fix(app): preserve timeline row heights when moving sessions

fix(ai): preserve provider-defined responses item ids (#45094 )
feat(ai): add native DeepInfra provider (#45108 )
2026-08-26 03:26:12 +00:00 · 2026-08-26 03:22:44 +00:00 · 2026-08-25 22:17:16 -05:00 · 2026-08-25 22:15:52 -05:00 · 2026-08-26 02:43:53 +00:00 · 2026-08-25 22:38:50 -04:00
549 changed files with 31153 additions and 4990 deletions
@@ -1,5 +0,0 @@
---
-"@opencode-ai/plugin": patch
---
-
-Export the Effect runtime used by Effect plugins and safely adapt its tool schemas across host module instances.
@@ -135,7 +135,16 @@ jobs:

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

-            if (linkedIssues === 0) {
+            // 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) {
              await addLabel('needs:issue');
              await comment('issue', `Thanks for your contribution!

@@ -22,6 +22,36 @@ 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:
@@ -41,6 +71,7 @@ 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
@@ -80,15 +111,35 @@ jobs:

      - name: Run unit tests
        timeout-minutes: 20
-        run: GITHUB_ACTIONS=false bun turbo test
+        run: |
+          if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
+            GITHUB_ACTIONS=false bun turbo test
+            exit 0
+          fi
+          GITHUB_ACTIONS=false bun turbo test --affected
        env:
          OPENCODE_EXPERIMENTAL_DISABLE_FILEWATCHER: ${{ runner.os == 'Windows' && 'true' || 'false' }}
+          TURBO_SCM_BASE: ${{ github.event_name == 'pull_request' && format('{0}^1', github.sha) || github.event.before }}
+          TURBO_SCM_HEAD: ${{ github.sha }}

      - name: Verify published codemode package
        if: runner.os == 'Linux'
        working-directory: packages/codemode
        run: bun run script/publish.ts --dry-run

+      - name: Verify packed workerd SDK
+        if: runner.os == 'Linux'
+        timeout-minutes: 15
+        run: |
+          if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
+            bun turbo verify:package --filter=@opencode-ai/sdk
+            exit 0
+          fi
+          bun turbo verify:package --affected --filter=@opencode-ai/sdk
+        env:
+          TURBO_SCM_BASE: ${{ github.event_name == 'pull_request' && format('{0}^1', github.sha) || github.event.before }}
+          TURBO_SCM_HEAD: ${{ github.sha }}
+
      - name: Verify compiled service lifecycle
        if: always()
        timeout-minutes: 10
@@ -127,7 +178,7 @@ jobs:

  e2e:
    name: e2e (${{ matrix.settings.name }})
-    if: github.ref_name != 'v2' && github.head_ref != 'v2'
+    needs: affected
    strategy:
      fail-fast: false
      matrix:
@@ -138,32 +189,38 @@ 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:
@@ -171,23 +228,24 @@ jobs:
          key: ${{ runner.os }}-${{ runner.arch }}-playwright-${{ steps.playwright-version.outputs.version }}-chromium

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

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

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

      - name: Upload Playwright artifacts
-        if: always()
+        if: always() && env.E2E_ENABLED == 'true'
        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
        with:
          name: playwright-${{ matrix.settings.name }}-${{ github.run_attempt }}
@@ -346,9 +346,7 @@
        "@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",
@@ -357,7 +355,6 @@
        "@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",
@@ -430,6 +427,7 @@
      },
      "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:*",
@@ -595,8 +593,8 @@
      },
      "peerDependencies": {
        "@opencode-ai/theme": "workspace:*",
-        "@opentui/core": ">=0.5.7",
-        "@opentui/solid": ">=0.5.7",
+        "@opentui/core": ">=0.5.8",
+        "@opentui/solid": ">=0.5.8",
        "solid-js": ">=1.9.0",
      },
      "optionalPeers": [
@@ -666,6 +664,7 @@
      "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:*",
@@ -685,6 +684,7 @@
      "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:*",
@@ -971,6 +971,7 @@
      "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",
@@ -1090,9 +1091,9 @@
    "@npmcli/arborist": "9.4.0",
    "@octokit/rest": "22.0.0",
    "@openauthjs/openauth": "0.0.0-20250322224806",
-    "@opentui/core": "0.5.7",
-    "@opentui/keymap": "0.5.7",
-    "@opentui/solid": "0.5.7",
+    "@opentui/core": "0.5.8",
+    "@opentui/keymap": "0.5.8",
+    "@opentui/solid": "0.5.8",
    "@pierre/diffs": "1.2.10",
    "@playwright/test": "1.59.1",
    "@sentry/solid": "10.36.0",
@@ -1176,8 +1177,6 @@

    "@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=="],
@@ -1204,8 +1203,6 @@

    "@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=="],
@@ -1610,6 +1607,8 @@

    "@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=="],
@@ -2218,27 +2217,27 @@

    "@opentelemetry/semantic-conventions": ["@opentelemetry/semantic-conventions@1.43.0", "", {}, "sha512-eSYWTm620tTk45EKSedaUL8MFYI8hW164hIXsgIHyxu3VobUB3fFCu5t0hQby6OoWRPsG1KkKUG2M5UadiLiVg=="],

-    "@opentui/core": ["@opentui/core@0.5.7", "", { "dependencies": { "bun-ffi-structs": "0.3.1", "diff": "9.0.0", "marked": "17.0.1", "string-width": "7.2.0", "strip-ansi": "7.1.2" }, "optionalDependencies": { "@opentui/core-darwin-arm64": "0.5.7", "@opentui/core-darwin-x64": "0.5.7", "@opentui/core-linux-arm64": "0.5.7", "@opentui/core-linux-arm64-musl": "0.5.7", "@opentui/core-linux-x64": "0.5.7", "@opentui/core-linux-x64-musl": "0.5.7", "@opentui/core-win32-arm64": "0.5.7", "@opentui/core-win32-x64": "0.5.7" }, "peerDependencies": { "web-tree-sitter": "0.25.10" } }, "sha512-/XDabTkfBs2Wy2FhlC4jvzbpphYAs4SlnCRKYEvi+metlXNTSAshSs43wqZ9O4IPd0E4EcQqdfeQ0sdc3yPpYw=="],
+    "@opentui/core": ["@opentui/core@0.5.8", "", { "dependencies": { "bun-ffi-structs": "0.3.1", "diff": "9.0.0", "marked": "17.0.1", "string-width": "7.2.0", "strip-ansi": "7.1.2" }, "optionalDependencies": { "@opentui/core-darwin-arm64": "0.5.8", "@opentui/core-darwin-x64": "0.5.8", "@opentui/core-linux-arm64": "0.5.8", "@opentui/core-linux-arm64-musl": "0.5.8", "@opentui/core-linux-x64": "0.5.8", "@opentui/core-linux-x64-musl": "0.5.8", "@opentui/core-win32-arm64": "0.5.8", "@opentui/core-win32-x64": "0.5.8" }, "peerDependencies": { "web-tree-sitter": "0.25.10" } }, "sha512-GbZ+nSLYZqxj2Z5TU19Mx6IsAvVsn2+7WEXz+6OlMGoootvt3TxP3vWwfGI0jWk9qp1ftRlML1JPNEkzy+9I8g=="],

-    "@opentui/core-darwin-arm64": ["@opentui/core-darwin-arm64@0.5.7", "", { "os": "darwin", "cpu": "arm64" }, "sha512-75TDJgFD6hDoCElIX35Yg3TIRF/jhrtVQO/9snhjGuTsZ7bd0W88jUlvSXSNhqB3CX431rwgi47B+asWPDI1lQ=="],
+    "@opentui/core-darwin-arm64": ["@opentui/core-darwin-arm64@0.5.8", "", { "os": "darwin", "cpu": "arm64" }, "sha512-c9Y1FBrSnA4sKUCMETsrLYOmsMTyJae8mU9cE6M4o9rXcr3ZLPA7o9AkPA3S0+kH+kZ0o1Fn0wLejgNxvEp5mg=="],

-    "@opentui/core-darwin-x64": ["@opentui/core-darwin-x64@0.5.7", "", { "os": "darwin", "cpu": "x64" }, "sha512-j+Dwu2yV8zahBFjnVIssYdPrHz/+pkd7xWxI+nHjV41V3c+scvH6vJY1U85VCVaoQs/zCH5xAJbmAAn7Sgeh8g=="],
+    "@opentui/core-darwin-x64": ["@opentui/core-darwin-x64@0.5.8", "", { "os": "darwin", "cpu": "x64" }, "sha512-oZ/6Iz1KN+4volMFKmmvziYJhMgyyJ99LfK1S+uPRyIRqzT3CESoJ58D4h04M1g0dHeHE4PvkU1p3uQKrXiP3g=="],

-    "@opentui/core-linux-arm64": ["@opentui/core-linux-arm64@0.5.7", "", { "os": "linux", "cpu": "arm64" }, "sha512-gWO9NWaivXRPc0XhEbxehj3ApyC1SW8Bf1cnHnME1zhw/EZNCNE3TQHgKwOnnNKtvlbhzLQjOI4OqyPYV8UcEQ=="],
+    "@opentui/core-linux-arm64": ["@opentui/core-linux-arm64@0.5.8", "", { "os": "linux", "cpu": "arm64" }, "sha512-N6i/ocrsTjIq9aUQyrfJkqUo+tc4P5ZS6xz38Cm1MhtDzwBE+CrNIJdHfoJfmRLhJkYHlw6Z0ToICn8ZNj4Bpw=="],

-    "@opentui/core-linux-arm64-musl": ["@opentui/core-linux-arm64-musl@0.5.7", "", { "os": "linux", "cpu": "arm64" }, "sha512-RumSHTasIAWU7jPBKiTuEZG/m+prfQKCHhL3JAQnBLgp7eAB20XZGfSLVWDBgxkyKq1rHCUgLtZktC5ZAyKfTA=="],
+    "@opentui/core-linux-arm64-musl": ["@opentui/core-linux-arm64-musl@0.5.8", "", { "os": "linux", "cpu": "arm64" }, "sha512-eFMB41AWODaYf8PsCx3vtTMX33tFgtSpR9tKNnUErCXlcnZ+WCRHVaJ9F6DrVNX3ir6HmgbwAmaWXvaVF+kLtg=="],

-    "@opentui/core-linux-x64": ["@opentui/core-linux-x64@0.5.7", "", { "os": "linux", "cpu": "x64" }, "sha512-fXDmrIlfp9xaoVkk3BNn3yUO/b7plBSOfUh2oWOezRA6E/g5z+hTO1alGFQCy+VV+1kfA6iuYadAww0xNtfACg=="],
+    "@opentui/core-linux-x64": ["@opentui/core-linux-x64@0.5.8", "", { "os": "linux", "cpu": "x64" }, "sha512-/2QM7/wMnML/sxchzbwgoU5tUu/7k836/kSOKMti8opjuecv1K+WWNKGXufhTNeRcXFZaba5rsCdYrf3VqnVsQ=="],

-    "@opentui/core-linux-x64-musl": ["@opentui/core-linux-x64-musl@0.5.7", "", { "os": "linux", "cpu": "x64" }, "sha512-1opDV+W7C1F4iWivCNKuYl5Hen+IeFXAuUPm3hVyN0UtdcP8LESpFJTr8QRwHpGSE9Jz1K2g2S2ipzI3u5mhgw=="],
+    "@opentui/core-linux-x64-musl": ["@opentui/core-linux-x64-musl@0.5.8", "", { "os": "linux", "cpu": "x64" }, "sha512-YXo+qUHYmep2uvv3ECvTeqr10aD7+lBsavYmsTLBzS5hHabbzlQ10oX/99nIRPC3Au1BMWD6d6zQP5czPx23eg=="],

-    "@opentui/core-win32-arm64": ["@opentui/core-win32-arm64@0.5.7", "", { "os": "win32", "cpu": "arm64" }, "sha512-gbiZUyttjq8s+bsnsaopygWN7LTS/RyzD6GOmpgOB2TVitvo0P6mxM+AebBGuu/7xri17Rqaxf3uqrgdKiJAeg=="],
+    "@opentui/core-win32-arm64": ["@opentui/core-win32-arm64@0.5.8", "", { "os": "win32", "cpu": "arm64" }, "sha512-7qBdhEAlh4tLFzW7nWLPlREtNiF6NZMDMi+4uDpUlAMhRavLr6wjcIcgfhNAF/puq06DjuZlPTwaMCgB3qOuwA=="],

-    "@opentui/core-win32-x64": ["@opentui/core-win32-x64@0.5.7", "", { "os": "win32", "cpu": "x64" }, "sha512-bDwon45lUxbV3rMcekTCg4mXcBu9uhdG6HLeLK5TVZf05/h+Ibkf2UeZ8A2jjQt+MK6jpYmxWwDCVCTzZ7EU7Q=="],
+    "@opentui/core-win32-x64": ["@opentui/core-win32-x64@0.5.8", "", { "os": "win32", "cpu": "x64" }, "sha512-Z76YaTKnmRDSHKdKa7iTBCXBQdInQLq7UG3qIE84nvyHCDtkhYtPWOaCzDisSgXC8Y6hJt6NiFbaNjQyxPkfpQ=="],

-    "@opentui/keymap": ["@opentui/keymap@0.5.7", "", { "dependencies": { "@opentui/core": "0.5.7" }, "peerDependencies": { "@opentui/react": "0.5.7", "@opentui/solid": "0.5.7", "react": ">=19.2.0", "solid-js": "1.9.12" }, "optionalPeers": ["@opentui/react", "@opentui/solid", "react", "solid-js"] }, "sha512-BqfSbjlLuctnew2aoPPdK3wpzJfEPp7ttLBkTu1wJTkps9AcfUgXqwloqvY4DUKjUBNFYYEpU/0CPxI4Blj4MA=="],
+    "@opentui/keymap": ["@opentui/keymap@0.5.8", "", { "dependencies": { "@opentui/core": "0.5.8" }, "peerDependencies": { "@opentui/react": "0.5.8", "@opentui/solid": "0.5.8", "react": ">=19.2.0", "solid-js": "1.9.12" }, "optionalPeers": ["@opentui/react", "@opentui/solid", "react", "solid-js"] }, "sha512-KQHKRnLZroSZIbHmGmSeDPsXi7Yykdu9x9ACmDzBFkgPGx0EIuXHqRcmrqmUaX485MQEZuCD9t6FdDsOMGyIFg=="],

-    "@opentui/solid": ["@opentui/solid@0.5.7", "", { "dependencies": { "@babel/core": "7.28.0", "@babel/preset-typescript": "7.27.1", "@opentui/core": "0.5.7", "babel-plugin-module-resolver": "5.0.2", "babel-preset-solid": "1.9.12", "entities": "7.0.1", "s-js": "^0.4.9" }, "peerDependencies": { "solid-js": "1.9.12" } }, "sha512-qrKAZd9xt4D67LXZUARg1Aw0dwEWi29GyOxpM6abqmyExGPMS7jih5Z9oMQJ9EXsouDKpW7Mejjq7WbX2AaecQ=="],
+    "@opentui/solid": ["@opentui/solid@0.5.8", "", { "dependencies": { "@babel/core": "7.28.0", "@babel/preset-typescript": "7.27.1", "@opentui/core": "0.5.8", "babel-plugin-module-resolver": "5.0.2", "babel-preset-solid": "1.9.12", "entities": "7.0.1", "s-js": "^0.4.9" }, "peerDependencies": { "solid-js": "1.9.12" } }, "sha512-L0NxuAU8XT+jlE5G90oA3kspqkof48b0hmzi5XLw+1gxnkxrkTb+YfKys+GzVK4UqhgwY9aW+TeDfrKbnfwCMw=="],

    "@oslojs/asn1": ["@oslojs/asn1@1.0.0", "", { "dependencies": { "@oslojs/binary": "1.0.0" } }, "sha512-zw/wn0sj0j0QKbIXfIlnEcTviaCzYOY3V5rAyjR6YtOByFtJiT574+8p9Wlach0lZH9fddD4yb9laEAIl4vXQA=="],

@@ -5908,10 +5907,6 @@

    "@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=="],
@@ -5950,10 +5945,6 @@

    "@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=="],
@@ -6132,6 +6123,8 @@

    "@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=="],
@@ -6956,6 +6949,10 @@

    "@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", "@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"]
+minimumReleaseAgeExcludes = ["@ai-sdk/amazon-bedrock", "@ai-sdk/anthropic", "@brendonovich/vite-plugin-opencode", "@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-LvDHCOm8OAZfvb0I0L6AbdOevRoQmEJnnqrSgAiNHv8=",
-    "aarch64-linux": "sha256-O0L0iHjb4cwl9xWHIna8VFHyQzoAKzfY8oMpVNayMOg=",
-    "aarch64-darwin": "sha256-ETP8FE71NqufYDUbR7tBdsMEOVQ44wLmsZBeZiSRBRY=",
-    "x86_64-darwin": "sha256-WUcoLldDriT3QxcdlnBQhuPrxDNub0EDvvZXk/pDMpY="
+    "x86_64-linux": "sha256-oQnV96kE3lIqsQaaUrH4tiEX8/5xvBWizXMGxayFSHo=",
+    "aarch64-linux": "sha256-Hkl1xdCQ7voAllwtyrm3TjQT9cfej29fvPIP3lH7zVo=",
+    "aarch64-darwin": "sha256-s0WHRB13qcD0KlgWNXO7gLpsDOnfB8xny81GnN5YeQc=",
+    "x86_64-darwin": "sha256-fnYi1AxCrnO3byW9keDfBH2ueCfGZdaweOrV6AerrGs="
  }
 }
@@ -49,9 +49,9 @@
      "@octokit/rest": "22.0.0",
      "@hono/standard-validator": "0.2.0",
      "@hono/zod-validator": "0.4.2",
-      "@opentui/core": "0.5.7",
-      "@opentui/keymap": "0.5.7",
-      "@opentui/solid": "0.5.7",
+      "@opentui/core": "0.5.8",
+      "@opentui/keymap": "0.5.8",
+      "@opentui/solid": "0.5.8",
      "@tanstack/solid-virtual": "3.13.37",
      "@shikijs/stream": "4.2.0",
      "@standard-schema/spec": "1.1.0",
@@ -157,9 +157,9 @@ const PROVIDERS: ReadonlyArray<Provider> = [
    id: "togetherai",
    label: "TogetherAI",
    tier: "compatible",
-    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)),
+    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)),
  },
  {
    id: "minimax",
@@ -200,8 +200,8 @@ const PROVIDERS: ReadonlyArray<Provider> = [
  {
    id: "cerebras",
    label: "Cerebras",
-    tier: "optional",
-    note: "OpenAI-compatible bridge",
+    tier: "compatible",
+    note: "Native Cerebras text/tool/tool-loop recorded tests",
    vars: [{ name: "CEREBRAS_API_KEY" }],
    validate: (env) => validateBearer("https://api.cerebras.ai/v1/models", Redacted.make(env.CEREBRAS_API_KEY)),
  },
@@ -36,7 +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", "bedrock-converse", "openrouter"])
+const RESPECTS_INLINE_HINTS = new Set(["anthropic-messages", "google-vertex-messages", "bedrock-converse", "openrouter"])

 const makeHint = (ttlSeconds: number | undefined): CacheHint =>
  ttlSeconds !== undefined ? new CacheHint({ type: "ephemeral", ttlSeconds }) : new CacheHint({ type: "ephemeral" })
@@ -1,7 +1,8 @@
 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"]

@@ -26,7 +27,18 @@ 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(request, executor.execute),
+      generate: (request) =>
+        request.model.route.generate(
+          {
+            ...sanitizeSurrogates({
+              ...request,
+              model: undefined,
+              http: mergeHttpOptions(request.model.http, request.http),
+            }),
+            model: request.model,
+          },
+          executor.execute,
+        ),
    })
  }),
 )
@@ -1,5 +1,5 @@
 import { Buffer } from "node:buffer"
-import { Effect, Schema } from "effect"
+import { Effect, Option, Schema } from "effect"
 import { Tool } from "@opencode-ai/schema/tool"
 import { Route } from "../route/client.js"
 import { Auth } from "../route/auth.js"
@@ -157,7 +157,7 @@ type AnthropicDocumentBlock = Schema.Schema.Type<typeof AnthropicDocumentBlock>
 const AnthropicThinkingBlock = Schema.Struct({
  type: Schema.tag("thinking"),
  thinking: Schema.String,
-  signature: Schema.optional(Schema.String),
+  signature: Schema.String,
  cache_control: Schema.optional(AnthropicCacheControl),
 })

@@ -361,6 +361,8 @@ 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),
@@ -371,13 +373,15 @@ 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(AnthropicStreamBlock),
-  delta: Schema.optional(AnthropicStreamDelta),
+  content_block: Schema.optional(Schema.Unknown),
+  delta: Schema.optional(Schema.Unknown),
  usage: Schema.optional(AnthropicUsage),
  // `type` and `message` are both required per Anthropic's spec, but
  // OpenAI-compatible proxies and gateway translations occasionally drop one
@@ -701,6 +705,26 @@ const lowerToolResultContent = Effect.fnUntraced(function* (part: ToolResultPart
  return yield* Effect.forEach(content, lowerToolResultContentItem)
 })

+const requireThinkingSignature = (request: LLMRequest) => {
+  if (request.model.compatibility?.requireSignature !== undefined)
+    return request.model.compatibility.requireSignature
+  const provider = request.model.provider.toLowerCase()
+  const model = request.model.id.toLowerCase()
+  const baseURL = (request.model.route.endpoint.baseURL ?? "").toLowerCase()
+  if (
+    provider === "kimi-for-coding" ||
+    provider === "moonshotai" ||
+    provider === "moonshotai-cn" ||
+    model.startsWith("kimi-") ||
+    baseURL.includes("api.kimi.com/coding") ||
+    baseURL.includes("api.moonshot.ai/anthropic") ||
+    baseURL.includes("api.moonshot.cn/anthropic")
+  )
+    return false
+  if (provider.includes("xiaomi") || model.includes("mimo") || baseURL.includes("xiaomimimo.com")) return false
+  return true
+}
+
 // Mid-conversation system messages became available with Opus 4.8 and version
 // 5 of the other supported Claude families. Treat later family versions as
 // compatible without assuming that every Anthropic Messages model is Claude.
@@ -720,9 +744,12 @@ const endsInServerToolUse = (message: LLMRequest["messages"][number]) => {
  return message.role === "assistant" && last?.type === "tool-call" && last.providerExecuted === true
 }

-const canUseNativeSystemUpdate = (messages: LLMRequest["messages"], index: number) => {
-  const previous = messages[index - 1]
-  const next = messages[index + 1]
+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
  return (
    previous !== undefined &&
    previous.role !== "system" &&
@@ -769,7 +796,7 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
    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.messages, index)) {
+      if (supportsNativeSystemUpdates(request) && canUseNativeSystemUpdate(request, index)) {
        messages.push(yield* lowerNativeSystemUpdate(message, breakpoints))
        continue
      }
@@ -807,15 +834,30 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
          continue
        }
        if (part.type === "reasoning") {
-          // Mirrors Vercel's @ai-sdk/anthropic: a signature marks visible
-          // thinking; only signature-less parts carrying redactedData
-          // round-trip as opaque redacted_thinking blocks.
+          // A signature marks visible thinking; only signature-less parts carrying
+          // redactedData round-trip as opaque redacted_thinking blocks.
          const signature = part.encrypted ?? signatureFromMetadata(part.providerMetadata)
          const redactedData = redactedDataFromMetadata(part.providerMetadata)
          if (signature === undefined && redactedData !== undefined) {
            content.push({ type: "redacted_thinking", data: redactedData })
            continue
          }
+          if (typeof signature !== "string" || signature.trim().length === 0) {
+            if (part.text.trim().length === 0) continue
+            if (!requireThinkingSignature(request)) {
+              content.push({ type: "thinking", thinking: part.text, signature: "" })
+              continue
+            }
+            // Without a signature this cannot be a valid thinking block per
+            // the SDK ThinkingBlockParam:3217 — demote to text so the
+            // conversation remains sendable.
+            content.push({
+              type: "text",
+              text: part.text,
+              cache_control: cacheControl(breakpoints, part.cache),
+            })
+            continue
+          }
          content.push({ type: "thinking", thinking: part.text, signature })
          continue
        }
@@ -1071,7 +1113,7 @@ const SERVER_TOOL_RESULT_NAMES: Record<AnthropicServerToolResultType, string> =

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

-const serverToolResultEvent = (block: NonNullable<AnthropicEvent["content_block"]>): LLMEvent | undefined => {
+const serverToolResultEvent = (block: AnthropicStreamBlock): LLMEvent | undefined => {
  if (!block.type || !isServerToolResultType(block.type)) return undefined
  const errorPayload =
    typeof block.content === "object" && block.content !== null && "type" in block.content
@@ -1098,7 +1140,10 @@ const onMessageStart = (state: ParserState, event: AnthropicEvent): StepResult =
  return [usage ? { ...state, usage: mergeUsage(state.usage, usage) } : state, NO_EVENTS]
 }

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

@@ -1189,11 +1234,12 @@ const onContentBlockStart = (state: ParserState, event: AnthropicEvent): StepRes

 const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(function* (
  state: ParserState,
-  event: AnthropicEvent,
+  event: AnthropicEvent & { readonly delta: AnthropicStreamDelta },
 ) {
  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) },
@@ -1202,6 +1248,7 @@ 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 [
      {
@@ -1214,6 +1261,7 @@ 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,
@@ -1266,7 +1314,10 @@ const onContentBlockStop = Effect.fn("AnthropicMessages.onContentBlockStop")(fun
  return [{ ...state, lifecycle, tools: result.tools, reasoningSignatures }, events] satisfies StepResult
 })

