chore: update nix node_modules hashes

fix(tui): clarify code mode tool call rendering
feat(ai): add native Groq provider with provider-specific options (#45288 )
2026-08-26 19:46:34 +00:00 · 2026-08-26 16:44:38 +00:00 · 2026-08-26 11:27:17 -04:00 · 2026-08-26 10:12:52 -05:00 · 2026-08-26 10:48:44 -04:00 · 2026-08-26 10:47:16 -04:00
376 changed files with 14959 additions and 2210 deletions
@@ -42,7 +42,7 @@ jobs:
      - name: Find affected packages
        id: packages
        env:
-          TURBO_SCM_BASE: ${{ github.event.pull_request.base.sha || github.event.before }}
+          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
@@ -71,6 +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
@@ -110,9 +111,16 @@ 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'
@@ -122,8 +130,15 @@ jobs:
      - name: Verify packed workerd SDK
        if: runner.os == 'Linux'
        timeout-minutes: 15
-        working-directory: packages/sdk
-        run: bun run verify:package
+        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()
@@ -164,7 +179,6 @@ jobs:
  e2e:
    name: e2e (${{ matrix.settings.name }})
    needs: affected
-    if: needs.affected.outputs.app == 'true' && github.ref_name != 'v2' && github.head_ref != 'v2'
    strategy:
      fail-fast: false
      matrix:
@@ -175,32 +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:
@@ -208,23 +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 }}
@@ -125,6 +125,7 @@
        "@effect/platform-node": "catalog:",
        "@opencode-ai/client": "workspace:*",
        "@opencode-ai/plugin": "workspace:*",
+        "@opencode-ai/pty": "0.1.9",
        "@opencode-ai/schema": "workspace:*",
        "@opencode-ai/server": "workspace:*",
        "@opencode-ai/tui": "workspace:*",
@@ -346,18 +347,14 @@
        "@ai-sdk/amazon-bedrock": "4.0.112",
        "@ai-sdk/anthropic": "3.0.82",
        "@ai-sdk/azure": "3.0.88",
-        "@ai-sdk/cerebras": "2.0.41",
        "@ai-sdk/cohere": "3.0.27",
-        "@ai-sdk/deepinfra": "2.0.41",
        "@ai-sdk/gateway": "3.0.104",
        "@ai-sdk/google-vertex": "4.0.128",
-        "@ai-sdk/groq": "3.0.31",
        "@ai-sdk/mistral": "3.0.51",
        "@ai-sdk/openai-compatible": "2.0.41",
        "@ai-sdk/perplexity": "3.0.26",
        "@ai-sdk/provider": "3.0.8",
        "@ai-sdk/provider-utils": "4.0.23",
-        "@ai-sdk/togetherai": "2.0.41",
        "@ai-sdk/vercel": "2.0.39",
        "@aws-sdk/credential-providers": "3.1057.0",
        "@ff-labs/fff-bun": "0.10.5",
@@ -367,6 +364,7 @@
        "@opencode-ai/ai": "workspace:*",
        "@opencode-ai/codemode": "workspace:*",
        "@opencode-ai/plugin": "workspace:*",
+        "@opencode-ai/pty": "0.1.9",
        "@opencode-ai/schema": "workspace:*",
        "@opencode-ai/util": "workspace:*",
        "@parcel/watcher": "2.5.1",
@@ -667,6 +665,7 @@
      "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:*",
@@ -973,6 +972,7 @@
      "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",
@@ -1178,8 +1178,6 @@

    "@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=="],
@@ -1206,8 +1204,6 @@

    "@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=="],
@@ -2160,6 +2156,20 @@

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

+    "@opencode-ai/pty": ["@opencode-ai/pty@0.1.9", "", { "optionalDependencies": { "@opencode-ai/pty-darwin-arm64": "0.1.9", "@opencode-ai/pty-darwin-x64": "0.1.9", "@opencode-ai/pty-linux-arm64-gnu": "0.1.9", "@opencode-ai/pty-linux-arm64-musl": "0.1.9", "@opencode-ai/pty-linux-x64-gnu": "0.1.9", "@opencode-ai/pty-linux-x64-musl": "0.1.9" }, "bin": { "opencode-pty": "bin/opencode-pty.js" } }, "sha512-9WysQgX9J3RXfZy/t/8MGqf1IGLeckyHQsT/eVnKiBnp+GnzOXtZePryj08CN/3GT2BuP5tWqMRMH0JzCMsrgg=="],
+
+    "@opencode-ai/pty-darwin-arm64": ["@opencode-ai/pty-darwin-arm64@0.1.9", "", { "os": "darwin", "cpu": "arm64" }, "sha512-a2OZGutBdVGDO+X4t37L2K8wD1phTcLJBcNLj8j3LqCmaHbJmQyNYYFo6i8loZCTjjhb+8Wq0wMxSLny4tqdwA=="],
+
+    "@opencode-ai/pty-darwin-x64": ["@opencode-ai/pty-darwin-x64@0.1.9", "", { "os": "darwin", "cpu": "x64" }, "sha512-+vyRLwzNMzP/JFtYEIkMHxRC9Lkd7sUeuAqI5medtdgZmvTwS3NdkHbq6zQQG3rTQ9asObPKilvcpQDf+LYHRQ=="],
+
+    "@opencode-ai/pty-linux-arm64-gnu": ["@opencode-ai/pty-linux-arm64-gnu@0.1.9", "", { "os": "linux", "cpu": "arm64" }, "sha512-vjNKhCsw6mI6w+9bITCyCmb5nS+XiWA8xMGn02GOKE9AahcVX9JIbjUF1zAmPXZ1QD7nxWoanh4FvnqukbYPHw=="],
+
+    "@opencode-ai/pty-linux-arm64-musl": ["@opencode-ai/pty-linux-arm64-musl@0.1.9", "", { "os": "linux", "cpu": "arm64" }, "sha512-xpS0N6/uEiJPabv6Ib9BpOlfyZdcUES7sMVa4bCrgcxy6y4bnQMeGF/Ju4u7LBABD/rjBKZM0XhPMnvZ6qoX1w=="],
+
+    "@opencode-ai/pty-linux-x64-gnu": ["@opencode-ai/pty-linux-x64-gnu@0.1.9", "", { "os": "linux", "cpu": "x64" }, "sha512-N1Dx8rOLkpJd2DSetZZW9dPnpL3mxbRQagk/7K7TbS7M1rDnLazhq8/T9vulLET8KTwu6lmWcZYuSDfPLNWXnA=="],
+
+    "@opencode-ai/pty-linux-x64-musl": ["@opencode-ai/pty-linux-x64-musl@0.1.9", "", { "os": "linux", "cpu": "x64" }, "sha512-osb203LrlwXpQKQABrCfD5QRVK9Ajiq5atulC/VuMye6KkErcetDnjQvobjhaEjgnSEGPCH4U9sM84okGqip5Q=="],
+
    "@opencode-ai/schema": ["@opencode-ai/schema@workspace:packages/schema"],

    "@opencode-ai/script": ["@opencode-ai/script@workspace:packages/script"],
@@ -5912,10 +5922,6 @@

    "@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=="],
@@ -5954,10 +5960,6 @@

    "@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=="],
@@ -2,7 +2,7 @@
 exact = true
 # Only install newly resolved package versions published at least 3 days ago.
 minimumReleaseAge = 259200
-minimumReleaseAgeExcludes = ["@ai-sdk/amazon-bedrock", "@ai-sdk/anthropic", "@brendonovich/vite-plugin-opencode", "@opencode-ai/sdk", "@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", "@opencode-ai/pty", "@opencode-ai/pty-darwin-arm64", "@opencode-ai/pty-darwin-x64", "@opencode-ai/pty-linux-arm64-gnu", "@opencode-ai/pty-linux-arm64-musl", "@opencode-ai/pty-linux-x64-gnu", "@opencode-ai/pty-linux-x64-musl", "@opentui/core", "@opentui/core-darwin-arm64", "@opentui/core-darwin-x64", "@opentui/core-linux-arm64", "@opentui/core-linux-arm64-musl", "@opentui/core-linux-x64", "@opentui/core-linux-x64-musl", "@opentui/core-win32-arm64", "@opentui/core-win32-x64", "@opentui/keymap", "@opentui/solid", "opentui-spinner", "gitlab-ai-provider", "opencode-gitlab-auth", "@ff-labs/fff-node", "@ff-labs/fff-bun", "@ff-labs/fff-bin-darwin-arm64", "@ff-labs/fff-bin-darwin-x64", "@ff-labs/fff-bin-linux-arm64-gnu", "@ff-labs/fff-bin-linux-arm64-musl", "@ff-labs/fff-bin-linux-x64-gnu", "@ff-labs/fff-bin-linux-x64-musl", "@ff-labs/fff-bin-win32-arm64", "@ff-labs/fff-bin-win32-x64", "@pierre/diffs", "@pierre/theming", "app-builder-lib", "dmg-builder", "electron-builder", "electron-publish", "blume"]

 [test]
 root = "./do-not-run-tests-from-root"
@@ -1,8 +1,8 @@
 {
  "nodeModules": {
-    "x86_64-linux": "sha256-3Jx1Q7hl+Y0Log/k2vd5y6dzBpzFKWlhShPESxn1Rm4=",
-    "aarch64-linux": "sha256-EiiI6g01oBIrExCMAUgT3w82P0fvu4FAJhI32C+ze0I=",
-    "aarch64-darwin": "sha256-s+w49HRp1+ewtiTaU65tPWjUiO1NQw3kzfemMEEQZb0=",
-    "x86_64-darwin": "sha256-/Ee5V7pnL/qm3c4ZHeWEjH7FhGVXArXryOugbG5vsz8="
+    "x86_64-linux": "sha256-Q7BQ46mKePJtaKzhHxahIXy/pZczPmm5cQuBDrgd2Bc=",
+    "aarch64-linux": "sha256-pqk4iUhXzEc4ei9zpeGpPjX7Q6pxH1K5rgotD5Wf91s=",
+    "aarch64-darwin": "sha256-1q3mK5zLqQA0vz7KErDOkjeAnmsTReI0lhBJfIobC/E=",
+    "x86_64-darwin": "sha256-dBMQ6tZxt5VjgWTZELHgPk6fVhBfNYfmY+AnQ3iJ88Q="
  }
 }
@@ -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 +36,12 @@ 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" })
@@ -69,14 +69,22 @@ export interface OptionsInput {
  // SDK Metadata:2649 {user_id?: string | null}
  readonly metadata?: { readonly user_id?: string | null }
  // SDK MessageCreateParamsContainer:2596 ContainerParams|string
-  readonly container?: string | { readonly id?: string | null; readonly skills?: ReadonlyArray<Record<string, unknown>> | null }
+  readonly container?:
+    | string
+    | { readonly id?: string | null; readonly skills?: ReadonlyArray<Record<string, unknown>> | null }
  readonly inference_geo?: string | null
  readonly inferenceGeo?: string | null
  readonly cache_control?: { readonly type: "ephemeral"; readonly ttl?: "5m" | "1h" }
  readonly cacheControl?: { readonly type: "ephemeral"; readonly ttl?: "5m" | "1h" }
  // SDK OutputConfig:2684 {effort, format: JSONOutputFormat}
-  readonly output_config?: { readonly effort?: string | null; readonly format?: { readonly type: "json_schema"; readonly schema: Record<string, unknown> } | null }
-  readonly outputConfig?: { readonly effort?: string | null; readonly format?: { readonly type: "json_schema"; readonly schema: Record<string, unknown> } | null }
+  readonly output_config?: {
+    readonly effort?: string | null
+    readonly format?: { readonly type: "json_schema"; readonly schema: Record<string, unknown> } | null
+  }
+  readonly outputConfig?: {
+    readonly effort?: string | null
+    readonly format?: { readonly type: "json_schema"; readonly schema: Record<string, unknown> } | null
+  }
 }

 export type ProviderOptionsInput = OptionsInput
@@ -259,7 +267,11 @@ const AnthropicToolChoice = Schema.Union([
    type: Schema.Literals(["auto", "any", "none"]),
    disable_parallel_tool_use: Schema.optional(Schema.Boolean),
  }),
-  Schema.Struct({ type: Schema.tag("tool"), name: Schema.String, disable_parallel_tool_use: Schema.optional(Schema.Boolean) }),
+  Schema.Struct({
+    type: Schema.tag("tool"),
+    name: Schema.String,
+    disable_parallel_tool_use: Schema.optional(Schema.Boolean),
+  }),
 ])

 const AnthropicThinking = Schema.Union([
@@ -506,7 +518,11 @@ const lowerServerToolResult = Effect.fn("AnthropicMessages.lowerServerToolResult
  // Prefer the provider-owned replay payload; fall back to the result value for
  // histories constructed directly from provider events.
  const payload = part.providerMetadata?.anthropic?.["result"] ?? part.result.value
-  return { type: wireType, tool_use_id: scrubToolCallID(part.id), content: payload } satisfies AnthropicServerToolResultBlock
+  return {
+    type: wireType,
+    tool_use_id: scrubToolCallID(part.id),
+    content: payload,
+  } satisfies AnthropicServerToolResultBlock
 })

 const fileIdFromMetadata = (metadata: MediaPart["metadata"]): string | undefined => {
@@ -554,9 +570,7 @@ const documentContextFromMetadata = (metadata: MediaPart["metadata"]): string |
  return undefined
 }

