", "tool-call"],
  },
  {
    name: "Groq Llama 3.3 70B",
@@ -225,13 +203,6 @@ describeRecordedGoldenScenarios([
    requires: ["GROQ_API_KEY"],
    scenarios: ["text", "tool-call", { id: "tool-loop", timeout: 30_000 }],
  },
-  {
-    name: "DeepInfra Llama 3.3 70B",
-    prefix: "deepinfra-chat",
-    model: deepInfra,
-    requires: ["DEEPINFRA_API_KEY"],
-    scenarios: ["text", "tool-call", { id: "tool-loop", timeout: 30_000 }],
-  },
  {
    name: "OpenRouter gpt-4o-mini",
    prefix: "openai-compatible-chat",
@@ -26,7 +26,9 @@ const recorded = recordedTests({
 describe("Google Vertex Gemini recorded", () => {
  recorded.effect("streams text", () =>
    Effect.gen(function* () {
-      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Reply with exactly one word: hello" }))
+      const response = yield* LLMClient.generate(
+        LLM.request({ model, prompt: "Reply with exactly one word: hello" }),
+      )

      expect(response.text.toLowerCase()).toContain("hello")
    }),
@@ -6,7 +6,7 @@ import { GoogleVertex, GoogleVertexChat, GoogleVertexMessages, GoogleVertexRespo
 import { LLMClient } from "../../src/route.js"
 import { compileRequest } from "../../src/route/client.js"
 import { it } from "../lib/effect.js"
-import { dynamicResponse, fixedResponse } from "../lib/http.js"
+import { dynamicResponse } from "../lib/http.js"
 import { deltaChunk, finishChunk } from "../lib/openai-chunks.js"
 import { sseEvents } from "../lib/sse.js"

@@ -89,7 +89,7 @@ describe("Google Vertex providers", () => {
                id: "call_1",
                name: "lookup",
                input: { query: "weather" },
-                providerMetadata: { vertex: { functionCallId: "provider_call_1" } },
+                providerMetadata: { google: { functionCallId: "provider_call_1" } },
              }),
            ]),
            Message.tool({
@@ -97,7 +97,7 @@ describe("Google Vertex providers", () => {
              name: "lookup",
              result: "sunny",
              resultType: "text",
-              providerMetadata: { vertex: { functionCallId: "provider_call_1" } },
+              providerMetadata: { google: { functionCallId: "provider_call_1" } },
            }),
          ],
        }),
@@ -122,91 +122,6 @@ describe("Google Vertex providers", () => {
    }),
  )