-const onMessageDelta = (state: ParserState, event: AnthropicEvent): StepResult => {
+const onMessageDelta = (
+  state: ParserState,
+  event: AnthropicEvent & { readonly delta?: AnthropicStreamDelta },
+): StepResult => {
  const usage = mergeUsage(state.usage, mapUsage(event.usage))
  return [
    {
@@ -1321,11 +1372,49 @@ 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") {
-    const block = event.content_block
-    if (block && (block.type === "tool_use" || block.type === "server_tool_use")) {
+    if (!ProviderShared.isRecord(event.content_block) || typeof event.content_block.type !== "string")
+      return invalidStreamEvent(event)
+    if (!isKnownStreamBlockType(event.content_block.type)) return Effect.succeed<StepResult>([state, NO_EVENTS])
+    const decoded = decodeAnthropicStreamBlock(event.content_block)
+    if (Option.isNone(decoded)) return invalidStreamEvent(event)
+    const block = decoded.value
+    if (block.type === "tool_use" || block.type === "server_tool_use") {
      if (event.index === undefined)
        return Effect.fail(ProviderShared.eventError(ADAPTER, `Anthropic ${block.type} missing index`))
      if (!block.id)
@@ -1333,11 +1422,22 @@ const step = (state: ParserState, event: AnthropicEvent) => {
          ProviderShared.eventError(ADAPTER, `Anthropic tool_use missing id at index ${event.index}`),
        )
    }
-    return Effect.succeed(onContentBlockStart(state, event))
+    return Effect.succeed(onContentBlockStart(state, { ...event, content_block: block }))
+  }
+  if (event.type === "content_block_delta") {
+    if (!ProviderShared.isRecord(event.delta)) return invalidStreamEvent(event)
+    if (typeof event.delta.type === "string" && !isKnownStreamDeltaType(event.delta.type))
+      return Effect.succeed<StepResult>([state, NO_EVENTS])
+    const decoded = decodeAnthropicStreamDelta(event.delta)
+    if (Option.isNone(decoded)) return invalidStreamEvent(event)
+    return onContentBlockDelta(state, { ...event, delta: decoded.value })
  }
-  if (event.type === "content_block_delta") return onContentBlockDelta(state, event)
  if (event.type === "content_block_stop") return onContentBlockStop(state, event)
-  if (event.type === "message_delta") return Effect.succeed(onMessageDelta(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_stop") return onMessageStop(state)
  if (event.type === "error") return onError(event)
  return Effect.succeed<StepResult>([state, NO_EVENTS])
@@ -212,11 +212,7 @@ const BedrockEvent = Schema.Struct({
      metrics: Schema.optional(Schema.Unknown),
    }),
  ),
-  internalServerException: Schema.optional(BedrockStreamException),
-  modelStreamErrorException: Schema.optional(BedrockStreamException),
-  validationException: Schema.optional(BedrockStreamException),
-  throttlingException: Schema.optional(BedrockStreamException),
-  serviceUnavailableException: Schema.optional(BedrockStreamException),
+  exception: Schema.optional(Schema.Struct({ type: Schema.String, details: BedrockStreamException })),
 })
 type BedrockEvent = Schema.Schema.Type<typeof BedrockEvent>

@@ -650,22 +646,14 @@ const step = (state: ParserState, event: BedrockEvent) =>
      ] as const
    }

-    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) {
+    if (event.exception) {
      return yield* new AIError({
        module: ADAPTER,
        method: "stream",
        reason: classifyProviderFailure({
-          message: exception[1]?.message ?? exception[1]?.originalMessage ?? "Bedrock Converse stream error",
-          code: exception[0],
+          message:
+            event.exception.details.message ?? event.exception.details.originalMessage ?? "Bedrock Converse stream error",
+          code: event.exception.type,
        }),
      })
    }
@@ -716,7 +704,7 @@ export const protocol = Protocol.make({
      reasoningSignatures: {},
    }),
    step,
-    onHalt,
+    onHalt: (state) => Effect.succeed(onHalt(state)),
  },
 })

@@ -82,7 +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({ [eventType]: parsed })
+      out.push(messageType === "exception" ? { exception: { type: eventType, details: parsed } } : { [eventType]: parsed })
    }
    return [cursor, out] as const
  })
@@ -1,4 +1,4 @@
-import { Effect, Schema } from "effect"
+import { Effect, Option, Schema } from "effect"
 import { Tool } from "@opencode-ai/schema/tool"
 import { Route } from "../route/client.js"
 import { Auth } from "../route/auth.js"
@@ -125,6 +125,7 @@ const GeminiContentPart = Schema.Union([
  GeminiFunctionCallPart,
  GeminiFunctionResponsePart,
 ])
+const decodeGeminiContentPart = Schema.decodeUnknownOption(GeminiContentPart)

 const GeminiContent = Schema.Struct({
  role: optionalNull(Schema.Literals(["user", "model"])),
@@ -132,6 +133,11 @@ const GeminiContent = Schema.Struct({
 })
 type GeminiContent = Schema.Schema.Type<typeof GeminiContent>

+const GeminiResponseContent = Schema.Struct({
+  role: optionalNull(Schema.Literals(["user", "model"])),
+  parts: optionalNull(Schema.Array(Schema.Unknown)),
+})
+
 const GeminiSystemInstruction = Schema.Struct({
  parts: Schema.Array(Schema.Struct({ text: Schema.String })),
 })
@@ -200,7 +206,7 @@ const GeminiUsage = Schema.Struct({
 type GeminiUsage = Schema.Schema.Type<typeof GeminiUsage>

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

@@ -222,6 +228,7 @@ const GeminiEvent = Schema.Struct({
 type GeminiEvent = Schema.Schema.Type<typeof GeminiEvent>

 interface ParserState {
+  readonly route: string
  readonly finishReason?: string
  readonly hasToolCalls: boolean
  readonly promptFeedback?: GeminiPromptFeedback
@@ -598,7 +605,21 @@ const step = (state: ParserState, event: GeminiEvent) => {
  // Supplier ids must be tracked across chunks of the same response, not just within one event's parts.
  const seenCallIds = new Set(nextState.seenCallIds)

-  for (const part of candidate.content.parts ?? []) {
+  for (const input of candidate.content.parts ?? []) {
+    if (
+      ProviderShared.isRecord(input) &&
+      !("text" in input) &&
+      !("inlineData" in input) &&
+      !("functionCall" in input) &&
+      !("functionResponse" in input)
+    )
+      continue
+    const decoded = decodeGeminiContentPart(input)
+    if (Option.isNone(decoded))
+      return Effect.fail(
+        ProviderShared.eventError(ADAPTER, `Invalid ${state.route} stream event`, ProviderShared.encodeJson(event)),
+      )
+    const part = decoded.value
    const signature = "thoughtSignature" in part && part.thoughtSignature ? part.thoughtSignature : undefined
    // Gemini attaches replay signatures to thought parts, visible text, or function calls;
    // each block kind must retain the signature attached to its own parts.
@@ -691,9 +712,13 @@ export const protocol = Protocol.make({
  },
  stream: {
    event: Protocol.jsonEvent(GeminiEvent),
-    initial: () => ({ hasToolCalls: false, lifecycle: Lifecycle.initial() }),
+    initial: (request) => ({
+      route: `${request.model.provider}/${request.model.route.id}`,
+      hasToolCalls: false,
+      lifecycle: Lifecycle.initial(),
+    }),
    step,
-    onHalt: finish,
+    onHalt: (state) => Effect.succeed(finish(state)),
  },
 })

@@ -154,7 +154,12 @@ export const driver = (input: DriverInput): WebSocketChannelDriver => {
          ...observation,
          checkpoint: {
            protocol: PROTOCOL,
-            value: { version: VERSION, responseID, request, output: output.slice() } satisfies CheckpointValue,
+            value: {
+              version: VERSION,
+              responseID,
+              request,
+              output: event.response?.output ? [...event.response.output] : output.slice(),
+            } satisfies CheckpointValue,
          },
        }
      }),
@@ -79,10 +79,60 @@ const OpenResponsesReasoningItem = Schema.Struct({
  encrypted_content: optionalNull(Schema.String),
 })

-const OpenResponsesItemReference = Schema.Struct({
-  type: Schema.tag("item_reference"),
-  id: 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>

 // `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.
@@ -111,7 +161,6 @@ export const InputItem = Schema.Union([
    phase: Schema.optionalKey(MessagePhase),
  }),
  OpenResponsesReasoningItem,
-  OpenResponsesItemReference,
  Schema.Struct({
    type: Schema.tag("function_call"),
    id: Schema.optionalKey(Schema.String),
@@ -124,10 +173,17 @@ 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
@@ -140,7 +196,7 @@ 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
 }
@@ -285,8 +341,10 @@ export const Event = Schema.StructWithRest(
  Schema.Struct({
    type: Schema.String,
    delta: Schema.optional(Schema.String),
+    arguments: Schema.optional(Schema.String),
    text: Schema.optional(Schema.String),
    item_id: Schema.optional(Schema.String),
+    output_index: Schema.optional(Schema.Number),
    summary_index: Schema.optional(Schema.Number),
    item: Schema.optional(StreamItem),
    response: Schema.optional(
@@ -295,6 +353,7 @@ export const Event = Schema.StructWithRest(
          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),
        }),
@@ -313,9 +372,6 @@ 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
@@ -324,10 +380,7 @@ export interface Extension {
    readonly media: ProviderShared.NormalizedMedia
    readonly request: LLMRequest
  }) => MediaInput | 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
+  readonly lowerHostedToolItem?: (item: unknown) => ExtendedHostedToolItem | undefined
 }

 const BASE: Extension = { id: ADAPTER, name: NAME }
@@ -339,10 +392,10 @@ 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"
@@ -387,53 +440,37 @@ export const lowerToolChoice = (protocolName: string, toolChoice: NonNullable<LL
    tool: (toolName) => ({ type: "function" as const, name: toolName }),
  })

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

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

-const lowerReasoning = (
-  part: ReasoningPart,
-  providerMetadataKey: string,
-  extension: Extension,
-): OpenResponsesReasoningInput | undefined => {
+const lowerReasoning = (part: ReasoningPart, providerMetadataKey: string): OpenResponsesReasoningInput | undefined => {
  const metadata = part.providerMetadata?.[providerMetadataKey]
+  if (!ProviderShared.isRecord(metadata)) return undefined
  const id = itemID(part.providerMetadata, providerMetadataKey)
-  if (!ProviderShared.isRecord(metadata) || !acceptsItemID(extension, "reasoning", id)) return undefined
  const encryptedContent =
    typeof metadata.reasoningEncryptedContent === "string" || metadata.reasoningEncryptedContent === null
      ? metadata.reasoningEncryptedContent
      : undefined
  return {
    type: "reasoning",
-    id,
+    ...(id === undefined ? {} : { id }),
    summary: part.text.length > 0 ? [{ type: "summary_text", text: part.text }] : [],
    encrypted_content: encryptedContent,
  }
@@ -524,10 +561,7 @@ const lowerToolResultOutput = Effect.fnUntraced(function* (
 })

 const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (request: LLMRequest, extension: Extension) {
-  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 input: LoweredInputItem[] = []
  const providerMetadataKey = request.model.route.providerMetadataKey ?? "openresponses"

  for (const message of request.messages) {
@@ -550,16 +584,14 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
    if (message.role === "assistant") {
      const content: TextPart[] = []
      const reasoningItems: Record<string, OpenResponsesReasoningInput> = {}
-      const reasoningReferences = new Set<string>()
-      const hostedToolReferences = new Set<string>()
+      const hostedToolItems = 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 rawID = itemID(part.providerMetadata, providerMetadataKey)
-          const id = acceptsItemID(extension, "message", rawID) ? rawID : undefined
+          const id = itemID(part.providerMetadata, providerMetadataKey)
          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)
@@ -584,51 +616,51 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
        }
        if (part.type === "reasoning") {
          flushText()
-          const reasoning = lowerReasoning(part, providerMetadataKey, extension)
+          const reasoning = lowerReasoning(part, providerMetadataKey)
          if (!reasoning) continue
-          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]
+          const existing = reasoning.id === undefined ? undefined : reasoningItems[reasoning.id]
          if (existing) {
            existing.summary.push(...reasoning.summary)
            if (typeof reasoning.encrypted_content === "string")
              existing.encrypted_content = reasoning.encrypted_content
            continue
          }
-          reasoningItems[reasoning.id] = reasoning
+          if (reasoning.id !== undefined) reasoningItems[reasoning.id] = reasoning
          input.push(reasoning)
          continue
        }
        if (part.type === "tool-call") {
          flushText()
          if (part.providerExecuted === true) continue
-          input.push(lowerToolCall(part, providerMetadataKey, extension))
+          input.push(lowerToolCall(part, providerMetadataKey))
          continue
        }
        if (part.type === "tool-result" && part.providerExecuted === true) {
          flushText()
          const id = itemID(part.providerMetadata, providerMetadataKey)
-          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"
+          const hosted =
+            part.result.type !== "json"
+              ? undefined
+              : Schema.is(HostedToolItem)(part.result.value)
                ? part.result.value
-                : [{ type: "text", text: ProviderShared.toolResultText(part) }]
-            input.push({
-              role: "user",
-              content: yield* Effect.forEach(content, (item) =>
-                lowerHostedToolResultContentItem(item, request, extension),
-              ),
-            })
+                : extension.lowerHostedToolItem?.(part.result.value)
+          if (id !== undefined && hosted?.id === id) {
+            if (!hostedToolItems.has(id)) {
+              input.push(hosted)
+              hostedToolItems.add(id)
+            }
+            continue
          }
-          if (reference) hostedToolReferences.add(reference)
+          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),
+            ),
+          })
          continue
        }
        return yield* ProviderShared.unsupportedContent(extension.name, "assistant", [
@@ -653,21 +685,16 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
    }
  }

-  // 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
+  return input
 })

 const lowerOptions = (request: LLMRequest) => {
  const options = OpenResponsesOptions.resolve(request)
-  const cacheKey = ProviderShared.clampPromptCacheKey(request.promptCacheKey)
+  const instructions = ProviderShared.joinText(request.system)
+  const cacheKey = ProviderShared.promptCacheKey(request)
+  const parallelToolCalls = resolveParallelToolCalls(request)
  return {
-    ...(options.instructions ? { instructions: options.instructions } : {}),
+    ...(instructions ? { instructions } : {}),
    ...(options.store !== undefined ? { store: options.store } : {}),
    ...(options.metadata ? { metadata: options.metadata } : {}),
    ...(options.safetyIdentifier ? { safety_identifier: options.safetyIdentifier } : {}),
@@ -683,11 +710,18 @@ const lowerOptions = (request: LLMRequest) => {
    ...(options.textVerbosity ? { text: { verbosity: options.textVerbosity } } : {}),
    ...(options.serviceTier ? { service_tier: options.serviceTier } : {}),
    ...(options.maxToolCalls !== undefined ? { max_tool_calls: options.maxToolCalls } : {}),
-    ...(options.parallelToolCalls !== undefined ? { parallel_tool_calls: options.parallelToolCalls } : {}),
+    ...(parallelToolCalls !== undefined ? { parallel_tool_calls: parallelToolCalls } : {}),
    ...(options.truncation ? { truncation: options.truncation } : {}),
  }
 }

+export const resolveParallelToolCalls = (request: LLMRequest) => {
+  const configured = OpenResponsesOptions.resolve(request).parallelToolCalls
+  if (configured !== undefined) return configured
+  const disabled = request.toolChoice?.disableParallelToolUse
+  return disabled === undefined ? undefined : !disabled
+}
+
 const allowedToolChoice = (request: LLMRequest) => {
  const allowed = OpenResponsesOptions.resolve(request).allowedTools
  if (!allowed) return undefined
@@ -808,6 +842,9 @@ 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]
@@ -883,25 +920,23 @@ const onOutputItemAdded = (state: ParserState, event: Event): StepResult => {
      events,
    ]
  }
-  if (item?.type !== "function_call" || !item.id) return [state, NO_EVENTS]
-  const metadata = providerMetadata(state, { itemId: item.id })
+  if (item?.type !== "function_call" || !item.call_id) return [state, NO_EVENTS]
+  const id = item.id ?? item.call_id
+  const metadata = item.id ? providerMetadata(state, { itemId: item.id }) : undefined
  const events: LLMEvent[] = []
  const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
  return [
    {
      ...state,
      lifecycle,
-      tools: ToolStream.start(state.tools, item.id, {
-        id: item.call_id ?? item.id,
+      tools: ToolStream.start(state.tools, id, {
+        id: item.call_id,
        name: item.name ?? "",
        input: item.arguments ?? "",
        providerMetadata: metadata,
      }),
    },
-    [
-      ...events,
-      LLMEvent.toolInputStart({ id: item.call_id ?? item.id, name: item.name ?? "", providerMetadata: metadata }),
-    ],
+    [...events, LLMEvent.toolInputStart({ id: item.call_id, name: item.name ?? "", providerMetadata: metadata })],
  ]
 }

@@ -956,31 +991,21 @@ const onReasoningSummaryPartDone = (state: ParserState, event: Event): StepResul
  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]: state.store !== false ? "concluded" : "can-conclude",
+            [event.summary_index]: "can-conclude",
          },
        },
      },
    },
-    events,
+    NO_EVENTS,
  ]
 }

@@ -988,12 +1013,24 @@ const onFunctionCallArgumentsDelta = Effect.fn("OpenResponses.onFunctionCallArgu
  state: ParserState,
  event: Event,
 ) {
-  if (!event.item_id || !event.delta || !state.tools[event.item_id]) 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
+  if (event.type === "response.function_call_arguments.done" && final === undefined)
+    return [state, NO_EVENTS] satisfies StepResult
+  if (final !== undefined && !final.startsWith(tool.input))
+    return [
+      { ...state, tools: ToolStream.start(state.tools, event.item_id, { ...tool, input: final }) },
+      NO_EVENTS,
+    ] satisfies StepResult
+  const delta = final === undefined ? event.delta : final.slice(tool.input.length)
+  if (!delta) return [state, NO_EVENTS] satisfies StepResult
  const result = ToolStream.appendExisting(
    state.id,
    state.tools,
    event.item_id,
-    event.delta,
+    delta,
    `${state.name} tool argument delta is missing its tool call`,
  )
  if (ToolStream.isError(result)) return yield* result
@@ -1031,18 +1068,19 @@ const onOutputItemDone = Effect.fn("OpenResponses.onOutputItemDone")(function* (
  }

  if (item.type === "function_call") {
-    if (!item.id || !item.call_id || !item.name) return [state, NO_EVENTS] satisfies StepResult
-    const tools = state.tools[item.id]
+    if (!item.call_id || !item.name) return [state, NO_EVENTS] satisfies StepResult
+    const id = item.id ?? item.call_id
+    const tools = state.tools[id]
      ? state.tools
-      : ToolStream.start(state.tools, item.id, {
+      : ToolStream.start(state.tools, id, {
          id: item.call_id,
          name: item.name,
-          providerMetadata: providerMetadata(state, { itemId: item.id }),
+          providerMetadata: item.id ? providerMetadata(state, { itemId: item.id }) : undefined,
        })
    const result =
      item.arguments === undefined
-        ? yield* ToolStream.finish(state.id, tools, item.id)
-        : yield* ToolStream.finishWithInput(state.id, tools, item.id, item.arguments)
+        ? yield* ToolStream.finish(state.id, tools, id)
+        : yield* ToolStream.finishWithInput(state.id, tools, id, item.arguments)
    const events: LLMEvent[] = []
    const resultEvents = result.events ?? []
    const lifecycle = resultEvents.length ? Lifecycle.stepStart(state.lifecycle, events) : state.lifecycle
@@ -1090,30 +1128,50 @@ 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 ||
+              ((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(state.id, state.tools)
-      : { tools: state.tools, events: NO_EVENTS }
-  const events: LLMEvent[] = [...pending.events]
+      ? yield* ToolStream.finishAll(current.id, current.tools)
+      : { tools: current.tools, events: NO_EVENTS }
+  const events: LLMEvent[] = [...reconciled[1], ...pending.events]
  const hasFunctionCall =
    pending.events.some((event) => LLMEvent.is.toolCall(event) || LLMEvent.is.toolInputError(event)) ||
-    state.hasFunctionCall
-  const lifecycle = Lifecycle.finish(state.lifecycle, events, {
+    current.hasFunctionCall
+  const lifecycle = Lifecycle.finish(current.lifecycle, events, {
    reason: {
      normalized: mapFinishReason(event, hasFunctionCall),
      raw: event.response?.incomplete_details?.reason,
    },
-    usage: mapUsage(event.response?.usage, state.providerMetadataKey),
+    usage: mapUsage(event.response?.usage, current.providerMetadataKey),
    providerMetadata:
      event.response?.id || event.response?.service_tier
-        ? providerMetadata(state, {
+        ? providerMetadata(current, {
            responseId: event.response.id,
            serviceTier: event.response.service_tier,
          })
        : undefined,
  })
-  return [{ ...state, lifecycle, hasFunctionCall, tools: pending.tools }, events] satisfies StepResult
+  return [{ ...current, lifecycle, hasFunctionCall, tools: pending.tools }, events] satisfies StepResult
 })

 // Build the prettiest summary available from whatever the provider supplied.
@@ -1160,7 +1218,11 @@ 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, event: Event) => {
+export const step = (state: ParserState, input: Event) => {
+  const event =
+    input.item_id && outputItemID(state, input) !== input.item_id
+      ? { ...input, item_id: outputItemID(state, input) }
+      : input
  if (event.type === "response.output_text.delta" || event.type === "response.output_text.done") {
    if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
    return Effect.succeed(
@@ -1186,7 +1248,6 @@ export const step = (state: ParserState, event: Event) => {
  if (
    event.type === "response.reasoning.done" ||
    event.type === "response.reasoning_summary_text.done" ||
-    event.type === "response.reasoning_summary.done" ||
    event.type === "response.reasoning_text.done"
  ) {
    if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
@@ -1203,9 +1264,17 @@ export const step = (state: ParserState, event: Event) => {
  if (event.type === "response.output_item.added") {
    if (event.item?.type === "message" && !event.item.id)
      return ProviderShared.eventError(state.id, `${event.type} message is missing id`)
-    return Effect.succeed(onOutputItemAdded(state, event))
+    const id = event.item?.id ?? (event.item?.type === "function_call" ? event.item.call_id : undefined)
+    return Effect.succeed(
+      onOutputItemAdded(
+        event.output_index !== undefined && id
+          ? { ...state, outputItems: { ...state.outputItems, [event.output_index]: id } }
+          : state,
+        event,
+      ),
+    )
  }
-  if (event.type === "response.function_call_arguments.delta")
+  if (event.type === "response.function_call_arguments.delta" || event.type === "response.function_call_arguments.done")
    return event.item_id
      ? onFunctionCallArgumentsDelta(state, event)
      : ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
@@ -1238,10 +1307,10 @@ 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({
@@ -7,7 +7,10 @@ import { HttpTransport } from "../route/transport/index.js"
 import { Protocol } from "../route/protocol.js"
 import {
  AIError,
+  InvalidProviderOutputReason,
  LLMEvent,
+  ProviderInternalReason,
+  UnknownProviderReason,
  Usage,
  type FinishReason,
  type FinishReasonDetails,
@@ -51,7 +54,12 @@ const OpenAIChatFunction = Schema.Struct({

 const OpenAIChatTool = Schema.Struct({
  type: Schema.tag("function"),
-  function: OpenAIChatFunction,
+  function: Schema.Struct({
+    name: Schema.String,
+    description: Schema.String,
+    parameters: JsonObject,
+    strict: Schema.optional(Schema.Boolean),
+  }),
  cache_control: Schema.optional(OpenAIChatCacheControl),
 })
 type OpenAIChatTool = Schema.Schema.Type<typeof OpenAIChatTool>
@@ -133,6 +141,7 @@ export const bodyFields = {
  store: Schema.optional(Schema.Boolean),
  prompt_cache_key: Schema.optional(Schema.String),
  reasoning_effort: Schema.optional(OpenAIOptions.OpenAIReasoningEffort),
+  tool_stream: Schema.optional(Schema.Boolean),
  max_completion_tokens: Schema.optional(Schema.Number),
  max_tokens: Schema.optional(Schema.Number),
  temperature: Schema.optional(Schema.Number),
@@ -218,16 +227,22 @@ const OpenAIChatChoice = Schema.StructWithRest(
  [Schema.Record(Schema.String, Schema.Unknown)],
 )

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

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

@@ -250,6 +265,7 @@ export interface ParserState {
  readonly reasoningEmitted: boolean
  readonly latestToolIndex?: number
  readonly nextToolIndex: number
+  readonly requireFinishReason: boolean
 }

 // =============================================================================
@@ -262,14 +278,21 @@ interface LoweringOptions {
  readonly cacheControl?: (
    cache: CacheHint | undefined,
  ) => Schema.Schema.Type<typeof OpenAIChatCacheControl> | undefined
+  readonly toolCallID?: (id: string) => string
 }

-const lowerTool = (tool: ToolDefinition, inputSchema: JsonSchema, options: LoweringOptions): OpenAIChatTool => ({
+const lowerTool = (
+  tool: ToolDefinition,
+  inputSchema: JsonSchema,
+  options: LoweringOptions,
+  supportsStrictMode: boolean,
+): OpenAIChatTool => ({
  type: "function",
  function: {
    name: tool.name,
    description: tool.description,
    parameters: inputSchema,
+    ...(supportsStrictMode ? { strict: false } : {}),
  },
  cache_control: options.cacheControl?.(tool.cache),
 })
@@ -282,8 +305,8 @@ const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>) =>
    tool: (name) => ({ type: "function" as const, function: { name } }),
  })