-const citationsFromMetadata = (
-  metadata: MediaPart["metadata"],
-): AnthropicDocumentBlock["citations"] | undefined => {
+const citationsFromMetadata = (metadata: MediaPart["metadata"]): AnthropicDocumentBlock["citations"] | undefined => {
  if (!ProviderShared.isRecord(metadata)) return undefined
  const raw = ProviderShared.isRecord(metadata.anthropic)
    ? (metadata.anthropic.citations ?? metadata.citations)
@@ -706,8 +720,7 @@ const lowerToolResultContent = Effect.fnUntraced(function* (part: ToolResultPart
 })

 const requireThinkingSignature = (request: LLMRequest) => {
-  if (request.model.compatibility?.requireSignature !== undefined)
-    return request.model.compatibility.requireSignature
+  if (request.model.compatibility?.requireSignature !== undefined) return request.model.compatibility.requireSignature
  const provider = request.model.provider.toLowerCase()
  const model = request.model.id.toLowerCase()
  const baseURL = (request.model.route.endpoint.baseURL ?? "").toLowerCase()
@@ -744,9 +757,12 @@ 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" &&
@@ -793,7 +809,7 @@ 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
      }
@@ -897,21 +913,24 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (

 const resolveOptions = Effect.fn("AnthropicMessages.resolveOptions")(function* (request: LLMRequest) {
  const input = request.providerOptions as Record<string, unknown> | undefined
-  const rawServiceTier = (input as Record<string, unknown> | undefined)?.service_tier ?? (input as Record<string, unknown> | undefined)?.serviceTier
+  const rawServiceTier =
+    (input as Record<string, unknown> | undefined)?.service_tier ??
+    (input as Record<string, unknown> | undefined)?.serviceTier
  const service_tier =
    rawServiceTier === "auto" || rawServiceTier === "standard_only"
      ? (rawServiceTier as "auto" | "standard_only")
      : undefined
  const rawMetadata = (input as Record<string, unknown> | undefined)?.metadata
  const metadata =
-    ProviderShared.isRecord(rawMetadata) &&
-    (typeof rawMetadata.user_id === "string" || rawMetadata.user_id === null)
+    ProviderShared.isRecord(rawMetadata) && (typeof rawMetadata.user_id === "string" || rawMetadata.user_id === null)
      ? { user_id: rawMetadata.user_id as string | null }
      : undefined
  const container =
    typeof (input as Record<string, unknown> | undefined)?.container === "string" ||
    ProviderShared.isRecord((input as Record<string, unknown> | undefined)?.container)
-      ? ((input as Record<string, unknown>).container as string | { id?: string | null; skills?: ReadonlyArray<Record<string, unknown>> | null })
+      ? ((input as Record<string, unknown>).container as
+          | string
+          | { id?: string | null; skills?: ReadonlyArray<Record<string, unknown>> | null })
      : undefined
  const rawInferenceGeo =
    (input as Record<string, unknown> | undefined)?.inference_geo ??
@@ -962,8 +981,7 @@ const resolveThinking = Effect.fn("AnthropicMessages.resolveThinking")(function*
    input.display === "summarized" || input.display === "omitted"
      ? (input.display as "summarized" | "omitted")
      : undefined
-  if (input.type === "adaptive")
-    return { type: "adaptive" as const, ...(display === undefined ? {} : { display }) }
+  if (input.type === "adaptive") return { type: "adaptive" as const, ...(display === undefined ? {} : { display }) }
  if (input.type === "disabled") return { type: "disabled" as const }
  if (input.type !== "enabled") return undefined
  const budget =
@@ -1415,9 +1433,7 @@ const step = (state: ParserState, event: AnthropicEvent) => {
      if (event.index === undefined)
        return Effect.fail(ProviderShared.eventError(ADAPTER, `Anthropic ${block.type} missing index`))
      if (!block.id)
-        return Effect.fail(
-          ProviderShared.eventError(ADAPTER, `Anthropic tool_use missing id at index ${event.index}`),
-        )
+        return Effect.fail(ProviderShared.eventError(ADAPTER, `Anthropic tool_use missing id at index ${event.index}`))
    }
    return Effect.succeed(onContentBlockStart(state, { ...event, content_block: block }))
  }
@@ -1470,10 +1486,9 @@ export const route = Route.make({
  provider: "anthropic",
  providerMetadataKey: "anthropic",
  protocol,
-  endpoint: Endpoint.path(
-    (input) => (input.request.model.provider === "anthropic" ? `${PATH}?beta=true` : PATH),
-    { baseURL: DEFAULT_BASE_URL },
-  ),
+  endpoint: Endpoint.path((input) => (input.request.model.provider === "anthropic" ? `${PATH}?beta=true` : PATH), {
+    baseURL: DEFAULT_BASE_URL,
+  }),
  auth: Auth.none,
  framing,
  headers: () => ({ "anthropic-version": "2023-06-01" }),
@@ -652,7 +652,9 @@ const step = (state: ParserState, event: BedrockEvent) =>
        method: "stream",
        reason: classifyProviderFailure({
          message:
-            event.exception.details.message ?? event.exception.details.originalMessage ?? "Bedrock Converse stream error",
+            event.exception.details.message ??
+            event.exception.details.originalMessage ??
+            "Bedrock Converse stream error",
          code: event.exception.type,
        }),
      })
@@ -82,7 +82,9 @@ const consumeFrames = (route: string) => (state: FrameBufferState, chunk: Uint8A
        "Failed to parse Bedrock Converse event-stream payload",
      )) as Record<string, unknown>
      delete parsed.p
-      out.push(messageType === "exception" ? { exception: { type: eventType, details: parsed } } : { [eventType]: parsed })
+      out.push(
+        messageType === "exception" ? { exception: { type: eventType, details: parsed } } : { [eventType]: parsed },
+      )
    }
    return [cursor, out] as const
  })
@@ -570,7 +570,12 @@ const finish = (state: ParserState): ReadonlyArray<LLMEvent> => {
      googleMetadata({ thoughtSignature: state.reasoningSignature }),
    )
  if (state.textSignature !== undefined)
-    lifecycle = Lifecycle.textEnd(lifecycle, events, "text-0", googleMetadata({ thoughtSignature: state.textSignature }))
+    lifecycle = Lifecycle.textEnd(
+      lifecycle,
+      events,
+      "text-0",
+      googleMetadata({ thoughtSignature: state.textSignature }),
+    )
  Lifecycle.finish(lifecycle, events, {
    reason: {
      normalized:
@@ -675,8 +680,9 @@ const step = (state: ParserState, event: GeminiEvent) => {
          id,
          name: part.functionCall.name,
          input,
-          providerMetadata:
-            part.thoughtSignature ? googleMetadata({ thoughtSignature: part.thoughtSignature }) : undefined,
+          providerMetadata: part.thoughtSignature
+            ? googleMetadata({ thoughtSignature: part.thoughtSignature })
+            : undefined,
        }),
      )
      hasToolCalls = true
@@ -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
 }
@@ -316,9 +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
@@ -327,10 +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 }
@@ -346,7 +396,6 @@ export interface ParserState {
  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"
@@ -391,53 +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,
  }
@@ -528,10 +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) {
@@ -554,16 +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)
@@ -588,51 +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", [
@@ -662,10 +690,11 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques

 const lowerOptions = (request: LLMRequest) => {
  const options = OpenResponsesOptions.resolve(request)
+  const instructions = ProviderShared.joinText(request.system)
  const cacheKey = ProviderShared.promptCacheKey(request)
  const parallelToolCalls = resolveParallelToolCalls(request)
  return {
-    ...(options.instructions ? { instructions: options.instructions } : {}),
+    ...(instructions ? { instructions } : {}),
    ...(options.store !== undefined ? { store: options.store } : {}),
    ...(options.metadata ? { metadata: options.metadata } : {}),
    ...(options.safetyIdentifier ? { safety_identifier: options.safetyIdentifier } : {}),
@@ -891,25 +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 })],
  ]
 }

@@ -964,31 +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,
  ]
 }

@@ -1051,18 +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
@@ -1116,10 +1134,11 @@ const onResponseFinish = Effect.fn("OpenResponses.onResponseFinish")(function* (
          event.response?.output ?? [],
          () => [state, NO_EVENTS] satisfies StepResult,
          ([current, events], item) => {
+            const id = item.id ?? (item.type === "function_call" ? item.call_id : undefined)
            if (
-              !item.id ||
-              ((item.type !== "function_call" || !current.tools[item.id]) &&
-                (item.type !== "reasoning" || !current.reasoningItems[item.id]))
+              !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(
@@ -1245,10 +1264,11 @@ export const step = (state: ParserState, input: 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`)
+    const id = event.item?.id ?? (event.item?.type === "function_call" ? event.item.call_id : undefined)
    return Effect.succeed(
      onOutputItemAdded(
-        event.output_index !== undefined && event.item?.id
-          ? { ...state, outputItems: { ...state.outputItems, [event.output_index]: event.item.id } }
+        event.output_index !== undefined && id
+          ? { ...state, outputItems: { ...state.outputItems, [event.output_index]: id } }
          : state,
        event,
      ),
@@ -1291,7 +1311,6 @@ export const initial = (request: LLMRequest, extension: Extension = BASE): Parse
  messageItems: new Set<string>(),
  messagePhases: {},
  reasoningItems: {},
-  store: OpenResponsesOptions.resolve(request).store,
 })

 export const protocol = Protocol.make({
@@ -278,6 +278,7 @@ interface LoweringOptions {
  readonly cacheControl?: (
    cache: CacheHint | undefined,
  ) => Schema.Schema.Type<typeof OpenAIChatCacheControl> | undefined
+  readonly toolCallID?: (id: string) => string
 }

 const lowerTool = (
@@ -304,8 +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,
@@ -363,8 +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[] = []
@@ -381,7 +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
    }
  }
@@ -391,15 +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)
@@ -427,7 +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),
      })
@@ -437,7 +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),
    })
@@ -453,11 +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
 })

@@ -478,12 +484,42 @@ 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) {
@@ -526,14 +562,19 @@ 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
@@ -555,7 +596,10 @@ const hasToolHistory = (messages: ReadonlyArray<LLMRequest["messages"][number]>)
 // models.dev provider naming: DeepSeek, Moonshot AI, Together AI, ZAI
 // (Zhipu + Coding Plan variants), Nvidia, Cerebras, Chutes, etc. still
 // require `max_tokens`.
-const detectMaxTokensField = (provider: string, baseURL: string | undefined): "max_tokens" | "max_completion_tokens" => {
+const detectMaxTokensField = (
+  provider: string,
+  baseURL: string | undefined,
+): "max_tokens" | "max_completion_tokens" => {
  const p = provider.toLowerCase()
  const url = (baseURL ?? "").toLowerCase()
  if (
@@ -605,7 +649,8 @@ const detectSupportsStore = (provider: string, baseURL: string | undefined): boo
  const isChutes = p === "chutes" || url.includes("chutes.ai")
  const isCloudflareWorkersAI = p === "cloudflare-workers-ai" || url.includes("api.cloudflare.com")
  const isCloudflareAiGateway = p === "cloudflare-ai-gateway" || url.includes("gateway.ai.cloudflare.com")
-  const isVercelAiGateway = p === "vercel-ai-gateway" || url.includes("ai-gateway.vercel.sh") || url.includes("vercel.sh")
+  const isVercelAiGateway =
+    p === "vercel-ai-gateway" || url.includes("ai-gateway.vercel.sh") || url.includes("vercel.sh")
  const isAntLing = p === "ant-ling" || url.includes("api.ant-ling.com")
  const isOpencode = p === "opencode" || url.includes("opencode.ai")
  const isNonStandard =
@@ -637,11 +682,7 @@ const detectSupportsStrictMode = (provider: string, baseURL: string | undefined)
  return !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia
 }