-  it.effect("round-trips Vertex Gemini metadata through signed content, tool calls, and usage", () =>
-    Effect.gen(function* () {
-      const model = GoogleVertex.configure({
-        accessToken: "vertex-token",
-        project: "vertex-project",
-      }).model("gemini-3.5-flash")
-      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Check the weather." })).pipe(
-        Effect.provide(
-          fixedResponse(
-            sseEvents({
-              candidates: [
-                {
-                  content: {
-                    role: "model",
-                    parts: [
-                      { text: "Thinking.", thought: true, thoughtSignature: "reasoning_sig" },
-                      { text: "Checking.", thoughtSignature: "text_sig" },
-                      {
-                        functionCall: { id: "provider_call_1", name: "lookup", args: { query: "weather" } },
-                        thoughtSignature: "tool_sig",
-                      },
-                    ],
-                  },
-                  finishReason: "STOP",
-                },
-              ],
-              promptFeedback: { blockReasonMessage: "Reviewed" },
-              usageMetadata: { promptTokenCount: 5, candidatesTokenCount: 2, thoughtsTokenCount: 1 },
-            }),
-          ),
-        ),
-      )
-      const reasoning = response.events.find((event) => event.type === "reasoning-end")
-      const text = response.events.find((event) => event.type === "text-delta")
-      const toolCall = response.toolCalls[0]
-
-      expect(reasoning?.providerMetadata).toEqual({ vertex: { thoughtSignature: "reasoning_sig" } })
-      expect(text?.providerMetadata).toEqual({ vertex: { thoughtSignature: "text_sig" } })
-      expect(toolCall).toMatchObject({
-        id: "provider_call_1",
-        providerMetadata: { vertex: { thoughtSignature: "tool_sig" } },
-      })
-      expect(response.usage?.providerMetadata).toEqual({
-        vertex: { promptTokenCount: 5, candidatesTokenCount: 2, thoughtsTokenCount: 1 },
-      })
-      expect(response.events.at(-1)?.providerMetadata).toEqual({
-        vertex: { promptFeedback: { blockReasonMessage: "Reviewed" } },
-      })
-
-      const prepared = yield* compileRequest(
-        LLM.request({
-          model,
-          messages: [
-            Message.assistant([
-              { type: "reasoning", text: "Thinking.", providerMetadata: reasoning?.providerMetadata },
-              { type: "text", text: "Checking.", providerMetadata: text?.providerMetadata },
-              ToolCallPart.make({
-                id: toolCall.id,
-                name: toolCall.name,
-                input: toolCall.input,
-                providerMetadata: toolCall.providerMetadata,
-              }),
-            ]),
-            Message.tool({ id: toolCall.id, name: toolCall.name, result: "sunny", resultType: "text" }),
-          ],
-        }),
-      )
-
-      expect(prepared.body.contents).toEqual([
-        {
-          role: "model",
-          parts: [
-            { text: "Thinking.", thought: true, thoughtSignature: "reasoning_sig" },
-            { text: "Checking.", thoughtSignature: "text_sig" },
-            { functionCall: { name: "lookup", args: { query: "weather" } }, thoughtSignature: "tool_sig" },
-          ],
-        },
-        {
-          role: "user",
-          parts: [{ functionResponse: { name: "lookup", response: { name: "lookup", content: "sunny" } } }],
-        },
-      ])
-    }),
-  )
-
  it.effect("projects Anthropic Messages onto the Vertex raw-predict API", () =>
    Effect.gen(function* () {
      const model = GoogleVertexMessages.configure({
@@ -1,185 +0,0 @@
-import { configure } from "@opencode-ai/ai/providers/groq"
-import { describe, expect } from "bun:test"
-import { Effect } from "effect"
-import { LLM, LLMEvent, LLMRequest, LLMResponse, Message, ToolChoice, ToolDefinition } from "../../src/index.js"
-import { LLMClient } from "../../src/route.js"
-import { compileRequest } from "../../src/route/client.js"
-import { recordedTests } from "../recorded-test.js"
-
-const apiKey = process.env.GROQ_API_KEY ?? "fixture"
-const recorded = recordedTests({
-  prefix: "groq-chat",
-  provider: "groq",
-  protocol: "groq-chat",
-  requires: ["GROQ_API_KEY"],
-})
-
-const weather = ToolDefinition.make({
-  name: "lookup_weather",
-  description: "Look up the current weather for a city",
-  inputSchema: {
-    type: "object",
-    properties: { city: { type: "string", enum: ["Paris", "London"] } },
-    required: ["city"],
-    additionalProperties: false,
-  },
-})
-
-describe("Groq recorded", () => {
-  recorded.effect.with(
-    "streams text with usage",
-    { tags: ["text", "usage"], metadata: { model: "openai/gpt-oss-20b" } },
-    () =>
-      Effect.gen(function* () {
-        const request = LLM.request({
-          model: configure({
-            apiKey,
-            providerOptions: {
-              includeReasoning: false,
-              reasoningEffort: "low",
-              serviceTier: "on_demand",
-              user: "recorded-test",
-            },
-          }).model("openai/gpt-oss-20b"),
-          prompt: "Reply with exactly one word: hello",
-          generation: { maxTokens: 512 },
-        })
-        const compiled = yield* compileRequest(request)
-        expect(compiled.body).toMatchObject({
-          max_completion_tokens: 512,
-          stream_options: { include_usage: true },
-          include_reasoning: false,
-          service_tier: "on_demand",
-          user: "recorded-test",
-        })
-        expect(compiled.body.max_tokens).toBeUndefined()
-        expect(compiled.body.store).toBeUndefined()
-        expect(compiled.body.reasoning_format).toBeUndefined()
-
-        const response = yield* LLMClient.generate(request)
-        expect(response.text.toLowerCase().trim()).toBe("hello")
-        expect(response.reasoning).toBe("")
-        expect(response.events.some(LLMEvent.is.textDelta)).toBe(true)
-        expectUsage(response)
-      }),
-    60_000,
-  )
-
-  for (const item of [
-    {
-      name: "continues Qwen parallel tool calls",
-      model: configure({ apiKey, providerOptions: { parallelToolCalls: true, reasoningEffort: "none" } }).model(
-        "qwen/qwen3.6-27b",
-      ),
-      cities: ["Paris", "London"],
-      reasoning: false,
-    },
-    {
-      name: "replays GPT OSS reasoning through a tool loop",
-      model: configure({ apiKey, providerOptions: { includeReasoning: true, reasoningEffort: "low" } }).model(
-        "openai/gpt-oss-20b",
-      ),
-      cities: ["Paris"],
-      reasoning: true,
-    },
-  ]) {
-    recorded.effect.with(
-      item.name,
-      {
-        tags: ["tool", "tool-loop", "usage", item.reasoning ? "reasoning" : "parallel"],
-        metadata: { model: item.model.id },
-      },
-      () =>
-        Effect.gen(function* () {
-          const request = LLM.request({
-            model: item.model,
-            prompt: `Look up the current weather in ${item.cities.join(" and ")}. Call lookup_weather once for each city in the same response before answering. After receiving all results, report each city's weather in one short sentence.`,
-            tools: [weather],
-            toolChoice: "required",
-            generation: { maxTokens: 1536 },
-          })
-          const compiled = yield* compileRequest(request)
-          expect(compiled.body.stream_options).toEqual({ include_usage: true })
-          expect(compiled.body.store).toBeUndefined()
-          expect(compiled.body.reasoning_format).toBe(item.reasoning ? undefined : "parsed")
-          expect(compiled.body.tools[0].function.strict).toBeUndefined()
-          if (!item.reasoning) expect(compiled.body.parallel_tool_calls).toBe(true)
-
-          const first = yield* LLMClient.generate(request)
-          expect(first.finishReason.normalized).toBe("tool-calls")
-          expect(first.toolCalls).toHaveLength(item.cities.length)
-          expect(new Set(first.toolCalls.map((call) => call.id)).size).toBe(item.cities.length)
-          expect(first.toolCalls.map((call) => call.input)).toEqual(
-            expect.arrayContaining(item.cities.map((city) => ({ city }))),
-          )
-          expect(first.toolCalls.every((call) => call.name === "lookup_weather")).toBe(true)
-          expectUsage(first)
-          if (item.reasoning) {
-            expect(first.reasoning.length).toBeGreaterThan(0)
-            expect(first.events.some(LLMEvent.is.reasoningDelta)).toBe(true)
-          }
-
-          const followUp = LLMRequest.update(request, {
-            toolChoice: ToolChoice.make("none"),
-            messages: [
-              ...request.messages,
-              first.message,
-              ...first.toolCalls.map((call) =>
-                Message.tool({ id: call.id, name: call.name, result: { condition: "sunny", temperature: "18C" } }),
-              ),
-            ],
-          })
-          const replay = yield* compileRequest(followUp)
-          if (item.reasoning) {
-            expect(replay.body.messages).toEqual(
-              expect.arrayContaining([expect.objectContaining({ role: "assistant", reasoning: first.reasoning })]),
-            )
-          }
-          expect(replay.body.reasoning_format).toBe(item.reasoning ? undefined : "parsed")
-
-          const second = yield* LLMClient.generate(followUp)
-          expect(second.finishReason.normalized).toBe("stop")
-          expect(second.toolCalls).toHaveLength(0)
-          expect(second.text.toLowerCase()).toContain("sunny")
-          item.cities.forEach((city) => expect(second.text).toContain(city))
-          expectUsage(second)
-        }),