-const lowerToolCall = (part: ToolCallPart): OpenAIChatAssistantToolCall => ({
-  id: part.id,
+const lowerToolCall = (part: ToolCallPart, options: LoweringOptions): OpenAIChatAssistantToolCall => ({
+  id: options.toolCallID?.(part.id) ?? part.id,
  type: "function",
  function: {
    name: part.name,
@@ -341,8 +364,9 @@ const lowerUserMessage = Effect.fn("OpenAIChat.lowerUserMessage")(function* (

 const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(function* (
  message: OpenAIChatRequestMessage,
-  configuredField?: string,
-  options: LoweringOptions = {},
+  configuredField: string | undefined,
+  requireReasoning: boolean,
+  options: LoweringOptions,
 ) {
  const content: TextPart[] = []
  const reasoning: ReasoningPart[] = []
@@ -359,7 +383,7 @@ const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(func
      continue
    }
    if (part.type === "tool-call") {
-      toolCalls.push(lowerToolCall(part))
+      toolCalls.push(lowerToolCall(part, options))
      continue
    }
  }
@@ -369,15 +393,17 @@ const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(func
  const nativeReasoning = openAICompatibleReasoningContent(message.native?.openaiCompatible)
  const fullyStructured = reasoning.every((part) => Array.isArray(part.providerMetadata?.openai?.reasoningDetails))
  const field = (() => {
-    if (configuredField !== undefined) return configuredField
-    if (reasoning.length === 0) return undefined
+    if (configuredField !== undefined && (requireReasoning || reasoning.length > 0 || nativeReasoning !== undefined))
+      return configuredField
+    if (reasoning.length === 0) return requireReasoning ? "reasoning_content" : undefined
    if (observedField !== undefined) return observedField
    if (nativeReasoning !== undefined) return "reasoning_content"
-    if (!fullyStructured) return "reasoning_content"
+    if (!fullyStructured || requireReasoning) return "reasoning_content"
  })()
  const reasoningText = (() => {
-    if (configuredField !== undefined) return reasoning.length === 0 ? (nativeReasoning ?? "") : text
-    if (reasoning.length === 0) return nativeReasoning
+    if (configuredField !== undefined)
+      return reasoning.length === 0 ? (nativeReasoning ?? (requireReasoning ? "" : undefined)) : text
+    if (reasoning.length === 0) return nativeReasoning ?? (requireReasoning ? "" : undefined)
    return text
  })()
  const cached = message.content.findLast((part) => "cache" in part && part.cache !== undefined)
@@ -405,7 +431,7 @@ const lowerToolMessages = Effect.fn("OpenAIChat.lowerToolMessages")(function* (
    if (part.result.type !== "content") {
      messages.push({
        role: "tool",
-        tool_call_id: part.id,
+        tool_call_id: options.toolCallID?.(part.id) ?? part.id,
        content: ProviderShared.toolResultText(part),
        cache_control: options.cacheControl?.(part.cache),
      })
@@ -415,7 +441,7 @@ 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: part.id,
+      tool_call_id: options.toolCallID?.(part.id) ?? part.id,
      content: text.join("\n"),
      cache_control: options.cacheControl?.(part.cache),
    })
@@ -431,11 +457,13 @@ const lowerToolMessages = Effect.fn("OpenAIChat.lowerToolMessages")(function* (

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

@@ -456,12 +484,37 @@ 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,
+    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) {
@@ -504,14 +557,16 @@ 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, options)
+      const lowered = yield* lowerToolMessages(message, lowering)
      messages.push(...lowered.messages)
      pendingImages.push(...lowered.images)
      continue
    }
    flushImages()
-    messages.push(...(yield* lowerMessage(message, request.model.compatibility?.reasoningField, options)))
+    messages.push(...(yield* lowerMessage(message, reasoningField, requireReasoning, lowering)))
  }
  flushImages()
  return messages
@@ -528,11 +583,122 @@ const hasToolHistory = (messages: ReadonlyArray<LLMRequest["messages"][number]>)
  return false
 }

-const lowerOptions = (request: LLMRequest) => {
+// Derive `max_tokens` vs `max_completion_tokens` from provider/baseURL when
+// explicit `compatibility.maxTokensField` is not set. Aligned with
+// models.dev provider naming: DeepSeek, Moonshot AI, Together AI, ZAI
+// (Zhipu + Coding Plan variants), Nvidia, Cerebras, Chutes, etc. still
+// require `max_tokens`.
+const detectMaxTokensField = (provider: string, baseURL: string | undefined): "max_tokens" | "max_completion_tokens" => {
+  const p = provider.toLowerCase()
+  const url = (baseURL ?? "").toLowerCase()
+  if (
+    p === "deepseek" ||
+    url.includes("deepseek.com") ||
+    p === "moonshotai" ||
+    url.includes("api.moonshot.ai") ||
+    p === "togetherai" ||
+    url.includes("api.together.") ||
+    p === "zai" ||
+    p === "zai-coding-plan" ||
+    p === "zhipuai" ||
+    p === "zhipuai-coding-plan" ||
+    url.includes("api.z.ai") ||
+    url.includes("open.bigmodel.cn") ||
+    p === "nvidia" ||
+    url.includes("integrate.api.nvidia.com") ||
+    p === "cerebras" ||
+    url.includes("cerebras.ai") ||
+    url.includes("llm.chutes.ai") ||
+    p === "chutes" ||
+    p === "cloudflare-ai-gateway" ||
+    url.includes("gateway.ai.cloudflare.com") ||
+    p === "cloudflare-workers-ai" ||
+    url.includes("api.cloudflare.com")
+  )
+    return "max_tokens"
+  return "max_completion_tokens"
+}
+
+const detectSupportsStore = (provider: string, baseURL: string | undefined): boolean => {
+  const p = provider.toLowerCase()
+  const url = (baseURL ?? "").toLowerCase()
+  const isNvidia = p === "nvidia" || url.includes("integrate.api.nvidia.com")
+  const isMoonshot = p === "moonshotai" || p === "moonshotai-cn" || url.includes("api.moonshot.")
+  const isTogether = p === "togetherai" || p === "together" || url.includes("api.together.")
+  const isZai =
+    p === "zai" ||
+    p === "zai-coding-plan" ||
+    p === "zhipuai" ||
+    p === "zhipuai-coding-plan" ||
+    url.includes("api.z.ai") ||
+    url.includes("open.bigmodel.cn")
+  const isDeepSeek = p === "deepseek" || url.includes("deepseek.com")
+  const isCerebras = p === "cerebras" || url.includes("cerebras.ai")
+  const isXai = p === "xai" || url.includes("api.x.ai")
+  const isChutes = p === "chutes" || url.includes("chutes.ai")
+  const isCloudflareWorkersAI = p === "cloudflare-workers-ai" || url.includes("api.cloudflare.com")
+  const isCloudflareAiGateway = p === "cloudflare-ai-gateway" || url.includes("gateway.ai.cloudflare.com")
+  const isVercelAiGateway = p === "vercel-ai-gateway" || url.includes("ai-gateway.vercel.sh") || url.includes("vercel.sh")
+  const isAntLing = p === "ant-ling" || url.includes("api.ant-ling.com")
+  const isOpencode = p === "opencode" || url.includes("opencode.ai")
+  const isNonStandard =
+    isNvidia ||
+    isCerebras ||
+    isXai ||
+    isTogether ||
+    isChutes ||
+    isDeepSeek ||
+    isZai ||
+    isMoonshot ||
+    isOpencode ||
+    isCloudflareWorkersAI ||
+    isCloudflareAiGateway ||
+    isVercelAiGateway ||
+    isAntLing
+  return !isNonStandard
+}
+
+const detectSupportsUsageInStreaming = (): boolean => true
+
+const detectSupportsStrictMode = (provider: string, baseURL: string | undefined): boolean => {
+  const p = provider.toLowerCase()
+  const url = (baseURL ?? "").toLowerCase()
+  const isMoonshot = p === "moonshotai" || p === "moonshotai-cn" || url.includes("api.moonshot.")
+  const isTogether = p === "togetherai" || p === "together" || url.includes("api.together.")
+  const isCloudflareAiGateway = p === "cloudflare-ai-gateway" || url.includes("gateway.ai.cloudflare.com")
+  const isNvidia = p === "nvidia" || url.includes("integrate.api.nvidia.com")
+  return !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia
+}
+
+const detectZaiToolStream = (
+  provider: string,
+  baseURL: string | undefined,
+  modelID: string,
+): boolean => {
+  const p = provider.toLowerCase()
+  const url = (baseURL ?? "").toLowerCase()
+  const isZai =
+    p === "zai" ||
+    p === "zai-coding-plan" ||
+    p === "zhipuai" ||
+    p === "zhipuai-coding-plan" ||
+    url.includes("api.z.ai") ||
+    url.includes("open.bigmodel.cn")
+  if (!isZai) return false
+  const id = modelID.toLowerCase()
+  if (id === "glm-4.5" || id === "glm-4.5-air" || id === "glm-4.5-flash" || id === "glm-4.5v") return false
+  return true
+}
+
+const lowerOptions = (request: LLMRequest, supportsStore: boolean) => {
  const options = OpenAIOptions.resolve(request)
-  const cacheKey = ProviderShared.clampPromptCacheKey(request.promptCacheKey)
+  const cacheKey = ProviderShared.promptCacheKey(request)
  return {
-    ...(options.store !== undefined ? { store: options.store } : {}),
+    ...(supportsStore && options.store !== undefined ? { store: options.store } : {}),
+    // For providers that support `store`, ensure stateless `store:false` is sent
+    // even when no explicit `providerOptions.store` was supplied, mirroring the
+    // native OpenAI Chat default. Non-standard providers omit `store` entirely.
+    ...(supportsStore && options.store === undefined ? { store: false } : {}),
    ...(cacheKey ? { prompt_cache_key: cacheKey } : {}),
    ...(options.reasoningEffort ? { reasoning_effort: options.reasoningEffort } : {}),
  }
@@ -551,8 +717,19 @@ export const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (
    )
  const generation = request.generation
  const toolSchemaCompatibility = request.model.compatibility?.toolSchema
-  const maxTokensField = request.model.compatibility?.maxTokensField ?? "max_tokens"
+  const provider = String(request.model.provider)
+  const baseURL = request.model.route.endpoint.baseURL
+  const detectedMaxTokensField = detectMaxTokensField(provider, baseURL)
+  const maxTokensField = request.model.compatibility?.maxTokensField ?? detectedMaxTokensField
+  const supportsStore = request.model.compatibility?.supportsStore ?? detectSupportsStore(provider, baseURL)
+  const supportsUsageInStreaming =
+    request.model.compatibility?.supportsUsageInStreaming ?? detectSupportsUsageInStreaming()
+  const supportsStrictMode = request.model.compatibility?.supportsStrictMode ?? detectSupportsStrictMode(provider, baseURL)
+  const zaiToolStream =
+    request.model.compatibility?.zaiToolStream ??
+    detectZaiToolStream(provider, baseURL, request.model.id)
  const hasHistory = hasToolHistory(request.messages)
+  const hasActiveTools = request.tools.length > 0
  return {
    model: request.model.id,
    messages: yield* lowerMessages(request, options),
@@ -566,11 +743,13 @@ export const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (
              tool,
              ToolSchemaProjection.modelCompatibility(tool.inputSchema, toolSchemaCompatibility),
              options,
+              supportsStrictMode,
            ),
          ),
    tool_choice: request.toolChoice ? yield* lowerToolChoice(request.toolChoice) : undefined,
    stream: true as const,
-    stream_options: { include_usage: true },
+    ...(supportsUsageInStreaming ? { stream_options: { include_usage: true } } : {}),
+    ...(zaiToolStream && hasActiveTools ? { tool_stream: true } : {}),
    ...(maxTokensField === "max_completion_tokens"
      ? { max_completion_tokens: generation?.maxTokens }
      : { max_tokens: generation?.maxTokens }),
@@ -580,7 +759,7 @@ export const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (
    presence_penalty: generation?.presencePenalty,
    seed: generation?.seed,
    stop: generation?.stop,
-    ...lowerOptions(request),
+    ...lowerOptions(request, supportsStore),
  }
 })

@@ -590,14 +769,40 @@ export const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (
 // Streaming parsers are small state machines: every event returns a new state
 // plus the common `LLMEvent`s produced by that event. Tool calls are accumulated
 // because OpenAI streams JSON arguments across multiple deltas.
-const mapFinishReason = (reason: string | null | undefined): FinishReason => {
-  if (reason === "stop") return "stop"
-  if (reason === "length") return "length"
-  if (reason === "content_filter") return "content-filter"
-  if (reason === "function_call" || reason === "tool_calls") return "tool-calls"
-  if (reason === "error") return "error"
-  return "unknown"
-}
+const finishReasonError = (event: OpenAIChatEvent, reason: AIError["reason"]) =>
+  new AIError({
+    module: ADAPTER,
+    method: "stream",
+    body: ProviderShared.encodeJson(event),
+    reason,
+  })
+
+const mapFinishReason = Effect.fn("OpenAIChat.mapFinishReason")(function* (event: OpenAIChatEvent, reason: string) {
+  switch (reason) {
+    case "error":
+      return yield* finishReasonError(
+        event,
+        new UnknownProviderReason({ message: "Provider reported an error (finish_reason: error)" }),
+      )
+    case "network_error":
+      return yield* finishReasonError(
+        event,
+        new ProviderInternalReason({ message: "Provider reported a network error (finish_reason: network_error)" }),
+      )
+    case "stop":
+    case "end":
+      return "stop" as const
+    case "length":
+      return "length" as const
+    case "content_filter":
+      return "content-filter" as const
+    case "function_call":
+    case "tool_calls":
+      return "tool-calls" as const
+    default:
+      return "unknown" as const
+  }
+})

 // OpenAI Chat reports `prompt_tokens` (inclusive total) with a
 // cached-read and cache-write subsets, and `completion_tokens` (inclusive
@@ -710,16 +915,20 @@ const reasoningMetadata = (field: ParserState["reasoningField"], details?: Reado

 const step = (state: ParserState, event: OpenAIChatEvent) =>
  Effect.gen(function* () {
-    if (event.error)
+    if (event.error) {
+      const body = ProviderShared.encodeJson(event)
      return yield* new AIError({
        module: ADAPTER,
        method: "stream",
+        body,
        reason: classifyProviderFailure({
          message: event.error.message,
          code: event.error.code === undefined || event.error.code === null ? undefined : String(event.error.code),
          status: typeof event.error.code === "number" ? event.error.code : undefined,
+          rawBody: body,
        }),
      })
+    }
    const events: LLMEvent[] = []
    const choice = event.choices?.[0]
    // Moonshot (and a few other OpenAI-compatible providers) attach usage to
@@ -728,8 +937,11 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
    const usage = mapUsage(event.usage) ?? (choiceUsage ? mapUsage(choiceUsage) : undefined) ?? state.usage
    const rawFinishReason = choice?.finish_reason
    const finishReason =
-      rawFinishReason !== undefined && rawFinishReason !== null
-        ? { normalized: mapFinishReason(rawFinishReason), raw: choice?.native_finish_reason ?? rawFinishReason }
+      rawFinishReason
+        ? {
+            normalized: yield* mapFinishReason(event, rawFinishReason),
+            raw: choice?.native_finish_reason ?? rawFinishReason,
+          }
        : state.finishReason
    const delta = choice?.delta
    const toolDeltas = delta?.tool_calls ?? []
@@ -749,7 +961,11 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
      toolDeltas.some((tool) => Boolean(tool.id) || Boolean(tool.function?.name) || Boolean(tool.function?.arguments))
    if (state.finishReason !== undefined) {
      if (hasLateContent)
-        return yield* ProviderShared.eventError(ADAPTER, "OpenAI Chat received content after the finish reason")
+        return yield* ProviderShared.eventError(
+          ADAPTER,
+          "OpenAI Chat received content after the finish reason",
+          ProviderShared.encodeJson(event),
+        )
      return [{ ...state, usage }, events] as const
    }

@@ -821,14 +1037,19 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
        { id: id || undefined, name: name || undefined, text },
        "OpenAI Chat tool call delta is missing id or name",
      )
-      if (ToolStream.isError(result)) return yield* result
+      if (ToolStream.isError(result))
+        return yield* ProviderShared.eventError(ADAPTER, result.reason.message, ProviderShared.encodeJson(event))
      tools = result.tools
      if (result.events.length) lifecycle = Lifecycle.stepStart(lifecycle, events)
      events.push(...result.events)
    }

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

    // Finalize accumulated tool inputs eagerly when finish_reason arrives so
    // valid calls and malformed local calls settle independently.
@@ -851,16 +1072,27 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
        reasoningEmitted,
        latestToolIndex,
        nextToolIndex,
+        requireFinishReason: state.requireFinishReason,
      },
      events,
    ] as const
  })

-const finishEvents = (state: ParserState): ReadonlyArray<LLMEvent> => {
+const finishEvents = Effect.fn("OpenAIChat.finishEvents")(function* (state: ParserState) {
+  if (state.finishReason === undefined && state.requireFinishReason)
+    return yield* new AIError({
+      module: ADAPTER,
+      method: "stream",
+      reason: new InvalidProviderOutputReason({
+        classification: "incomplete-stream",
+        message: "OpenAI Chat stream ended without finish_reason",
+        route: ADAPTER,
+      }),
+    })
  const events: LLMEvent[] = []
  const toolCallEvents =
    state.finishReason === undefined && Object.keys(state.tools).length > 0
-      ? Effect.runSync(ToolStream.finishAll(ADAPTER, state.tools)).events
+      ? (yield* ToolStream.finishAll(ADAPTER, state.tools)).events
      : state.toolCallEvents
  const hasToolCalls = toolCallEvents.length > 0
  const reason = state.finishReason
@@ -869,7 +1101,7 @@ const finishEvents = (state: ParserState): ReadonlyArray<LLMEvent> => {
        normalized:
          state.finishReason.normalized === "stop" && hasToolCalls ? "tool-calls" : state.finishReason.normalized,
      }
-    : { normalized: hasToolCalls ? ("tool-calls" as const) : ("unknown" as const) }
+    : { normalized: hasToolCalls ? ("tool-calls" as const) : ("stop" as const) }
  const metadata = reasoningMetadata(
    state.reasoningField,
    state.reasoningDetailsObserved ? state.reasoningDetails : undefined,
@@ -883,7 +1115,7 @@ const finishEvents = (state: ParserState): ReadonlyArray<LLMEvent> => {
  events.push(...toolCallEvents)
  Lifecycle.finish(lifecycle, events, { reason, usage: state.usage })
  return events
-}
+})

 // =============================================================================
 // Protocol And OpenAI Route
@@ -912,6 +1144,7 @@ export const protocol = Protocol.make({
      reasoningDetailsObserved: false,
      reasoningEmitted: false,
      nextToolIndex: 0,
+      requireFinishReason: request.model.compatibility?.requireFinishReason ?? true,
    }),
    step,
    onHalt: finishEvents,
@@ -17,6 +17,7 @@ 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 { optionalArray, ProviderShared } from "./shared.js"
+import { JsonObject, optionalArray, optionalNull, 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"
@@ -32,6 +32,40 @@ 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([
@@ -41,6 +75,7 @@ const OpenAIResponsesToolChoice = Schema.Union([

 const OpenAIResponsesCoreFields = {
  ...OpenResponses.coreFields,
+  input: Schema.Array(Schema.Union([OpenResponses.InputItem, OpenAIResponsesHostedToolItem])),
  tools: optionalArray(OpenAIResponsesTools),
  tool_choice: Schema.optional(OpenAIResponsesToolChoice),
 }
@@ -51,28 +86,10 @@ 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,
-  acceptsItemID: (kind: OpenResponses.ItemKind, id: string) => {
-    const prefixes = ITEM_ID_PREFIXES[kind]
-    return prefixes.length === 0 || prefixes.some((prefix) => id.startsWith(prefix))
-  },
+  lowerHostedToolItem: (item: unknown) => (Schema.is(OpenAIResponsesHostedToolItem)(item) ? item : undefined),
 } satisfies OpenResponses.Extension

 const nativeImageToolInput = (tool: ToolDefinition) => {
@@ -105,14 +122,18 @@ const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>, tool
        : { type: "function" as const, name },
  })

+const decodeBody = ProviderShared.validateWith(Schema.decodeUnknownEffect(OpenAIResponsesBody))
+
 const fromRequest = Effect.fn("OpenAIResponses.fromRequest")(function* (request: LLMRequest) {
  const body = yield* OpenResponses.fromRequestWithExtension(
    LLMRequest.update(request, { tools: [], toolChoice: undefined }),
    extension,
  )
  const toolSchemaCompatibility = request.model.compatibility?.toolSchema
-  return {
+  const parallelToolCalls = OpenResponses.resolveParallelToolCalls(request)
+  return yield* decodeBody({
    ...body,
+    ...(parallelToolCalls === undefined ? {} : { parallel_tool_calls: parallelToolCalls }),
    tools:
      request.tools.length === 0
        ? undefined
@@ -121,7 +142,7 @@ 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) {
@@ -162,9 +183,11 @@ const HOSTED_TOOLS = {
 } as const satisfies ResponsesHostedTools.Definitions

 const step = (state: OpenResponses.ParserState, event: OpenResponses.Event) => {
-  if (event.type === "response.reasoning_text.delta" || event.type === "response.reasoning_summary.delta")
+  if (event.type === "response.reasoning_text.delta")
    return event.item_id
-      ? Effect.succeed(OpenResponses.onReasoningDelta(state, event, event.item_id))
+      ? Effect.succeed(
+          OpenResponses.onReasoningDelta(state, event, OpenResponses.outputItemID(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)
@@ -205,7 +228,7 @@ export const route = Route.make({
  endpoint,
  auth,
  transport,
-  defaults: { providerOptions: { store: false } },
+  defaults: { providerOptions: { store: false, include: ["reasoning.encrypted_content"] } },
 })

 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 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
+export const promptCacheKey = (request: LLMRequest): string | undefined => {
+  if (request.cache === "none" || request.promptCacheKey === undefined) return undefined
+  const chars = Array.from(request.promptCacheKey)
+  if (chars.length <= OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH) return request.promptCacheKey
  return chars.slice(0, OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH).join("")
 }