-const detectZaiToolStream = (
-  provider: string,
-  baseURL: string | undefined,
-  modelID: string,
-): boolean => {
+const detectZaiToolStream = (provider: string, baseURL: string | undefined, modelID: string): boolean => {
  const p = provider.toLowerCase()
  const url = (baseURL ?? "").toLowerCase()
  const isZai =
@@ -691,10 +732,10 @@ export const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (
  const supportsStore = request.model.compatibility?.supportsStore ?? detectSupportsStore(provider, baseURL)
  const supportsUsageInStreaming =
    request.model.compatibility?.supportsUsageInStreaming ?? detectSupportsUsageInStreaming()
-  const supportsStrictMode = request.model.compatibility?.supportsStrictMode ?? detectSupportsStrictMode(provider, baseURL)
+  const supportsStrictMode =
+    request.model.compatibility?.supportsStrictMode ?? detectSupportsStrictMode(provider, baseURL)
  const zaiToolStream =
-    request.model.compatibility?.zaiToolStream ??
-    detectZaiToolStream(provider, baseURL, request.model.id)
+    request.model.compatibility?.zaiToolStream ?? detectZaiToolStream(provider, baseURL, request.model.id)
  const hasHistory = hasToolHistory(request.messages)
  const hasActiveTools = request.tools.length > 0
  return {
@@ -783,11 +824,10 @@ const mapUsage = (usage: OpenAIChatEvent["usage"]): Usage | undefined => {
  if (!usage) return undefined
  const input = usage.prompt_tokens ?? undefined
  const output = usage.completion_tokens ?? undefined
-  const cached =
-    (usage.prompt_tokens_details?.cached_tokens ??
-      (usage as { prompt_cache_hit_tokens?: number | null }).prompt_cache_hit_tokens ??
-      (usage as { cached_tokens?: number | null }).cached_tokens ??
-      undefined) as number | undefined
+  const cached = (usage.prompt_tokens_details?.cached_tokens ??
+    (usage as { prompt_cache_hit_tokens?: number | null }).prompt_cache_hit_tokens ??
+    (usage as { cached_tokens?: number | null }).cached_tokens ??
+    undefined) as number | undefined
  const cacheWrite = usage.prompt_tokens_details?.cache_write_tokens ?? undefined
  const reasoning = usage.completion_tokens_details?.reasoning_tokens ?? undefined
  const nonCached = ProviderShared.subtractTokens(input, ProviderShared.sumTokens(cached, cacheWrite))
@@ -903,13 +943,12 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
    const choiceUsage = (choice as unknown as { usage?: OpenAIChatEvent["usage"] })?.usage
    const usage = mapUsage(event.usage) ?? (choiceUsage ? mapUsage(choiceUsage) : undefined) ?? state.usage
    const rawFinishReason = choice?.finish_reason
-    const finishReason =
-      rawFinishReason
-        ? {
-            normalized: yield* mapFinishReason(event, rawFinishReason),
-            raw: choice?.native_finish_reason ?? rawFinishReason,
-          }
-        : state.finishReason
+    const finishReason = rawFinishReason
+      ? {
+          normalized: yield* mapFinishReason(event, rawFinishReason),
+          raw: choice?.native_finish_reason ?? rawFinishReason,
+        }
+      : state.finishReason
    const delta = choice?.delta
    const toolDeltas = delta?.tool_calls ?? []
    let tools = state.tools
@@ -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) => {
@@ -105,6 +122,8 @@ 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 }),
@@ -112,7 +131,7 @@ const fromRequest = Effect.fn("OpenAIResponses.fromRequest")(function* (request:
  )
  const toolSchemaCompatibility = request.model.compatibility?.toolSchema
  const parallelToolCalls = OpenResponses.resolveParallelToolCalls(request)
-  return {
+  return yield* decodeBody({
    ...body,
    ...(parallelToolCalls === undefined ? {} : { parallel_tool_calls: parallelToolCalls }),
    tools:
@@ -123,7 +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) {
@@ -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)) &&
@@ -29,10 +29,9 @@ export type ResponseIncludable = (typeof ResponseIncludables)[number] | (string

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

 export const Truncations = ["auto", "disabled"] as const
 export type Truncation = (typeof Truncations)[number]
@@ -56,7 +55,6 @@ 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),
@@ -34,37 +34,35 @@ export const onDone: (
  state: OpenResponses.ParserState,
  item: Item,
  tools: Definitions,
-) => Effect.Effect<OpenResponses.StepResult, AIError> = Effect.fn("ResponsesHostedTools.onDone")(function* (
-  state,
-  item,
-  tools,
-) {
-  const tool = tools[item.type]
-  if (!tool) return [state, []] satisfies OpenResponses.StepResult
-  const providerMetadata = OpenResponses.providerMetadata(state, { itemId: item.id })
-  const events: LLMEvent[] = []
-  const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
-  events.push(
-    LLMEvent.toolCall({
-      id: item.id,
-      name: tool.name,
-      input: tool.input(item),
-      providerExecuted: true,
-      providerMetadata,
-    }),
-    LLMEvent.toolResult({
-      id: item.id,
-      name: tool.name,
-      result: tool.result
-        ? yield* tool.result(item)
-        : item.error !== undefined && item.error !== null
-          ? { type: "error", value: item.error }
-          : { type: "json", value: item },
-      providerExecuted: true,
-      providerMetadata,
-    }),
-  )
-  return [{ ...state, lifecycle }, events] satisfies OpenResponses.StepResult
-})
+) => Effect.Effect<OpenResponses.StepResult, AIError> = Effect.fn("ResponsesHostedTools.onDone")(
+  function* (state, item, tools) {
+    const tool = tools[item.type]
+    if (!tool) return [state, []] satisfies OpenResponses.StepResult
+    const providerMetadata = OpenResponses.providerMetadata(state, { itemId: item.id })
+    const events: LLMEvent[] = []
+    const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
+    events.push(
+      LLMEvent.toolCall({
+        id: item.id,
+        name: tool.name,
+        input: tool.input(item),
+        providerExecuted: true,
+        providerMetadata,
+      }),
+      LLMEvent.toolResult({
+        id: item.id,
+        name: tool.name,
+        result: tool.result
+          ? yield* tool.result(item)
+          : item.error !== undefined && item.error !== null
+            ? { type: "error", value: item.error }
+            : { type: "json", value: item },
+        providerExecuted: true,
+        providerMetadata,
+      }),
+    )
+    return [{ ...state, lifecycle }, events] satisfies OpenResponses.StepResult
+  },
+)

 export * as ResponsesHostedTools from "./responses-hosted-tools.js"
@@ -1,15 +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),
@@ -40,15 +40,16 @@ const patterns = [
  /model_context_window_exceeded/i,
  /too many tokens/i,
  /token limit exceeded/i,
+  /request_too_large/i,
 ]

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

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

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

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

@@ -106,6 +107,7 @@ export function classifyProviderFailure(input: ProviderFailure): AIError["reason
    clientScoped &&
    (codes.includes("context_length_exceeded") ||
      codes.includes("model_context_window_exceeded") ||
+      codes.includes("request_too_large") ||
      isContextOverflow(text))
  )
    return new InvalidRequestReason({ ...common, classification: "context-overflow" })
@@ -1,5 +1,5 @@
 import { Auth } from "../route/auth.js"
-import type { Route as RouteDef, RouteDefaultsInput } from "../route/client.js"
+import { Route, type RouteDefaultsInput } from "../route/client.js"
 import type { ProviderPackage } from "../provider-package.js"
 import { OpenAIChat } from "../protocols/openai-chat.js"
 import { OpenAIResponses } from "../protocols/openai-responses.js"
@@ -26,9 +26,15 @@ export interface Settings extends ProviderPackage.Settings {
  readonly providerOptions?: OpenAIProviderOptionsInput
 }

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

 const chatRoute = OpenAIChat.route.with({
@@ -38,7 +44,7 @@ const chatRoute = OpenAIChat.route.with({

 export const routes = [responsesRoute, chatRoute]

-const configuredRoute = <Body, Prepared>(route: RouteDef<Body, Prepared>, input: Config) => {
+const configuredRoute = <Body, Prepared>(route: Route<Body, Prepared>, input: Config) => {
  const region = input.region ?? input.credentials?.region ?? "us-east-1"
  const credentials = input.credentials === undefined ? undefined : { ...input.credentials, region }
  return route.with({
@@ -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)
@@ -0,0 +1,116 @@
+import { Effect, Schema } from "effect"
+import type { ProviderPackage } from "../provider-package.js"
+import { OpenAIChat } from "../protocols/openai-chat.js"
+import { ProviderShared } from "../protocols/shared.js"
+import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
+import { Route, type RouteDefaultsInput } from "../route/client.js"
+import { Endpoint } from "../route/endpoint.js"
+import { Framing } from "../route/framing.js"
+import { Protocol } from "../route/protocol.js"
+import { ProviderID, type ModelID, type LLMRequest } from "../schema/index.js"
+import { profiles } from "./openai-compatible-profile.js"
+import type { OpenAIProviderOptionsInput } from "./openai-options.js"
+
+export const id = ProviderID.make("groq")
+
+export type ProviderOptions = Pick<OpenAIProviderOptionsInput, "reasoningEffort"> & {
+  /** Controls visible reasoning on GPT-OSS; other models always use parsed reasoning. */
+  readonly includeReasoning?: boolean
+  readonly parallelToolCalls?: boolean
+  readonly serviceTier?: "on_demand" | "flex" | "auto" | "performance" | (string & {})
+  readonly user?: string
+}
+
+export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
+  ProviderAuthOption<"optional"> & {
+    readonly baseURL?: string
+    readonly providerOptions?: ProviderOptions
+  }
+
+export interface Settings extends ProviderPackage.Settings {
+  readonly apiKey?: string
+  readonly baseURL?: string
+  readonly providerOptions?: ProviderOptions
+}
+
+const Options = Schema.Struct({
+  includeReasoning: Schema.optional(Schema.Boolean),
+  parallelToolCalls: Schema.optional(Schema.Boolean),
+  serviceTier: Schema.optional(Schema.String),
+  user: Schema.optional(Schema.String),
+})
+
+export const protocol = Protocol.make({
+  id: "groq-chat",
+  body: {
+    schema: Schema.Struct({
+      ...OpenAIChat.bodyFields,
+      reasoning_format: Schema.optional(Schema.Literal("parsed")),
+      include_reasoning: Schema.optional(Schema.Boolean),
+      parallel_tool_calls: Schema.optional(Schema.Boolean),
+      service_tier: Schema.optional(Schema.String),
+      user: Schema.optional(Schema.String),
+    }),
+    from: Effect.fn("Groq.fromRequest")(function* (request: LLMRequest) {
+      const options = yield* ProviderShared.validateWith(Schema.decodeUnknownEffect(Options))(
+        request.providerOptions ?? {},
+      )
+      const gptOSS = request.model.id.startsWith("openai/gpt-oss-")
+      return {
+        ...(yield* OpenAIChat.fromRequest(request)),
+        reasoning_format: gptOSS ? undefined : ("parsed" as const),
+        include_reasoning: gptOSS ? options.includeReasoning : undefined,
+        parallel_tool_calls: options.parallelToolCalls,
+        service_tier: options.serviceTier,
+        user: options.user,
+      }
+    }),
+  },
+  stream: OpenAIChat.protocol.stream,
+})
+
+export const route = Route.make({
+  id: "groq-chat",
+  provider: id,
+  providerMetadataKey: "openai",
+  protocol,
+  endpoint: Endpoint.path("/chat/completions", { baseURL: profiles.groq.baseURL }),
+  framing: Framing.sse,
+})
+
+export const configure = (input: LanguageModelOptions = {}) => {
+  const { apiKey: _apiKey, auth: _auth, baseURL, ...defaults } = input
+  const configured = route.with({
+    ...defaults,
+    endpoint: { baseURL: baseURL ?? profiles.groq.baseURL },
+    auth: AuthOptions.bearer(input, "GROQ_API_KEY"),
+  })
+  return {
+    id,
+    model: (modelID: string | ModelID) =>
+      configured.model<ProviderOptions>({
+        id: modelID,
+        compatibility: {
+          maxTokensField: "max_completion_tokens",
+          reasoningField: "reasoning",
+          requireReasoning: false,
+          supportsStore: false,
+          supportsStrictMode: false,
+        },
+      }),
+    configure,
+  }
+}
+
+export const provider = configure()
+
+export const model: ProviderPackage.Definition<Settings, ProviderOptions>["model"] = (modelID, settings) =>
+  configure({
+    apiKey: settings.apiKey,
+    baseURL: settings.baseURL,
+    headers: settings.headers === undefined ? undefined : { ...settings.headers },
+    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
+    providerOptions: settings.providerOptions,
+  }).model(modelID)
+
+export * as Groq from "./groq.js"
@@ -3,16 +3,20 @@ export * as AnthropicCompatible from "./anthropic-compatible.js"
 export * as AmazonBedrock from "./amazon-bedrock.js"
 export * as AmazonBedrockMantle from "./amazon-bedrock-mantle.js"
 export * as Azure from "./azure.js"
+export * as Cerebras from "./cerebras.js"
 export * as Cloudflare from "./cloudflare.js"
 export { CloudflareAIGateway, CloudflareWorkersAI } from "./cloudflare.js"
+export * as DeepInfra from "./deepinfra.js"
 export * as Google from "./google.js"
 export * as GoogleVertex from "./google-vertex.js"
 export * as GoogleVertexChat from "./google-vertex-chat.js"
 export * as GoogleVertexMessages from "./google-vertex-messages.js"
 export * as GoogleVertexResponses from "./google-vertex-responses.js"
+export * as Groq from "./groq.js"
 export * as OpenAI from "./openai.js"
 export * as OpenAICompatible from "./openai-compatible.js"
 export * as OpenAICompatibleResponses from "./openai-compatible-responses.js"
 export * as OpenRouter from "./openrouter.js"
+export * as TogetherAI from "./togetherai.js"
 export * as XAI from "./xai.js"
 export * as ZAI from "./zai.js"
@@ -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)
@@ -339,9 +339,7 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
                return onHalt
                  ? parsed.pipe(
                      Stream.concat(
-                        Stream.suspend(() =>
-                          Stream.unwrap(onHalt(state).pipe(Effect.map(Stream.fromIterable))),
-                        ),
+                        Stream.suspend(() => Stream.unwrap(onHalt(state).pipe(Effect.map(Stream.fromIterable)))),
                      ),
                    )
                  : parsed
@@ -153,8 +153,11 @@ 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),
@@ -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(
@@ -211,6 +211,28 @@ describe("RequestExecutor", () => {
    }).pipe(Effect.provide(responsesLayer([new Response("request too large", { status: 413 })]))),
  )