-      60_000,
-    )
-  }
-
-  recorded.effect.with(
-    "streams Qwen parsed reasoning",
-    { tags: ["reasoning", "usage"], metadata: { model: "qwen/qwen3.6-27b" } },
-    () =>
-      Effect.gen(function* () {
-        const request = LLM.request({
-          model: configure({
-            apiKey,
-            providerOptions: { reasoningEffort: "default" },
-          }).model("qwen/qwen3.6-27b"),
-          prompt:
-            "What is 173 multiplied by 219? Think through the arithmetic, then reply with only the final integer.",
-          generation: { maxTokens: 2048 },
-        })
-        const compiled = yield* compileRequest(request)
-        expect(compiled.body).toMatchObject({ reasoning_format: "parsed", reasoning_effort: "default" })
-        expect(compiled.body.include_reasoning).toBeUndefined()
-
-        const response = yield* LLMClient.generate(request)
-        expect(response.text.replaceAll(",", "").trim()).toBe("37887")
-        expect(response.text).not.toContain("<think>")
-        expect(response.reasoning.length).toBeGreaterThan(0)
-        expect(response.events.some(LLMEvent.is.reasoningDelta)).toBe(true)
-        expectUsage(response)
-      }),
-    60_000,
-  )
-})
-
-function expectUsage(response: LLMResponse) {
-  expect(response.usage).toBeDefined()
-  expect(response.usage?.inputTokens).toBeGreaterThan(0)
-  expect(response.usage?.outputTokens).toBeGreaterThan(0)
-  expect(response.events.filter(LLMEvent.is.finish)).toHaveLength(1)
-}
@@ -1,112 +0,0 @@
-import { expect } from "bun:test"
-import { Effect } from "effect"
-import { LanguageModel, LLM, Message } from "../../src/index.js"
-import { OpenAIChat } from "../../src/protocols/openai-chat.js"
-import { Groq } from "../../src/providers/groq.js"
-import { compileRequest } from "../../src/route/client.js"
-import { it } from "../lib/effect.js"
-import { weatherTool } from "../recorded-scenarios.js"
-
-it.effect("Groq reuses Chat streaming and defaults to parsed reasoning", () =>
-  Effect.gen(function* () {
-    expect(Groq.protocol.stream).toBe(OpenAIChat.protocol.stream)
-    const model = Groq.configure({ apiKey: "fixture" }).model("llama-3.3-70b-versatile")
-    expect(model.route.endpoint.baseURL).toBe("https://api.groq.com/openai/v1")
-    const compiled = yield* compileRequest(
-      LLM.request({ model, prompt: "Hello", tools: [weatherTool], generation: { maxTokens: 64 } }),
-    )
-    expect(compiled.body).toMatchObject({
-      max_completion_tokens: 64,
-      stream_options: { include_usage: true },
-      reasoning_format: "parsed",
-    })
-    for (const key of ["store", "max_tokens", "include_reasoning", "parallel_tool_calls", "service_tier", "user"])
-      expect(compiled.body[key]).toBeUndefined()
-    expect(compiled.body.tools?.[0]?.function).not.toHaveProperty("strict")
-  }),
-)
-
-it.effect("Groq lowers its own options for custom catalog identities and endpoints", () =>
-  Effect.gen(function* () {
-    const model = LanguageModel.update(
-      Groq.model("qwen/qwen3.6-27b", {
-        apiKey: "fixture",
-        baseURL: "https://gateway.example/v1",
-        headers: { "x-client": "test" },
-        body: { custom: "value" },
-        providerOptions: {
-          reasoningEffort: "default",
-          parallelToolCalls: true,
-          serviceTier: "flex",
-          user: "test-user",
-        },
-      }),
-      { provider: "custom-groq" },
-    )
-    const compiled = yield* compileRequest(
-      LLM.request({ model, prompt: "Hello", providerOptions: { parallelToolCalls: false, includeReasoning: false } }),
-    )
-    expect(model.route.endpoint.baseURL).toBe("https://gateway.example/v1")
-    expect(model.route.defaults.headers).toEqual({ "x-client": "test" })
-    expect(model.route.defaults.http?.body).toEqual({ custom: "value" })
-    expect(compiled.body).toMatchObject({
-      reasoning_effort: "default",
-      reasoning_format: "parsed",
-      parallel_tool_calls: false,
-      service_tier: "flex",
-      user: "test-user",
-    })
-    expect(compiled.body.include_reasoning).toBeUndefined()
-    for (const key of ["reasoningFormat", "reasoningEffort", "parallelToolCalls", "serviceTier"])
-      expect(compiled.body).not.toHaveProperty(key)
-  }),
-)
-
-it.effect("Groq replays reasoning only when present and preserves explicit reasoning exclusion", () =>
-  Effect.gen(function* () {
-    const compiled = yield* compileRequest(
-      LLM.request({
-        model: Groq.configure({ apiKey: "fixture" }).model("openai/gpt-oss-20b"),
-        messages: [
-          Message.user("Think"),
-          Message.assistant([
-            { type: "reasoning", text: "Thinking" },
-            { type: "text", text: "Answer" },
-          ]),
-          Message.user("Again"),
-          Message.assistant("Answer only"),
-          Message.user("Continue"),
-        ],
-        providerOptions: { reasoningEffort: "low", includeReasoning: false },
-      }),
-    )
-    expect(compiled.body).toMatchObject({ reasoning_effort: "low", include_reasoning: false })
-    expect(compiled.body.reasoning_format).toBeUndefined()
-    expect(compiled.body.messages[1]).toMatchObject({ reasoning: "Thinking", content: "Answer" })
-    expect(compiled.body.messages[1]).not.toHaveProperty("reasoning_content")
-    expect(compiled.body.messages[3]).not.toHaveProperty("reasoning")
-  }),
-)
-
-it.effect("Groq omits reasoning_format for the GPT-OSS family by default", () =>
-  Effect.gen(function* () {
-    for (const id of ["openai/gpt-oss-20b", "openai/gpt-oss-120b", "openai/gpt-oss-safeguard-20b"]) {
-      const compiled = yield* compileRequest(
-        LLM.request({ model: Groq.configure({ apiKey: "fixture" }).model(id), prompt: "Hello" }),
-      )
-      expect(compiled.body.reasoning_format).toBeUndefined()
-      expect(compiled.body.include_reasoning).toBeUndefined()
-    }
-  }),
-)
-
-it.effect("Groq validates option types", () =>
-  Effect.gen(function* () {
-    for (const providerOptions of [{ includeReasoning: "false" }, { parallelToolCalls: "false" }]) {
-      const error = yield* compileRequest(
-        LLM.request({ model: Groq.configure({ apiKey: "fixture" }).model("qwen"), prompt: "Hello", providerOptions }),
-      ).pipe(Effect.flip)
-      expect(error.reason._tag).toBe("InvalidRequest")
-    }
-  }),
-)
@@ -1,257 +0,0 @@
-import { describe, expect } from "bun:test"
-import { ConfigProvider, Effect } from "effect"
-import { HttpClientRequest } from "effect/unstable/http"
-import { LLM, Message, ToolDefinition } from "../../src/index.js"
-import {
-  AmazonBedrock,
-  AmazonBedrockMantle,
-  Anthropic,
-  AnthropicCompatible,
-  Azure,
-  Cerebras,
-  CloudflareAIGateway,
-  CloudflareWorkersAI,
-  DeepInfra,
-  Google,
-  GoogleVertex,
-  GoogleVertexChat,
-  GoogleVertexMessages,
-  GoogleVertexResponses,
-  Groq,
-  OpenAI,
-  OpenAICompatible,
-  OpenAICompatibleResponses,
-  OpenRouter,
-  TogetherAI,
-  XAI,
-} from "../../src/providers/index.js"
-import { compileRequest } from "../../src/route/client.js"
-import { it } from "../lib/effect.js"
-import { dynamicResponse } from "../lib/http.js"
-import { sseEvents } from "../lib/sse.js"
-
-describe("native OpenAI-compatible providers", () => {
-  it.effect("assigns provider-owned metadata namespaces across native routes", () =>
-    Effect.gen(function* () {
-      const vertex = { project: "project", accessToken: "token" }
-      const providers = [
-        [OpenAI.configure({ apiKey: "test" }).chat("model"), "openai"],
-        [OpenAI.configure({ apiKey: "test" }).responses("model"), "openai"],
-        [Azure.configure({ resourceName: "resource", apiKey: "test" }).chat("model"), "azure"],
-        [Azure.configure({ resourceName: "resource", apiKey: "test" }).responses("model"), "azure"],
-        [AmazonBedrock.configure({ apiKey: "test" }).model("model"), "bedrock"],
-        [AmazonBedrockMantle.configure({ apiKey: "test" }).chat("model"), "mantle"],
-        [AmazonBedrockMantle.configure({ apiKey: "test" }).responses("model"), "mantle"],
-        [Google.configure({ apiKey: "test" }).model("model"), "google"],
-        [GoogleVertex.configure(vertex).model("model"), "vertex"],
-        [GoogleVertexChat.configure(vertex).model("model"), "vertex"],
-        [GoogleVertexResponses.configure(vertex).model("model"), "vertex"],
-        [GoogleVertexMessages.configure(vertex).model("model"), "anthropic"],
-        [Anthropic.configure({ apiKey: "test" }).model("model"), "anthropic"],
-        [
-          AnthropicCompatible.configure({ baseURL: "https://example.test/v1", provider: "minimax" }).model("model"),
-          "minimax",
-        ],
-        [
-          OpenAICompatible.configure({ baseURL: "https://example.test/v1", provider: "custom" }).model("model"),
-          "custom",
-        ],
-        [
-          OpenAICompatibleResponses.configure({ baseURL: "https://example.test/v1", provider: "custom" }).model(
-            "model",
-          ),
-          "custom",
-        ],
-        [Cerebras.configure({ apiKey: "test" }).model("model"), "cerebras"],
-        [DeepInfra.configure({ apiKey: "test" }).model("model"), "deepinfra"],
-        [TogetherAI.configure({ apiKey: "test" }).model("model"), "togetherai"],
-        [CloudflareAIGateway.configure({ accountId: "account" }).model("model"), "cloudflare-ai-gateway"],
-        [CloudflareWorkersAI.configure({ accountId: "account" }).model("model"), "cloudflare-workers-ai"],
-        [OpenRouter.configure({ apiKey: "test" }).model("model"), "openrouter"],
-        [XAI.configure({ apiKey: "test" }).chat("model"), "xai"],
-        [XAI.configure({ apiKey: "test" }).responses("model"), "xai"],