@@ -210,10 +210,9 @@ 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. 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`.
+ * element. Retry control events are ignored without interrupting the stream.
+ * Decoder failures become provider output errors so the public error channel
+ * stays `AIError`.
 */
 export const sseFraming = (
  bytes: Stream.Stream<Uint8Array, AIError>,
@@ -221,9 +220,23 @@ export const sseFraming = (
 ): Stream.Stream<string, AIError> =>
  bytes.pipe(
    Stream.decodeText(),
-    Stream.pipeThroughChannel(Sse.decode()),
-    Stream.catchTag("Retry", () => Stream.empty),
-    Stream.catchTag("SseError", (error) => Stream.fail(eventError("sse", error.message))),
+    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.filter(
      (event) =>
        (events === undefined || events.has(event.event)) &&
@@ -56,7 +56,6 @@ 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),
@@ -0,0 +1,64 @@
+/*
+ * Adapted from partial-json by the Promplate Dev Team:
+ * https://github.com/promplate/partial-json-parser-js/blob/main/src/options.ts
+ * Licensed under the MIT License; see partial-json.ts for the complete notice.
+ */
+
+/**
+ * allow partial strings like `"hello \u12` to be parsed as `"hello `
+ */
+export const STR = 0b000000001
+
+/**
+ * allow partial numbers like `123.` to be parsed as `123`
+ */
+export const NUM = 0b000000010
+
+/**
+ * allow partial arrays like `[1, 2,` to be parsed as `[1, 2]`
+ */
+export const ARR = 0b000000100
+
+/**
+ * allow partial objects like `{"a": 1, "b":` to be parsed as `{"a": 1}`
+ */
+export const OBJ = 0b000001000
+
+/**
+ * allow `nu` to be parsed as `null`
+ */
+export const NULL = 0b000010000
+
+/**
+ * allow `tr` to be parsed as `true`, and `fa` to be parsed as `false`
+ */
+export const BOOL = 0b000100000
+
+/**
+ * allow `Na` to be parsed as `NaN`
+ */
+export const NAN = 0b001000000
+
+/**
+ * allow `Inf` to be parsed as `Infinity`
+ */
+export const INFINITY = 0b010000000
+
+/**
+ * allow `-Inf` to be parsed as `-Infinity`
+ */
+export const _INFINITY = 0b100000000
+
+export const INF = INFINITY | _INFINITY
+export const SPECIAL = NULL | BOOL | INF | NAN
+export const ATOM = STR | NUM | SPECIAL
+export const COLLECTION = ARR | OBJ
+export const ALL = ATOM | COLLECTION
+
+/**
+ * Control what types you allow to be partially parsed.
+ * The default is to allow all types to be partially parsed, which in most cases is the best option.
+ */
+export const Allow = { STR, NUM, ARR, OBJ, NULL, BOOL, NAN, INFINITY, _INFINITY, INF, SPECIAL, ATOM, COLLECTION, ALL }
+
+export default Allow
@@ -0,0 +1,282 @@
+/*
+ * Adapted from partial-json by the Promplate Dev Team:
+ * https://github.com/promplate/partial-json-parser-js
+ *
+ * MIT License
+ *
+ * Copyright (c) 2023 Promplate Dev Team
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+import { Schema } from "effect"
+import { Allow } from "./partial-json-options.js"
+export * from "./partial-json-options.js"
+
+export class PartialJSON extends Error {}
+export class MalformedJSON extends Error {}
+
+const decodeJson = Schema.decodeUnknownSync(Schema.fromJsonString(Schema.Unknown))
+
+/** Parse complete or incomplete JSON, restricted by the supplied partial-value flags. */
+export function parseJSON(jsonString: string, allowPartial = Allow.ALL): unknown {
+  if (typeof jsonString !== "string") throw new TypeError(`expecting str, got ${typeof jsonString}`)
+  const input = jsonString.trim()
+  if (!input) throw new Error(`${jsonString} is empty`)
+  try {
+    return decodeJson(input)
+  } catch {}
+
+  const repaired = repairJSON(input)
+  if (repaired !== input) {
+    try {
+      return decodeJson(repaired)
+    } catch {}
+  }
+
+  try {
+    return _parseJSON(input, allowPartial)
+  } catch (error) {
+    if (repaired !== input) return _parseJSON(repaired, allowPartial)
+    throw error
+  }
+}
+
+const repairJSON = (input: string) => {
+  let repaired = ""
+  let quoted = false
+
+  for (let index = 0; index < input.length; index++) {
+    const character = input[index]
+    if (!quoted) {
+      repaired += character
+      if (character === '"') quoted = true
+      continue
+    }
+
+    if (character === '"') {
+      repaired += character
+      quoted = false
+      continue
+    }
+
+    if (character === "\\") {
+      const next = input[index + 1]
+      if (next === "u" && /^[0-9a-fA-F]{4}$/.test(input.slice(index + 2, index + 6))) {
+        repaired += input.slice(index, index + 6)
+        index += 5
+        continue
+      }
+      if (next !== undefined && '"\\/bfnrtu'.includes(next)) {
+        repaired += `\\${next}`
+        index++
+        continue
+      }
+      repaired += "\\\\"
+      continue
+    }
+
+    const code = character.charCodeAt(0)
+    repaired += code <= 0x1f ? `\\u${code.toString(16).padStart(4, "0")}` : character
+  }
+
+  return repaired
+}
+
+const _parseJSON = (jsonString: string, allow: number) => {
+  const length = jsonString.length
+  let index = 0
+
+  const markPartialJSON = (message: string): never => {
+    throw new PartialJSON(`${message} at position ${index}`)
+  }
+
+  const throwMalformedError = (message: string): never => {
+    throw new MalformedJSON(`${message} at position ${index}`)
+  }
+
+  const parseAny = (): unknown => {
+    skipBlank()
+    if (index >= length) markPartialJSON("Unexpected end of input")
+    if (jsonString[index] === '"') return parseStr()
+    if (jsonString[index] === "{") return parseObj()
+    if (jsonString[index] === "[") return parseArr()
+    if (
+      jsonString.substring(index, index + 4) === "null" ||
+      (Allow.NULL & allow && length - index < 4 && "null".startsWith(jsonString.substring(index)))
+    ) {
+      index += 4
+      return null
+    }
+    if (
+      jsonString.substring(index, index + 4) === "true" ||
+      (Allow.BOOL & allow && length - index < 4 && "true".startsWith(jsonString.substring(index)))
+    ) {
+      index += 4
+      return true
+    }
+    if (
+      jsonString.substring(index, index + 5) === "false" ||
+      (Allow.BOOL & allow && length - index < 5 && "false".startsWith(jsonString.substring(index)))
+    ) {
+      index += 5
+      return false
+    }
+    if (
+      jsonString.substring(index, index + 8) === "Infinity" ||
+      (Allow.INFINITY & allow && length - index < 8 && "Infinity".startsWith(jsonString.substring(index)))
+    ) {
+      index += 8
+      return Infinity
+    }
+    if (
+      jsonString.substring(index, index + 9) === "-Infinity" ||
+      (Allow._INFINITY & allow &&
+        1 < length - index &&
+        length - index < 9 &&
+        "-Infinity".startsWith(jsonString.substring(index)))
+    ) {
+      index += 9
+      return -Infinity
+    }
+    if (
+      jsonString.substring(index, index + 3) === "NaN" ||
+      (Allow.NAN & allow && length - index < 3 && "NaN".startsWith(jsonString.substring(index)))
+    ) {
+      index += 3
+      return NaN
+    }
+    return parseNum()
+  }
+
+  const parseStr = (): string => {
+    const start = index
+    let escape = false
+    index++
+    while (index < length && (jsonString[index] !== '"' || (escape && jsonString[index - 1] === "\\"))) {
+      escape = jsonString[index] === "\\" ? !escape : false
+      index++
+    }
+    if (jsonString.charAt(index) === '"') {
+      try {
+        return decodeJson(jsonString.substring(start, ++index - Number(escape))) as string
+      } catch (error) {
+        throwMalformedError(String(error))
+      }
+    }
+    if (Allow.STR & allow) {
+      try {
+        return decodeJson(`${jsonString.substring(start, index - Number(escape))}"`) as string
+      } catch {
+        return decodeJson(`${jsonString.substring(start, jsonString.lastIndexOf("\\"))}"`) as string
+      }
+    }
+    return markPartialJSON("Unterminated string literal")
+  }
+
+  const parseObj = (): Record<string, unknown> => {
+    index++
+    skipBlank()
+    const object: Record<string, unknown> = {}
+    try {
+      while (jsonString[index] !== "}") {
+        skipBlank()
+        if (index >= length && Allow.OBJ & allow) return object
+        const key = parseStr()
+        skipBlank()
+        index++
+        try {
+          Object.defineProperty(object, key, {
+            value: parseAny(),
+            enumerable: true,
+            configurable: true,
+            writable: true,
+          })
+        } catch (error) {
+          if (Allow.OBJ & allow) return object
+          throw error
+        }
+        skipBlank()
+        if (jsonString[index] === ",") index++
+      }
+    } catch {
+      if (Allow.OBJ & allow) return object
+      return markPartialJSON("Expected '}' at end of object")
+    }
+    index++
+    return object
+  }
+
+  const parseArr = (): unknown[] => {
+    index++
+    const array: unknown[] = []
+    try {
+      while (jsonString[index] !== "]") {
+        array.push(parseAny())
+        skipBlank()
+        if (jsonString[index] === ",") index++
+      }
+    } catch {
+      if (Allow.ARR & allow) return array
+      return markPartialJSON("Expected ']' at end of array")
+    }
+    index++
+    return array
+  }
+
+  const parseNum = (): unknown => {
+    if (index === 0) {
+      if (jsonString === "-") throwMalformedError("Not sure what '-' is")
+      try {
+        return decodeJson(jsonString)
+      } catch (error) {
+        if (Allow.NUM & allow) {
+          try {
+            return decodeJson(jsonString.substring(0, jsonString.lastIndexOf("e")))
+          } catch {}
+        }
+        throwMalformedError(String(error))
+      }
+    }
+
+    const start = index
+    if (jsonString[index] === "-") index++
+    while (jsonString[index] && !",]}".includes(jsonString[index])) index++
+    if (index === length && !(Allow.NUM & allow)) markPartialJSON("Unterminated number literal")
+
+    try {
+      return decodeJson(jsonString.substring(start, index))
+    } catch (error) {
+      if (jsonString.substring(start, index) === "-") markPartialJSON("Not sure what '-' is")
+      try {
+        return decodeJson(jsonString.substring(start, jsonString.lastIndexOf("e")))
+      } catch {
+        throwMalformedError(String(error))
+      }
+    }
+  }
+
+  const skipBlank = () => {
+    while (index < length && " \n\r\t".includes(jsonString[index])) index++
+  }
+
+  return parseAny()
+}
+
+export const parse = parseJSON
@@ -1,8 +1,10 @@
-import { Effect } from "effect"
-import { AIError, LLMEvent, type ProviderMetadata, type ToolCall, type ToolInputError } from "../../schema/index.js"
+import { Effect, Option } from "effect"
+import { AIError, LLMEvent, type ProviderMetadata, type ToolCall } from "../../schema/index.js"
 import { eventError, parseToolInput, type ToolAccumulator } from "../shared.js"
+import { parse } from "./partial-json.js"

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

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

 const toolCall = (route: string, tool: PendingTool, inputOverride?: string) => {
  const raw = inputOverride ?? tool.input
  return parseToolInput(route, tool.name, raw).pipe(
-    Effect.map((input): ToolCall | ToolInputError =>
-      LLMEvent.toolCall({
-        id: tool.id,
-        name: tool.name,
-        input,
-        providerExecuted: tool.providerExecuted ? true : undefined,
-        providerMetadata: tool.providerMetadata,
-      }),
-    ),
    Effect.catch((error) =>
      tool.providerExecuted
        ? Effect.fail(error)
        : Effect.succeed(
-            LLMEvent.toolInputError({
-              id: tool.id,
-              name: tool.name,
-              raw,
-            }),
+            Option.getOrElse(
+              Option.map(parsePartialInput(raw), (input) => input ?? {}),
+              () => ({}),
+            ),
          ),
    ),
+    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 | ToolInputError): ReadonlyArray<LLMEvent> =>
-  event.type === "tool-input-error"
-    ? [event]
-    : [LLMEvent.toolInputEnd({ id: tool.id, name: tool.name, providerMetadata: tool.providerMetadata }), event]
+const finishEvents = (tool: PendingTool, event: ToolCall): ReadonlyArray<LLMEvent> => [
+  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>(
@@ -176,7 +179,7 @@ export const appendExisting = <K extends StreamKey>(

 /**
 * Finalize one pending tool call: parse the accumulated raw JSON, remove it
- * from state, and return either a call or a non-executable local input error.
+ * from state, and recover incomplete local arguments when needed.
 * Missing keys are a no-op because some providers emit stop events for
 * non-tool content blocks.
 */
@@ -1,15 +1,52 @@
+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 ?? {} },
@@ -35,7 +72,10 @@ const step = (state: OpenResponses.ParserState, event: OpenResponses.Event) => {

 export const protocol = Protocol.make({
  id: ADAPTER,
-  body: OpenResponses.protocol.body,
+  body: {
+    schema: XAIResponsesBody,
+    from: fromRequest,
+  },
  stream: {
    event: OpenResponses.protocol.stream.event,
    initial: (request) => OpenResponses.initial(request, extension),
@@ -0,0 +1,58 @@
+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)
@@ -0,0 +1,61 @@
+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)
@@ -3,8 +3,10 @@ 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"
@@ -14,5 +16,6 @@ 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, ProviderShared } from "../protocols/shared.js"
+import { isRecord } from "../protocols/shared.js"

 export const profile = OpenAICompatibleProfiles.profiles.openrouter
 export const id = ProviderID.make(profile.provider)
@@ -115,12 +115,10 @@ 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
        }),
      ),
@@ -0,0 +1,58 @@
+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 } },
+  defaults: { providerOptions: { store: false, include: ["reasoning.encrypted_content"] } },
 })

 const chatRoute = Route.make({
@@ -7,6 +7,7 @@ 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 {
@@ -321,12 +322,30 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
                Stream.mapEffect(decodeEvent(route)),
                protocol.stream.terminal ? Stream.takeUntil(protocol.stream.terminal) : (stream) => stream,
              )
-              const stream = events.pipe(
-                Stream.mapAccumEffect(
-                  () => protocol.stream.initial(request),
-                  protocol.stream.step,
-                  protocol.stream.onHalt ? { onHalt: protocol.stream.onHalt } : undefined,
-                ),
+              const stream = Stream.suspend(() => {
+                let state = protocol.stream.initial(request)
+                const parsed = events.pipe(
+                  Stream.mapEffect((event) =>
+                    protocol.stream.step(state, event).pipe(
+                      Effect.map(([next, output]) => {
+                        state = next
+                        return output
+                      }),
+                    ),
+                  ),
+                  Stream.flatMap(Stream.fromIterable),
+                )
+                const onHalt = protocol.stream.onHalt
+                return onHalt
+                  ? parsed.pipe(
+                      Stream.concat(
+                        Stream.suspend(() =>
+                          Stream.unwrap(onHalt(state).pipe(Effect.map(Stream.fromIterable))),
+                        ),
+                      ),
+                    )
+                  : parsed
+              }).pipe(
                Stream.catchCause((cause) => Stream.fail(streamError(route, `Failed to read ${route} stream`, cause))),
                requireTerminalEvent(route),
              )
@@ -382,7 +401,8 @@ export function make<Body, Prepared, Frame, Event, State>(
 }

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

  const body = yield* route.body
@@ -59,8 +59,8 @@ export interface ProtocolStream<Frame, Event, State> {
  readonly step: (state: State, event: Event) => Effect.Effect<readonly [State, ReadonlyArray<LLMEvent>], AIError>
  /** Optional request-completion signal for transports that do not end naturally. */
  readonly terminal?: (event: Event) => boolean
-  /** Optional flush emitted when the framed stream ends. */
-  readonly onHalt?: (state: State) => ReadonlyArray<LLMEvent>
+  /** Optional effectful flush emitted when the framed stream ends. */
+  readonly onHalt?: (state: State) => Effect.Effect<ReadonlyArray<LLMEvent>, AIError>
 }

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

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

 export namespace LanguageModelCompatibility {
@@ -0,0 +1,12 @@
+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 } from "../src/providers.js"
+import { AmazonBedrock, GoogleVertexMessages } 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"
@@ -86,6 +86,27 @@ 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, mergeProviderOptions } from "../src/index.js"
+import { LLM, Message, ToolCallPart, mergeProviderOptions } from "../src/index.js"
 import { AnthropicMessages, OpenAIChat } from "../src/protocols.js"
 import { Auth, LLMClient } from "../src/route.js"
 import { compileRequest } from "../src/route/client.js"
@@ -66,7 +66,7 @@ describe("request option precedence", () => {
      expect(prepared.body).toMatchObject({
        model: "gpt-4o-mini",
        stream: true,
-        max_tokens: 30,
+        max_completion_tokens: 30,
        temperature: 0.5,
        top_p: 0.9,
        frequency_penalty: 0.25,
@@ -247,6 +247,73 @@ 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({
@@ -0,0 +1,32 @@
+{
+  "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"
+      }
+    }
+  ]
+}
@@ -0,0 +1,32 @@
+{
+  "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": [
    {
@@ -18,7 +24,7 @@
        "headers": {
          "content-type": "application/json"
        },
-        "body": "{\"model\":\"@cf/openai/gpt-oss-20b\",\"messages\":[{\"role\":\"system\",\"content\":\"Call tools exactly as requested.\"},{\"role\":\"user\",\"content\":\"Call get_weather with city exactly Paris.\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false}}}],\"tool_choice\":{\"type\":\"function\",\"function\":{\"name\":\"get_weather\"}},\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":120,\"temperature\":0}"
+        "body": "{\"model\": \"@cf/openai/gpt-oss-20b\", \"messages\": [{\"role\": \"system\", \"content\": \"Call tools exactly as requested.\"}, {\"role\": \"user\", \"content\": \"Call get_weather with city exactly Paris.\"}], \"tools\": [{\"type\": \"function\", \"function\": {\"name\": \"get_weather\", \"description\": \"Get current weather for a city.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"city\": {\"type\": \"string\"}}, \"required\": [\"city\"], \"additionalProperties\": false}, \"strict\": false}}], \"tool_choice\": {\"type\": \"function\", \"function\": {\"name\": \"get_weather\"}}, \"stream\": true, \"stream_options\": {\"include_usage\": true}, \"max_tokens\": 120, \"temperature\": 0}"
      },
      "response": {
        "status": 200,
@@ -29,4 +35,4 @@
      }
    }
  ]
-}
+}
@@ -0,0 +1,32 @@
+{
+  "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"
+      }
+    }
+  ]
+}
@@ -0,0 +1,32 @@
+{
+  "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"
+      }
+    }
+  ]
+}
@@ -0,0 +1,50 @@
+{
+  "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"
+      }
+    }
+  ]
+}
@@ -32,7 +32,7 @@
        {
          "direction": "client",
          "kind": "text",
-          "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\"}}"
+          "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.\"}"
        },
        {
          "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\"}}"
+          "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.\"}"
        },
        {
          "direction": "server",
@@ -32,7 +32,7 @@
        {
          "direction": "client",
          "kind": "text",
-          "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\"}}"
+          "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.\"}"
        },
        {
          "direction": "server",
@@ -81,7 +81,7 @@
        {
          "direction": "client",
          "kind": "text",
-          "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\"}}"
+          "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.\"}"
        },
        {
          "direction": "server",
@@ -32,7 +32,7 @@
        {
          "direction": "client",
          "kind": "text",
-          "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\"}}"
+          "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.\"}"
        },
        {
          "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\"}}"
+          "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.\"}"
        },
        {
          "direction": "server",
@@ -91,7 +91,7 @@
        {
          "direction": "client",
          "kind": "text",
-          "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\"}}"
+          "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.\"}"
        },
        {
          "direction": "server",
@@ -26,7 +26,7 @@
        "headers": {
          "content-type": "application/json"
        },
-        "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}"
+        "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.\"}"
      },
      "response": {
        "status": 200,
@@ -18,7 +18,7 @@
        "headers": {
          "content-type": "application/json"
        },
-        "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}"
+        "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.\"}"
      },
      "response": {
        "status": 200,
@@ -25,7 +25,7 @@
        "headers": {
          "content-type": "application/json"
        },
-        "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}"
+        "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.\"}"
      },
      "response": {
        "status": 200,
@@ -43,7 +43,7 @@
        "headers": {
          "content-type": "application/json"
        },
-        "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}"
+        "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.\"}"
      },
      "response": {
        "status": 200,
@@ -0,0 +1,58 @@
+import { describe, expect, test } from "bun:test"
+import { Allow, MalformedJSON, PartialJSON, parse } from "../src/protocols/utils/partial-json.js"
+
+describe("partial JSON", () => {
+  test("parses complete JSON", () => {
+    expect(parse('{"key":"value","items":[1,true,null]}')).toEqual({
+      key: "value",
+      items: [1, true, null],
+    })
+
+    const object = parse('{"__proto__":{"safe":true}}') as Record<string, unknown>
+    expect(Object.hasOwn(object, "__proto__")).toBe(true)
+  })
+
+  test("parses partial strings", () => {
+    expect(parse('"hello')).toBe("hello")
+    expect(parse('"hello \\u12')).toBe("hello ")
+    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([""])
+    expect(parse('{"key":"', Allow.OBJ)).toEqual({})
+    expect(parse('{"key":"', Allow.OBJ | Allow.STR)).toEqual({ key: "" })
+  })
+
+  test("parses partial literals and numbers", () => {
+    expect(parse("nu", Allow.NULL)).toBeNull()
+    expect(parse("tr", Allow.BOOL)).toBe(true)
+    expect(parse("fa", Allow.BOOL)).toBe(false)
+    expect(parse("1e", Allow.NUM)).toBe(1)
+  })
+
+  test("distinguishes disallowed partial values from malformed values", () => {
+    expect(() => parse("[", Allow.STR)).toThrow(PartialJSON)
+    expect(() => parse("n", ~Allow.NULL)).toThrow(MalformedJSON)
+  })
+
+  test("rejects empty input", () => {
+    expect(() => parse("  ")).toThrow("is empty")
+  })
+})
@@ -26,6 +26,9 @@ 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"),
    ])

    for (const module of modules) expect(module.model).toBeFunction()
@@ -35,6 +38,24 @@ 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")
@@ -95,7 +116,11 @@ 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 })
+    expect(selected.route.defaults.providerOptions).toEqual({
+      reasoningEffort: "low",
+      store: true,
+      include: ["reasoning.encrypted_content"],
+    })
  })

  test("maps Anthropic-compatible settings onto the executable model", async () => {
@@ -285,7 +310,10 @@ 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 })
+    expect(responses.route.defaults.providerOptions).toEqual({
+      store: false,
+      include: ["reasoning.encrypted_content"],
+    })
  })

  test("rejects conflicting Vertex auth settings at runtime", async () => {
@@ -5,6 +5,7 @@ import { CacheHint, LLM, AIError, LLMRequest, Message, ToolCallPart, ToolDefinit
 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"
@@ -18,6 +19,21 @@ const opus48 = AnthropicMessages.route
  .with({ endpoint: { baseURL: "https://api.anthropic.test/v1/" }, auth: Auth.header("x-api-key", "test") })
  .model({ id: "claude-opus-4-8" })

+const compileUnsignedReasoning = (model: LLMRequest["model"]) =>
+  compileRequest(
+    LLM.request({
+      model,
+      messages: [Message.assistant([{ type: "reasoning", text: "unsigned reasoning" }])],
+      cache: "none",
+    }),
+  )
+
+const vertexOpus48 = GoogleVertexMessages.configure({
+  accessToken: "test",
+  location: "global",
+  project: "test",
+}).model("claude-opus-4-8")
+
 const request = LLM.request({
  id: "req_1",
  model,
@@ -277,6 +293,149 @@ 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(
@@ -564,6 +723,65 @@ describe("Anthropic Messages route", () => {
    }),
  )

+  it.effect("demotes unsigned reasoning when signatures are required", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileUnsignedReasoning(model)
+
+      expect(prepared.body.messages).toEqual([
+        { role: "assistant", content: [{ type: "text", text: "unsigned reasoning" }] },
+      ])
+    }),
+  )
+
+  it.effect("infers empty-signature compatibility across Kimi providers", () =>
+    Effect.gen(function* () {
+      const coding = AnthropicMessages.route.with({
+        provider: "kimi-for-coding",
+        endpoint: { baseURL: "https://compatible.test/v1/" },
+        auth: Auth.header("x-api-key", "test"),
+      })
+      const moonshot = AnthropicMessages.route
+        .with({
+          provider: "moonshotai",
+          endpoint: { baseURL: "https://api.moonshot.ai/anthropic" },
+          auth: Auth.bearer("test"),
+        })
+        .model({ id: "kimi-k2.6" })
+      const codingPrepared = yield* compileUnsignedReasoning(coding.model({ id: "k3" }))
+      const moonshotPrepared = yield* compileUnsignedReasoning(moonshot)
+
+      expect(codingPrepared.body.messages).toEqual([
+        {
+          role: "assistant",
+          content: [{ type: "thinking", thinking: "unsigned reasoning", signature: "" }],
+        },
+      ])
+      expect(moonshotPrepared.body.messages).toEqual([
+        {
+          role: "assistant",
+          content: [{ type: "thinking", thinking: "unsigned reasoning", signature: "" }],
+        },
+      ])
+    }),
+  )
+
+  it.effect("lets an explicit signature requirement override inference", () =>
+    Effect.gen(function* () {
+      const compatible = AnthropicMessages.route
+        .with({
+          provider: "kimi-for-coding",
+          endpoint: { baseURL: "https://api.kimi.com/coding/v1/" },
+          auth: Auth.header("x-api-key", "test"),
+        })
+        .model({ id: "k3", compatibility: { requireSignature: true } })
+      const prepared = yield* compileUnsignedReasoning(compatible)
+
+      expect(prepared.body.messages).toEqual([
+        { role: "assistant", content: [{ type: "text", text: "unsigned reasoning" }] },
+      ])
+    }),
+  )
+
  it.effect("round-trips redacted thinking as redacted_thinking blocks", () =>
    Effect.gen(function* () {
      const prepared = yield* compileRequest(
@@ -702,6 +920,108 @@ 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(
@@ -1060,8 +1380,14 @@ describe("Anthropic Messages route", () => {
      expect(response.events).toEqual([
        { type: "step-start", index: 0 },
        { type: "tool-input-start", id: "call_1", name: "lookup" },
-        { type: "tool-input-delta", id: "call_1", name: "lookup", text: '{"query"' },
-        { type: "tool-input-delta", id: "call_1", name: "lookup", text: ':"weather"}' },
+        { type: "tool-input-delta", id: "call_1", name: "lookup", text: '{"query"', input: {} },
+        {
+          type: "tool-input-delta",
+          id: "call_1",
+          name: "lookup",
+          text: ':"weather"}',
+          input: { query: "weather" },
+        },
        { type: "tool-input-end", id: "call_1", name: "lookup", providerMetadata: undefined },
        {
          type: "tool-call",
@@ -475,8 +475,14 @@ describe("Bedrock Converse route", () => {
      ])
      const events = response.events.filter((event) => event.type === "tool-input-delta")
      expect(events).toEqual([
-        { type: "tool-input-delta", id: "tool_1", name: "lookup", text: '{"query"' },
-        { type: "tool-input-delta", id: "tool_1", name: "lookup", text: ':"weather"}' },
+        { type: "tool-input-delta", id: "tool_1", name: "lookup", text: '{"query"', input: {} },
+        {
+          type: "tool-input-delta",
+          id: "tool_1",
+          name: "lookup",
+          text: ':"weather"}',
+          input: { query: "weather" },
+        },
      ])
      expect(response.events.at(-1)).toMatchObject({
        type: "finish",
@@ -485,7 +491,7 @@ describe("Bedrock Converse route", () => {
    }),
  )

-  it.effect("emits malformed tool input as an unexecuted tool error", () =>
+  it.effect("recovers incomplete tool input at finalization", () =>
    Effect.gen(function* () {
      const body = eventStreamBody(
        ["messageStart", { role: "assistant" }],
@@ -502,10 +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-input-error")).toMatchObject({
+      expect(response.events.find((event) => event.type === "tool-call")).toMatchObject({
        id: "tool_1",
        name: "lookup",
-        raw: '{"query":"partial',
+        input: { query: "partial" },
      })
      expect(response.finishReason).toEqual({ normalized: "tool-calls", raw: "end_turn" })
    }),
@@ -710,6 +716,32 @@ 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([
@@ -1,11 +1,12 @@
 import { describe, expect } from "bun:test"
 import { Effect } from "effect"
 import { HttpClientRequest } from "effect/unstable/http"
-import { LLM } from "../../src/index.js"
+import { LLM, Message } from "../../src/index.js"
 import { AmazonBedrockMantle } from "../../src/providers.js"
 import { compileRequest, LLMClient } from "../../src/route/client.js"
 import { it } from "../lib/effect.js"
-import { dynamicResponse } from "../lib/http.js"
+import { dynamicResponse, fixedResponse } from "../lib/http.js"
+import { sseEvents } from "../lib/sse.js"
 import { recordedTests } from "../recorded-test.js"

 const credentials = {
@@ -71,7 +72,9 @@ describe("Amazon Bedrock Mantle provider", () => {
            Effect.gen(function* () {
              const request = yield* HttpClientRequest.toWeb(input.request)
              seen.push({ url: request.url, authorization: request.headers.get("authorization") ?? undefined })
-              return input.respond("", { headers: { "content-type": "text/event-stream" } })
+              return input.respond(sseEvents({ choices: [{ delta: {}, finish_reason: "stop" }] }), {
+                headers: { "content-type": "text/event-stream" },
+              })
            }),
          ),
        ),
@@ -80,6 +83,39 @@ 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(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({
@@ -906,6 +906,54 @@ describe("Gemini route", () => {
    }),
  )