+  it.effect("classifies Anthropic request_too_large as context overflow", () =>
+    Effect.gen(function* () {
+      const executor = yield* RequestExecutor.Service
+      const error = yield* executor.execute(request).pipe(Effect.flip)
+
+      expectAIError(error)
+      expect(error.reason).toMatchObject({
+        _tag: "InvalidRequest",
+        classification: "context-overflow",
+        http: { response: { status: 413 } },
+      })
+    }).pipe(
+      Effect.provide(
+        responsesLayer([
+          new Response('{"error":{"type":"request_too_large","message":"Request exceeds the maximum size"}}', {
+            status: 413,
+          }),
+        ]),
+      ),
+    ),
+  )
+
  it.effect("does not classify ordinary invalid requests as context overflow", () =>
    Effect.gen(function* () {
      const executor = yield* RequestExecutor.Service
@@ -1,10 +1,7 @@
 {
  "version": 1,
  "metadata": {
-    "tags": [
-      "prefix:azure",
-      "provider:azure"
-    ],
+    "tags": ["prefix:azure", "provider:azure"],
    "name": "azure/chat-streams-text",
    "recordedAt": "2026-08-23T17:21:53.198Z"
  },
@@ -1,10 +1,7 @@
 {
  "version": 1,
  "metadata": {
-    "tags": [
-      "prefix:azure",
-      "provider:azure"
-    ],
+    "tags": ["prefix:azure", "provider:azure"],
    "name": "azure/responses-calls-a-tool",
    "recordedAt": "2026-08-23T17:21:55.170Z"
  },
@@ -1,10 +1,7 @@
 {
  "version": 1,
  "metadata": {
-    "tags": [
-      "prefix:azure",
-      "provider:azure"
-    ],
+    "tags": ["prefix:azure", "provider:azure"],
    "name": "azure/responses-continues-after-a-tool-result",
    "recordedAt": "2026-08-23T17:21:56.397Z"
  },
@@ -1,10 +1,7 @@
 {
  "version": 1,
  "metadata": {
-    "tags": [
-      "prefix:azure",
-      "provider:azure"
-    ],
+    "tags": ["prefix:azure", "provider:azure"],
    "name": "azure/responses-streams-text",
    "recordedAt": "2026-08-23T17:21:54.158Z"
  },
@@ -2,11 +2,7 @@
  "version": 1,
  "metadata": {
    "model": "openai.gpt-oss-120b",
-    "tags": [
-      "prefix:bedrock-mantle",
-      "provider:amazon-bedrock",
-      "protocol:openai-responses"
-    ],
+    "tags": ["prefix:bedrock-mantle", "provider:amazon-bedrock", "protocol:openai-responses"],
    "name": "bedrock-mantle/streams-text",
    "recordedAt": "2026-08-25T03:29:02.968Z"
  },
@@ -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"
+      }
+    }
+  ]
+}
@@ -7,13 +7,7 @@
    "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": [
    {
@@ -35,4 +29,4 @@
      }
    }
  ]
-}
+}
@@ -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"
+      }
+    }
+  ]
+}
@@ -1,11 +1,7 @@
 {
  "version": 1,
  "metadata": {
-    "tags": [
-      "prefix:google-vertex",
-      "provider:google-vertex",
-      "protocol:gemini"
-    ],
+    "tags": ["prefix:google-vertex", "provider:google-vertex", "protocol:gemini"],
    "name": "google-vertex/calls-a-tool",
    "recordedAt": "2026-08-23T17:21:51.036Z"
  },
@@ -1,11 +1,7 @@
 {
  "version": 1,
  "metadata": {
-    "tags": [
-      "prefix:google-vertex",
-      "provider:google-vertex",
-      "protocol:gemini"
-    ],
+    "tags": ["prefix:google-vertex", "provider:google-vertex", "protocol:gemini"],
    "name": "google-vertex/continues-after-a-tool-result",
    "recordedAt": "2026-08-23T17:21:51.853Z"
  },
@@ -1,11 +1,7 @@
 {
  "version": 1,
  "metadata": {
-    "tags": [
-      "prefix:google-vertex",
-      "provider:google-vertex",
-      "protocol:gemini"
-    ],
+    "tags": ["prefix:google-vertex", "provider:google-vertex", "protocol:gemini"],
    "name": "google-vertex/streams-text",
    "recordedAt": "2026-08-23T17:21:50.112Z"
  },
@@ -0,0 +1,29 @@
+{
+  "version": 1,
+  "metadata": {
+    "model": "openai/gpt-oss-20b",
+    "tags": ["prefix:groq-chat", "provider:groq", "protocol:groq-chat", "text", "usage"],
+    "name": "groq-chat/streams-text-with-usage",
+    "recordedAt": "2026-08-26T14:40:09.833Z"
+  },
+  "interactions": [
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://api.groq.com/openai/v1/chat/completions",
+        "headers": {
+          "content-type": "application/json"
+        },
+        "body": "{\"model\":\"openai/gpt-oss-20b\",\"messages\":[{\"role\":\"user\",\"content\":\"Reply with exactly one word: hello\"}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"reasoning_effort\":\"low\",\"max_completion_tokens\":512,\"include_reasoning\":false,\"service_tier\":\"on_demand\",\"user\":\"recorded-test\"}"
+      },
+      "response": {
+        "status": 200,
+        "headers": {
+          "content-type": "text/event-stream"
+        },
+        "body": "data: {\"id\":\"chatcmpl-660048b0-3c99-4215-a582-e116b77eb881\",\"object\":\"chat.completion.chunk\",\"created\":1787755209,\"model\":\"openai/gpt-oss-20b\",\"system_fingerprint\":\"fp_66891002f6\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\"},\"logprobs\":null,\"finish_reason\":null}],\"x_groq\":{\"id\":\"req_01m0z87915eep9bpf10gg7331e\",\"seed\":94036161}}\n\ndata: {\"id\":\"chatcmpl-660048b0-3c99-4215-a582-e116b77eb881\",\"object\":\"chat.completion.chunk\",\"created\":1787755209,\"model\":\"openai/gpt-oss-20b\",\"system_fingerprint\":\"fp_66891002f6\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\"hello\"},\"logprobs\":null,\"finish_reason\":null}]}\n\ndata: {\"id\":\"chatcmpl-660048b0-3c99-4215-a582-e116b77eb881\",\"object\":\"chat.completion.chunk\",\"created\":1787755209,\"model\":\"openai/gpt-oss-20b\",\"system_fingerprint\":\"fp_66891002f6\",\"choices\":[{\"index\":0,\"delta\":{},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"x_groq\":{\"id\":\"req_01m0z87915eep9bpf10gg7331e\",\"usage\":{\"queue_time\":0.10886435,\"prompt_tokens\":78,\"prompt_time\":0.003693734,\"completion_tokens\":20,\"completion_time\":0.020459983,\"total_tokens\":98,\"total_time\":0.024153717,\"completion_tokens_details\":{\"reasoning_tokens\":10}}},\"usage\":{\"queue_time\":0.10886435,\"prompt_tokens\":78,\"prompt_time\":0.003693734,\"completion_tokens\":20,\"completion_time\":0.020459983,\"total_tokens\":98,\"total_time\":0.024153717,\"completion_tokens_details\":{\"reasoning_tokens\":10}}}\n\ndata: {\"id\":\"chatcmpl-660048b0-3c99-4215-a582-e116b77eb881\",\"object\":\"chat.completion.chunk\",\"created\":1787755209,\"model\":\"openai/gpt-oss-20b\",\"system_fingerprint\":\"fp_66891002f6\",\"choices\":[],\"usage\":{\"queue_time\":0.10886435,\"prompt_tokens\":78,\"prompt_time\":0.003693734,\"completion_tokens\":20,\"completion_time\":0.020459983,\"total_tokens\":98,\"total_time\":0.024153717,\"completion_tokens_details\":{\"reasoning_tokens\":10}},\"service_tier\":\"on_demand\"}\n\ndata: [DONE]\n\n"
+      }
+    }
+  ]
+}
@@ -32,7 +32,7 @@
        {
          "direction": "client",
          "kind": "text",
-          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"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,
@@ -5,14 +5,7 @@
    "route": "openai-responses",
    "transport": "http",
    "model": "gpt-5.5",
-    "tags": [
-      "prefix:openai-responses",
-      "provider:openai",
-      "flagship",
-      "tool",
-      "tool-loop",
-      "golden"
-    ],
+    "tags": ["prefix:openai-responses", "provider:openai", "flagship", "tool", "tool-loop", "golden"],
    "name": "openai-responses/openai-responses-gpt-5-5-tool-loop",
    "recordedAt": "2026-08-20T06:30:22.262Z"
  },
@@ -25,7 +18,7 @@
        "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 +36,7 @@
        "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,
@@ -1,14 +1,7 @@
 {
  "version": 1,
  "metadata": {
-    "tags": [
-      "prefix:pdf",
-      "pdf",
-      "provider:openai",
-      "protocol:openai-responses",
-      "tool",
-      "tool-result"
-    ],
+    "tags": ["prefix:pdf", "pdf", "provider:openai", "protocol:openai-responses", "tool", "tool-result"],
    "name": "pdf/openai-tool-result",
    "recordedAt": "2026-08-25T03:29:08.297Z"
  },
@@ -21,7 +14,7 @@
        "headers": {
          "content-type": "application/json"
        },
-        "body": "{\"model\":\"gpt-4o-mini\",\"input\":[{\"role\":\"system\",\"content\":\"Read the PDF returned by the tool and follow the user's response format exactly.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Return only the verification code from the PDF.\"}]},{\"type\":\"function_call\",\"call_id\":\"call_pdf_1\",\"name\":\"read_pdf\",\"arguments\":\"{}\"},{\"type\":\"function_call_output\",\"call_id\":\"call_pdf_1\",\"output\":[{\"type\":\"input_text\",\"text\":\"PDF read successfully\"},{\"type\":\"input_file\",\"filename\":\"verification.pdf\",\"file_data\":\"data:application/pdf;base64,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\"}]}],\"tools\":[{\"type\":\"function\",\"name\":\"read_pdf\",\"description\":\"Read the attached PDF.\",\"parameters\":{\"type\":\"object\",\"properties\":{},\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"max_output_tokens\":40,\"temperature\":0,\"stream\":true}"
+        "body": "{\"model\":\"gpt-4o-mini\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Return only the verification code from the PDF.\"}]},{\"type\":\"function_call\",\"call_id\":\"call_pdf_1\",\"name\":\"read_pdf\",\"arguments\":\"{}\"},{\"type\":\"function_call_output\",\"call_id\":\"call_pdf_1\",\"output\":[{\"type\":\"input_text\",\"text\":\"PDF read successfully\"},{\"type\":\"input_file\",\"filename\":\"verification.pdf\",\"file_data\":\"data:application/pdf;base64,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\"}]}],\"tools\":[{\"type\":\"function\",\"name\":\"read_pdf\",\"description\":\"Read the attached PDF.\",\"parameters\":{\"type\":\"object\",\"properties\":{},\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"max_output_tokens\":40,\"temperature\":0,\"stream\":true,\"instructions\":\"Read the PDF returned by the tool and follow the user's response format exactly.\"}"
      },
      "response": {
        "status": 200,
@@ -1,13 +1,7 @@
 {
  "version": 1,
  "metadata": {
-    "tags": [
-      "prefix:pdf",
-      "pdf",
-      "provider:openai",
-      "protocol:openai-responses",
-      "user-input"
-    ],
+    "tags": ["prefix:pdf", "pdf", "provider:openai", "protocol:openai-responses", "user-input"],
    "name": "pdf/openai-user-input",
    "recordedAt": "2026-08-25T03:29:05.645Z"
  },
@@ -1,14 +1,7 @@
 {
  "version": 1,
  "metadata": {
-    "tags": [
-      "prefix:pdf",
-      "pdf",
-      "provider:xai",
-      "protocol:xai-responses",
-      "tool",
-      "tool-result"
-    ],
+    "tags": ["prefix:pdf", "pdf", "provider:xai", "protocol:xai-responses", "tool", "tool-result"],
    "name": "pdf/xai-tool-result",
    "recordedAt": "2026-08-25T03:29:11.774Z"
  },
@@ -21,7 +14,7 @@
        "headers": {
          "content-type": "application/json"
        },
-        "body": "{\"model\":\"grok-4.5\",\"input\":[{\"role\":\"system\",\"content\":\"Read the PDF returned by the tool and follow the user's response format exactly.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Return only the verification code from the PDF.\"}]},{\"type\":\"function_call\",\"call_id\":\"call_pdf_1\",\"name\":\"read_pdf\",\"arguments\":\"{}\"},{\"type\":\"function_call_output\",\"call_id\":\"call_pdf_1\",\"output\":[{\"type\":\"input_text\",\"text\":\"PDF read successfully\"},{\"type\":\"input_file\",\"filename\":\"verification.pdf\",\"file_data\":\"data:application/pdf;base64,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\"}]}],\"tools\":[{\"type\":\"function\",\"name\":\"read_pdf\",\"description\":\"Read the attached PDF.\",\"parameters\":{\"type\":\"object\",\"properties\":{},\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"max_output_tokens\":40,\"temperature\":0,\"stream\":true}"
+        "body": "{\"model\":\"grok-4.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Return only the verification code from the PDF.\"}]},{\"type\":\"function_call\",\"call_id\":\"call_pdf_1\",\"name\":\"read_pdf\",\"arguments\":\"{}\"},{\"type\":\"function_call_output\",\"call_id\":\"call_pdf_1\",\"output\":[{\"type\":\"input_text\",\"text\":\"PDF read successfully\"},{\"type\":\"input_file\",\"filename\":\"verification.pdf\",\"file_data\":\"data:application/pdf;base64,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\"}]}],\"tools\":[{\"type\":\"function\",\"name\":\"read_pdf\",\"description\":\"Read the attached PDF.\",\"parameters\":{\"type\":\"object\",\"properties\":{},\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"max_output_tokens\":40,\"temperature\":0,\"stream\":true,\"instructions\":\"Read the PDF returned by the tool and follow the user's response format exactly.\"}"
      },
      "response": {
        "status": 200,
@@ -1,13 +1,7 @@
 {
  "version": 1,
  "metadata": {
-    "tags": [
-      "prefix:pdf",
-      "pdf",
-      "provider:xai",
-      "protocol:xai-responses",
-      "user-input"
-    ],
+    "tags": ["prefix:pdf", "pdf", "provider:xai", "protocol:xai-responses", "user-input"],
    "name": "pdf/xai-user-input",
    "recordedAt": "2026-08-25T03:29:10.612Z"
  },
@@ -52,4 +52,4 @@
      }
    }
  ]
-}
+}
@@ -2,12 +2,7 @@
  "version": 1,
  "metadata": {
    "model": "anthropic/claude-sonnet-4.6",
-    "tags": [
-      "prefix:openai-compatible-chat",
-      "provider:vercel-ai-gateway",
-      "protocol:openai-chat",
-      "reasoning"
-    ],
+    "tags": ["prefix:openai-compatible-chat", "provider:vercel-ai-gateway", "protocol:openai-chat", "reasoning"],
    "name": "vercel-ai-gateway-reasoning",
    "recordedAt": "2026-07-18T11:28:42.077Z"
  },
@@ -31,4 +26,4 @@
      }
    }
  ]
-}
+}
@@ -18,13 +18,19 @@ describe("provider error classification", () => {
    expect(messages.every(isContextOverflow)).toBe(true)
  })