-      ] as const
-
-      for (const [model, key] of providers) expect(model.route.providerMetadataKey).toBe(key)
-    }),
-  )
-
-  it.effect("preserves native Together AI and Cerebras provider and route identities", () =>
-    Effect.gen(function* () {
-      const together = TogetherAI.configure({ apiKey: "fixture" }).model("meta-llama/Llama-3.3-70B")
-      const cerebras = Cerebras.configure({ apiKey: "fixture" }).model("qwen-3-235b-a22b")
-
-      expect(together).toMatchObject({
-        provider: "togetherai",
-        compatibility: { maxTokensField: "max_tokens", supportsStore: false, supportsStrictMode: false },
-        route: { id: "togetherai-chat", protocol: "openai-chat" },
-      })
-      expect(together.route.endpoint.baseURL).toBe("https://api.together.xyz/v1")
-      expect(cerebras).toMatchObject({
-        provider: "cerebras",
-        compatibility: { maxTokensField: "max_tokens", reasoningField: "reasoning", supportsStore: false },
-        route: { id: "cerebras-chat", protocol: "openai-chat" },
-      })
-      expect(cerebras.route.endpoint.baseURL).toBe("https://api.cerebras.ai/v1")
-    }),
-  )
-
-  it.effect("preserves native DeepInfra provider and route identity", () =>
-    Effect.gen(function* () {
-      const deepinfra = DeepInfra.configure({ apiKey: "fixture" }).model("google/gemma-3-27b-it")
-      expect(deepinfra).toMatchObject({
-        provider: "deepinfra",
-        compatibility: { maxTokensField: "max_tokens", reasoningField: "reasoning_content", supportsStore: false },
-        route: { id: "deepinfra-chat", protocol: "openai-chat" },
-      })
-      expect(deepinfra.route.endpoint.baseURL).toBe("https://api.deepinfra.com/v1/openai")
-    }),
-  )
-
-  it.effect("applies native provider request defaults even with a custom gateway URL", () =>
-    Effect.gen(function* () {
-      const together = yield* compileRequest(
-        LLM.request({
-          model: TogetherAI.configure({ apiKey: "fixture", baseURL: "https://gateway.example/v1" }).model("llama"),
-          prompt: "Use a tool.",
-          generation: { maxTokens: 32 },
-          tools: [
-            ToolDefinition.make({ name: "lookup", description: "Look up data", inputSchema: { type: "object" } }),
-          ],
-          providerOptions: { store: true },
-        }),
-      )
-
-      expect(together.body).toMatchObject({
-        max_tokens: 32,
-        stream_options: { include_usage: true },
-        tools: [{ function: { name: "lookup" } }],
-      })
-      expect(together.body).not.toHaveProperty("max_completion_tokens")
-      expect(together.body).not.toHaveProperty("store")
-      expect(together.body.tools?.[0]?.function).not.toHaveProperty("strict")
-
-      const cerebras = yield* compileRequest(
-        LLM.request({
-          model: Cerebras.configure({ apiKey: "fixture", baseURL: "https://gateway.example/v1" }).model("qwen"),
-          generation: { maxTokens: 48 },
-          messages: [
-            Message.user("Think first."),
-            Message.assistant([
-              { type: "reasoning", text: "A deliberate thought." },
-              { type: "text", text: "An answer." },
-            ]),
-            Message.user("Continue."),
-          ],
-          providerOptions: { store: true },
-        }),
-      )
-
-      expect(cerebras.body).toMatchObject({
-        max_tokens: 48,
-        messages: [
-          { role: "user", content: "Think first." },
-          { role: "assistant", content: "An answer.", reasoning: "A deliberate thought." },
-          { role: "user", content: "Continue." },
-        ],
-      })
-      expect(cerebras.body).not.toHaveProperty("max_completion_tokens")
-      expect(cerebras.body).not.toHaveProperty("store")
-      expect(cerebras.body.messages[1]).not.toHaveProperty("reasoning_content")
-    }),
-  )
-
-  it.effect("normalizes DeepInfra API roots without duplicating the OpenAI path", () =>
-    Effect.gen(function* () {
-      for (const baseURL of [
-        "https://gateway.example/v1",
-        "https://gateway.example/v1/",
-        "https://gateway.example/v1/openai",
-        "https://gateway.example/v1/openai/",
-      ]) {
-        expect(DeepInfra.configure({ apiKey: "fixture", baseURL }).model("gemma").route.endpoint.baseURL).toBe(
-          "https://gateway.example/v1/openai",
-        )
-      }
-    }),
-  )
-
-  it.effect("maps package settings onto native executable models", () =>
-    Effect.gen(function* () {
-      for (const native of [TogetherAI, Cerebras]) {
-        const selected = native.model("provider-model", {
-          apiKey: "fixture",
-          baseURL: "https://gateway.example/v1",
-          headers: { "x-application": "opencode" },
-          body: { service_tier: "priority" },
-          providerOptions: { reasoningEffort: "high" },
-        })
-
-        expect(selected.route.endpoint.baseURL).toBe("https://gateway.example/v1")
-        expect(selected.route.defaults.headers).toEqual({ "x-application": "opencode" })
-        expect(selected.route.defaults.http?.body).toEqual({ service_tier: "priority" })
-        expect(selected.route.defaults.providerOptions).toEqual({ reasoningEffort: "high" })
-      }
-    }),
-  )
-
-  it.effect("resolves provider environment credentials and preserves deprecated Together credentials", () =>
-    Effect.gen(function* () {
-      const scenarios = [
-        {
-          model: TogetherAI.configure().model("llama"),
-          env: { TOGETHER_API_KEY: "together-primary", TOGETHER_AI_API_KEY: "together-legacy" },
-          token: "together-primary",
-          url: "https://api.together.xyz/v1/chat/completions",
-        },
-        {
-          model: TogetherAI.configure().model("llama"),
-          env: { TOGETHER_AI_API_KEY: "together-legacy" },
-          token: "together-legacy",
-          url: "https://api.together.xyz/v1/chat/completions",
-        },
-        {
-          model: Cerebras.configure().model("qwen"),
-          env: { CEREBRAS_API_KEY: "cerebras-secret" },
-          token: "cerebras-secret",
-          url: "https://api.cerebras.ai/v1/chat/completions",
-        },
-        {
-          model: DeepInfra.configure().model("gemma"),
-          env: { DEEPINFRA_API_KEY: "deepinfra-secret" },
-          token: "deepinfra-secret",
-          url: "https://api.deepinfra.com/v1/openai/chat/completions",
-        },
-        {
-          model: Groq.configure().model("llama"),
-          env: { GROQ_API_KEY: "groq-secret" },
-          token: "groq-secret",
-          url: "https://api.groq.com/openai/v1/chat/completions",
-        },
-      ]
-
-      yield* Effect.forEach(scenarios, (scenario) =>
-        LLM.generate(LLM.request({ model: scenario.model, prompt: "Say hello." })).pipe(
-          Effect.provide(
-            dynamicResponse((input) =>
-              Effect.gen(function* () {
-                const request = yield* HttpClientRequest.toWeb(input.request).pipe(Effect.orDie)
-                expect(request.url).toBe(scenario.url)
-                expect(request.headers.get("authorization")).toBe(`Bearer ${scenario.token}`)
-                return input.respond(
-                  sseEvents(
-                    { id: "chatcmpl_fixture", choices: [{ delta: { content: "Hello" }, finish_reason: null }] },
-                    { id: "chatcmpl_fixture", choices: [{ delta: {}, finish_reason: "stop" }] },
-                  ),
-                  { headers: { "content-type": "text/event-stream" } },
-                )
-              }),
-            ),
-          ),
-          Effect.provide(ConfigProvider.layer(ConfigProvider.fromEnv({ env: scenario.env }))),
-          Effect.tap((response) => Effect.sync(() => expect(response.text).toBe("Hello"))),
-        ),
-      )
-    }),
-  )
-})
@@ -68,13 +68,11 @@ for (const item of cases) {
          expect(response.text.replaceAll(",", "").trim()).toBe("37887")
          expect(response.reasoning.length).toBeGreaterThan(0)
          expect(response.events.some(LLMEvent.is.reasoningDelta)).toBe(true)
-          const metadata = response.message.content.find((part) => part.type === "reasoning")?.providerMetadata?.[
-            item.model.route.providerMetadataKey ?? String(item.model.provider)
-          ]
-          expect(metadata?.reasoningField).toBe(item.structured ? "reasoning" : "reasoning_content")
-          expect(Array.isArray(metadata?.reasoningDetails)).toBe(item.structured)
+          const metadata = response.message.content.find((part) => part.type === "reasoning")?.providerMetadata
+          expect(metadata?.openai?.reasoningField).toBe(item.structured ? "reasoning" : "reasoning_content")
+          expect(Array.isArray(metadata?.openai?.reasoningDetails)).toBe(item.structured)
          if (!item.structured) return
-          const details = metadata?.reasoningDetails
+          const details = metadata?.openai?.reasoningDetails
          if (!Array.isArray(details)) return
          expect(
            details.some(
@@ -128,11 +126,7 @@ for (const item of cases) {
          ).toMatch(/^Paris is sunny\.?$/)
          const details = events
            .filter(LLMEvent.is.reasoningEnd)
-            .map(
-              (event) =>
-                event.providerMetadata?.[item.model.route.providerMetadataKey ?? String(item.model.provider)]
-                  ?.reasoningDetails,
-            )
+            .map((event) => event.providerMetadata?.openai?.reasoningDetails)
            .find(Array.isArray)
          expect(Array.isArray(details)).toBe(item.structured)
          if (!item.structured || !Array.isArray(details)) return
@@ -85,28 +85,6 @@ describe("OpenAI Chat route", () => {
    }),
  )