+  it.effect("ignores unknown response parts", () =>
+    Effect.gen(function* () {
+      const response = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents({
+              candidates: [
+                {
+                  content: {
+                    role: "model",
+                    parts: [
+                      { text: "Hello " },
+                      { executableCode: { language: "PYTHON", code: "print('ignored')" } },
+                      { text: "world" },
+                    ],
+                  },
+                  finishReason: "STOP",
+                },
+              ],
+            }),
+          ),
+        ),
+      )
+
+      expect(response.text).toBe("Hello world")
+      expect(response.finishReason).toEqual({ normalized: "stop", raw: "STOP" })
+    }),
+  )
+
+  it.effect("rejects malformed recognized response parts", () =>
+    Effect.gen(function* () {
+      const error = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents({
+              candidates: [{ content: { role: "model", parts: [{ text: 42 }] } }],
+            }),
+          ),
+        ),
+        Effect.flip,
+      )
+
+      expect(error).toBeInstanceOf(AIError)
+      expect(error.reason).toMatchObject({ _tag: "InvalidProviderOutput" })
+      expect(error.message).toContain("Invalid google/gemini stream event")
+    }),
+  )
+
  it.effect("preserves thoughtSignature for reasoning and tool-call continuation", () =>
    Effect.gen(function* () {
      const body = sseEvents({
@@ -1,5 +1,6 @@
 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"
@@ -47,14 +48,16 @@ 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 = OpenAICompatible.togetherai
-  .configure({
-    apiKey: process.env.TOGETHER_AI_API_KEY ?? "fixture",
-  })
-  .model("meta-llama/Llama-3.3-70B-Instruct-Turbo")
+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 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")
@@ -193,8 +196,27 @@ describeRecordedGoldenScenarios([
    name: "TogetherAI Llama 3.3 70B",
    prefix: "openai-compatible-chat",
    model: together,
-    requires: ["TOGETHER_AI_API_KEY"],
-    scenarios: ["text", "tool-call"],
+    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 },
+    ],
  },
  {
    name: "Groq Llama 3.3 70B",
@@ -203,6 +225,13 @@ 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",
@@ -0,0 +1,184 @@
+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 { Cerebras, DeepInfra, TogetherAI } 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("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",
+        },
+      ]
+
+      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"))),
+        ),
+      )
+    }),
+  )
+})
@@ -47,7 +47,7 @@ describe("OpenAI Chat route", () => {
    Effect.gen(function* () {
      const prepared = yield* compileRequest(request)

-      expect(prepared.body).toEqual({
+      expect(prepared.body).toMatchObject({
        model: "gpt-4o-mini",
        messages: [
          { role: "system", content: "You are concise." },
@@ -55,7 +55,8 @@ describe("OpenAI Chat route", () => {
        ],
        stream: true,
        stream_options: { include_usage: true },
-        max_tokens: 20,
+        store: false,
+        max_completion_tokens: 20,
        temperature: 0,
      })
    }),
@@ -84,6 +85,28 @@ 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(
@@ -146,6 +169,56 @@ 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(
@@ -191,6 +264,21 @@ 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({
@@ -325,7 +413,7 @@ describe("OpenAI Chat route", () => {
        }),
      )

-      expect(prepared.body).toEqual({
+      expect(prepared.body).toMatchObject({
        model: "gpt-4o-mini",
        messages: [
          { role: "user", content: "What is the weather?" },
@@ -345,10 +433,40 @@ describe("OpenAI Chat route", () => {
        tools: [],
        stream: true,
        stream_options: { include_usage: true },
+        store: false,
      })
    }),
  )

+  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" } }
@@ -414,6 +532,30 @@ 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(
@@ -1162,8 +1304,14 @@ describe("OpenAI Chat route", () => {
      expect(response.events).toEqual([
        { type: "step-start", index: 0 },
        { type: "tool-input-start", id: "call_1", name: "lookup", providerMetadata: undefined },
-        { type: "tool-input-delta", id: "call_1", name: "lookup", text: '{"query"' },
-        { type: "tool-input-delta", id: "call_1", name: "lookup", text: ':"weather"}' },
+        { type: "tool-input-delta", id: "call_1", name: "lookup", text: '{"query"', input: {} },
+        {
+          type: "tool-input-delta",
+          id: "call_1",
+          name: "lookup",
+          text: ':"weather"}',
+          input: { query: "weather" },
+        },
        { type: "tool-input-end", id: "call_1", name: "lookup", providerMetadata: undefined },
        {
          type: "tool-call",
@@ -1243,6 +1391,11 @@ describe("OpenAI Chat route", () => {
      ).pipe(Effect.provide(fixedResponse(body)), Effect.flip)

      expect(error.message).toContain("OpenAI Chat tool call delta is missing id or name")
+      expect(error.reason._tag).toBe("InvalidProviderOutput")
+      if (error.reason._tag !== "InvalidProviderOutput") return
+      expect(decodeJson(error.reason.raw ?? "")).toMatchObject({
+        choices: [{ finish_reason: "tool_calls" }],
+      })
    }),
  )

@@ -1256,6 +1409,7 @@ describe("OpenAI Chat route", () => {
        deltaChunk({ tool_calls: [{ index: 0, function: { arguments: ':"weather"}' } }] }),
      )
      const input = LLMRequest.update(request, {
+        model: LanguageModel.update(model, { compatibility: { requireFinishReason: false } }),
        tools: [ToolDefinition.make({ name: "lookup", description: "Lookup data", inputSchema: { type: "object" } })],
      })
      const response = yield* LLMClient.generate(input).pipe(Effect.provide(fixedResponse(body)))
@@ -1263,8 +1417,14 @@ describe("OpenAI Chat route", () => {
      expect(response.events).toEqual([
        { type: "step-start", index: 0 },
        { type: "tool-input-start", id: "call_1", name: "lookup", providerMetadata: undefined },
-        { type: "tool-input-delta", id: "call_1", name: "lookup", text: '{"query"' },
-        { type: "tool-input-delta", id: "call_1", name: "lookup", text: ':"weather"}' },
+        { type: "tool-input-delta", id: "call_1", name: "lookup", text: '{"query"', input: {} },
+        {
+          type: "tool-input-delta",
+          id: "call_1",
+          name: "lookup",
+          text: ':"weather"}',
+          input: { query: "weather" },
+        },
        { type: "tool-input-end", id: "call_1", name: "lookup", providerMetadata: undefined },
        {
          type: "tool-call",
@@ -70,7 +70,7 @@ describe("OpenAI-compatible Chat route", () => {
        baseURL: "https://api.deepseek.test/v1/",
        query: { "api-version": "2026-01-01" },
      })
-      expect(prepared.body).toEqual({
+      expect(prepared.body).toMatchObject({
        model: "deepseek-chat",
        messages: [
          { role: "system", content: "You are concise." },
@@ -79,7 +79,7 @@ describe("OpenAI-compatible Chat route", () => {
        tools: [
          {
            type: "function",
-            function: { name: "lookup", description: "Lookup data", parameters: { type: "object" } },
+            function: { name: "lookup", description: "Lookup data", parameters: { type: "object" }, strict: false },
          },
        ],
        tool_choice: "required",
@@ -130,7 +130,7 @@ describe("OpenAI-compatible Chat route", () => {
    Effect.gen(function* () {
      const prepared = yield* compileRequest(request)

-      expect(prepared.body).toEqual({
+      expect(prepared.body).toMatchObject({
        model: "deepseek-chat",
        messages: [
          { role: "system", content: "You are concise." },
@@ -158,6 +158,29 @@ describe("OpenAI-compatible Chat route", () => {
    }),
  )

+  it.effect("enables ZAI tool streaming except for GLM 4.5 models", () =>
+    Effect.gen(function* () {
+      const prepare = (provider: string, baseURL: string, id: string) =>
+        compileRequest(
+          LLM.request({
+            model: OpenAICompatibleChat.route.with({ provider, endpoint: { baseURL } }).model({ id }),
+            prompt: "Use a tool.",
+            tools: [ToolDefinition.make({ name: "lookup", description: "Lookup data", inputSchema: {} })],
+          }),
+        )
+
+      const current = yield* prepare("zai", "https://api.z.ai/api/paas/v4", "glm-4.7")
+      expect(current.body).toMatchObject({ tool_stream: true })
+
+      const legacy = yield* Effect.all(
+        ["glm-4.5", "glm-4.5-air", "glm-4.5-flash", "glm-4.5v"].map((id) =>
+          prepare("zhipuai", "https://open.bigmodel.cn/api/paas/v4", id),
+        ),
+      )
+      legacy.forEach((item) => expect(item.body).not.toHaveProperty("tool_stream"))
+    }),
+  )
+
  it.effect("matches AI SDK compatible tool request body fixture", () =>
    Effect.gen(function* () {
      const prepared = yield* compileRequest(
@@ -180,7 +203,7 @@ describe("OpenAI-compatible Chat route", () => {
        }),
      )

-      expect(prepared.body).toEqual({
+      expect(prepared.body).toMatchObject({
        model: "deepseek-chat",
        messages: [
          { role: "user", content: "What is the weather?" },
@@ -204,6 +227,7 @@ describe("OpenAI-compatible Chat route", () => {
              name: "lookup",
              description: "Lookup data",
              parameters: { type: "object", properties: { query: { type: "string" } }, required: ["query"] },
+              strict: false,
            },
          },
        ],
@@ -214,6 +238,135 @@ 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(
@@ -329,13 +482,106 @@ describe("OpenAI-compatible Chat route", () => {
    }),
  )

-  it.effect("treats an empty finish reason as terminal", () =>
+  it.effect("rejects a stream without a required finish reason", () =>
    Effect.gen(function* () {
-      const response = yield* LLMClient.generate(request).pipe(
+      const error = yield* LLMClient.generate(request).pipe(
+        Effect.provide(fixedResponse(sseEvents(deltaChunk({ content: "Hello" }), deltaChunk({}, "")))),
+        Effect.flip,
+      )
+
+      expect(error.reason).toMatchObject({
+        _tag: "InvalidProviderOutput",
+        classification: "incomplete-stream",
+        message: "OpenAI Chat stream ended without finish_reason",
+      })
+    }),
+  )
+
+  it.effect("infers stop when finish reasons are optional", () =>
+    Effect.gen(function* () {
+      const compatible = OpenAICompatibleChat.route
+        .with({ provider: "custom", endpoint: { baseURL: "https://api.custom.test/v1" } })
+        .model({ id: "custom-model", compatibility: { requireFinishReason: false } })
+      const response = yield* LLMClient.generate(LLMRequest.update(request, { model: compatible })).pipe(
        Effect.provide(fixedResponse(sseEvents(deltaChunk({ content: "Hello" }), deltaChunk({}, "")))),
      )

-      expect(response.finishReason).toEqual({ normalized: "unknown", raw: "" })
+      expect(response.finishReason).toEqual({ normalized: "stop" })
+    }),
+  )
+
+  it.effect("normalizes the end finish reason to stop", () =>
+    Effect.gen(function* () {
+      const response = yield* LLMClient.generate(request).pipe(
+        Effect.provide(fixedResponse(sseEvents(deltaChunk({ content: "Hello" }), deltaChunk({}, "end")))),
+      )
+
+      expect(response.finishReason).toEqual({ normalized: "stop", raw: "end" })
+    }),
+  )
+
+  it.effect("classifies provider error finish reasons", () =>
+    Effect.gen(function* () {
+      const error = yield* LLMClient.generate(request).pipe(
+        Effect.provide(fixedResponse(sseEvents(deltaChunk({}, "network_error")))),
+        Effect.flip,
+      )
+
+      expect(error.reason).toMatchObject({
+        _tag: "ProviderInternal",
+        message: "Provider reported a network error (finish_reason: network_error)",
+      })
+      expect(decodeJson(error.body ?? "")).toMatchObject({
+        id: "chatcmpl_fixture",
+        choices: [{ finish_reason: "network_error" }],
+      })
+
+      const generic = yield* LLMClient.generate(request).pipe(
+        Effect.provide(fixedResponse(sseEvents(deltaChunk({}, "error")))),
+        Effect.flip,
+      )
+      expect(generic.reason).toMatchObject({
+        _tag: "UnknownProvider",
+        message: "Provider reported an error (finish_reason: error)",
+      })
+    }),
+  )
+
+  it.effect("preserves explicit provider error events", () =>
+    Effect.gen(function* () {
+      const error = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents({
+              id: "chatcmpl_error",
+              error: { code: 502, message: "Provider disconnected", details: { upstream: "vendor" } },
+              trace_id: "trace_1",
+            }),
+          ),
+        ),
+        Effect.flip,
+      )
+
+      expect(error.reason).toMatchObject({ _tag: "ProviderInternal", message: "Provider disconnected", status: 502 })
+      expect(decodeJson(error.body ?? "")).toMatchObject({
+        id: "chatcmpl_error",
+        error: { code: 502, message: "Provider disconnected", details: { upstream: "vendor" } },
+        trace_id: "trace_1",
+      })
+    }),
+  )
+
+  it.effect("preserves provider finish outcomes in the common reason algebra", () =>
+    Effect.gen(function* () {
+      const filtered = yield* LLMClient.generate(request).pipe(
+        Effect.provide(fixedResponse(sseEvents(deltaChunk({}, "content_filter")))),
+      )
+      const future = yield* LLMClient.generate(request).pipe(
+        Effect.provide(fixedResponse(sseEvents(deltaChunk({}, "future_reason")))),
+      )
+
+      expect(filtered.finishReason).toEqual({ normalized: "content-filter", raw: "content_filter" })
+      expect(future.finishReason).toEqual({ normalized: "unknown", raw: "future_reason" })
    }),
  )

@@ -355,6 +601,11 @@ describe("OpenAI-compatible Chat route", () => {
      )

      expect(error.message).toContain("OpenAI Chat received content after the finish reason")
+      expect(error.reason._tag).toBe("InvalidProviderOutput")
+      if (error.reason._tag !== "InvalidProviderOutput") return
+      expect(decodeJson(error.reason.raw ?? "")).toMatchObject({
+        choices: [{ delta: { tool_calls: [{ id: "call_1" }] } }],
+      })
    }),
  )
 })
@@ -47,15 +47,31 @@ describe("Open Responses-compatible route", () => {
      })
      expect(prepared.body).toEqual({
        model: "example-model",
-        input: [
-          { role: "system", content: "You are concise." },
-          { role: "user", content: [{ type: "input_text", text: "Say hello." }] },
-        ],
+        input: [{ role: "user", content: [{ type: "input_text", text: "Say hello." }] }],
+        instructions: "You are concise.",
        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({
@@ -66,10 +82,12 @@ 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." },
@@ -132,6 +150,40 @@ describe("Open Responses-compatible route", () => {
    }),
  )

+  it.effect("lowers canonical parallel tool control", () =>
+    Effect.gen(function* () {
+      const model = configure({
+        apiKey: "test-key",
+        baseURL: "https://responses.example.test/v1",
+      }).model("example-model")
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model,
+          prompt: "Read the file.",
+          tools: [
+            ToolDefinition.make({
+              name: "read",
+              description: "Read a file.",
+              inputSchema: { type: "object" },
+            }),
+          ],
+          toolChoice: { type: "auto", disableParallelToolUse: true },
+        }),
+      )
+
+      expect(prepared.body.parallel_tool_calls).toBe(false)
+      expect(prepared.body.tools).toEqual([
+        {
+          type: "function",
+          name: "read",
+          description: "Read a file.",
+          parameters: { type: "object" },
+          strict: false,
+        },
+      ])
+    }),
+  )
+
  it.effect("keeps foreign item id grammars but drops malformed ids", () =>
    Effect.gen(function* () {
      const model = configure({
@@ -143,15 +195,19 @@ describe("Open Responses-compatible route", () => {
          model,
          messages: [
            Message.assistant([
-              // 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: "Dropped.",
-                providerMetadata: { openresponses: { itemId: `m${"a".repeat(64)}` } },
+                text: "Long.",
+                providerMetadata: { openresponses: { itemId: `history_${"a".repeat(64)}` } },
              },
+              {
+                type: "text",
+                text: "Opaque.",
+                providerMetadata: { openresponses: { itemId: "provider_value/with+symbols" } },
+              },
+              { type: "text", text: "No suffix.", providerMetadata: { openresponses: { itemId: "msg_" } } },
+              { type: "text", text: "No prefix.", providerMetadata: { openresponses: { itemId: "_item" } } },
            ]),
          ],
        }),
@@ -166,13 +222,202 @@ describe("Open Responses-compatible route", () => {
        },
        {
          type: "message",
+          id: `history_${"a".repeat(64)}`,
          role: "assistant",
-          content: [{ type: "output_text", text: "Dropped." }],
+          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." },
+          ],
        },
      ])
    }),
  )

+  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: { openresponses: { 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: { openresponses: { itemId: "msg_1" } } },
+      ])
+    }),
+  )
+
+  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: { openresponses: { 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: { openresponses: { itemId: "rs_raw", reasoningEncryptedContent: "raw-state" } },
+      })
+    }),
+  )
+
  it.effect("reconciles raw reasoning finals without streamed deltas", () =>
    Effect.gen(function* () {
      const model = configure({
@@ -136,6 +136,7 @@ describe("OpenAI Responses WebSocket recorded", () => {
      expect(channel.opens()).toBe(1)
      expect(channel.sent).toHaveLength(2)
      expect(channel.sent[1]).toMatchObject({
+        instructions: "Call get_weather once, then reply exactly: Paris is sunny.",
        previous_response_id: expect.any(String),
        input: [{ type: "function_call_output", call_id: call.id, output: expect.any(String) }],
      })
@@ -167,8 +168,8 @@ describe("OpenAI Responses WebSocket recorded", () => {
      expect(channel.opens()).toBe(2)
      expect(channel.sent[1]).not.toHaveProperty("previous_response_id")
      expect(channel.sent[1]).toMatchObject({
+        instructions: "Follow the user's exact reply instruction.",
        input: [
-          { role: "system", content: "Follow the user's exact reply instruction." },
          { role: "user", content: [{ type: "input_text", text: "Reply exactly: Alpha." }] },
          { role: "assistant", content: [{ type: "output_text", text: "Alpha." }] },
          { role: "user", content: [{ type: "input_text", text: "Reply exactly: Beta." }] },
@@ -204,8 +205,8 @@ describe("OpenAI Responses WebSocket recorded", () => {
      expect(channel.sent[1]).toHaveProperty("previous_response_id", expect.any(String))
      expect(channel.sent[2]).not.toHaveProperty("previous_response_id")
      expect(channel.sent[2]).toMatchObject({
+        instructions: "Follow the user's exact reply instruction.",
        input: [
-          { role: "system", content: "Follow the user's exact reply instruction." },
          { role: "user", content: [{ type: "input_text", text: "Reply exactly: Ready." }] },
          { role: "assistant", content: [{ type: "output_text", text: "Ready." }] },
          { role: "user", content: [{ type: "input_text", text: "Reply exactly: Recovered." }] },
@@ -190,6 +190,21 @@ describe("OpenRouter", () => {
    }),
  )

+  it.effect("omits the prompt cache key when caching is disabled", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: OpenRouter.configure({ apiKey: "test-key" }).model("openai/gpt-4o-mini"),
+          prompt: "Hello",
+          promptCacheKey: "session_123",
+          cache: "none",
+        }),
+      )
+
+      expect(prepared.body).not.toHaveProperty("prompt_cache_key")
+    }),
+  )
+
  it.effect("filters invalid known OpenRouter options while preserving extensions", () =>
    Effect.gen(function* () {
      const invalid: Record<string, unknown> = {
@@ -1,6 +1,6 @@
 import { describe, expect } from "bun:test"
 import { Effect } from "effect"
-import { LLM, LLMEvent } from "../../src/index.js"
+import { LLM, LLMEvent, Message } from "../../src/index.js"
 import { XAI } from "../../src/providers.js"
 import { OpenResponses } from "../../src/protocols/open-responses.js"
 import { OpenAIResponses } from "../../src/protocols/openai-responses.js"
@@ -14,13 +14,24 @@ import { sseEvents } from "../lib/sse.js"
 const model = XAI.configure({ apiKey: "test", baseURL: "https://api.x.ai/v1" }).responses("grok-4.6")

 describe("xAI Responses route", () => {
-  it.effect("extends the Open Responses baseline directly", () =>
+  it.effect("composes the Open Responses baseline with xAI extensions", () =>
    Effect.gen(function* () {
-      expect(XAIResponses.protocol.body).toBe(OpenResponses.protocol.body)
+      expect(XAIResponses.protocol.body).not.toBe(OpenResponses.protocol.body)
      expect(XAIResponses.protocol.body).not.toBe(OpenAIResponses.protocol.body)

      const prepared = yield* compileRequest(LLM.request({ model, prompt: "Hello" }))
      expect(prepared.protocol).toBe("xai-responses")
+      expect(prepared.body.store).toBe(false)
+      expect(prepared.body.include).toEqual(["reasoning.encrypted_content"])
+    }),
+  )
+
+  it.effect("allows callers to opt out of encrypted reasoning", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(LLM.request({ model, prompt: "Hello", providerOptions: { include: [] } }))
+
+      expect(prepared.body.store).toBe(false)
+      expect(prepared.body.include).toBeUndefined()
    }),
  )

@@ -59,16 +70,106 @@ describe("xAI Responses route", () => {
    }),
  )

-  it.effect("parses xAI hosted tool items", () =>
+  it.effect("routes xAI reasoning summaries by output index", () =>
    Effect.gen(function* () {
-      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Search X" })).pipe(
+      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Think" })).pipe(
        Effect.provide(
          fixedResponse(
            sseEvents(
              {
-                type: "response.output_item.done",
-                item: { type: "x_search_call", id: "x_search_1", status: "completed", action: { query: "news" } },
+                type: "response.output_item.added",
+                output_index: 3,
+                item: { type: "reasoning", id: "reasoning_1" },
              },
+              {
+                type: "response.reasoning_summary_text.delta",
+                output_index: 3,
+                item_id: "wrong_reasoning",
+                summary_index: 0,
+                delta: "Considering.",
+              },
+              {
+                type: "response.output_item.done",
+                output_index: 3,
+                item: { type: "reasoning", id: "reasoning_1", encrypted_content: "opaque" },
+              },
+              { type: "response.completed", response: { id: "response_1" } },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.reasoning).toBe("Considering.")
+      expect(response.message.content.find((part) => part.type === "reasoning")).toMatchObject({
+        providerMetadata: { xai: { itemId: "reasoning_1", reasoningEncryptedContent: "opaque" } },
+      })
+    }),
+  )
+
+  it.effect("replays xAI hosted tool items when continuing with the same provider", () =>
+    Effect.gen(function* () {
+      const item = { type: "x_search_call", id: "x_search_1", status: "completed", action: { query: "news" } }
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model,
+          messages: [
+            Message.assistant([
+              {
+                type: "tool-result",
+                id: "x_search_1",
+                name: "x_search",
+                result: { type: "json", value: item },
+                providerExecuted: true,
+                providerMetadata: { xai: { itemId: "x_search_1" } },
+              },
+            ]),
+          ],
+        }),
+      )
+
+      expect(prepared.body.input).toEqual([item])
+    }),
+  )
+
+  it.effect("replays shared and xAI hosted tool items but rejects OpenAI extensions", () =>
+    Effect.gen(function* () {
+      const items = [
+        { type: "web_search_call", id: "ws_1", status: "completed" },
+        { type: "image_generation_call", id: "ig_1", status: "completed", result: "AQID" },
+        { type: "computer_call", id: "computer_1", status: "completed" },
+      ]
+      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: { xai: { itemId: item.id } },
+            }),
+          ),
+        }),
+      )
+
+      expect(prepared.body.input).toEqual([
+        items[0],
+        items[1],
+        { role: "user", content: [{ type: "input_text", text: JSON.stringify(items[2]) }] },
+      ])
+    }),
+  )
+
+  it.effect("parses xAI hosted tool items", () =>
+    Effect.gen(function* () {
+      const item = { type: "x_search_call", id: "x_search_1", status: "completed", action: { query: "news" } }
+      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Search X" })).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              { type: "response.output_item.done", item },
              { type: "response.completed", response: { id: "response_1" } },
            ),
          ),
@@ -80,6 +181,11 @@ describe("xAI Responses route", () => {
        name: "x_search",
        input: { query: "news" },
        providerExecuted: true,
+        providerMetadata: { xai: { itemId: "x_search_1" } },
+      })
+      expect(response.events.find(LLMEvent.is.toolResult)).toMatchObject({
+        result: { type: "json", value: item },
+        providerMetadata: { xai: { itemId: "x_search_1" } },
      })
    }),
  )
@@ -78,6 +78,31 @@ describe("Z.ai Images", () => {
    ),
  )