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

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

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

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

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

    for (const module of modules) expect(module.model).toBeFunction()
@@ -35,6 +39,24 @@ 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")
@@ -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"
@@ -27,6 +28,12 @@ const compileUnsignedReasoning = (model: LLMRequest["model"]) =>
    }),
  )

+const vertexOpus48 = GoogleVertexMessages.configure({
+  accessToken: "test",
+  location: "global",
+  project: "test",
+}).model("claude-opus-4-8")
+
 const request = LLM.request({
  id: "req_1",
  model,
@@ -286,6 +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(
@@ -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 { OpenAIResponses } from "../../src/protocols/openai-responses.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"

@@ -19,6 +20,7 @@ describe("Amazon Bedrock Mantle provider", () => {
  it.effect("uses Chat by default and exposes Responses", () =>
    Effect.gen(function* () {
      const provider = AmazonBedrockMantle.configure({ credentials })
+      expect(provider.responses("openai.gpt-oss-120b").route.transport).toBe(OpenAIResponses.httpTransport)
      const chat = yield* compileRequest(LLM.request({ model: provider.model("openai.gpt-oss-120b"), prompt: "Hi" }))
      const responses = yield* compileRequest(
        LLM.request({ model: provider.responses("openai.gpt-oss-120b"), prompt: "Hi" }),
@@ -83,6 +85,39 @@ 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({
@@ -515,7 +515,10 @@ describe("Gemini route", () => {
        {
          role: "model",
          parts: [
-            { functionCall: { id: "call_image", name: "read", args: { path: "pixel.png" } }, thoughtSignature: "sig_1" },
+            {
+              functionCall: { id: "call_image", name: "read", args: { path: "pixel.png" } },
+              thoughtSignature: "sig_1",
+            },
          ],
        },
        {
@@ -606,10 +609,7 @@ describe("Gemini route", () => {
      expect(prepared.body.contents).toEqual([
        {
          role: "model",
-          parts: [
-            { functionCall: { name: "shot", args: {} } },
-            { functionCall: { name: "shot", args: {} } },
-          ],
+          parts: [{ functionCall: { name: "shot", args: {} } }, { functionCall: { name: "shot", args: {} } }],
        },
        {
          role: "user",
@@ -1071,7 +1071,9 @@ describe("Gemini route", () => {
      const prepared = yield* compileRequest(
        LLM.request({
          model,
-          messages: [Message.assistant([{ type: "text", text: "All done.", providerMetadata: delta?.providerMetadata }])],
+          messages: [
+            Message.assistant([{ type: "text", text: "All done.", providerMetadata: delta?.providerMetadata }]),
+          ],
        }),
      )
      expect(prepared.body.contents).toEqual([
@@ -1572,9 +1574,7 @@ describe("Gemini route", () => {
              { candidates: [{ content: { role: "model", parts: null } }] },
              { candidates: [{ content: null, finishReason: null }] },
              {
-                candidates: [
-                  { content: { role: "model", parts: [{ text: "Hello" }] }, finishReason: "STOP" as const },
-                ],
+                candidates: [{ content: { role: "model", parts: [{ text: "Hello" }] }, finishReason: "STOP" as const }],
              },
            ),
          ),
@@ -1,5 +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",
@@ -26,9 +26,7 @@ const recorded = recordedTests({
 describe("Google Vertex Gemini recorded", () => {
  recorded.effect("streams text", () =>
    Effect.gen(function* () {
-      const response = yield* LLMClient.generate(
-        LLM.request({ model, prompt: "Reply with exactly one word: hello" }),
-      )
+      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Reply with exactly one word: hello" }))