-  it.effect("omits empty and whitespace-only assistant messages", () =>
-    Effect.gen(function* () {
-      const prepared = yield* compileRequest(
-        LLM.request({
-          model,
-          messages: [
-            Message.user("Before."),
-            Message.assistant([]),
-            Message.assistant(""),
-            Message.assistant(" \n\t "),
-            Message.assistant("After."),
-          ],
-        }),
-      )
-
-      expect(prepared.body.messages).toEqual([
-        { role: "user", content: "Before." },
-        { role: "assistant", content: "After." },
-      ])
-    }),
-  )
-
  it.effect("replays canonical reasoning as OpenAI-compatible reasoning_content", () =>
    Effect.gen(function* () {
      const prepared = yield* compileRequest(
@@ -169,56 +147,6 @@ describe("OpenAI Chat route", () => {
    }),
  )

-  it.effect("preserves observed reasoning fields when reasoning is required", () =>
-    Effect.gen(function* () {
-      const prepared = yield* compileRequest(
-        LLM.request({
-          model: LanguageModel.update(model, { compatibility: { requireReasoning: true } }),
-          messages: [
-            Message.assistant([
-              {
-                type: "reasoning",
-                text: "thinking",
-                providerMetadata: { openai: { reasoningField: "reasoning_text" } },
-              },
-              { type: "text", text: "Hello" },
-            ]),
-            Message.assistant("Done"),
-          ],
-        }),
-      )
-
-      expect(prepared.body.messages).toEqual([
-        { role: "assistant", content: "Hello", reasoning_text: "thinking" },
-        { role: "assistant", content: "Done", reasoning_content: "" },
-      ])
-    }),
-  )
-
-  it.effect("omits empty configured reasoning fields when reasoning is explicitly optional", () =>
-    Effect.gen(function* () {
-      const prepared = yield* compileRequest(
-        LLM.request({
-          model: LanguageModel.update(model, {
-            compatibility: { reasoningField: "reasoning_text", requireReasoning: false },
-          }),
-          messages: [
-            Message.assistant([
-              { type: "reasoning", text: "thinking" },
-              { type: "text", text: "Hello" },
-            ]),
-            Message.assistant("Done"),
-          ],
-        }),
-      )
-
-      expect(prepared.body.messages).toEqual([
-        { role: "assistant", content: "Hello", reasoning_text: "thinking" },
-        { role: "assistant", content: "Done" },
-      ])
-    }),
-  )
-
  it.effect("rejects reasoning fields that conflict with assistant message fields", () =>
    Effect.gen(function* () {
      const error = yield* compileRequest(
@@ -264,21 +192,6 @@ describe("OpenAI Chat route", () => {
    }),
  )