+  it.effect("sanitizes unpaired surrogates in outbound image requests", () =>
+    Image.generate({
+      model: ZAI.configure({ apiKey: "test", http: { body: { metadata: { source: "default\uDC00" } } } }).image("model"),
+      prompt: "A red circle \uD800 on a white background \u{1F600}",
+    }).pipe(
+      Effect.provide(
+        ImageClient.layer.pipe(
+          Layer.provide(
+            dynamicResponse((input) => {
+              expect(JSON.parse(input.text)).toMatchObject({
+                prompt: "A red circle \uFFFD on a white background \u{1F600}",
+                metadata: { source: "default\uFFFD" },
+              })
+              return Effect.succeed(
+                input.respond(JSON.stringify({ data: [{ url: "https://example.test/image.jpg" }] }), {
+                  headers: { "content-type": "application/json" },
+                }),
+              )
+            }),
+          ),
+        ),
+      ),
+    ),
+  )
+
  it.effect("lets raw native options override aliases", () =>
    Image.generate({
      model: ZAI.configure({ apiKey: "test" }).image("model"),
@@ -20,6 +20,7 @@ type ScenarioInput =
      readonly name?: string
      readonly cassette?: string
      readonly tags?: ReadonlyArray<string>
+      readonly prompt?: string
      readonly maxTokens?: number
      readonly temperature?: number | false
      readonly timeout?: number
@@ -87,6 +88,7 @@ const runTarget = (target: TargetInput) => {
            yield* runGoldenScenario(input.id, {
              id: `recorded_${kebab(target.name).replaceAll("-", "_")}_${input.id.replaceAll("-", "_")}`,
              model: target.model,
+              prompt: input.prompt,
              maxTokens: input.maxTokens,
              temperature: input.temperature,
            })
@@ -164,6 +164,7 @@ export const expectGoldenWeatherToolLoop = (events: ReadonlyArray<LLMEvent>) =>
 export interface GoldenScenarioContext {
  readonly id: string
  readonly model: LanguageModel
+  readonly prompt?: string
  readonly maxTokens?: number
  readonly temperature?: number | false
 }
@@ -298,7 +299,7 @@ const runGeneratedConversation = (context: GoldenScenarioContext, steps: Readonl

 const runTextScenario = (context: GoldenScenarioContext) =>
  runGeneratedConversation(context, [
-    user("Reply exactly with: Hello!"),
+    user(context.prompt ?? "Reply exactly with: Hello!"),
    assistant.expectText(/^Hello!?$/, {
      system: "You are concise.",
      maxTokens: context.maxTokens ?? 40,
@@ -102,6 +102,38 @@ describe("AI.Usage", () => {
    expect(error.reason._tag).toBe("InvalidProviderOutput")
  })

+  test("sseFraming ignores retry directives without ending the stream", async () => {
+    const encoder = new TextEncoder()
+    const frames = await Effect.runPromise(
+      ProviderShared.sseFraming(
+        Stream.make(
+          encoder.encode("retry: 1000\n\n"),
+          encoder.encode('data: {"first":true}\n\n'),
+          encoder.encode("retry: 2000\n\n"),
+          encoder.encode('data: {"second":true}\n\n'),
+        ).pipe(Stream.rechunk(1)),
+      ).pipe(Stream.runCollect),
+    )
+
+    expect(Array.from(frames)).toEqual(['{"first":true}', '{"second":true}'])
+  })
+
+  test("sseFraming preserves event data around retry directives", async () => {
+    const encoder = new TextEncoder()
+    const frames = await Effect.runPromise(
+      ProviderShared.sseFraming(
+        Stream.make(
+          encoder.encode("event: update\ndata: first\n"),
+          encoder.encode("retry: 1000\n"),
+          encoder.encode("data: second\n\n"),
+        ).pipe(Stream.rechunk(1)),
+        new Set(["update"]),
+      ).pipe(Stream.runCollect),
+    )
+
+    expect(Array.from(frames)).toEqual(["first\nsecond"])
+  })
+
  test("visibleOutputTokens clamps reasoning > output to zero", () => {
    expect(new Usage({ outputTokens: 10, reasoningTokens: 4 }).visibleOutputTokens).toBe(6)
    expect(new Usage({ outputTokens: 10 }).visibleOutputTokens).toBe(10)
@@ -101,7 +101,7 @@ describe("LLMClient tools", () => {
      const messages = Reflect.get(second, "messages")
      const tools = Reflect.get(second, "tools")

-      expect(Reflect.get(second, "max_tokens")).toBe(50)
+      expect(Reflect.get(second, "max_completion_tokens")).toBe(50)
      expect(Reflect.get(second, "tool_choice")).toBe("auto")
      expect(tools).toHaveLength(1)
      expect(
@@ -23,9 +23,11 @@ describe("ToolStream", () => {

      expect(first.events).toEqual([
        { type: "tool-input-start", id: "call_1", name: "lookup" },
-        { type: "tool-input-delta", id: "call_1", name: "lookup", text: '{"query"' },
+        { type: "tool-input-delta", id: "call_1", name: "lookup", text: '{"query"', input: {} },
+      ])
+      expect(second.events).toEqual([
+        { type: "tool-input-delta", id: "call_1", name: "lookup", text: ':"weather"}', input: { query: "weather" } },
      ])
-      expect(second.events).toEqual([{ type: "tool-input-delta", id: "call_1", name: "lookup", text: ':"weather"}' }])
      expect(finished).toEqual({
        tools: {},
        events: [
@@ -36,6 +38,45 @@ describe("ToolStream", () => {
    }),
  )

+  it.effect("exposes cumulative partial string values", () =>
+    Effect.gen(function* () {
+      const result = ToolStream.appendOrStart(
+        ADAPTER,
+        ToolStream.empty<number>(),
+        0,
+        { id: "call_1", name: "lookup", text: '{"query":"wea' },
+        "missing tool",
+      )
+      if (ToolStream.isError(result)) return yield* result
+
+      expect(result.events.at(-1)).toEqual({
+        type: "tool-input-delta",
+        id: "call_1",
+        name: "lookup",
+        text: '{"query":"wea',
+        input: { query: "wea" },
+      })
+    }),
+  )
+
+  it.effect("defaults partial input to an empty object when the accumulated value cannot be parsed", () =>
+    Effect.gen(function* () {
+      const result = ToolStream.appendOrStart(
+        ADAPTER,
+        ToolStream.empty<number>(),
+        0,
+        { id: "call_1", name: "lookup", text: "x" },
+        "missing tool",
+      )
+      if (ToolStream.isError(result)) return yield* result
+
+      expect(result.events).toEqual([
+        { type: "tool-input-start", id: "call_1", name: "lookup" },
+        { type: "tool-input-delta", id: "call_1", name: "lookup", text: "x", input: {} },
+      ])
+    }),
+  )
+
  it.effect("keeps accumulated identity when later deltas contain empty strings", () =>
    Effect.gen(function* () {
      const first = ToolStream.appendOrStart(
@@ -91,7 +132,7 @@ describe("ToolStream", () => {
    }),
  )

-  it.effect("finalizes malformed local input as a non-executable tool error", () =>
+  it.effect("finalizes incomplete local input using the partial JSON parser", () =>
    Effect.gen(function* () {
      const tools = ToolStream.start(ToolStream.empty<string>(), "item_1", {
        id: "call_1",
@@ -103,18 +144,46 @@ describe("ToolStream", () => {
      expect(finished).toEqual({
        tools: {},
        events: [
-          {
-            type: "tool-input-error",
-            id: "call_1",
-            name: "lookup",
-            raw: '{"query":"partial',
-          },
+          { type: "tool-input-end", id: "call_1", name: "lookup" },
+          { type: "tool-call", id: "call_1", name: "lookup", input: { query: "partial" } },
        ],
      })
    }),
  )

-  it.effect("preserves valid siblings when one parallel input is malformed", () =>
+  it.effect("repairs malformed string escapes in final local input", () =>
+    Effect.gen(function* () {
+      const tools = ToolStream.start(ToolStream.empty<string>(), "item_1", {
+        id: "call_1",
+        name: "lookup",
+        input: '{"path":"A\\H","text":"first\tsecond"}',
+      })
+      const finished = yield* ToolStream.finish(ADAPTER, tools, "item_1")
+
+      expect(finished.events).toEqual([
+        { type: "tool-input-end", id: "call_1", name: "lookup" },
+        { type: "tool-call", id: "call_1", name: "lookup", input: { path: "A\\H", text: "first\tsecond" } },
+      ])
+    }),
+  )
+
+  it.effect("defaults unrecoverable local input to an empty object", () =>
+    Effect.gen(function* () {
+      const tools = ToolStream.start(ToolStream.empty<string>(), "item_1", {
+        id: "call_1",
+        name: "lookup",
+        input: "invalid",
+      })
+      const finished = yield* ToolStream.finish(ADAPTER, tools, "item_1")
+
+      expect(finished.events).toEqual([
+        { type: "tool-input-end", id: "call_1", name: "lookup" },
+        { type: "tool-call", id: "call_1", name: "lookup", input: {} },
+      ])
+    }),
+  )
+
+  it.effect("recovers incomplete input alongside valid parallel tool calls", () =>
    Effect.gen(function* () {
      const valid = ToolStream.start(ToolStream.empty<number>(), 0, {
        id: "call_valid",
@@ -133,12 +202,8 @@ describe("ToolStream", () => {
        events: [
          { type: "tool-input-end", id: "call_valid", name: "lookup" },
          { type: "tool-call", id: "call_valid", name: "lookup", input: { query: "weather" } },
-          {
-            type: "tool-input-error",
-            id: "call_invalid",
-            name: "lookup",
-            raw: '{"query":"partial',
-          },
+          { type: "tool-input-end", id: "call_invalid", name: "lookup" },
+          { type: "tool-call", id: "call_invalid", name: "lookup", input: { query: "partial" } },
        ],
      })
    }),
@@ -0,0 +1,53 @@
+import { expect, test } from "@playwright/test"
+import { mockOpenCodeServer } from "../utils/mock-server"
+import { expectAppVisible } from "../utils/waits"
+
+const draftID = "draft_new_session_workspace_branch"
+const directory = "C:/OpenCode/WorkspaceBranch"
+const server = `http://${process.env.PLAYWRIGHT_SERVER_HOST ?? "127.0.0.1"}:${process.env.PLAYWRIGHT_SERVER_PORT ?? "4096"}`
+
+test("selects a base branch for a new workspace", async ({ page }) => {
+  await mockOpenCodeServer(page, {
+    directory,
+    project: {
+      id: "proj_new_session_workspace_branch",
+      worktree: directory,
+      vcs: "git",
+      name: "workspace-branch",
+      time: { created: 1700000000000, updated: 1700000000000 },
+      sandboxes: [],
+    },
+    provider: { all: [], connected: [], default: {} },
+    sessions: [],
+    pageMessages: () => ({ items: [] }),
+    vcsBranches: ["feature/api", "main", "origin/release"],
+  })
+  await page.addInitScript(
+    ({ directory, draftID, server }) => {
+      localStorage.setItem(
+        "opencode.global.dat:server",
+        JSON.stringify({
+          projects: { local: [{ worktree: directory, expanded: true }] },
+          lastProject: { local: directory },
+        }),
+      )
+      localStorage.setItem(
+        "opencode.window.browser.dat:tabs",
+        JSON.stringify([{ type: "draft", draftID, server, directory }]),
+      )
+    },
+    { directory, draftID, server },
+  )
+
+  await page.goto(`/new-session?draftId=${draftID}`)
+  await expectAppVisible(page.locator('[data-component="composer-editor"]'))
+  await page.getByRole("button", { name: "Local", exact: true }).click()
+  await page.getByRole("menuitem", { name: "New workspace", exact: true }).click()
+  await page.getByRole("button", { name: "from main", exact: true }).click()
+  await page.getByRole("menuitemradio", { name: "feature/api", exact: true }).click()
+
+  const selected = page.getByRole("button", { name: "from feature/api", exact: true })
+  await expect(selected).toBeVisible()
+  await selected.click()
+  await expect(page.getByRole("menuitemradio", { name: "feature/api", exact: true })).toBeChecked()
+})
@@ -20,45 +20,47 @@ const messages = [
  },
  { id: "msg_second", type: "user", text: "Second prompt", time: { created: 4 } },
 ] satisfies SessionMessageInfo[]
+const session = {
+  id: sessionID,
+  slug: "session-message-revert",
+  projectID,
+  directory,
+  title: "Session message revert",
+  agent: "build",
+  model: { id: "test", providerID: "opencode" },
+  version: "dev",
+  time: { created: 1, updated: 4 },
+}
+const fixture = {
+  directory,
+  project: {
+    id: projectID,
+    worktree: directory,
+    canonical: directory,
+    vcs: "git",
+    name: "session-message-revert",
+    time: { created: 1, updated: 1 },
+    sandboxes: [],
+  },
+  provider: {
+    all: [
+      {
+        id: "opencode",
+        name: "OpenCode",
+        models: { test: { id: "test", name: "Test", variants: {}, limit: { context: 200_000 } } },
+      },
+    ],
+    connected: ["opencode"],
+    default: { providerID: "opencode", modelID: "test" },
+  },
+  pageMessages: () => ({ items: messages }),
+}

 test("reverts directly to the selected user message", async ({ page }) => {
  const staged: { sessionID: string; messageID: string }[] = []
  await mockOpenCodeServer(page, {
-    directory,
-    project: {
-      id: projectID,
-      worktree: directory,
-      canonical: directory,
-      vcs: "git",
-      name: "session-message-revert",
-      time: { created: 1, updated: 1 },
-      sandboxes: [],
-    },
-    provider: {
-      all: [
-        {
-          id: "opencode",
-          name: "OpenCode",
-          models: { test: { id: "test", name: "Test", variants: {}, limit: { context: 200_000 } } },
-        },
-      ],
-      connected: ["opencode"],
-      default: { providerID: "opencode", modelID: "test" },
-    },
-    sessions: [
-      {
-        id: sessionID,
-        slug: "session-message-revert",
-        projectID,
-        directory,
-        title: "Session message revert",
-        agent: "build",
-        model: { id: "test", providerID: "opencode" },
-        version: "dev",
-        time: { created: 1, updated: 4 },
-      },
-    ],
-    pageMessages: () => ({ items: messages }),
+    ...fixture,
+    sessions: [session],
    onRevertStage: (input) => staged.push(input),
  })
  await page.goto(`/server/${base64Encode(server)}/session/${sessionID}`)
@@ -77,3 +79,19 @@ test("reverts directly to the selected user message", async ({ page }) => {
  await expect(page.getByRole("textbox", { name: "Prompt" })).toHaveText("Second prompt")
  expect(staged).toEqual([{ sessionID, messageID: "msg_second" }])
 })
+
+test("hides revert actions in a child session", async ({ page }) => {
+  await mockOpenCodeServer(page, {
+    ...fixture,
+    sessions: [
+      { ...session, id: "ses_parent", slug: "parent", title: "Parent session" },
+      { ...session, parentID: "ses_parent" },
+    ],
+  })
+  await page.goto(`/server/${base64Encode(server)}/session/${sessionID}`)
+  await expectSessionTitle(page, "Session message revert")
+
+  const message = page.locator('[data-message-id="msg_second"]')
+  await message.hover()
+  await expect(message.getByRole("button", { name: "Revert message" })).toHaveCount(0)
+})
@@ -0,0 +1,226 @@
+import { expect, test, type Page } from "@playwright/test"
+import type { OpenCodeEvent } from "@opencode-ai/client/promise"
+import { base64Encode } from "@opencode-ai/util/encode"
+import { mockOpenCodeServer } from "../utils/mock-server"
+import { expectAppVisible } from "../utils/waits"
+
+const directory = "C:/OpenCode/SessionQueueRegression"
+const projectID = "proj_session_queue_regression"
+const sessionID = "ses_session_queue_regression"
+const server = `http://${process.env.PLAYWRIGHT_SERVER_HOST ?? "127.0.0.1"}:${process.env.PLAYWRIGHT_SERVER_PORT ?? "4096"}`
+
+type InboxRow = {
+  id: string
+  sessionID: string
+  timeCreated: number
+  type: "user"
+  payload: { text: string; metadata?: Record<string, unknown> }
+  delivery: "steer" | "queue"
+}
+
+function createQueueMock(seed: string[]) {
+  const rows: InboxRow[] = seed.map((text, index) => ({
+    id: `inb_seed_${index + 1}`,
+    sessionID,
+    timeCreated: 1700000000000 + index,
+    type: "user",
+    payload: { text },
+    delivery: "queue",
+  }))
+  const events: OpenCodeEvent[] = []
+  const prompts: Record<string, unknown>[] = []
+  const changes: { inboxID: string; action: "cancel" | "steer" }[] = []
+  const log: string[] = []
+  let sequence = 0
+  const emit = (type: OpenCodeEvent["type"], data: OpenCodeEvent["data"]) => {
+    sequence += 1
+    events.push({
+      id: `evt_queue_${sequence}`,
+      type,
+      created: Date.now(),
+      durable: { aggregateID: sessionID, seq: sequence, version: 1 },
+      data,
+    } as OpenCodeEvent)
+  }
+  return {
+    rows,
+    prompts,
+    changes,
+    log,
+    events: () => events.splice(0),
+    onPrompt: (input: { sessionID: string; body: Record<string, unknown> }) => {
+      prompts.push(input.body)
+      log.push(`prompt:${String(input.body.delivery ?? "steer")}`)
+      const row: InboxRow = {
+        id: typeof input.body.id === "string" ? input.body.id : `inb_mock_${sequence}`,
+        sessionID: input.sessionID,
+        timeCreated: Date.now(),
+        type: "user",
+        payload: {
+          text: typeof input.body.text === "string" ? input.body.text : "",
+          ...(input.body.metadata === undefined ? {} : { metadata: input.body.metadata as Record<string, unknown> }),
+        },
+        delivery: input.body.delivery === "queue" ? "queue" : "steer",
+      }
+      rows.push(row)
+      emit("session.inbox.enqueued", {
+        sessionID: input.sessionID,
+        inboxID: row.id,
+        item: { type: "user", payload: row.payload, delivery: row.delivery },
+      })
+    },
+    onInboxChange: (input: { sessionID: string; inboxID: string; action: "cancel" | "steer" }) => {
+      changes.push({ inboxID: input.inboxID, action: input.action })
+      log.push(`${input.action}:${input.inboxID}`)
+      const index = rows.findIndex((row) => row.id === input.inboxID)
+      const row = rows[index]
+      if (!row) return
+      if (input.action === "cancel") {
+        rows.splice(index, 1)
+        emit("session.inbox.cancelled", { sessionID: input.sessionID, inboxID: input.inboxID })
+        return
+      }
+      row.delivery = "steer"
+      emit("session.inbox.delivery.changed", {
+        sessionID: input.sessionID,
+        inboxID: input.inboxID,
+        delivery: "steer",
+      })
+    },
+  }
+}
+
+async function openSession(page: Page, mock: ReturnType<typeof createQueueMock>, followUpBehavior?: "queue" | "steer") {
+  if (followUpBehavior) {
+    await page.addInitScript(
+      (behavior) => localStorage.setItem("settings.v3", JSON.stringify({ general: { followUpBehavior: behavior } })),
+      followUpBehavior,
+    )
+  }
+  await mockOpenCodeServer(page, {
+    directory,
+    project: {
+      id: projectID,
+      worktree: directory,
+      vcs: "git",
+      name: "session-queue-regression",
+      time: { created: 1700000000000, updated: 1700000000000 },
+      sandboxes: [],
+    },
+    provider: {
+      all: [
+        {
+          id: "opencode",
+          name: "OpenCode",
+          models: { "queue-model": { id: "queue-model", name: "Queue Model", limit: { context: 200_000 } } },
+        },
+      ],
+      connected: ["opencode"],
+      default: { providerID: "opencode", modelID: "queue-model" },
+    },
+    sessions: [
+      {
+        id: sessionID,
+        slug: "session-queue-regression",
+        projectID,
+        directory,
+        title: "Session queue regression",
+        version: "dev",
+        time: { created: 1700000000000, updated: 1700000000000 },
+      },
+    ],
+    pageMessages: () => ({ items: [] }),
+    sessionStatus: () => ({ [sessionID]: { type: "running" } }),
+    inbox: () => mock.rows.map((row) => ({ ...row, payload: { ...row.payload } })),
+    onPrompt: mock.onPrompt,
+    onInboxChange: mock.onInboxChange,
+    events: mock.events,
+  })
+  await page.goto(`/server/${base64Encode(server)}/session/${sessionID}`)
+  const composer = page.locator('[data-component="composer"]')
+  await expectAppVisible(composer)
+  return {
+    composer,
+    input: composer.locator('[data-component="composer-editor"]'),
+    rows: page.locator('[data-component="session-queue-row"]'),
+  }
+}
+
+test("follow-up preference controls Enter while Mod+Enter uses the alternate delivery", async ({ page }) => {
+  const mock = createQueueMock([])
+  const view = await openSession(page, mock, "queue")
+
+  await view.input.fill("queue this follow-up")
+  await expect(view.composer.locator('[data-action="composer-alternate-delivery"]')).toContainText("Steer")
+  await view.input.press("Enter")
+  await expect(view.rows.getByText("queue this follow-up", { exact: true })).toBeVisible()
+
+  await view.input.fill("steer this correction")
+  await view.input.press("ControlOrMeta+Enter")
+  await expect.poll(() => mock.prompts.map((prompt) => prompt.delivery)).toEqual(["queue", "steer"])
+  await expect(view.input).toHaveText("")
+})
+
+test("dragging reorders queued prompts", async ({ page }) => {
+  const mock = createQueueMock(["first queued prompt", "second queued prompt", "third queued prompt"])
+  const view = await openSession(page, mock)
+  await expect(view.rows).toHaveCount(3)
+
+  const first = view.rows.filter({ hasText: "first queued prompt" })
+  const third = view.rows.filter({ hasText: "third queued prompt" })
+  await first.getByRole("button", { name: "Reorder queued prompt" }).hover()
+  await page.mouse.down()
+  const target = await third.boundingBox()
+  if (!target) throw new Error("The target queue row is not visible")
+  await page.mouse.move(target.x + target.width / 2, target.y + target.height / 2, { steps: 10 })
+  await page.mouse.up()
+
+  await expect(view.rows.locator('[data-action="session-queue-edit"]')).toHaveText([
+    "second queued prompt",
+    "third queued prompt",
+    "first queued prompt",
+  ])
+  expect(mock.prompts.map((prompt) => prompt.text)).toEqual([
+    "second queued prompt",
+    "third queued prompt",
+    "first queued prompt",
+  ])
+  expect(mock.changes).toEqual([
+    { inboxID: "inb_seed_1", action: "cancel" },
+    { inboxID: "inb_seed_2", action: "cancel" },
+    { inboxID: "inb_seed_3", action: "cancel" },
+  ])
+})
+
+test("editing restores the existing draft and replaces only the original queue position", async ({ page }) => {
+  const mock = createQueueMock(["first queued prompt", "tighten the error copy", "third queued prompt"])
+  const view = await openSession(page, mock)
+  const original = view.rows.getByText("tighten the error copy", { exact: true })
+  await expect(original).toBeVisible()
+
+  await view.input.fill("my in-progress draft")
+  await original.click()
+  await expect(view.input).toHaveText("tighten the error copy")
+  await view.input.press("Escape")
+  await expect(view.input).toHaveText("my in-progress draft")
+
+  await original.click()
+  await expect(view.input).toHaveText("tighten the error copy")
+  await view.input.fill("tighten the error copy and add a retry hint")
+  await view.input.press("Enter")
+
+  await expect(view.rows.locator('[data-action="session-queue-edit"]')).toHaveText([
+    "first queued prompt",
+    "tighten the error copy and add a retry hint",
+    "third queued prompt",
+  ])
+  await expect(view.input).toHaveText("my in-progress draft")
+  expect(mock.prompts.map((prompt) => prompt.text)).toEqual([
+    "tighten the error copy and add a retry hint",
+    "tighten the error copy and add a retry hint",
+    "third queued prompt",
+  ])
+  expect(mock.prompts.every((prompt) => prompt.delivery === "queue" && prompt.resume === false)).toBe(true)
+  expect(mock.changes.map((change) => change.action)).toEqual(["cancel", "cancel", "cancel"])
+  expect(mock.log[0]).toBe("prompt:queue")
+})
@@ -139,7 +139,7 @@ test.describe("regression: session timeline local row state", () => {
    expect(siblingProbe).toEqual({
      fileMarker: "before",
      frameMarker: "before",
-      rowKey: `assistant-part:part:${assistantMessageID}:${editPartID}`,
+      rowKey: `assistant-part:file:part:${assistantMessageID}:${editPartID}`,
      rowMarker: "before",
      shadowRoots: 0,
      toolMarker: "before",
@@ -59,12 +59,12 @@ test("transitions a streaming shell from writing through command execution", asy
  await expect(tool.locator('[data-slot="collapsible-trigger"]')).toHaveCSS("height", "28px")
  await expect(tool.locator('[data-component="tool-trigger"]')).toHaveCSS("gap", "6px")
  await expect(title).toHaveCSS("font-size", "13px")
-  await expect(title).toHaveCSS("font-family", "Inter, sans-serif")
+  await expect(title).toHaveCSS("font-family", /^Inter,/)
  await expect(title).toHaveCSS("font-weight", "530")
  await expect(title).toHaveCSS("line-height", "16px")
  await expect(title).toHaveCSS("color", "rgb(22, 22, 22)")
  await expect(subtitle).toHaveCSS("font-size", "13px")
-  await expect(subtitle).toHaveCSS("font-family", "Inter, sans-serif")
+  await expect(subtitle).toHaveCSS("font-family", /^Inter,/)
  await expect(subtitle).toHaveCSS("font-weight", "440")
  await expect(subtitle).toHaveCSS("line-height", "16px")
  await expect(subtitle).toHaveCSS("color", "rgb(92, 92, 92)")
@@ -64,6 +64,76 @@ test("transitions shell and question through running error outcomes", async ({ p
  await expect(page.locator(`[data-timeline-part-id="${questionID}"]`)).toContainText(/dismissed/i)
 })