      expect(response.text.toLowerCase()).toContain("hello")
    }),
@@ -0,0 +1,185 @@
+import { configure } from "@opencode-ai/ai/providers/groq"
+import { describe, expect } from "bun:test"
+import { Effect } from "effect"
+import { LLM, LLMEvent, LLMRequest, LLMResponse, Message, ToolChoice, ToolDefinition } from "../../src/index.js"
+import { LLMClient } from "../../src/route.js"
+import { compileRequest } from "../../src/route/client.js"
+import { recordedTests } from "../recorded-test.js"
+
+const apiKey = process.env.GROQ_API_KEY ?? "fixture"
+const recorded = recordedTests({
+  prefix: "groq-chat",
+  provider: "groq",
+  protocol: "groq-chat",
+  requires: ["GROQ_API_KEY"],
+})
+
+const weather = ToolDefinition.make({
+  name: "lookup_weather",
+  description: "Look up the current weather for a city",
+  inputSchema: {
+    type: "object",
+    properties: { city: { type: "string", enum: ["Paris", "London"] } },
+    required: ["city"],
+    additionalProperties: false,
+  },
+})
+
+describe("Groq recorded", () => {
+  recorded.effect.with(
+    "streams text with usage",
+    { tags: ["text", "usage"], metadata: { model: "openai/gpt-oss-20b" } },
+    () =>
+      Effect.gen(function* () {
+        const request = LLM.request({
+          model: configure({
+            apiKey,
+            providerOptions: {
+              includeReasoning: false,
+              reasoningEffort: "low",
+              serviceTier: "on_demand",
+              user: "recorded-test",
+            },
+          }).model("openai/gpt-oss-20b"),
+          prompt: "Reply with exactly one word: hello",
+          generation: { maxTokens: 512 },
+        })
+        const compiled = yield* compileRequest(request)
+        expect(compiled.body).toMatchObject({
+          max_completion_tokens: 512,
+          stream_options: { include_usage: true },
+          include_reasoning: false,
+          service_tier: "on_demand",
+          user: "recorded-test",
+        })
+        expect(compiled.body.max_tokens).toBeUndefined()
+        expect(compiled.body.store).toBeUndefined()
+        expect(compiled.body.reasoning_format).toBeUndefined()
+
+        const response = yield* LLMClient.generate(request)
+        expect(response.text.toLowerCase().trim()).toBe("hello")
+        expect(response.reasoning).toBe("")
+        expect(response.events.some(LLMEvent.is.textDelta)).toBe(true)
+        expectUsage(response)
+      }),
+    60_000,
+  )
+
+  for (const item of [
+    {
+      name: "continues Qwen parallel tool calls",
+      model: configure({ apiKey, providerOptions: { parallelToolCalls: true, reasoningEffort: "none" } }).model(
+        "qwen/qwen3.6-27b",
+      ),
+      cities: ["Paris", "London"],
+      reasoning: false,
+    },
+    {
+      name: "replays GPT OSS reasoning through a tool loop",
+      model: configure({ apiKey, providerOptions: { includeReasoning: true, reasoningEffort: "low" } }).model(
+        "openai/gpt-oss-20b",
+      ),
+      cities: ["Paris"],
+      reasoning: true,
+    },
+  ]) {
+    recorded.effect.with(
+      item.name,
+      {
+        tags: ["tool", "tool-loop", "usage", item.reasoning ? "reasoning" : "parallel"],
+        metadata: { model: item.model.id },
+      },
+      () =>
+        Effect.gen(function* () {
+          const request = LLM.request({
+            model: item.model,
+            prompt: `Look up the current weather in ${item.cities.join(" and ")}. Call lookup_weather once for each city in the same response before answering. After receiving all results, report each city's weather in one short sentence.`,
+            tools: [weather],
+            toolChoice: "required",
+            generation: { maxTokens: 1536 },
+          })
+          const compiled = yield* compileRequest(request)
+          expect(compiled.body.stream_options).toEqual({ include_usage: true })
+          expect(compiled.body.store).toBeUndefined()
+          expect(compiled.body.reasoning_format).toBe(item.reasoning ? undefined : "parsed")
+          expect(compiled.body.tools[0].function.strict).toBeUndefined()
+          if (!item.reasoning) expect(compiled.body.parallel_tool_calls).toBe(true)
+
+          const first = yield* LLMClient.generate(request)
+          expect(first.finishReason.normalized).toBe("tool-calls")
+          expect(first.toolCalls).toHaveLength(item.cities.length)
+          expect(new Set(first.toolCalls.map((call) => call.id)).size).toBe(item.cities.length)
+          expect(first.toolCalls.map((call) => call.input)).toEqual(
+            expect.arrayContaining(item.cities.map((city) => ({ city }))),
+          )
+          expect(first.toolCalls.every((call) => call.name === "lookup_weather")).toBe(true)
+          expectUsage(first)
+          if (item.reasoning) {
+            expect(first.reasoning.length).toBeGreaterThan(0)
+            expect(first.events.some(LLMEvent.is.reasoningDelta)).toBe(true)
+          }
+
+          const followUp = LLMRequest.update(request, {
+            toolChoice: ToolChoice.make("none"),
+            messages: [
+              ...request.messages,
+              first.message,
+              ...first.toolCalls.map((call) =>
+                Message.tool({ id: call.id, name: call.name, result: { condition: "sunny", temperature: "18C" } }),
+              ),
+            ],
+          })
+          const replay = yield* compileRequest(followUp)
+          if (item.reasoning) {
+            expect(replay.body.messages).toEqual(
+              expect.arrayContaining([expect.objectContaining({ role: "assistant", reasoning: first.reasoning })]),
+            )
+          }
+          expect(replay.body.reasoning_format).toBe(item.reasoning ? undefined : "parsed")
+
+          const second = yield* LLMClient.generate(followUp)
+          expect(second.finishReason.normalized).toBe("stop")
+          expect(second.toolCalls).toHaveLength(0)
+          expect(second.text.toLowerCase()).toContain("sunny")
+          item.cities.forEach((city) => expect(second.text).toContain(city))
+          expectUsage(second)
+        }),
+      60_000,
+    )
+  }
+
+  recorded.effect.with(
+    "streams Qwen parsed reasoning",
+    { tags: ["reasoning", "usage"], metadata: { model: "qwen/qwen3.6-27b" } },
+    () =>
+      Effect.gen(function* () {
+        const request = LLM.request({
+          model: configure({
+            apiKey,
+            providerOptions: { reasoningEffort: "default" },
+          }).model("qwen/qwen3.6-27b"),
+          prompt:
+            "What is 173 multiplied by 219? Think through the arithmetic, then reply with only the final integer.",
+          generation: { maxTokens: 2048 },
+        })
+        const compiled = yield* compileRequest(request)
+        expect(compiled.body).toMatchObject({ reasoning_format: "parsed", reasoning_effort: "default" })
+        expect(compiled.body.include_reasoning).toBeUndefined()
+
+        const response = yield* LLMClient.generate(request)
+        expect(response.text.replaceAll(",", "").trim()).toBe("37887")
+        expect(response.text).not.toContain("<think>")
+        expect(response.reasoning.length).toBeGreaterThan(0)
+        expect(response.events.some(LLMEvent.is.reasoningDelta)).toBe(true)
+        expectUsage(response)
+      }),
+    60_000,
+  )
+})
+
+function expectUsage(response: LLMResponse) {
+  expect(response.usage).toBeDefined()
+  expect(response.usage?.inputTokens).toBeGreaterThan(0)
+  expect(response.usage?.outputTokens).toBeGreaterThan(0)
+  expect(response.events.filter(LLMEvent.is.finish)).toHaveLength(1)
+}
@@ -0,0 +1,112 @@
+import { expect } from "bun:test"
+import { Effect } from "effect"
+import { LanguageModel, LLM, Message } from "../../src/index.js"
+import { OpenAIChat } from "../../src/protocols/openai-chat.js"
+import { Groq } from "../../src/providers/groq.js"
+import { compileRequest } from "../../src/route/client.js"
+import { it } from "../lib/effect.js"
+import { weatherTool } from "../recorded-scenarios.js"
+
+it.effect("Groq reuses Chat streaming and defaults to parsed reasoning", () =>
+  Effect.gen(function* () {
+    expect(Groq.protocol.stream).toBe(OpenAIChat.protocol.stream)
+    const model = Groq.configure({ apiKey: "fixture" }).model("llama-3.3-70b-versatile")
+    expect(model.route.endpoint.baseURL).toBe("https://api.groq.com/openai/v1")
+    const compiled = yield* compileRequest(
+      LLM.request({ model, prompt: "Hello", tools: [weatherTool], generation: { maxTokens: 64 } }),
+    )
+    expect(compiled.body).toMatchObject({
+      max_completion_tokens: 64,
+      stream_options: { include_usage: true },
+      reasoning_format: "parsed",
+    })
+    for (const key of ["store", "max_tokens", "include_reasoning", "parallel_tool_calls", "service_tier", "user"])
+      expect(compiled.body[key]).toBeUndefined()
+    expect(compiled.body.tools?.[0]?.function).not.toHaveProperty("strict")
+  }),
+)
+
+it.effect("Groq lowers its own options for custom catalog identities and endpoints", () =>
+  Effect.gen(function* () {
+    const model = LanguageModel.update(
+      Groq.model("qwen/qwen3.6-27b", {
+        apiKey: "fixture",
+        baseURL: "https://gateway.example/v1",
+        headers: { "x-client": "test" },
+        body: { custom: "value" },
+        providerOptions: {
+          reasoningEffort: "default",
+          parallelToolCalls: true,
+          serviceTier: "flex",
+          user: "test-user",
+        },
+      }),
+      { provider: "custom-groq" },
+    )
+    const compiled = yield* compileRequest(
+      LLM.request({ model, prompt: "Hello", providerOptions: { parallelToolCalls: false, includeReasoning: false } }),
+    )
+    expect(model.route.endpoint.baseURL).toBe("https://gateway.example/v1")
+    expect(model.route.defaults.headers).toEqual({ "x-client": "test" })
+    expect(model.route.defaults.http?.body).toEqual({ custom: "value" })
+    expect(compiled.body).toMatchObject({
+      reasoning_effort: "default",
+      reasoning_format: "parsed",
+      parallel_tool_calls: false,
+      service_tier: "flex",
+      user: "test-user",
+    })
+    expect(compiled.body.include_reasoning).toBeUndefined()
+    for (const key of ["reasoningFormat", "reasoningEffort", "parallelToolCalls", "serviceTier"])
+      expect(compiled.body).not.toHaveProperty(key)
+  }),
+)
+
+it.effect("Groq replays reasoning only when present and preserves explicit reasoning exclusion", () =>
+  Effect.gen(function* () {
+    const compiled = yield* compileRequest(
+      LLM.request({
+        model: Groq.configure({ apiKey: "fixture" }).model("openai/gpt-oss-20b"),
+        messages: [
+          Message.user("Think"),
+          Message.assistant([
+            { type: "reasoning", text: "Thinking" },
+            { type: "text", text: "Answer" },
+          ]),
+          Message.user("Again"),
+          Message.assistant("Answer only"),
+          Message.user("Continue"),
+        ],
+        providerOptions: { reasoningEffort: "low", includeReasoning: false },
+      }),
+    )
+    expect(compiled.body).toMatchObject({ reasoning_effort: "low", include_reasoning: false })
+    expect(compiled.body.reasoning_format).toBeUndefined()
+    expect(compiled.body.messages[1]).toMatchObject({ reasoning: "Thinking", content: "Answer" })
+    expect(compiled.body.messages[1]).not.toHaveProperty("reasoning_content")
+    expect(compiled.body.messages[3]).not.toHaveProperty("reasoning")
+  }),
+)
+
+it.effect("Groq omits reasoning_format for the GPT-OSS family by default", () =>
+  Effect.gen(function* () {
+    for (const id of ["openai/gpt-oss-20b", "openai/gpt-oss-120b", "openai/gpt-oss-safeguard-20b"]) {
+      const compiled = yield* compileRequest(
+        LLM.request({ model: Groq.configure({ apiKey: "fixture" }).model(id), prompt: "Hello" }),
+      )
+      expect(compiled.body.reasoning_format).toBeUndefined()
+      expect(compiled.body.include_reasoning).toBeUndefined()
+    }
+  }),
+)
+
+it.effect("Groq validates option types", () =>
+  Effect.gen(function* () {
+    for (const providerOptions of [{ includeReasoning: "false" }, { parallelToolCalls: "false" }]) {
+      const error = yield* compileRequest(
+        LLM.request({ model: Groq.configure({ apiKey: "fixture" }).model("qwen"), prompt: "Hello", providerOptions }),
+      ).pipe(Effect.flip)
+      expect(error.reason._tag).toBe("InvalidRequest")
+    }
+  }),
+)
@@ -0,0 +1,190 @@
+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, Groq, 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",
+        },
+        {
+          model: Groq.configure().model("llama"),
+          env: { GROQ_API_KEY: "groq-secret" },
+          token: "groq-secret",
+          url: "https://api.groq.com/openai/v1/chat/completions",
+        },
+      ]
+
+      yield* Effect.forEach(scenarios, (scenario) =>
+        LLM.generate(LLM.request({ model: scenario.model, prompt: "Say hello." })).pipe(
+          Effect.provide(
+            dynamicResponse((input) =>
+              Effect.gen(function* () {
+                const request = yield* HttpClientRequest.toWeb(input.request).pipe(Effect.orDie)
+                expect(request.url).toBe(scenario.url)
+                expect(request.headers.get("authorization")).toBe(`Bearer ${scenario.token}`)
+                return input.respond(
+                  sseEvents(
+                    { id: "chatcmpl_fixture", choices: [{ delta: { content: "Hello" }, finish_reason: null }] },
+                    { id: "chatcmpl_fixture", choices: [{ delta: {}, finish_reason: "stop" }] },
+                  ),
+                  { headers: { "content-type": "text/event-stream" } },
+                )
+              }),
+            ),
+          ),
+          Effect.provide(ConfigProvider.layer(ConfigProvider.fromEnv({ env: scenario.env }))),
+          Effect.tap((response) => Effect.sync(() => expect(response.text).toBe("Hello"))),
+        ),
+      )
+    }),
+  )
+})
@@ -85,6 +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(
@@ -147,6 +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(
@@ -366,6 +438,35 @@ describe("OpenAI Chat route", () => {
    }),
  )