-  it.effect("omits the prompt cache key when caching is disabled", () =>
-    Effect.gen(function* () {
-      const prepared = yield* compileRequest(
-        LLM.request({
-          model,
-          prompt: "Hello",
-          promptCacheKey: "session_123",
-          cache: "none",
-        }),
-      )
-
-      expect(prepared.body).not.toHaveProperty("prompt_cache_key")
-    }),
-  )
-
  it.effect("maps the xAI Chat prompt cache key to conversation affinity", () =>
    LLMClient.generate(
      LLM.request({
@@ -438,35 +351,6 @@ describe("OpenAI Chat route", () => {
    }),
  )

-  it.effect("limits OpenAI and Azure Chat tool call IDs to 40 characters", () =>
-    Effect.gen(function* () {
-      const id = `call_${"a".repeat(48)}`
-      const models = [
-        model,
-        Azure.configure({ baseURL: "https://opencode-test.openai.azure.com/openai/", apiKey: "test" }).chat("gpt-4o"),
-      ]
-
-      yield* Effect.forEach(models, (selected) =>
-        Effect.gen(function* () {
-          const prepared = yield* compileRequest(
-            LLM.request({
-              model: selected,
-              messages: [
-                Message.assistant([ToolCallPart.make({ id, name: "lookup", input: {} })]),
-                Message.tool({ id, name: "lookup", result: "Sunny" }),
-              ],
-            }),
-          )
-
-          expect(prepared.body.messages).toMatchObject([
-            { role: "assistant", tool_calls: [{ id: id.slice(0, 40) }] },
-            { role: "tool", tool_call_id: id.slice(0, 40) },
-          ])
-        }),
-      )
-    }),
-  )
-
  it.effect("preserves structured tool errors for the model", () =>
    Effect.gen(function* () {
      const error = { error: { type: "unknown", message: "Tool execution interrupted" } }
@@ -532,30 +416,6 @@ describe("OpenAI Chat route", () => {
    }),
  )

-  it.effect("bridges image tool results before their synthetic user message when required", () =>
-    Effect.gen(function* () {
-      const prepared = yield* compileRequest(
-        LLM.request({
-          model: LanguageModel.update(model, { compatibility: { requireAssistantAfterTool: true } }),
-          messages: [
-            Message.assistant([ToolCallPart.make({ id: "call_image", name: "read", input: {} })]),
-            Message.tool({
-              id: "call_image",
-              name: "read",
-              result: {
-                type: "content",
-                value: [{ type: "file", uri: "data:image/png;base64,AAECAw==", mime: "image/png", name: "pixel.png" }],
-              },
-            }),
-          ],
-        }),
-      )
-
-      expect(prepared.body.messages.map((message) => message.role)).toEqual(["assistant", "tool", "assistant", "user"])
-      expect(prepared.body.messages[2]).toEqual({ role: "assistant", content: "Done." })
-    }),
-  )
-
  it.effect("orders parallel tool responses before one aggregated vision message", () =>
    Effect.gen(function* () {
      const prepared = yield* compileRequest(
@@ -903,70 +763,6 @@ describe("OpenAI Chat route", () => {
    }),
  )

-  it.effect("uses the configured provider metadata namespace for reasoning and usage", () =>
-    Effect.gen(function* () {
-      const selected = LanguageModel.update(model, {
-        route: { ...model.route, providerMetadataKey: "vendor" },
-      })
-      const details = [{ type: "reasoning.text", text: "thinking", signature: "signed" }]
-      const response = yield* LLMClient.generate(LLMRequest.update(request, { model: selected })).pipe(
-        Effect.provide(
-          fixedResponse(
-            sseEvents(
-              { choices: [{ delta: { reasoning: "thinking", reasoning_details: details } }] },
-              deltaChunk({ content: "Hello" }),
-              deltaChunk({}, "stop"),
-              usageChunk({ prompt_tokens: 5, completion_tokens: 2, total_tokens: 7 }),
-            ),
-          ),
-        ),
-      )
-
-      expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
-        vendor: { reasoningField: "reasoning", reasoningDetails: details },
-      })
-      expect(response.usage?.providerMetadata).toEqual({
-        vendor: { prompt_tokens: 5, completion_tokens: 2, total_tokens: 7 },
-      })
-
-      const replay = yield* compileRequest(LLM.request({ model: selected, messages: [response.message] }))
-      expect(replay.body.messages).toEqual([
-        { role: "assistant", content: "Hello", reasoning: "thinking", reasoning_details: details },
-      ])
-    }),
-  )
-
-  it.effect("falls back to the selected provider for the metadata namespace", () =>
-    Effect.gen(function* () {
-      const compatible = model.route.with({ provider: "deepseek" }).model({ id: "deepseek-chat" })
-      const selected = LanguageModel.update(compatible, {
-        route: { ...compatible.route, providerMetadataKey: undefined },
-      })
-      const response = yield* LLMClient.generate(LLMRequest.update(request, { model: selected })).pipe(
-        Effect.provide(
-          fixedResponse(
-            sseEvents(
-              deltaChunk({ reasoning_content: "thinking" }),
-              deltaChunk({ content: "Hello" }),
-              deltaChunk({}, "stop"),
-              usageChunk({ prompt_tokens: 5, completion_tokens: 2, total_tokens: 7 }),
-            ),
-          ),
-        ),
-      )
-
-      expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
-        deepseek: { reasoningField: "reasoning_content" },
-      })
-      expect(response.usage?.providerMetadata).toEqual({
-        deepseek: { prompt_tokens: 5, completion_tokens: 2, total_tokens: 7 },
-      })
-
-      const replay = yield* compileRequest(LLM.request({ model: selected, messages: [response.message] }))
-      expect(replay.body.messages).toEqual([{ role: "assistant", content: "Hello", reasoning_content: "thinking" }])
-    }),
-  )
-
  it.effect("parses and replays a configured custom reasoning field", () =>
    Effect.gen(function* () {
      const custom = LanguageModel.update(model, { compatibility: { reasoningField: "vendor_reasoning" } })
@@ -238,135 +238,6 @@ describe("OpenAI-compatible Chat route", () => {
    }),
  )