+test("preserves surviving grouped patch state when its first patch fails", async ({ page }) => {
+  const failed = "prt_grouped_patch_failed"
+  const surviving = "prt_grouped_patch_surviving"
+  const timeline = await setupTimeline(page, {
+    messages: [
+      userMessage(),
+      assistantMessage(
+        [
+          toolPart(failed, "patch", "running", { patchText: "Update src/failed.ts" }),
+          toolPart(
+            surviving,
+            "patch",
+            "running",
+            { patchText: "Update src/surviving.ts" },
+            {
+              metadata: {
+                files: [
+                  {
+                    file: "src/surviving.ts",
+                    status: "modified",
+                    patch: "@@ -1 +1 @@\n-export const value = 1\n+export const value = 2",
+                    additions: 1,
+                    deletions: 1,
+                  },
+                ],
+              },
+            },
+          ),
+        ],
+        { completed: false },
+      ),
+    ],
+  })
+
+  const group = page.locator(`[data-timeline-part-ids="${failed},${surviving}"]`)
+  const file = group.locator('[data-scope="apply-patch"] button')
+  await expect(file).toBeVisible()
+  await file.click()
+  await expect(file).toHaveAttribute("aria-expanded", "true")
+  await group.evaluate((element) => {
+    const row = element.closest<HTMLElement>("[data-timeline-key]")
+    if (row) row.dataset.groupIdentity = "preserved"
+  })
+
+  await timeline.send(
+    partUpdated(
+      toolPart(failed, "patch", "error", { patchText: "Update src/failed.ts" }, { error: "Patch failed visibly" }),
+    ),
+  )
+
+  const failedRow = page.locator("[data-timeline-key]", {
+    has: page.locator(`[data-timeline-part-id="${failed}"]`),
+  })
+  const survivingRow = page.locator("[data-timeline-key]", {
+    has: page.locator(`[data-timeline-part-id="${surviving}"]`),
+  })
+  await expect(failedRow).toHaveAttribute("data-timeline-key", /^assistant-part:part:/)
+  await expect(survivingRow).toHaveAttribute("data-timeline-key", /^assistant-part:file:/)
+  await expect(failedRow.getByText("Patch failed visibly")).toBeVisible()
+  await expect(survivingRow).toHaveAttribute("data-group-identity", "preserved")
+  await expect(survivingRow.locator('[data-scope="apply-patch"] button')).toHaveAttribute("aria-expanded", "true")
+  await expect
+    .poll(async () => {
+      const previous = await failedRow.boundingBox()
+      const next = await survivingRow.boundingBox()
+      return previous && next ? next.y - (previous.y + previous.height) : Number.NEGATIVE_INFINITY
+    })
+    .toBeGreaterThanOrEqual(-0.5)
+})
+
 test("labels all web search provider variants", async ({ page }) => {
  const parts = [
    toolPart(
@@ -68,6 +68,41 @@ test("keyboard navigation follows the visible tab order", async ({ page }) => {
  await expect(page).toHaveURL(new RegExp(`${hrefC.replace(/[.*+?^${}()|[\]\\]/g, "\\$&")}$`))
 })

+test("cramped tabs only show the close button for the active tab", async ({ page }) => {
+  await page.setViewportSize({ width: 360, height: 720 })
+  await mockServer(page)
+  await page.addInitScript(
+    ({ server, sessionA, sessionB, sessionC }) => {
+      localStorage.setItem(
+        "opencode.window.browser.dat:tabs",
+        JSON.stringify([
+          { type: "session", server, sessionId: sessionA },
+          { type: "session", server, sessionId: sessionB },
+          { type: "session", server, sessionId: sessionC },
+        ]),
+      )
+    },
+    { server, sessionA: sessionA.id, sessionB: sessionB.id, sessionC: sessionC.id },
+  )
+
+  const hrefA = `/server/${base64Encode(server)}/session/${sessionA.id}`
+  const hrefB = `/server/${base64Encode(server)}/session/${sessionB.id}`
+  await page.goto(hrefA)
+
+  const tabA = page.locator(`[data-titlebar-tab-slot]:has(a[href="${hrefA}"])`)
+  const tabB = page.locator(`[data-titlebar-tab-slot]:has(a[href="${hrefB}"])`)
+  await expect(tabA).toHaveAttribute("data-active", "true")
+  await expect(tabB).toBeVisible()
+  await expect(tabA.locator('[data-slot="tab-close"]')).toBeVisible()
+  await expect(tabB.locator('[data-slot="tab-close"]')).toBeHidden()
+
+  await tabB.locator(`a[href="${hrefB}"]`).click()
+
+  await expect(page).toHaveURL(new RegExp(`${hrefB.replace(/[.*+?^${}()|[\]\\]/g, "\\$&")}$`))
+  await expect(tabA.locator('[data-slot="tab-close"]')).toBeHidden()
+  await expect(tabB.locator('[data-slot="tab-close"]')).toBeVisible()
+})
+
 function session(id: string, title: string) {
  return {
    id,
@@ -126,6 +126,8 @@ test("routes typing to the composer unless the open terminal is focused", async

  const composer = page.locator('[data-component="composer-editor"]')
  const terminal = page.locator('[data-component="terminal"]')
+  await composer.click()
+  await expect(composer).toBeFocused()
  await page.keyboard.press("Control+Backquote")
  await expect(terminal).toBeVisible()
  await expect.poll(() => terminal.evaluate((element) => element.contains(document.activeElement))).toBe(true)
@@ -100,6 +100,7 @@ const Group = HttpApiGroup.make("mock")
  .add(HttpApiEndpoint.get("formRequests", "/api/form/request", { success: Json }))
  .add(HttpApiEndpoint.get("vcs", "/api/vcs", { success: Json }))
  .add(HttpApiEndpoint.get("vcsStatus", "/api/vcs/status", { success: Json }))
+  .add(HttpApiEndpoint.get("vcsBranches", "/api/vcs/branches", { success: Json }))
  .add(HttpApiEndpoint.get("vcsDiff", "/api/vcs/diff", { success: Json }))
  .add(HttpApiEndpoint.get("fsList", "/api/fs/list", { query: Query, success: Json }))
  .add(
@@ -174,6 +175,39 @@ const Group = HttpApiGroup.make("mock")
      success: Json,
    }),
  )
+  .add(
+    HttpApiEndpoint.post("sessionPrompt", "/api/session/:sessionID/prompt", {
+      params: SessionParams,
+      payload: JsonPayload,
+      success: Json,
+    }),
+  )
+  .add(
+    HttpApiEndpoint.post("sessionSwitchAgent", "/api/session/:sessionID/agent", {
+      params: SessionParams,
+      payload: JsonPayload,
+      success: NoContent,
+    }),
+  )
+  .add(
+    HttpApiEndpoint.post("sessionSwitchModel", "/api/session/:sessionID/model", {
+      params: SessionParams,
+      payload: JsonPayload,
+      success: NoContent,
+    }),
+  )
+  .add(
+    HttpApiEndpoint.delete("sessionInboxCancel", "/api/session/:sessionID/inbox/:inboxID", {
+      params: { ...SessionParams, inboxID: Schema.String },
+      success: NoContent,
+    }),
+  )
+  .add(
+    HttpApiEndpoint.post("sessionInboxSteer", "/api/session/:sessionID/inbox/:inboxID/steer", {
+      params: { ...SessionParams, inboxID: Schema.String },
+      success: NoContent,
+    }),
+  )
  .add(
    HttpApiEndpoint.get("sessionPermission", "/api/session/:sessionID/permission", {
      params: SessionParams,
@@ -21,6 +21,7 @@ export interface MockServerConfig {
    cursor?: string
  }
  vcsDiff?: unknown[]
+  vcsBranches?: string[]
  messageDelay?: number
  beforeMessagesResponse?: (input: { sessionID: string; before?: string }) => Promise<void>
  onMessages?: (input: { sessionID: string; before?: string; phase: "start" | "end" }) => void
@@ -35,6 +36,9 @@ export interface MockServerConfig {
  fileContent?: (path: string) => unknown | Promise<unknown>
  findFiles?: (input: { query: string; dirs?: string; limit?: number }) => unknown
  sessionStatus?: Record<string, unknown> | (() => Record<string, unknown>)
+  inbox?: unknown[] | (() => unknown[])
+  onPrompt?: (input: { sessionID: string; body: Record<string, unknown> }) => void
+  onInboxChange?: (input: { sessionID: string; inboxID: string; action: "cancel" | "steer" }) => void
 }

 type MockStreamWindow = Window & {
@@ -293,6 +297,7 @@ function mockHandlers(config: MockServerConfig, state: { cursors: Map<string, st
        vcs: () =>
          Effect.succeed({ location: location(config), data: { branch: { current: "main", default: "main" } } }),
        vcsStatus: () => Effect.succeed({ location: location(config), data: [] }),
+        vcsBranches: () => Effect.succeed({ location: location(config), data: config.vcsBranches ?? ["main"] }),
        vcsDiff: () => Effect.succeed({ location: location(config), data: config.vcsDiff ?? [] }),
        fsList: (ctx) =>
          Effect.promise(() => Promise.resolve(config.fileList?.(ctx.query.path ?? ""))).pipe(
@@ -397,7 +402,39 @@ function mockHandlers(config: MockServerConfig, state: { cursors: Map<string, st
        sessionFormReply: () => noContent,
        sessionFormCancel: () => noContent,
        sessionBackground: () => noContent,
-        sessionInbox: () => Effect.succeed({ data: [] }),
+        sessionInbox: () =>
+          Effect.sync(() => ({ data: typeof config.inbox === "function" ? config.inbox() : (config.inbox ?? []) })),
+        sessionPrompt: (ctx) =>
+          Effect.sync(() => {
+            const body = record(ctx.payload) ? ctx.payload : {}
+            config.onPrompt?.({ sessionID: ctx.params.sessionID, body })
+            return {
+              data: {
+                id: typeof body.id === "string" ? body.id : `inb_mock_${Date.now()}`,
+                sessionID: ctx.params.sessionID,
+                timeCreated: Date.now(),
+                type: "user",
+                payload: {
+                  text: typeof body.text === "string" ? body.text : "",
+                  ...(body.files === undefined ? {} : { files: body.files }),
+                  ...(body.agents === undefined ? {} : { agents: body.agents }),
+                  ...(body.skills === undefined ? {} : { skills: body.skills }),
+                  ...(body.metadata === undefined ? {} : { metadata: body.metadata }),
+                },
+                delivery: body.delivery === "queue" ? "queue" : "steer",
+              },
+            }
+          }),
+        sessionInboxCancel: (ctx) =>
+          Effect.sync(() =>
+            config.onInboxChange?.({ sessionID: ctx.params.sessionID, inboxID: ctx.params.inboxID, action: "cancel" }),
+          ).pipe(Effect.andThen(noContent)),
+        sessionInboxSteer: (ctx) =>
+          Effect.sync(() =>
+            config.onInboxChange?.({ sessionID: ctx.params.sessionID, inboxID: ctx.params.inboxID, action: "steer" }),
+          ).pipe(Effect.andThen(noContent)),
+        sessionSwitchAgent: () => noContent,
+        sessionSwitchModel: () => noContent,
        sessionPermission: (ctx) => {
          const permissions =
            typeof config.permissions === "function" ? config.permissions() : (config.permissions ?? [])
@@ -39,6 +39,26 @@ export type ComposerSelection = {
  variant?: string
 }

+export type ComposerDelivery = "steer" | "queue"
+
+// Contract between the composer and the session prompt queue. The session
+// owns the queue (pending inbox items); the composer only asks which delivery
+// a submit should use and delegates edit confirmation while a queued prompt
+// is loaded in the editor.
+export type ComposerQueue = {
+  count: Accessor<number>
+  // Delivery a plain submit uses right now.
+  delivery: Accessor<ComposerDelivery>
+  // Delivery offered on Mod+Enter and the toolbar hint button; undefined hides the hint.
+  alternate: Accessor<ComposerDelivery | undefined>
+  // Inbox ID of the queued prompt currently loaded in the composer for editing.
+  editing: Accessor<string | undefined>
+  confirmEdit: (delivery: ComposerDelivery) => void
+  cancelEdit: () => void
+  // Loads the first queued prompt into the composer. Returns false when the queue is empty.
+  editFirst: () => boolean
+}
+
 export type ComposerSession = {
  id: string
  directory: string
@@ -8,7 +8,7 @@ import { Tooltip } from "@opencode-ai/ui/tooltip"
 import { ComposerEditor } from "./editor/editor"
 import { ModelSelectorPopover } from "@/providers/models/select-dialog"
 import { DialogSelectModelUnpaid } from "@/providers/models/unpaid"
-import { useCommand } from "@/shell/commands/command"
+import { formatKeybind, useCommand } from "@/shell/commands/command"
 import { useLanguage } from "@/runtime/i18n/language"
 import type { ComposerModel } from "./model"

@@ -32,6 +32,7 @@ export function Composer(props: {
        modelControlsVisible={!props.model.model.loading}
        attachKeybind={command.keybindParts("file.attach")}
        attachShortcut={command.keybind("file.attach")}
+        alternateKeybind={[formatKeybind("mod", language.t), formatKeybind("enter", language.t)]}
        modelControl={
          <ComposerModelControl
            loading={props.model.model.loading}
@@ -1,7 +1,8 @@
-import { createEffect, createMemo, For, Show, type JSX } from "solid-js"
+import { createEffect, createMemo, createSignal, For, Show, type JSX } from "solid-js"
 import { FileIcon } from "@opencode-ai/ui/file-icon"
 import { Icon } from "@opencode-ai/ui/icon"
 import { IconButton } from "@opencode-ai/ui/icon-button"
+import { createAnimatedPresence } from "@/runtime/animated-presence"
 import { ProviderIcon } from "@opencode-ai/ui/provider-icon"
 import { useI18n } from "@opencode-ai/ui/context/i18n"
 import { Button } from "@opencode-ai/ui/button"
@@ -45,6 +46,7 @@ export type ComposerEditorProps = {
  modelControlsVisible?: boolean
  attachKeybind?: string[]
  attachShortcut?: string
+  alternateKeybind?: string[]
 }

 export function ComposerEditor(props: ComposerEditorProps) {
@@ -177,10 +179,15 @@ export function ComposerEditor(props: ComposerEditorProps) {
            }}
            onKeyDown={(event) => {
              if (props.controller.onKeyDown(event)) return
+              const mod = event.metaKey || event.ctrlKey
+              if (mod && event.key === "ArrowUp" && !event.shiftKey && !event.altKey) {
+                if (view.submit.queue?.editFirst()) event.preventDefault()
+                return
+              }
              if (event.key === "Enter" && !event.shiftKey && !event.isComposing) {
                event.preventDefault()
                if (event.repeat) return
-                props.controller.submit()
+                props.controller.submit(mod ? { alternate: true } : undefined)
              }
            }}
            onKeyUp={updateCursor}
@@ -248,6 +255,12 @@ export function ComposerEditor(props: ComposerEditorProps) {
              </Show>
            </Show>
          </div>
+          <Show when={state.mode === "normal"}>
+            <ComposerEditorAlternateDelivery
+              controller={props.controller}
+              keybind={props.alternateKeybind ?? ["Mod", "Enter"]}
+            />
+          </Show>
          <ComposerEditorSubmitButton
            mode={state.mode}
            stopping={view.submit.stopping()}
@@ -255,7 +268,7 @@ export function ComposerEditor(props: ComposerEditorProps) {
            accent={props.accentSubmit}
            sendLabel={i18n.t("ui.promptInput.send")}
            stopLabel={i18n.t("ui.promptInput.stop")}
-            onSubmit={props.controller.submit}
+            onSubmit={() => props.controller.submit()}
            onStop={props.controller.stop}
          />
        </div>
@@ -691,6 +704,49 @@ export function ComposerEditorPopover(props: {
  )
 }

+// "Steer ⌘⏎" / "Queue ⌘⏎" hint next to the submit button: submits with the
+// delivery opposite to what plain Enter does. Visible only while the queue
+// exposes an alternate (turn running and composer holding a value), so it
+// disappears on its own when the current turn ends.
+function ComposerEditorAlternateDelivery(props: { controller: ComposerEditorModel; keybind: string[] }) {
+  const i18n = useI18n()
+  const view = props.controller.view
+  const action = createMemo(() => {
+    const queue = view.submit.queue
+    if (!queue || !props.controller.canSubmit()) return undefined
+    if (queue.editing()) return "steer" as const
+    return queue.alternate()
+  })
+  const [button, setButton] = createSignal<HTMLButtonElement>()
+  const presence = createAnimatedPresence(action, () => button() ?? null)
+  return (
+    <Show when={presence.present() && presence.value()} keyed>
+      {(delivery) => (
+        <Tooltip placement="top" inactive={delivery !== "steer"} value={i18n.t("ui.promptInput.steerHint")}>
+          <Button
+            ref={setButton}
+            data-action="composer-alternate-delivery"
+            type="button"
+            variant="ghost-muted"
+            size="small"
+            class="me-3 gap-1.5 px-1.5 text-v2-text-text-muted ![font-weight:530] duration-150 motion-reduce:animate-none"
+            classList={{
+              "animate-in fade-in": presence.animate() && presence.show(),
+              "animate-out fade-out fill-mode-forwards": presence.animate() && !presence.show(),
+            }}
+            onClick={() => props.controller.submit({ alternate: true })}
+          >
+            {delivery === "steer" ? i18n.t("ui.promptInput.steer") : i18n.t("ui.promptInput.queue")}
+            <span class="hidden sm:block">
+              <Keybind keys={props.keybind} variant="neutral" />
+            </span>
+          </Button>
+        </Tooltip>
+      )}
+    </Show>
+  )
+}
+
 export function ComposerEditorSubmitButton(props: {
  mode: ComposerMode
  stopping: boolean
@@ -19,6 +19,7 @@ import {
  type ComposerInteractionEvent,
 } from "../suggestions/machine"
 import { clonePrompt, promptLength } from "../prompt-parts"
+import type { ComposerQueue } from "../adapter"

 export type ComposerSelectControl = {
  options: Accessor<ComposerOption[]>
@@ -37,7 +38,8 @@ export type ComposerEditorView = {
  submit: {
    stopping: Accessor<boolean>
    working?: Accessor<boolean>
-    onSubmit: () => void
+    queue?: ComposerQueue
+    onSubmit: (options?: { alternate?: boolean }) => void
    onStop: () => void
  }
  shell?: {
@@ -212,6 +214,11 @@ export function createComposerEditor(input: {
      )
    }
    if (handled) return true
+    if (event.key === "Escape" && input.view.submit.queue?.editing()) {
+      event.preventDefault()
+      input.view.submit.queue.cancelEdit()
+      return true
+    }
    const stop =
      input.view.submit.working?.() &&
      ((event.ctrlKey && !event.metaKey && !event.altKey && !event.shiftKey && event.key.toLowerCase() === "g") ||
@@ -354,8 +361,8 @@ export function createComposerEditor(input: {
    openShell() {
      dispatch({ type: "mode.shell" })
    },
-    submit() {
-      input.view.submit.onSubmit()
+    submit(options?: { alternate?: boolean }) {
+      input.view.submit.onSubmit(options)
      dispatch({ type: "popover.close" })
    },
    stop() {
@@ -16,7 +16,7 @@ import { createSessionTabs } from "@/session/helpers"
 import { showToast } from "@/shell/notifications/toast"
 import { formatServerError } from "@/runtime/server/errors"
 import { Skill } from "@opencode-ai/schema/skill"
-import type { ComposerAdapter, ComposerControls } from "./adapter"
+import type { ComposerAdapter, ComposerControls, ComposerQueue } from "./adapter"
 import type { ImageAttachmentPart } from "./state"
 import type { PromptHistoryComment } from "./history/entry"
 import { createComposerHistory } from "./history/store"
@@ -27,7 +27,7 @@ export type ComposerModel = ComposerEditorModel & {
  readonly model: ComposerControls["model"]
 }

-export function createComposerModel(adapter: ComposerAdapter): ComposerModel {
+export function createComposerModel(adapter: ComposerAdapter, options?: { queue?: ComposerQueue }): ComposerModel {
  const sdk = useWorkspaceLocation()
  const data = useData()
  const files = useFile()
@@ -80,7 +80,11 @@ export function createComposerModel(adapter: ComposerAdapter): ComposerModel {
  })
  const stopping = createMemo(() => adapter.working() && blank())
  const placeholder = () =>
-    composerPlaceholder(mode(), (key, params) => language.t(key as Parameters<typeof language.t>[0], params as never))
+    composerPlaceholder(
+      mode(),
+      (key, params) => language.t(key as Parameters<typeof language.t>[0], params as never),
+      adapter.working() || (options?.queue?.count() ?? 0) > 0,
+    )

  const historyComments = () => {
    const byID = new Map(comments.all().map((item) => [`${item.file}\n${item.id}`, item] as const))
@@ -253,6 +257,11 @@ export function createComposerModel(adapter: ComposerAdapter): ComposerModel {
    resetHistory: () => controller.resetHistory(),
    setMode: (next) => controller.dispatch({ type: next === "shell" ? "mode.shell" : "mode.normal" }),
    closePopover: () => controller.dispatch({ type: "popover.close" }),
+    delivery: (alternate) => {
+      const queue = options?.queue
+      if (!queue) return "steer"
+      return (alternate ? queue.alternate() : queue.delivery()) ?? "steer"
+    },
    notify: {
      missingSelection: () =>
        showToast({
@@ -360,7 +369,18 @@ export function createComposerModel(adapter: ComposerAdapter): ComposerModel {
      submit: {
        stopping,
        working: adapter.working,
-        onSubmit: () => void submission.submit(new Event("submit")),
+        queue: options?.queue,
+        onSubmit: (submitOptions) => {
+          const queue = options?.queue
+          // Confirming an edit re-admits the queued prompt instead of sending
+          // the composer value as a new prompt. Enter keeps it queued in
+          // place; the alternate action sends it as a steer.
+          if (queue?.editing()) {
+            queue.confirmEdit(submitOptions?.alternate ? "steer" : "queue")
+            return
+          }
+          void submission.submit(new Event("submit"), submitOptions)
+        },
        onStop: () => void submission.stop(),
      },
    },
@@ -12,4 +12,12 @@ describe("Composer placeholder", () => {
  test("uses the command and context hint in normal mode", () => {
    expect(composerPlaceholder("normal", t)).toBe("ui.promptInput.placeholder.normal/@")
  })
+
+  test("uses the follow-up copy while a turn runs or prompts are queued", () => {
+    expect(composerPlaceholder("normal", t, true)).toBe("ui.promptInput.placeholder.followUp/@")
+  })
+
+  test("keeps the shell placeholder while a turn runs", () => {
+    expect(composerPlaceholder("shell", t, true)).toBe("prompt.placeholder.shell:git status")
+  })
 })
@@ -1,7 +1,9 @@
 export function composerPlaceholder(
  mode: "normal" | "shell",
  t: (key: string, params?: Record<string, string>) => string,
+  followUp?: boolean,
 ) {
  if (mode === "shell") return t("prompt.placeholder.shell", { example: "git status" })
+  if (followUp) return t("ui.promptInput.placeholder.followUp", { slash: "/", at: "@" })
  return t("ui.promptInput.placeholder.normal", { slash: "/", at: "@" })
 }
@@ -5,7 +5,7 @@ import type { Accessor } from "solid-js"
 import type { PromptHistoryComment } from "./history/entry"
 import type { ImageAttachmentPart, Prompt } from "./state"
 import { clonePrompt, promptLength } from "./prompt-parts"
-import type { ComposerAdapter, ComposerSelection, ComposerSession } from "./adapter"
+import type { ComposerAdapter, ComposerDelivery, ComposerSelection, ComposerSession } from "./adapter"
 import { createComposerSubmission } from "./submission-state"
 import { buildPromptRequest } from "./request"
 import { setCursorPosition } from "./editor/dom"
@@ -21,7 +21,7 @@ type ComposerSubmission = {
  text: string
  images: ImageAttachmentPart[]
  selection: ComposerSelection
-  delivery: "steer"
+  delivery: ComposerDelivery
 }

 type ComposerSubmitInput = {
@@ -33,6 +33,7 @@ type ComposerSubmitInput = {
  resetHistory: () => void
  setMode: (mode: "normal" | "shell") => void
  closePopover: () => void
+  delivery?: (alternate: boolean) => ComposerDelivery
  notify: {
    missingSelection: () => void
    failed: (kind: "shell" | "command" | "prompt", error: unknown) => void
@@ -45,7 +46,7 @@ type ComposerSubmitInput = {
 }

 export function createComposerSubmit(input: ComposerSubmitInput) {
-  const submit = async (event: globalThis.Event) => {
+  const submit = async (event: globalThis.Event, options?: { alternate?: boolean }) => {
    event.preventDefault()

    const submission = createComposerSubmission({
@@ -56,7 +57,7 @@ export function createComposerSubmit(input: ComposerSubmitInput) {
        selection: item.selection ? { ...item.selection } : undefined,
      })),
    })
-    const value = readSubmission(input, submission.prompt, submission.context)
+    const value = readSubmission(input, submission.prompt, submission.context, options?.alternate ?? false)
    if (!value) {
      if (input.adapter.working() && input.adapter.kind === "active-session") void input.adapter.interrupt()
      return
@@ -113,7 +114,10 @@ export function createComposerSubmit(input: ComposerSubmitInput) {

      if (command) {
        clearSubmission(input, submission)
-        void sendCommand(session, value, command).catch((error) =>
+        // Commands always steer: the server applies a command's configured
+        // agent and model immediately at admission, so queueing one would
+        // reconfigure the turn it is supposed to wait behind.
+        void sendCommand(session, { ...value, delivery: "steer" }, command).catch((error) =>
          failSubmission(input, session, "command", error, restore, value.id),
        )
        return
@@ -157,6 +161,7 @@ function readSubmission(
  input: ComposerSubmitInput,
  prompt: Prompt,
  context: ComposerSubmission["context"],
+  alternate: boolean,
 ): ComposerSubmission | undefined {
  const text = prompt.map((part) => ("content" in part ? part.content : "")).join("")
  const mode = input.mode()
@@ -195,7 +200,7 @@ function readSubmission(
      model: { modelID: model.id, providerID: model.provider.id },
      variant,
    },
-    delivery: "steer",
+    delivery: input.delivery?.(alternate) ?? "steer",
  }
 }