+  it.effect("limits OpenAI and Azure Chat tool call IDs to 40 characters", () =>
+    Effect.gen(function* () {
+      const id = `call_${"a".repeat(48)}`
+      const models = [
+        model,
+        Azure.configure({ baseURL: "https://opencode-test.openai.azure.com/openai/", apiKey: "test" }).chat("gpt-4o"),
+      ]
+
+      yield* Effect.forEach(models, (selected) =>
+        Effect.gen(function* () {
+          const prepared = yield* compileRequest(
+            LLM.request({
+              model: selected,
+              messages: [
+                Message.assistant([ToolCallPart.make({ id, name: "lookup", input: {} })]),
+                Message.tool({ id, name: "lookup", result: "Sunny" }),
+              ],
+            }),
+          )
+
+          expect(prepared.body.messages).toMatchObject([
+            { role: "assistant", tool_calls: [{ id: id.slice(0, 40) }] },
+            { role: "tool", tool_call_id: id.slice(0, 40) },
+          ])
+        }),
+      )
+    }),
+  )
+
  it.effect("preserves structured tool errors for the model", () =>
    Effect.gen(function* () {
      const error = { error: { type: "unknown", message: "Tool execution interrupted" } }
@@ -431,6 +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(
@@ -238,6 +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(
@@ -47,10 +47,8 @@ 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"],
@@ -84,10 +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." },
@@ -195,15 +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" } } },
            ]),
          ],
        }),
@@ -218,9 +222,62 @@ 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]) }] },
      ])
    }),
  )
@@ -250,6 +307,43 @@ describe("Open Responses-compatible route", () => {
    }),
  )

+  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({
@@ -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." }] },
@@ -112,10 +112,8 @@ describe("OpenAI Responses route", () => {

      expect(prepared.body).toEqual({
        model: "gpt-4.1-mini",
-        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.",
        store: false,
        include: ["reasoning.encrypted_content"],
        stream: true,
@@ -469,7 +467,7 @@ describe("OpenAI Responses route", () => {
    }),
  )

-  it.effect("continues a tool call with only the new tool output", () =>
+  it.effect("continues an item-id-less tool call with only the new tool output", () =>
    Effect.gen(function* () {
      const firstRequest = {
        type: "response.create",
@@ -485,7 +483,6 @@ describe("OpenAI Responses route", () => {
          type: "response.output_item.done",
          item: {
            type: "function_call",
-            id: "fc_1",
            status: "completed",
            call_id: "call_1",
            name: "weather",
@@ -1597,8 +1594,8 @@ describe("OpenAI Responses route", () => {
      )

      expect(prepared.body).toMatchObject({
+        instructions: "You are concise. Continue from the provided history.",
        input: [
-          { role: "system", content: "You are concise. Continue from the provided history." },
          {
            role: "user",
            content: [
@@ -2120,6 +2117,47 @@ describe("OpenAI Responses route", () => {
    }),
  )

+  it.effect("routes item-id-less function arguments by output index and prefers item completion", () =>
+    Effect.gen(function* () {
+      const item = { type: "function_call", call_id: "call_1", name: "lookup", arguments: "" }
+      const response = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              { type: "response.output_item.added", output_index: 2, item },
+              {
+                type: "response.function_call_arguments.delta",
+                output_index: 2,
+                item_id: "opaque_delta",
+                delta: '{"query":"streamed"}',
+              },
+              {
+                type: "response.function_call_arguments.done",
+                output_index: 2,
+                item_id: "opaque_done",
+                arguments: '{"query":"arguments-done"}',
+              },
+              {
+                type: "response.output_item.done",
+                output_index: 2,
+                item: { ...item, arguments: '{"query":"output-item-done"}' },
+              },
+              { type: "response.completed", response: { id: "resp_1" } },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.events.filter((event) => event.type === "tool-input-delta")).toMatchObject([
+        { id: "call_1", text: '{"query":"streamed"}' },
+      ])
+      expect(response.events.filter(LLMEvent.is.toolCall)).toEqual([
+        expect.objectContaining({ id: "call_1", name: "lookup", input: { query: "output-item-done" } }),
+      ])
+      expect(response.events.find(LLMEvent.is.toolCall)?.providerMetadata).toBeUndefined()
+    }),
+  )
+
  it.effect("routes reasoning summary events by output index", () =>
    Effect.gen(function* () {
      const response = yield* LLMClient.generate(request).pipe(
@@ -2298,13 +2336,25 @@ describe("OpenAI Responses route", () => {
  it.effect("rejects function argument events without the spec-required item id", () =>
    Effect.gen(function* () {
      const events = [
-        { type: "response.function_call_arguments.delta", delta: "{}" },
-        { type: "response.function_call_arguments.done", arguments: "{}" },
+        { type: "response.function_call_arguments.delta", output_index: 0, delta: "{}" },
+        { type: "response.function_call_arguments.done", output_index: 0, arguments: "{}" },
      ]

      for (const event of events) {
        const error = yield* LLMClient.generate(request).pipe(
-          Effect.provide(fixedResponse(sseEvents(event, { type: "response.completed", response: { id: "resp_1" } }))),
+          Effect.provide(
+            fixedResponse(
+              sseEvents(
+                {
+                  type: "response.output_item.added",
+                  output_index: 0,
+                  item: { type: "function_call", call_id: "call_1", name: "lookup", arguments: "" },
+                },
+                event,
+                { type: "response.completed", response: { id: "resp_1" } },
+              ),
+            ),
+          ),
          Effect.flip,
        )

@@ -2758,7 +2808,7 @@ describe("OpenAI Responses route", () => {
    }),
  )

-  it.effect("closes reasoning summary parts when storage is not disabled", () =>
+  it.effect("preserves final reasoning metadata when storage is enabled", () =>
    Effect.gen(function* () {
      const response = yield* LLMClient.generate(LLMRequest.update(request, { providerOptions: { store: true } })).pipe(
        Effect.provide(
@@ -2776,7 +2826,7 @@ describe("OpenAI Responses route", () => {
              { type: "response.reasoning_summary_part.done", item_id: "rs_1", summary_index: 1 },
              {
                type: "response.output_item.done",
-                item: { type: "reasoning", id: "rs_1", encrypted_content: null },
+                item: { type: "reasoning", id: "rs_1", encrypted_content: "encrypted-state" },
              },
              { type: "response.completed", response: { id: "resp_1" } },
            ),
@@ -2786,7 +2836,11 @@ describe("OpenAI Responses route", () => {

      expect(response.events.filter((event) => event.type === "reasoning-end")).toEqual([
        { type: "reasoning-end", id: "rs_1:0", providerMetadata: { openai: { itemId: "rs_1" } } },
-        { type: "reasoning-end", id: "rs_1:1", providerMetadata: { openai: { itemId: "rs_1" } } },
+        {
+          type: "reasoning-end",
+          id: "rs_1:1",
+          providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
+        },
      ])
    }),
  )
@@ -2891,7 +2945,7 @@ describe("OpenAI Responses route", () => {
    }),
  )

-  it.effect("references stored reasoning items by id", () =>
+  it.effect("replays complete reasoning items when storage is enabled", () =>
    Effect.gen(function* () {
      const prepared = yield* compileRequest(
        LLM.request({
@@ -2901,7 +2955,7 @@ describe("OpenAI Responses route", () => {
              {
                type: "reasoning",
                text: "Checked the previous diff.",
-                providerMetadata: { openai: { itemId: "rs_1" } },
+                providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
              },
            ]),
          ],
@@ -2909,12 +2963,20 @@ describe("OpenAI Responses route", () => {
        }),
      )

-      expect(prepared.body.input).toEqual([{ type: "item_reference", id: "rs_1" }])
+      expect(prepared.body.input).toEqual([
+        {
+          type: "reasoning",
+          id: "rs_1",
+          summary: [{ type: "summary_text", text: "Checked the previous diff." }],
+          encrypted_content: "encrypted-state",
+        },
+      ])
    }),
  )

-  it.effect("references stored provider-executed hosted tool results by id", () =>
+  it.effect("replays complete hosted tool items when storage is enabled", () =>
    Effect.gen(function* () {
+      const item = { type: "web_search_call", id: "ws_1", status: "completed" }
      const prepared = yield* compileRequest(
        LLM.request({
          model,
@@ -2931,7 +2993,7 @@ describe("OpenAI Responses route", () => {
                type: "tool-result",
                id: "ws_1",
                name: "web_search",
-                result: { type: "json", value: { type: "web_search_call", id: "ws_1", status: "completed" } },
+                result: { type: "json", value: item },
                providerExecuted: true,
                providerMetadata: { openai: { itemId: "ws_1" } },
              },
@@ -2943,14 +3005,15 @@ describe("OpenAI Responses route", () => {
      )

      expect(prepared.body.input).toEqual([
-        { type: "item_reference", id: "ws_1" },
+        item,
        { role: "user", content: [{ type: "input_text", text: "Continue." }] },
      ])
    }),
  )

-  it.effect("continues stateless hosted tool results with their text form", () =>
+  it.effect("replays stateless hosted tool results as native provider items", () =>
    Effect.gen(function* () {
+      const item = { type: "web_search_call", id: "ws_1", status: "completed" }
      const prepared = yield* compileRequest(
        LLM.request({
          model,
@@ -2968,7 +3031,7 @@ describe("OpenAI Responses route", () => {
                type: "tool-result",
                id: "ws_1",
                name: "web_search",
-                result: { type: "json", value: { type: "web_search_call", id: "ws_1", status: "completed" } },
+                result: { type: "json", value: item },
                providerExecuted: true,
                providerMetadata: { openai: { itemId: "ws_1" } },
              },
@@ -2981,6 +3044,74 @@ describe("OpenAI Responses route", () => {

      expect(prepared.body.input).toEqual([
        { role: "user", content: [{ type: "input_text", text: "Search." }] },
+        { type: "web_search_call", id: "ws_1", status: "completed" },
+        { role: "user", content: [{ type: "input_text", text: "Continue." }] },
+      ])
+    }),
+  )
+
+  it.effect("replays OpenAI hosted tool extensions but rejects foreign and unknown items", () =>
+    Effect.gen(function* () {
+      const items = [
+        { type: "computer_call", id: "computer_1", status: "completed", action: { type: "click", x: 1, y: 2 } },
+        { type: "x_search_call", id: "x_search_1", status: "completed" },
+        { type: "future_call", id: "future_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: { openai: { 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]) }] },
+      ])
+    }),
+  )
+
+  it.effect("preserves foreign hosted tool results as portable message content when storage is enabled", () =>
+    Effect.gen(function* () {
+      const item = { type: "web_search_call", id: "ws_1", status: "completed" }
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: xaiModel,
+          messages: [
+            Message.assistant([
+              ToolCallPart.make({
+                id: "ws_1",
+                name: "web_search",
+                input: { query: "effect 4" },
+                providerExecuted: true,
+                providerMetadata: { openai: { itemId: "ws_1" } },
+              }),
+              {
+                type: "tool-result",
+                id: "ws_1",
+                name: "web_search",
+                result: { type: "json", value: item },
+                providerExecuted: true,
+                providerMetadata: { openai: { itemId: "ws_1" } },
+              },
+            ]),
+            Message.user("Continue."),
+          ],
+          providerOptions: { store: true },
+        }),
+      )
+
+      expect(prepared.body.input).toEqual([
        {
          role: "user",
          content: [{ type: "input_text", text: '{"type":"web_search_call","id":"ws_1","status":"completed"}' }],
@@ -2990,38 +3121,79 @@ describe("OpenAI Responses route", () => {
    }),
  )