-  it.effect("normalizes tool call IDs for the selected model family", () =>
-    Effect.gen(function* () {
-      const longID = `call_${"a".repeat(48)}`
-      const cases = [
-        { provider: "custom", model: "mistral-small", id: "toolu_01CBhTTz95qkd9LJMdC9sf8t", expected: "toolu01CB" },
-        { provider: "custom", model: "devstral-small", id: "abc", expected: "abc000000" },
-        { provider: "custom", model: "codestral-latest", id: "toolu_01CBhTTz95", expected: "toolu01CB" },
-        { provider: "custom", model: "pixtral-large", id: "toolu_01CBhTTz95", expected: "toolu01CB" },
-        { provider: "custom", model: "open-mixtral-8x22b", id: "toolu_01CBhTTz95", expected: "toolu01CB" },
-        { provider: "gateway", model: "anthropic/claude-sonnet-4", id: "call|item/+", expected: "call_item__" },
-        { provider: "gateway", model: "openai/gpt-4o", id: longID, expected: longID.slice(0, 40) },
-        { provider: "custom", model: "ordinary-model", id: "call|item/+", expected: "call|item/+" },
-        { provider: "mistral", model: "zai-glm-5-2", id: "call_long_identifier", expected: "call_long_identifier" },
-      ]
-
-      yield* Effect.forEach(cases, (item) =>
-        Effect.gen(function* () {
-          const prepared = yield* compileRequest(
-            LLM.request({
-              model: OpenAICompatibleChat.route
-                .with({ provider: item.provider, endpoint: { baseURL: "https://api.custom.test/v1" } })
-                .model({ id: item.model }),
-              messages: [
-                Message.assistant([ToolCallPart.make({ id: item.id, name: "lookup", input: {} })]),
-                Message.tool({ id: item.id, name: "lookup", result: { type: "content", value: [] } }),
-              ],
-            }),
-          )
-
-          expect(prepared.body.messages).toMatchObject([
-            { role: "assistant", tool_calls: [{ id: item.expected }] },
-            { role: "tool", tool_call_id: item.expected },
-          ])
-        }),
-      )
-    }),
-  )
-
-  it.effect("bridges tool results for Mistral-family models and honors compatibility overrides", () =>
-    Effect.gen(function* () {
-      const cases = [
-        { id: "mistral-small", bridge: true },
-        { id: "devstral-small", bridge: true },
-        { id: "codestral-latest", bridge: true },
-        { id: "pixtral-large", bridge: true },
-        { id: "open-mixtral-8x22b", bridge: true },
-        { id: "ordinary-model", bridge: false },
-        { id: "ordinary-model", override: true, bridge: true },
-        { id: "mistral-small", override: false, bridge: false },
-      ] as const
-
-      yield* Effect.forEach(cases, (item) =>
-        Effect.gen(function* () {
-          const selected = OpenAICompatibleChat.route
-            .with({ provider: "custom", endpoint: { baseURL: "https://api.custom.test/v1" } })
-            .model({
-              id: item.id,
-              compatibility: "override" in item ? { requireAssistantAfterTool: item.override } : undefined,
-            })
-          const prepared = yield* compileRequest(
-            LLM.request({
-              model: selected,
-              messages: [
-                Message.assistant([ToolCallPart.make({ id: "call_1", name: "lookup", input: {} })]),
-                Message.tool({ id: "call_1", name: "lookup", result: "Sunny" }),
-                Message.user("What next?"),
-              ],
-            }),
-          )
-
-          expect(prepared.body.messages.map((message) => message.role)).toEqual(
-            item.bridge ? ["assistant", "tool", "assistant", "user"] : ["assistant", "tool", "user"],
-          )
-          if (item.bridge) expect(prepared.body.messages[2]).toEqual({ role: "assistant", content: "Done." })
-        }),
-      )
-    }),
-  )
-
-  it.effect("requires reasoning for DeepSeek models, providers, and endpoints unless explicitly overridden", () =>
-    Effect.gen(function* () {
-      const cases = [
-        { id: "DeepSeek-V3", provider: "custom", baseURL: "https://api.custom.test/v1", required: true },
-        { id: "custom-model", provider: "deepseek", baseURL: "https://api.custom.test/v1", required: true },
-        { id: "custom-model", provider: "custom", baseURL: "https://API.DeepSeek.COM/v1", required: true },
-        { id: "ordinary-model", provider: "custom", baseURL: "https://api.custom.test/v1", required: false },
-        {
-          id: "ordinary-model",
-          provider: "custom",
-          baseURL: "https://api.custom.test/v1",
-          compatibility: { requireReasoning: true, reasoningField: "reasoning" },
-          required: true,
-          field: "reasoning",
-        },
-        {
-          id: "deepseek-chat",
-          provider: "deepseek",
-          baseURL: "https://api.deepseek.com/v1",
-          compatibility: { requireReasoning: false },
-          required: false,
-        },
-      ] as const
-
-      yield* Effect.forEach(cases, (item) =>
-        Effect.gen(function* () {
-          const selected = OpenAICompatibleChat.route
-            .with({ provider: item.provider, endpoint: { baseURL: item.baseURL } })
-            .model({ id: item.id, compatibility: "compatibility" in item ? item.compatibility : undefined })
-          const prepared = yield* compileRequest(
-            LLM.request({
-              model: selected,
-              messages: [
-                Message.assistant("Hello"),
-                Message.assistant([ToolCallPart.make({ id: "call_1", name: "lookup", input: {} })]),
-                Message.tool({ id: "call_1", name: "lookup", result: "Sunny" }),
-              ],
-            }),
-          )
-          const field = "field" in item ? item.field : "reasoning_content"
-
-          for (const message of prepared.body.messages.filter((message) => message.role === "assistant")) {
-            if (item.required) expect(message).toHaveProperty(field, "")
-            else expect(message).not.toHaveProperty(field)
-          }
-        }),
-      )
-    }),
-  )
-
  it.effect("posts to the configured compatible endpoint and parses text usage", () =>
    Effect.gen(function* () {
      const response = yield* LLMClient.generate(request).pipe(
@@ -437,7 +308,7 @@ describe("OpenAI-compatible Chat route", () => {
        outputTokens: undefined,
        totalTokens: undefined,
        providerMetadata: {
-          deepseek: {
+          openai: {
            prompt_tokens: null,
            completion_tokens: null,
            total_tokens: null,
@@ -47,31 +47,15 @@ describe("Open Responses-compatible route", () => {
      })
      expect(prepared.body).toEqual({
        model: "example-model",
-        input: [{ role: "user", content: [{ type: "input_text", text: "Say hello." }] }],
-        instructions: "You are concise.",
+        input: [
+          { role: "system", content: "You are concise." },
+          { role: "user", content: [{ type: "input_text", text: "Say hello." }] },
+        ],
        stream: true,
-        store: false,
-        include: ["reasoning.encrypted_content"],
      })
    }),
  )

-  it.effect("allows callers to override stateless encrypted reasoning defaults", () =>
-    Effect.gen(function* () {
-      const model = configure({
-        apiKey: "test-key",
-        baseURL: "https://responses.example.test/v1",
-        provider: "example",
-      }).model("example-model")
-      const prepared = yield* compileRequest(
-        LLM.request({ model, prompt: "Say hello.", providerOptions: { store: true, include: [] } }),
-      )
-
-      expect(prepared.body.store).toBe(true)
-      expect(prepared.body.include).toBeUndefined()
-    }),
-  )
-
  it.effect("lowers chronological system updates as standard developer messages", () =>
    Effect.gen(function* () {
      const model = configure({
@@ -82,12 +66,10 @@ describe("Open Responses-compatible route", () => {
      const prepared = yield* compileRequest(
        LLM.request({
          model,
-          system: "Initial instructions.",
          messages: [Message.user("Before."), Message.system("Operator update."), Message.assistant("After.")],
        }),
      )

-      expect(prepared.body.instructions).toBe("Initial instructions.")
      expect(prepared.body.input).toEqual([
        { role: "user", content: [{ type: "input_text", text: "Before." }] },
        { role: "developer", content: "Operator update." },
@@ -195,19 +177,15 @@ describe("Open Responses-compatible route", () => {
          model,
          messages: [
            Message.assistant([
-              { type: "text", text: "Kept.", providerMetadata: { "openai-compatible": { itemId: "history_1" } } },
+              // The baseline does not enforce a provider id grammar, so a
+              // non-OpenAI but well-formed token is resent as-is.
+              { type: "text", text: "Kept.", providerMetadata: { openresponses: { itemId: "history_1" } } },
+              // Shape violations are dropped even without a grammar policy.
              {
                type: "text",
-                text: "Long.",
-                providerMetadata: { "openai-compatible": { itemId: `history_${"a".repeat(64)}` } },
+                text: "Dropped.",
+                providerMetadata: { openresponses: { itemId: `m${"a".repeat(64)}` } },
              },
-              {
-                type: "text",
-                text: "Opaque.",
-                providerMetadata: { "openai-compatible": { itemId: "provider_value/with+symbols" } },
-              },
-              { type: "text", text: "No suffix.", providerMetadata: { "openai-compatible": { itemId: "msg_" } } },
-              { type: "text", text: "No prefix.", providerMetadata: { "openai-compatible": { itemId: "_item" } } },
            ]),
          ],
        }),
@@ -222,487 +200,13 @@ describe("Open Responses-compatible route", () => {
        },
        {
          type: "message",
-          id: `history_${"a".repeat(64)}`,
          role: "assistant",
-          content: [{ type: "output_text", text: "Long." }],
-        },
-        {
-          type: "message",
-          id: "provider_value/with+symbols",
-          role: "assistant",
-          content: [{ type: "output_text", text: "Opaque." }],
-        },
-        {
-          type: "message",
-          role: "assistant",
-          content: [
-            { type: "output_text", text: "No suffix." },
-            { type: "output_text", text: "No prefix." },
-          ],
+          content: [{ type: "output_text", text: "Dropped." }],
        },
      ])
    }),
  )