@@ -274,15 +279,8 @@ async function sendCommand(
  const request = await buildSubmissionRequest(session, value)
  await session.api.command({
    sessionID: session.id,
-    id: value.id,
    command: command.command,
-    arguments: command.arguments,
-    agent: value.selection.agent,
-    model: {
-      id: value.selection.model.modelID,
-      providerID: value.selection.model.providerID,
-      variant: value.selection.variant,
-    },
+    text: command.arguments,
    files: request.files.map((file) => ({ uri: file.uri, name: file.name, mention: file.mention })),
    agents: request.agents,
    skills: request.skills,
@@ -292,23 +290,30 @@ async function sendCommand(

 async function sendPrompt(session: ComposerSession, value: ComposerSubmission) {
  const request = await buildSubmissionRequest(session, value)
-  const current = session.current()
-  if (current?.agent !== value.selection.agent) {
-    await session.api.switchAgent({ sessionID: session.id, agent: value.selection.agent })
-  }
-  if (
-    current?.model?.providerID !== value.selection.model.providerID ||
-    current.model.id !== value.selection.model.modelID ||
-    (current.model.variant ?? "default") !== (value.selection.variant ?? "default")
-  ) {
-    await session.api.switchModel({
-      sessionID: session.id,
-      model: {
-        id: value.selection.model.modelID,
-        providerID: value.selection.model.providerID,
-        variant: value.selection.variant,
-      },
-    })
+  // Switching agent or model reconfigures the session immediately, and with it
+  // the remainder of a running turn. A steer targets that turn, so its
+  // selection applies now; a queued follow-up must not reconfigure the turn it
+  // waits behind, so it runs with the session selection at delivery time (the
+  // intended selection stays recorded in its metadata).
+  if (value.delivery === "steer") {
+    const current = session.current()
+    if (current?.agent !== value.selection.agent) {
+      await session.api.switchAgent({ sessionID: session.id, agent: value.selection.agent })
+    }
+    if (
+      current?.model?.providerID !== value.selection.model.providerID ||
+      current.model.id !== value.selection.model.modelID ||
+      (current.model.variant ?? "default") !== (value.selection.variant ?? "default")
+    ) {
+      await session.api.switchModel({
+        sessionID: session.id,
+        model: {
+          id: value.selection.model.modelID,
+          providerID: value.selection.model.providerID,
+          variant: value.selection.variant,
+        },
+      })
+    }
  }

  const admission = {
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Brendonovich	019da0f2bf	fix(app): preserve timeline row heights when moving sessions	2026-08-26 03:22:44 +00:00
Aiden Cline	5add6a8e19	fix(ai): preserve provider-defined responses item ids (#45094 )	2026-08-25 22:17:16 -05:00
Aiden Cline	f3c390b89e	feat(ai): add native DeepInfra provider (#45108 )	2026-08-25 22:15:52 -05:00
opencode-agent[bot]andBrendonovich	185c3e5136	fix(app): show project logos throughout settings (#45134 ) Co-authored-by: Brendonovich <14191578+Brendonovich@users.noreply.github.com>	2026-08-26 02:43:53 +00:00
Kit Langton	cb863b8ea5	fix(tui): dismiss the active interaction with ctrl-c (#45111 )	2026-08-25 22:38:50 -04:00
Kit Langton	695c043e6b	feat(core): refresh unpinned plugins on startup (#45118 )	2026-08-25 22:32:05 -04:00
opencode-agent[bot]	d53456da3b	chore: update nix node_modules hashes	2026-08-26 02:16:23 +00:00
opencode-agent[bot]andBrendonovich	64e930628d	fix(ci): compare affected packages against actual PR base (#45130 ) Co-authored-by: Brendonovich <Brendonovich@users.noreply.github.com>	2026-08-26 10:13:53 +08:00
Kit Langton	3c73ce1dc7	fix(core): materialize mentioned skills on prompts (#44840 )	2026-08-25 22:04:09 -04:00
Kit Langton	4b71ae6a0d	feat(core): support git plugin packages (#45110 )	2026-08-25 22:02:49 -04:00
opencode-agent[bot]andBrendonovich	e211b6f30e	test: run only affected unit suites (#45034 ) Co-authored-by: Brendonovich <14191578+Brendonovich@users.noreply.github.com>	2026-08-26 09:51:44 +08:00
Kit Langton	ab4621e437	fix(sdk): keep packaged Effect runtime coherent (#45122 )	2026-08-26 01:51:36 +00:00
Kit Langton	97ba700fac	fix(tui): preserve prompt metadata visibility across sessions (#45116 )	2026-08-25 21:37:58 -04:00
Luke Parker	03fb5c6c67	fix(app): prevent clipped virtual timeline rows (#45115 )	2026-08-26 01:33:32 +00:00
Luke Parker	5a54eb4afc	fix(app): stream running shell tool output (#45106 )	2026-08-26 00:58:15 +00:00
Kit Langton	c4dcf72e13	fix(tui): detect clipped transcript bottom (#45100 )	2026-08-25 20:46:52 -04:00
opencode-agent[bot]	27e0de6b23	chore: update nix node_modules hashes	2026-08-26 00:41:39 +00:00
Kit Langton	73d7b1d4c1	fix(tui): preserve interrupted Mermaid diagrams (#45102 )	2026-08-26 00:25:00 +00:00
Kit Langton	690ad8e8bd	test(core): isolate host configuration and credentials (#44845 )	2026-08-25 20:22:21 -04:00
Aiden Cline	6c97be6974	feat(ai): add native Cerebras and Together AI providers (#45098 )	2026-08-25 19:20:47 -05:00
Dax Raad	6cd1ffac50	chore: synchronize bun lockfile	2026-08-25 19:18:17 -04:00
Aiden Cline	f08c234890	fix(ai): ignore SSE retry directives without ending streams (#45093 )	2026-08-25 18:17:45 -05:00
Kit Langton	7f2b052db6	refactor(core): remove unused Drizzle migration framework	2026-08-25 19:16:26 -04:00
Dax Raad	297a3328c6	feat(tui): group MCP integrations in connection dialog	2026-08-25 19:14:12 -04:00
Kit Langton	7f9e5e91ab	feat(tui): add experimental session preview tabs (#45021 )	2026-08-25 18:51:36 -04:00
Aiden Cline	3726e3254d	fix(ai): enable Vertex Anthropic prompt caching (#45088 )	2026-08-25 17:44:32 -05:00
Aiden Cline	24605d048f	fix(ai): send responses instructions at top level (#45085 )	2026-08-25 17:34:40 -05:00
Aiden Cline	0a84625618	fix(ai): accept responses calls without item ids (#45081 )	2026-08-25 17:29:57 -05:00
Major Hayden	2e7f06a155	fix(ai): preserve Vertex Anthropic tool continuations (#43498 ) Signed-off-by: Major Hayden <major@mhtx.net>	2026-08-25 17:22:51 -05:00
Aiden Cline	c2a3b813a0	fix(ai): require reasoning fields for deepseek assistants (#45075 )	2026-08-25 17:13:17 -05:00
opencode-agent[bot]andrekram1-node	b79cad5ec8	fix(core): respect automatic compaction opt-out on overflow (#45036 ) Co-authored-by: rekram1-node <rekram1-node@users.noreply.github.com>	2026-08-25 16:58:39 -05:00
opencode-agent[bot]andneriousy	61dd296161	fix(www): generate CLI schema without tracking it (#45080 ) Co-authored-by: neriousy <34747899+neriousy@users.noreply.github.com>	2026-08-25 23:54:07 +02:00
Filip	66a790c624	chore(www): regenerate documentation artifacts (#45077 )	2026-08-25 23:33:01 +02:00
Aiden Cline	0ae3aac317	fix(ai): replay responses history independently of storage (#45050 )	2026-08-25 16:28:22 -05:00
Dax	eb1ac54d73	feat(tui): support multiple integration accounts (#45072 )	2026-08-25 17:03:16 -04:00
opencode-agent[bot]andrekram1-node	d2100a51f1	feat(cli): restore debug paths command (#45063 ) Co-authored-by: rekram1-node <rekram1-node@users.noreply.github.com>	2026-08-25 23:01:43 +02:00
opencode-agent[bot]andneriousy	d4803ffe38	feat(cli): publish generated CLI configuration schema (#45070 ) Co-authored-by: neriousy <34747899+neriousy@users.noreply.github.com>	2026-08-25 22:49:45 +02:00
Kit Langton	db7837814c	fix(project): refresh renamed projects across live clients (#45043 )	2026-08-25 20:13:10 +00:00
opencode-agent[bot]andneriousy	543a4f4912	fix(ci): preserve required e2e matrix checks (#45065 ) Co-authored-by: neriousy <34747899+neriousy@users.noreply.github.com>	2026-08-25 22:11:43 +02:00
opencode-agent[bot]andneriousy	d562f6df1e	fix(core): recover legacy database migration history (#45062 ) Co-authored-by: neriousy <34747899+neriousy@users.noreply.github.com>	2026-08-25 22:02:26 +02:00
Aiden Cline	263a442a6c	fix(ai): normalize chat tool call ids (#45056 )	2026-08-25 14:39:00 -05:00
Filip	82947af8e3	fix(core): ignore SSE comment heartbeats (#43626 )	2026-08-25 21:34:08 +02:00
Filip	e1afdaac52	fix(core): fall back on oversized websocket requests v2 (#43100 )	2026-08-25 21:27:08 +02:00
Aiden Cline	a5829431b0	fix(ai): bridge tool results for mistral family models (#45051 )	2026-08-25 13:40:35 -05:00
Aiden Cline	1ca82d154c	fix(ai): omit empty chat assistant messages (#45046 )	2026-08-25 12:56:24 -05:00
James Long	bc1f67e518	feat(cli): embed persistent PTY service binaries (#44970 )	2026-08-25 13:19:48 -04:00
Kit Langton	1f45962c84	fix(tui): support inline session rename and title regeneration (#45023 )	2026-08-25 13:10:22 -04:00
Kit Langton	88242e21a8	fix(tui): stop completed exploration groups from spinning	2026-08-25 13:09:03 -04:00
opencode-agent[bot]R44VC0RP nexxeln thdxr	0fd719067d	feat(plugin): add permission review hooks (#45003 ) Co-authored-by: R44VC0RP <R44VC0RP@users.noreply.github.com> Co-authored-by: nexxeln <nexxeln@users.noreply.github.com> Co-authored-by: thdxr <thdxr@users.noreply.github.com>	2026-08-25 16:57:17 +00:00
Aiden Cline	c94a4913c0	fix(ai): preserve unencrypted reasoning history (#45032 )	2026-08-25 11:44:44 -05:00
opencode-agent[bot]andBrendonovich	6c32ba81e2	test: skip unaffected app e2e tests (#45030 ) Co-authored-by: Brendonovich <14191578+Brendonovich@users.noreply.github.com>	2026-08-26 00:33:14 +08:00
Kit Langton	938a82226a	test(tui): synchronize tab close frame before pointer moves (#45017 )	2026-08-25 12:30:20 -04:00
Aiden Cline	c46b76b58e	fix(ai): route response events by output index (#45013 )	2026-08-25 11:27:40 -05:00
opencode-agent[bot]	fbb3730fdd	chore: update nix node_modules hashes	2026-08-25 16:26:05 +00:00
opencode-agent[bot]andBrendonovich	96cff7bb7a	fix(app): clarify context usage label (#45012 ) Co-authored-by: Brendonovich <14191578+Brendonovich@users.noreply.github.com>	2026-08-26 00:20:00 +08:00
Kit Langton	578f8d637a	fix(server): pin shared Effect platform runtime (#45016 )	2026-08-25 12:19:24 -04:00
Kit Langton	e5308a988f	feat(merman): refine diagram styling (#44815 )	2026-08-25 16:07:12 +00:00
opencode-agent[bot]andnexxeln	d177f29dba	fix(tui): keep autocomplete selection visible (#44983 ) Co-authored-by: nexxeln <95541290+nexxeln@users.noreply.github.com>	2026-08-25 21:30:53 +05:30
Aiden Cline	7f51da509b	fix(ai): preserve terminal reasoning metadata (#44997 )	2026-08-25 10:37:49 -05:00
James Long	004b647311	feat(server): add persistent PTY daemon API (#44969 )	2026-08-25 11:31:47 -04:00
Aiden Cline	cce86ac166	fix(ai): fail unknown Bedrock stream exceptions (#45004 )	2026-08-25 10:08:21 -05:00
Shoubhit Dash	b71291c05a	refactor(core): move mercurial vcs into internal plugin (#44993 )	2026-08-25 20:35:21 +05:30
Shoubhit Dash	0ab2d783e8	refactor(core): move git vcs into internal plugin (#44992 )	2026-08-25 20:26:43 +05:30
Aiden Cline	3deac93d27	fix(ai): sanitize outbound provider request surrogates (#44880 )	2026-08-25 09:53:47 -05:00
opencode-agent[bot]Brendonovich Brendonovich	ed582d1bdb	feat(core): run worktree setup scripts (#44455 ) Co-authored-by: Brendonovich <14191578+Brendonovich@users.noreply.github.com> Co-authored-by: Brendonovich <Brendonovich@users.noreply.github.com>	2026-08-25 22:45:01 +08:00
opencode-agent[bot]andBrendonovich	db5a10dad1	fix(app): hide close buttons on cramped inactive tabs (#44995 ) Co-authored-by: Brendonovich <14191578+Brendonovich@users.noreply.github.com>	2026-08-25 22:35:41 +08:00
Aiden Cline	ff59a22ff4	fix(ai): reconcile completed response output items (#44893 )	2026-08-25 09:32:42 -05:00
Shoubhit Dash	9c1787617c	feat(plugin): add vcs provider api (#44979 )	2026-08-25 19:37:06 +05:30
opencode-agent[bot]andBrendonovich	7601ab9fc4	fix(desktop): hide sessions during removal (#44913 ) Co-authored-by: Brendonovich <14191578+Brendonovich@users.noreply.github.com>	2026-08-25 21:09:20 +08:00
opencode-agent[bot]Brendonovich Brendan Allan	4fb8a6038a	feat(app): select worktree base branch (#44906 ) Co-authored-by: Brendonovich <14191578+Brendonovich@users.noreply.github.com> Co-authored-by: Brendan Allan <git@brendonovich.dev>	2026-08-25 21:05:52 +08:00
opencode-agent[bot]andrekram1-node	5c25c38961	fix(tui): paste into active custom answers (#44849 ) Co-authored-by: rekram1-node <rekram1-node@users.noreply.github.com>	2026-08-25 18:19:28 +05:30
Brendan Allan	e3aa13c7d0	fix(app): show pending states when switching servers (#44947 )	2026-08-25 20:34:53 +08:00
opencode-agent[bot]andBrendan	5963a30621	fix(ui): restore stacked dialog focus (#44941 ) Co-authored-by: Brendan <14191578+Brendonovich@users.noreply.github.com>	2026-08-25 19:10:09 +08:00
opencode-agent[bot]	bcd1769521	chore: update nix node_modules hashes	2026-08-25 09:05:45 +00:00
Brendan Allan	691cb456ae	feat(desktop): add development component picker (#44922 )	2026-08-25 16:48:22 +08:00
opencode-agent[bot]andBrendonovich	190f189fbe	fix(app): route notification clicks through tabs (#44897 ) Co-authored-by: Brendonovich <14191578+Brendonovich@users.noreply.github.com>	2026-08-25 14:11:59 +08:00
Major Hayden	8c126e98da	fix(ci): check PR body for linked issue on non-default branches (#43964 ) Signed-off-by: Major Hayden <major@mhtx.net>	2026-08-25 00:38:02 -05:00
Aiden Cline	ce8a489aaa	fix(ai): respect prompt cache opt-out (#44891 )	2026-08-25 00:36:18 -05:00
Brendan Allan	1f7ae3f638	fix(app): animate composer delivery controls (#44886 )	2026-08-25 13:34:53 +08:00
Luke Parker	5ad0f0dc5a	fix(app): prevent timeline row identity collisions (#44878 )	2026-08-25 15:33:45 +10:00
opencode-agent[bot]rekram1-node Aiden Cline	e589969398	fix(core): resolve compatible shells for commands (#44485 ) Co-authored-by: rekram1-node <rekram1-node@users.noreply.github.com> Co-authored-by: Aiden Cline <aidenpcline@gmail.com>	2026-08-25 00:24:08 -05:00
Aiden Cline	42867d3bbc	fix(ai): recover incomplete streamed tool arguments (#44875 )	2026-08-25 00:22:03 -05:00
Brendan Allan	d4cdb99e4c	fix(desktop): suppress resize observer loop warnings (#44883 )	2026-08-25 13:17:29 +08:00
Aiden Cline	0a78b11222	fix(ai): ignore unknown Anthropic stream variants (#44817 )	2026-08-25 00:03:08 -05:00
Aiden Cline	28c1806950	fix(core): route Copilot fallback models through AI SDK (#44882 )	2026-08-25 00:01:42 -05:00
opencode-agent[bot]andrekram1-node	683f5fdee0	fix(tui): inherit model for new sessions (#44879 ) Co-authored-by: rekram1-node <rekram1-node@users.noreply.github.com>	2026-08-24 23:58:09 -05:00
Dax Raad	e9b5e055f5	fix(docs): refine header links and copy feedback	2026-08-25 00:53:10 -04:00
Aiden Cline	f327adb0f2	fix(core): support Zod tool schemas (#44861 )	2026-08-24 23:50:01 -05:00
Dax Raad	442bc92a21	feat(docs): add markdown copy button to page headings	2026-08-25 00:48:39 -04:00
Dax Raad	f2ff93a5b7	fix(www): use official favicon	2026-08-25 00:44:42 -04:00
Aiden Cline	1b30098e8d	fix(ai): default responses to encrypted reasoning (#44863 )	2026-08-24 23:43:04 -05:00
Dax Raad	1144ef6c5d	feat(www): serve cached markdown documentation	2026-08-25 00:34:34 -04:00
Luke Parker	d6deb62379	fix(app): default typography to Inter and IBM Plex Mono (#44876 )	2026-08-25 14:30:21 +10:00
Aiden Cline	895eff09b0	fix(ai): classify missing chat finish reasons as incomplete (#44864 )	2026-08-24 22:37:47 -05:00
Aiden Cline	d0252f7179	fix(ai): handle completed response function arguments (#44862 )	2026-08-24 22:36:43 -05:00
Aiden Cline	d78c13fce3	refactor(core): normalize tool input errors (#44818 )	2026-08-24 22:23:25 -05:00
Aiden Cline	6bb5200464	fix(ai): enforce chat finish reasons (#44743 )	2026-08-24 22:10:21 -05:00
Luke Parker	a02a2f5799	feat(app): queue and steer follow-up prompts (#44683 )	2026-08-25 13:01:04 +10:00
Aiden Cline	63c23c98de	feat(ai): parse partial tool input (#44830 )	2026-08-24 21:56:53 -05:00
Dax Raad	9a90b94921	docs(sdk): align Effect and Cloudflare customization guides	2026-08-24 22:52:30 -04:00
Dax Raad	f03418afde	docs: improve build and SDK customization guides	2026-08-24 22:52:30 -04:00
opencode-agent[bot]thdxr Dax	19d0009891	docs: clarify prompt data handling (#44854 ) Co-authored-by: thdxr <826656+thdxr@users.noreply.github.com> Co-authored-by: Dax <mail@thdxr.com>	2026-08-24 22:29:36 -04:00
Luke Parker	9fc85ae9db	fix(cli): honor notification-only automatic updates (#44820 )	2026-08-24 19:34:15 -05:00
Dax	2e4b2c82f4	fix(tui): resolve plugin SDK imports at runtime (#44822 )	2026-08-24 20:29:38 -04:00
Aiden Cline	a9042a58ab	feat(ai): add partial JSON parser (#44792 )	2026-08-24 18:59:06 -05:00
Aiden Cline	244ec6c8f7	fix(core): validate JSON schema tool input (#44789 )	2026-08-24 18:56:57 -05:00
Dax Raad	e28471e0ad	docs: rename build sidebar intro	2026-08-24 19:04:37 -04:00
Dax Raad	778d5b675c	docs: split client and sdk guides	2026-08-24 19:03:42 -04:00
Filip	127113188e	docs(github): correct action token configuration (#44795 )	2026-08-25 00:29:41 +02:00
Dax Raad	ce16b7cc12	docs: simplify plugin guide routes	2026-08-24 18:28:23 -04:00
Aiden Cline	eda6d774bf	fix(ai): ignore unknown Gemini response parts (#44745 )	2026-08-24 17:23:35 -05:00
Dax Raad	e11b3d08b6	docs: clarify plugin skill guidance	2026-08-24 18:10:40 -04:00
Dax Raad	0cdd711abf	docs: expand plugin guides	2026-08-24 18:10:40 -04:00
Kit Langton	22c63833d2	feat(workspace): support caller-supplied IDs (#44771 )	2026-08-24 18:08:42 -04:00
Kit Langton	42d160f4a0	test: stabilize asynchronous integration checks (#44787 )	2026-08-24 18:02:13 -04:00
opencode-agent[bot]andrekram1-node	8be467de8d	fix(core): respect disabled Plan agent config (#44761 ) Co-authored-by: rekram1-node <63023139+rekram1-node@users.noreply.github.com>	2026-08-24 16:55:37 -05:00
opencode-agent[bot]nexxeln rekram1-node	50c5218bca	fix(core): clarify integration auth errors (#44786 ) Co-authored-by: nexxeln <95541290+nexxeln@users.noreply.github.com> Co-authored-by: rekram1-node <rekram1-node@users.noreply.github.com>	2026-08-24 16:53:14 -05:00
Kit Langton	c1763e2b64	fix(server): make text generation locationless (#44773 )	2026-08-24 17:36:02 -04:00
Kit Langton	34bd7c220c	feat(session): report interrupt result (#44766 )	2026-08-24 17:34:37 -04:00
Dax	7f5ea1889c	test(core): provide command render services	2026-08-24 17:34:06 -04:00
Kit Langton	a02b0a4729	fix(core): make workspace destroy idempotent (#44769 )	2026-08-24 17:27:01 -04:00
Filip	d8ce27fa29	fix(github): support immutable OIDC subjects (#44778 )	2026-08-24 23:23:44 +02:00
Kit Langton	2f740cec5d	fix(tui): highlight shell script files (#44772 )	2026-08-24 17:21:19 -04:00
opencode-agent[bot]andFilip	162c3fcebd	fix(app): hide revert for child sessions (#44774 ) Co-authored-by: Filip <34747899+neriousy@users.noreply.github.com>	2026-08-24 23:20:09 +02:00
Dax	71f81dc0fe	fix(tui): simplify subagent continuation rows (#44763 )	2026-08-24 16:12:56 -04:00
Dax	de388dede4	feat(command): execute plugin callbacks	2026-08-24 16:01:58 -04:00
opencode-agent[bot]	c936acd3fe	chore: update nix node_modules hashes	2026-08-24 19:42:01 +00:00
Kit Langton	c19186ee54	feat(sdk): add Promise-first embedded host (#44746 )	2026-08-24 15:26:05 -04:00
Dax Raad	563943c52e	feat(www): restore native API reference	2026-08-24 15:21:26 -04:00
Dax	575bbd6ea1	fix(www): update docs section navigation (#44753 )	2026-08-24 14:48:26 -04:00
Aiden Cline	5d9b53b2c7	fix(ai): honor Responses parallel tool control (#44744 )	2026-08-24 13:29:43 -05:00
Aiden Cline	4780248e84	fix(ai): require thinking signature with provider-aware fallback (#44713 )	2026-08-24 13:13:48 -05:00
opencode-agent[bot]andBrendonovich	43d4968356	fix(app): focus terminal after opening panel (#44737 ) Co-authored-by: Brendonovich <14191578+Brendonovich@users.noreply.github.com>	2026-08-24 17:50:53 +00:00
Aiden Cline	0164c1c8bc	fix(ai): remove obsolete reasoning summary aliases (#44733 )	2026-08-24 12:42:36 -05:00
Aiden Cline	6a687398eb	feat(ai): gate Chat Completions body fields by provider (#44710 )	2026-08-24 12:40:56 -05:00
Kit Langton	f4cb9d06c8	fix(tui): keep remotely closed tabs closed (#44715 )	2026-08-24 13:06:02 -04:00
opencode-agent[bot]andrekram1-node	3e82b1a9fd	fix(tui): provide location context in MCP tests (#44734 ) Co-authored-by: rekram1-node <rekram1-node@users.noreply.github.com>	2026-08-24 12:05:25 -05:00
Simon Klee	e2a7600a2a	tui: preserve close target after tab removal (#44720 )	2026-08-24 19:04:50 +02:00
Dax Raad	aa8c1f6dac	fix(tui): scope MCP dialog to location	2026-08-24 12:17:12 -04:00
Kit Langton	23c3a1461c	fix(merman): harden responsive diagram layouts (#44714 )	2026-08-24 12:09:00 -04:00
opencode-agent[bot]	50a8539e4b	chore: update nix node_modules hashes	2026-08-24 14:29:37 +00:00
Shoubhit Dash	7b47589225	refactor(core): optimize stats tool summary (#44690 )	2026-08-24 19:39:04 +05:30
OpeOginniandShoubhit Dash	84275c6e9d	fix(cli): restore print-logs flag for `--standalone` runs (#44112 ) Co-authored-by: Shoubhit Dash <shoubhit2005@gmail.com>	2026-08-24 19:32:38 +05:30
Kit Langton	793ea52fa7	fix(core): make packed SDK boot on workerd (#44703 )	2026-08-24 14:02:26 +00:00
Simon Klee	40380ad9b5	tui: update OpenTUI to 0.5.8 (#44698 )	2026-08-24 15:42:27 +02:00