-  it.effect("replays only shared hosted tool items", () =>
-    Effect.gen(function* () {
-      const model = configure({
-        apiKey: "test-key",
-        baseURL: "https://responses.example.test/v1",
-        provider: "example",
-      }).model("example-model")
-      const items = [
-        { type: "web_search_call", id: "ws_1", status: "completed" },
-        { type: "x_search_call", id: "x_search_1", status: "completed" },
-        { type: "future_call", id: "future_1", status: "completed" },
-        { type: "file_search_call", id: "fs_1", queries: "not-an-array" },
-      ]
-      const prepared = yield* compileRequest(
-        LLM.request({
-          model,
-          messages: items.map((item) =>
-            Message.assistant({
-              type: "tool-result",
-              id: item.id,
-              name: item.type,
-              result: { type: "json", value: item },
-              providerExecuted: true,
-              providerMetadata: { example: { itemId: item.id } },
-            }),
-          ),
-        }),
-      )
-
-      expect(prepared.body.input).toEqual([
-        items[0],
-        { role: "user", content: [{ type: "input_text", text: JSON.stringify(items[1]) }] },
-        { role: "user", content: [{ type: "input_text", text: JSON.stringify(items[2]) }] },
-        { role: "user", content: [{ type: "input_text", text: JSON.stringify(items[3]) }] },
-      ])
-    }),
-  )
-
-  it.effect("routes response deltas by output index", () =>
-    Effect.gen(function* () {
-      const model = configure({
-        apiKey: "test-key",
-        baseURL: "https://responses.example.test/v1",
-      }).model("example-model")
-      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Say hello." })).pipe(
-        Effect.provide(
-          fixedResponse(
-            sseEvents(
-              { type: "response.output_item.added", output_index: 2, item: { type: "message", id: "msg_1" } },
-              { type: "response.output_text.delta", output_index: 2, item_id: "wrong_message", delta: "Indexed" },
-              { type: "response.output_item.done", output_index: 2, item: { type: "message", id: "msg_1" } },
-              { type: "response.completed", response: { id: "resp_1" } },
-            ),
-          ),
-        ),
-      )
-
-      expect(response.message.content).toEqual([
-        { type: "text", text: "Indexed", providerMetadata: { "openai-compatible": { itemId: "msg_1" } } },
-      ])
-    }),
-  )
-
-  describe("stream validation", () => {
-    const request = LLM.request({
-      model: configure({ apiKey: "test-key", baseURL: "https://responses.example.test/v1" }).model("example-model"),
-      prompt: "Respond.",
-    })
-
-    const fixtures = [
-      {
-        item: { type: "message" },
-        events: [
-          { type: "response.output_text.delta", delta: "Preserved" },
-          { type: "response.output_text.done", text: "Preserved" },
-          { type: "response.refusal.delta", delta: "Preserved" },
-          { type: "response.refusal.done", refusal: "Preserved" },
-        ],
-      },
-      {
-        item: { type: "reasoning", encrypted_content: "encrypted-state" },
-        events: [
-          { type: "response.reasoning.delta", delta: "Preserved" },
-          { type: "response.reasoning.done", text: "Preserved" },
-          { type: "response.reasoning_summary_text.delta", delta: "Preserved" },
-          { type: "response.reasoning_summary_text.done", text: "Preserved" },
-          { type: "response.reasoning_text.done", text: "Preserved" },
-        ],
-      },
-      {
-        item: { type: "function_call", call_id: "call_1", name: "lookup" },
-        events: [
-          { type: "response.function_call_arguments.delta", delta: '{"query":"Preserved"}' },
-          { type: "response.function_call_arguments.done", arguments: '{"query":"Preserved"}' },
-        ],
-      },
-    ]
-
-    const routings = [
-      { name: "empty item and event IDs", id: "", item_id: "" },
-      { name: "empty event ID with registered index", id: "item_1", item_id: "", output_index: 2 },
-      { name: "empty stored ID with registered index", id: "", item_id: "wrong_item", output_index: 2 },
-      { name: "empty item and event IDs with registered index", id: "", item_id: "", output_index: 2 },
-    ]
-
-    fixtures.forEach((fixture) => {
-      fixture.events.forEach((event) => {
-        routings.forEach((routing) => {
-          it.effect(`${event.type} preserves content with ${routing.name}`, () =>
-            Effect.gen(function* () {
-              const item = { ...fixture.item, id: routing.id }
-              const response = yield* LLMClient.generate(request).pipe(
-                Effect.provide(
-                  fixedResponse(
-                    sseEvents(
-                      { type: "response.output_item.added", output_index: routing.output_index, item },
-                      { ...event, item_id: routing.item_id, output_index: routing.output_index },
-                      { type: "response.output_item.done", output_index: routing.output_index, item },
-                      { type: "response.completed", response: { id: "resp_1" } },
-                    ),
-                  ),
-                ),
-              )
-
-              const metadata = { "openai-compatible": { itemId: routing.id } }
-              if (fixture.item.type === "function_call") {
-                expect(response.toolCalls).toEqual([
-                  expect.objectContaining({
-                    id: "call_1",
-                    name: "lookup",
-                    input: { query: "Preserved" },
-                    providerMetadata: metadata,
-                  }),
-                ])
-                return
-              }
-              if (fixture.item.type === "reasoning") {
-                expect(response.message.content).toEqual([
-                  {
-                    type: "reasoning",
-                    text: "Preserved",
-                    providerMetadata: {
-                      "openai-compatible": { itemId: routing.id, reasoningEncryptedContent: "encrypted-state" },
-                    },
-                  },
-                ])
-                expect(response.events.filter(LLMEvent.is.reasoningEnd)).toHaveLength(1)
-                return
-              }
-              expect(response.message.content).toEqual([
-                { type: "text", text: "Preserved", providerMetadata: metadata },
-              ])
-              expect(response.events.filter(LLMEvent.is.textEnd)).toEqual([
-                expect.objectContaining({ id: routing.id, providerMetadata: metadata }),
-              ])
-            }),
-          )
-        })
-      })
-    })
-
-    routings.forEach((routing) => {
-      it.effect(`preserves reasoning summary boundaries and terminal metadata with ${routing.name}`, () =>
-        Effect.gen(function* () {
-          const address = { item_id: routing.item_id, output_index: routing.output_index }
-          const response = yield* LLMClient.generate(request).pipe(
-            Effect.provide(
-              fixedResponse(
-                sseEvents(
-                  {
-                    type: "response.output_item.added",
-                    output_index: routing.output_index,
-                    item: { type: "reasoning", id: routing.id },
-                  },
-                  { type: "response.reasoning_summary_part.added", ...address, summary_index: 0 },
-                  { type: "response.reasoning_summary_text.delta", ...address, summary_index: 0, delta: "First." },
-                  { type: "response.reasoning_summary_text.done", ...address, summary_index: 0, text: "First." },
-                  { type: "response.reasoning_summary_part.done", ...address, summary_index: 0 },
-                  { type: "response.reasoning_summary_part.added", ...address, summary_index: 1 },
-                  { type: "response.reasoning_summary_text.done", ...address, summary_index: 1, text: "Second." },
-                  { type: "response.reasoning_summary_part.done", ...address, summary_index: 1 },
-                  {
-                    type: "response.completed",
-                    response: { output: [{ type: "reasoning", id: routing.id, encrypted_content: "final-state" }] },
-                  },
-                ),
-              ),
-            ),
-          )
-
-          expect(response.message.content).toEqual([
-            {
-              type: "reasoning",
-              text: "First.",
-              providerMetadata: { "openai-compatible": { itemId: routing.id } },
-            },
-            {
-              type: "reasoning",
-              text: "Second.",
-              providerMetadata: {
-                "openai-compatible": { itemId: routing.id, reasoningEncryptedContent: "final-state" },
-              },
-            },
-          ])
-          expect(response.events.filter(LLMEvent.is.reasoningEnd)).toEqual([
-            expect.objectContaining({
-              id: `${routing.id}:0`,
-              providerMetadata: { "openai-compatible": { itemId: routing.id } },
-            }),
-            expect.objectContaining({
-              id: `${routing.id}:1`,
-              providerMetadata: {
-                "openai-compatible": { itemId: routing.id, reasoningEncryptedContent: "final-state" },
-              },
-            }),
-          ])
-        }),
-      )
-    })
-
-    it.effect("reconciles pending empty-ID function arguments from completed output", () =>
-      Effect.gen(function* () {
-        const item = { type: "function_call", id: "", call_id: "call_1", name: "lookup" }
-        const response = yield* LLMClient.generate(request).pipe(
-          Effect.provide(
-            fixedResponse(
-              sseEvents(
-                { type: "response.output_item.added", item },
-                { type: "response.function_call_arguments.delta", item_id: "", delta: '{"query":"partial' },
-                {
-                  type: "response.completed",
-                  response: { output: [{ ...item, arguments: '{"query":"complete"}' }] },
-                },
-              ),
-            ),
-          ),
-        )
-
-        expect(response.toolCalls).toEqual([
-          expect.objectContaining({
-            id: "call_1",
-            name: "lookup",
-            input: { query: "complete" },
-            providerMetadata: { "openai-compatible": { itemId: "" } },
-          }),
-        ])
-      }),
-    )
-
-    it.effect("treats null output items as no-ops without disturbing registered items", () =>
-      Effect.gen(function* () {
-        const response = yield* LLMClient.generate(request).pipe(
-          Effect.provide(
-            fixedResponse(
-              sseEvents(
-                { type: "response.output_item.added", output_index: 0, item: null },
-                { type: "response.output_item.done", output_index: 0, item: null },
-                { type: "response.output_item.added", output_index: 0, item: { type: "message", id: "msg_1" } },
-                { type: "response.output_text.delta", output_index: 0, item_id: "wrong_item", delta: "Before " },
-                { type: "response.output_item.added", output_index: 0, item: null },
-                { type: "response.output_item.done", output_index: 0, item: null },
-                { type: "response.output_text.delta", output_index: 0, item_id: "wrong_item", delta: "after" },
-                { type: "response.output_item.done", output_index: 0, item: { type: "message", id: "msg_1" } },
-                { type: "response.completed", response: { id: "resp_1" } },
-              ),
-            ),
-          ),
-        )
-
-        expect(response.message.content).toEqual([
-          { type: "text", text: "Before after", providerMetadata: { "openai-compatible": { itemId: "msg_1" } } },
-        ])
-        expect(response.events.map((event) => event.type)).toEqual([
-          "step-start",
-          "text-start",
-          "text-delta",
-          "text-delta",
-          "text-end",
-          "step-finish",
-          "finish",
-        ])
-      }),
-    )
-
-    it.effect("rejects missing, null, and non-string event IDs even with a registered output index", () =>
-      Effect.gen(function* () {
-        yield* Effect.forEach(
-          [
-            ...fixtures.flatMap((fixture) => fixture.events.map((event) => ({ item: fixture.item, event }))),
-            ...["response.reasoning_summary_part.added", "response.reasoning_summary_part.done"].map((type) => ({
-              item: { type: "reasoning" },
-              event: { type, summary_index: 0 },
-            })),
-          ],
-          (fixture) =>
-            Effect.forEach([undefined, null, 0, false, {}, []], (item_id) =>
-              Effect.gen(function* () {
-                const error = yield* LLMClient.generate(request).pipe(
-                  Effect.provide(
-                    fixedResponse(
-                      sseEvents(
-                        {
-                          type: "response.output_item.added",
-                          output_index: 0,
-                          item: { ...fixture.item, id: "item_1" },
-                        },
-                        { ...fixture.event, output_index: 0, item_id },
-                        { type: "response.completed", response: { id: "resp_1" } },
-                      ),
-                    ),
-                  ),
-                  Effect.flip,
-                )
-                expect(error.reason._tag).toBe("InvalidProviderOutput")
-              }),
-            ),
-        )
-      }),
-    )
-
-    it.effect("keeps malformed output item IDs invalid", () =>
-      Effect.gen(function* () {
-        yield* Effect.forEach(["response.output_item.added", "response.output_item.done"], (type) =>
-          Effect.forEach(fixtures, (fixture) =>
-            Effect.forEach(
-              fixture.item.type === "message" ? [undefined, null, 0, false, {}, []] : [null, 0, false, {}, []],
-              (id) =>
-                Effect.gen(function* () {
-                  const error = yield* LLMClient.generate(request).pipe(
-                    Effect.provide(
-                      fixedResponse(
-                        sseEvents(
-                          { type, item: { ...fixture.item, id } },
-                          { type: "response.completed", response: { id: "resp_1" } },
-                        ),
-                      ),
-                    ),
-                    Effect.flip,
-                  )
-                  expect(error.reason._tag).toBe("InvalidProviderOutput")
-                }),
-            ),
-          ),
-        )
-      }),
-    )
-  })
-
-  it.effect("streams function calls without optional item ids through the shared baseline", () =>
-    Effect.gen(function* () {
-      const model = configure({
-        apiKey: "test-key",
-        baseURL: "https://responses.example.test/v1",
-        provider: "example",
-      }).model("example-model")
-      const item = { type: "function_call", call_id: "call_1", name: "lookup", arguments: "" }
-      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Look it up." })).pipe(
-        Effect.provide(
-          fixedResponse(
-            sseEvents(
-              { type: "response.output_item.added", output_index: 1, item },
-              {
-                type: "response.function_call_arguments.delta",
-                output_index: 1,
-                item_id: "opaque_item",
-                delta: '{"query":"shared"}',
-              },
-              {
-                type: "response.output_item.done",
-                output_index: 1,
-                item: { ...item, arguments: '{"query":"complete"}' },
-              },
-              { type: "response.completed", response: { id: "resp_1" } },
-            ),
-          ),
-        ),
-      )
-
-      expect(response.events.filter(LLMEvent.is.toolCall)).toEqual([
-        expect.objectContaining({ id: "call_1", name: "lookup", input: { query: "complete" } }),
-      ])
-      expect(response.events.find(LLMEvent.is.toolCall)?.providerMetadata).toBeUndefined()
-    }),
-  )
-
-  it.effect("finalizes pending function calls from completed response output", () =>
-    Effect.gen(function* () {
-      const model = configure({
-        apiKey: "test-key",
-        baseURL: "https://responses.example.test/v1",
-        provider: "example",
-      }).model("example-model")
-      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Look it up." })).pipe(
-        Effect.provide(
-          fixedResponse(
-            sseEvents(
-              {
-                type: "response.output_item.added",
-                item: { type: "function_call", id: "item_1", call_id: "call_1", name: "lookup", arguments: "" },
-              },
-              { type: "response.function_call_arguments.delta", item_id: "item_1", delta: '{"query":"par' },
-              {
-                type: "response.completed",
-                response: {
-                  output: [
-                    {
-                      type: "function_call",
-                      id: "item_1",
-                      call_id: "call_1",
-                      name: "lookup",
-                      arguments: '{"query":"complete"}',
-                    },
-                  ],
-                },
-              },
-            ),
-          ),
-        ),
-      )
-
-      expect(response.events.find(LLMEvent.is.toolCall)).toMatchObject({
-        input: { query: "complete" },
-        providerMetadata: { example: { itemId: "item_1" } },
-      })
-    }),
-  )
-
-  it.effect("preserves terminal reasoning metadata when item completion is missing", () =>
-    Effect.gen(function* () {
-      const model = configure({
-        apiKey: "test-key",
-        baseURL: "https://responses.example.test/v1",
-      }).model("example-model")
-      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Think it through." })).pipe(
-        Effect.provide(
-          fixedResponse(
-            sseEvents(
-              {
-                type: "response.output_item.added",
-                item: { type: "reasoning", id: "rs_raw", encrypted_content: null },
-              },
-              { type: "response.reasoning_summary_text.delta", item_id: "rs_raw", delta: "Thinking" },
-              {
-                type: "response.completed",
-                response: {
-                  output: [{ type: "reasoning", id: "rs_raw", encrypted_content: "raw-state" }],
-                },
-              },
-            ),
-          ),
-        ),
-      )
-
-      expect(response.events.find((event) => event.type === "reasoning-end")).toMatchObject({
-        providerMetadata: { "openai-compatible": { itemId: "rs_raw", reasoningEncryptedContent: "raw-state" } },
-      })
-    }),
-  )
-
  it.effect("reconciles raw reasoning finals without streamed deltas", () =>
    Effect.gen(function* () {
      const model = configure({
@@ -747,7 +251,7 @@ describe("Open Responses-compatible route", () => {
            Message.assistant({
              type: "text",
              text: "Unclassified.",
-              providerMetadata: { "openai-compatible": { phase: null } },
+              providerMetadata: { openresponses: { phase: null } },
            }),
          ],
        }),
@@ -806,7 +310,7 @@ describe("Open Responses-compatible route", () => {
        {
          type: "text",
          text: "I can't help with that.",
-          providerMetadata: { example: { itemId: "msg_refusal" } },
+          providerMetadata: { openresponses: { itemId: "msg_refusal" } },
        },
      ])

@@ -895,7 +399,7 @@ describe("Open Responses-compatible route", () => {

      expect(response.toolCalls).toEqual([])
      expect(response.events.find(LLMEvent.is.finish)).toMatchObject({
-        providerMetadata: { example: { responseId: "resp_1" } },
+        providerMetadata: { openresponses: { responseId: "resp_1" } },
      })
    }),
  )
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Brendan Allan	c92d311161	fix(session-ui): animate appended inline code chunks	2026-08-25 14:07:29 +08:00
Brendan Allan	23faebc5bb	test(session-ui): stream inline code in partial chunks	2026-08-25 14:01:20 +08:00
Brendan Allan	95cbdd7cc9	test(session-ui): demonstrate streamed inline code fade	2026-08-25 13:59:37 +08:00
Brendan Allan	80a22790d0	fix(session-ui): fade in streamed inline code	2026-08-25 13:55:47 +08:00