-  it.effect("drops replayed item ids outside the server's grammar", () =>
+  it.effect("does not replay hosted tool items whose result id differs from provider metadata", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model,
+          messages: [
+            Message.assistant([
+              {
+                type: "tool-result",
+                id: "ws_1",
+                name: "web_search",
+                result: { type: "json", value: { type: "web_search_call", id: "ws_other", status: "completed" } },
+                providerExecuted: true,
+                providerMetadata: { openai: { itemId: "ws_1" } },
+              },
+            ]),
+          ],
+        }),
+      )
+
+      expect(prepared.body.input).toEqual([
+        {
+          role: "user",
+          content: [{ type: "input_text", text: '{"type":"web_search_call","id":"ws_other","status":"completed"}' }],
+        },
+      ])
+    }),
+  )
+
+  it.effect("preserves provider-issued item ids and removes malformed ids without dropping items", () =>
    Effect.gen(function* () {
      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
            Message.assistant([
-              // Fails the message id prefix.
              {
                type: "text",
                text: "Hello",
                providerMetadata: { openai: { itemId: "history_1" } },
              },
-              // Oversized for the Responses item id limit.
              {
                type: "text",
                text: "World",
-                providerMetadata: { openai: { itemId: `m${"a".repeat(64)}` } },
+                providerMetadata: { openai: { itemId: `message_${"a".repeat(64)}` } },
              },
-              // Fails the reasoning id prefix, so the whole item is unreplayable
-              // statelessly and is skipped rather than sent malformed.
              {
                type: "reasoning",
                text: "Checked the diff.",
                providerMetadata: { openai: { itemId: "thinking_1", reasoningEncryptedContent: "encrypted-state" } },
              },
+              {
+                type: "reasoning",
+                text: "Missing suffix.",
+                providerMetadata: { openai: { itemId: "rs_", reasoningEncryptedContent: "another-state" } },
+              },
+              {
+                type: "reasoning",
+                text: "No prefix separator.",
+                providerMetadata: { openai: { itemId: "550e8400-e29b-41d4-a716-446655440000" } },
+              },
              ToolCallPart.make({
                id: "call_1",
                name: "lookup",
                input: { query: "weather" },
                providerMetadata: { openai: { itemId: "toolu_01A" } },
              }),
+              ToolCallPart.make({
+                id: "call_2",
+                name: "lookup",
+                input: { query: "news" },
+                providerMetadata: { openai: { itemId: "fc_" } },
+              }),
            ]),
          ],
        }),
@@ -3030,41 +3202,70 @@ describe("OpenAI Responses route", () => {
      expect(prepared.body.input).toEqual([
        {
          type: "message",
+          id: "history_1",
          role: "assistant",
-          content: [
-            { type: "output_text", text: "Hello" },
-            { type: "output_text", text: "World" },
-          ],
+          content: [{ type: "output_text", text: "Hello" }],
+        },
+        {
+          type: "message",
+          id: `message_${"a".repeat(64)}`,
+          role: "assistant",
+          content: [{ type: "output_text", text: "World" }],
+        },
+        {
+          type: "reasoning",
+          id: "thinking_1",
+          summary: [{ type: "summary_text", text: "Checked the diff." }],
+          encrypted_content: "encrypted-state",
+        },
+        {
+          type: "reasoning",
+          summary: [{ type: "summary_text", text: "Missing suffix." }],
+          encrypted_content: "another-state",
+        },
+        {
+          type: "reasoning",
+          summary: [{ type: "summary_text", text: "No prefix separator." }],
        },
        {
          type: "function_call",
+          id: "toolu_01A",
          call_id: "call_1",
          name: "lookup",
          arguments: '{"query":"weather"}',
        },
+        {
+          type: "function_call",
+          call_id: "call_2",
+          name: "lookup",
+          arguments: '{"query":"news"}',
+        },
      ])
    }),
  )

-  it.effect("keeps well-formed hosted references and drops malformed ones under storage", () =>
+  it.effect("falls back to portable hosted results when stored item metadata is malformed", () =>
    Effect.gen(function* () {
-      const hostedResult = (itemId: string) => [
-        ToolCallPart.make({
-          id: itemId,
-          name: "web_search",
-          input: { query: "effect 4" },
-          providerExecuted: true,
-          providerMetadata: { openai: { itemId } },
-        }),
-        {
-          type: "tool-result" as const,
-          id: itemId,
-          name: "web_search",
-          result: { type: "json" as const, value: { status: "completed" } },
-          providerExecuted: true as const,
-          providerMetadata: { openai: { itemId } },
-        },
-      ]
+      const hostedResult = (itemId: string) => {
+        const item = { type: "web_search_call", id: itemId, status: "completed" }
+        return [
+          ToolCallPart.make({
+            id: itemId,
+            name: "web_search",
+            input: { query: "effect 4" },
+            providerExecuted: true,
+            providerMetadata: { openai: { itemId } },
+          }),
+          {
+            type: "tool-result" as const,
+            id: itemId,
+            name: "web_search",
+            result: { type: "json" as const, value: item },
+            providerExecuted: true as const,
+            providerMetadata: { openai: { itemId } },
+          },
+        ]
+      }
      const prepared = yield* compileRequest(
        LLM.request({
          model,
@@ -3073,7 +3274,13 @@ describe("OpenAI Responses route", () => {
        }),
      )

-      expect(prepared.body.input).toEqual([{ type: "item_reference", id: "ws_1" }])
+      expect(prepared.body.input).toEqual([
+        { type: "web_search_call", id: "ws_1", status: "completed" },
+        {
+          role: "user",
+          content: [{ type: "input_text", text: '{"type":"web_search_call","id":"bad ref","status":"completed"}' }],
+        },
+      ])
    }),
  )

@@ -3119,6 +3326,43 @@ describe("OpenAI Responses route", () => {
    }),
  )

+  it.effect("preserves foreign hosted images as portable image content when storage is enabled", () =>
+    Effect.gen(function* () {
+      const item = { type: "image_generation_call", id: "ig_1", status: "completed", result: "AQID" }
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: xaiModel,
+          messages: [
+            Message.assistant([
+              ToolCallPart.make({
+                id: "ig_1",
+                name: "image_generation",
+                input: {},
+                providerExecuted: true,
+                providerMetadata: { openai: { itemId: "ig_1" } },
+              }),
+              ToolResultPart.make({
+                id: "ig_1",
+                name: "image_generation",
+                result: {
+                  type: "content",
+                  value: [{ type: "file", uri: "data:image/png;base64,AQID", mime: "image/png" }],
+                },
+                providerExecuted: true,
+                providerMetadata: { openai: { itemId: "ig_1" } },
+              }),
+            ]),
+          ],
+          providerOptions: { store: true },
+        }),
+      )
+
+      expect(prepared.body.input).toEqual([
+        { role: "user", content: [{ type: "input_image", image_url: "data:image/png;base64,AQID" }] },
+      ])
+    }),
+  )
+
  it.effect("joins streamed summary blocks into one continuation reasoning item", () =>
    Effect.gen(function* () {
      const prepared = yield* compileRequest(
@@ -3301,6 +3545,43 @@ describe("OpenAI Responses route", () => {
    }),
  )

+  it.effect("finalizes and replays a completed function call without an optional item id", () =>
+    Effect.gen(function* () {
+      const response = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              {
+                type: "response.output_item.done",
+                item: { type: "function_call", call_id: "call_1", name: "lookup", arguments: '{"query":"weather"}' },
+              },
+              { type: "response.completed", response: { id: "resp_1" } },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.events.filter(LLMEvent.is.toolCall)).toEqual([
+        expect.objectContaining({ id: "call_1", name: "lookup", input: { query: "weather" } }),
+      ])
+      expect(response.events.find(LLMEvent.is.toolCall)?.providerMetadata).toBeUndefined()
+
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model,
+          messages: [
+            response.message,
+            Message.tool({ id: "call_1", name: "lookup", resultType: "json", result: { forecast: "sunny" } }),
+          ],
+        }),
+      )
+      expect(prepared.body.input).toEqual([
+        { type: "function_call", call_id: "call_1", name: "lookup", arguments: '{"query":"weather"}' },
+        { type: "function_call_output", call_id: "call_1", output: '{"forecast":"sunny"}' },
+      ])
+    }),
+  )
+
  it.effect("emits only missing function arguments from the arguments done event", () =>
    Effect.gen(function* () {
      const body = sseEvents(
@@ -3525,6 +3806,37 @@ describe("OpenAI Responses route", () => {
    }),
  )

+  it.effect("reconciles an item-id-less pending function call from completed response output", () =>
+    Effect.gen(function* () {
+      const item = { type: "function_call", call_id: "call_1", name: "lookup", arguments: "" }
+      const response = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              { type: "response.output_item.added", output_index: 0, item },
+              {
+                type: "response.function_call_arguments.delta",
+                output_index: 0,
+                item_id: "opaque_delta",
+                delta: '{"query":"partial',
+              },
+              {
+                type: "response.completed",
+                response: { id: "resp_1", output: [{ ...item, arguments: '{"query":"complete"}' }] },
+              },
+            ),
+          ),
+        ),
+      )
+
+      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()
+      expect(response.events.filter(LLMEvent.is.toolInputEnd)).toHaveLength(1)
+    }),
+  )
+
  it.effect("lets completed response output override arguments done", () =>
    Effect.gen(function* () {
      const body = sseEvents(
@@ -3828,13 +4140,15 @@ describe("OpenAI Responses route", () => {
    }),
  )

-  it.effect("decodes image generation output as image content", () =>
+  it.effect("replays hosted image results as portable content regardless of storage", () =>
    Effect.gen(function* () {
      const item = {
        type: "image_generation_call",
        id: "ig_1",
        status: "completed",
        result: "AQID",
+        action: "generate",
+        output_format: "png",
      }
      const response = yield* LLMClient.generate(request).pipe(
        Effect.provide(
@@ -3847,6 +4161,9 @@ describe("OpenAI Responses route", () => {
        ),
      )

+      expect(response.events.find(LLMEvent.is.toolCall)).toMatchObject({
+        providerMetadata: { openai: { itemId: "ig_1" } },
+      })
      expect(response.events.find(LLMEvent.is.toolResult)).toMatchObject({
        id: "ig_1",
        name: "image_generation",
@@ -3855,7 +4172,52 @@ describe("OpenAI Responses route", () => {
          type: "content",
          value: [{ type: "file", uri: "data:image/png;base64,AQID", mime: "image/png" }],
        },
+        providerMetadata: { openai: { itemId: "ig_1" } },
      })
+
+      const prepared = yield* Effect.forEach([false, true], (store) =>
+        compileRequest(LLM.request({ model, messages: [response.message], providerOptions: { store } })),
+      )
+      expect(prepared.map((request) => request.body.input)).toEqual([
+        [{ role: "user", content: [{ type: "input_image", image_url: "data:image/png;base64,AQID" }] }],
+        [{ role: "user", content: [{ type: "input_image", image_url: "data:image/png;base64,AQID" }] }],
+      ])
+    }),
+  )
+
+  it.effect("preserves failed hosted tool results as portable error content", () =>
+    Effect.gen(function* () {
+      const item = {
+        type: "web_search_call",
+        id: "ws_failed",
+        status: "failed",
+        error: { code: "search_failed", message: "Search unavailable" },
+      }
+      const response = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              { type: "response.output_item.done", item },
+              { type: "response.completed", response: { id: "resp_1" } },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.events.find(LLMEvent.is.toolResult)).toMatchObject({
+        result: { type: "error", value: item.error },
+        providerMetadata: { openai: { itemId: "ws_failed" } },
+      })
+
+      const prepared = yield* compileRequest(
+        LLM.request({ model, messages: [response.message], providerOptions: { store: true } }),
+      )
+      expect(prepared.body.input).toEqual([
+        {
+          role: "user",
+          content: [{ type: "input_text", text: '{"code":"search_failed","message":"Search unavailable"}' }],
+        },
+      ])
    }),
  )

@@ -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"
@@ -14,9 +14,9 @@ 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" }))
@@ -106,16 +106,70 @@ describe("xAI Responses route", () => {
    }),
  )

+  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: "x_search_call", id: "x_search_1", status: "completed", action: { query: "news" } },
-              },
+              { type: "response.output_item.done", item },
              { type: "response.completed", response: { id: "response_1" } },
            ),
          ),
@@ -127,6 +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" } },
      })
    }),
  )
@@ -80,7 +80,9 @@ 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"),
+      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(
@@ -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)
@@ -24,11 +24,17 @@ test("session settings use the remote server context", async ({ page }) => {
  await configureServers(page)

  await page.goto(`/server/${base64Encode(serverB)}/session/${sessionB.id}`)
-  await expect(page.getByRole("heading", { name: sessionB.title, exact: true })).toBeVisible()
+  const sessionHeading = page.getByRole("heading", { name: sessionB.title, exact: true, includeHidden: true })
+  await expect(sessionHeading).toBeVisible()
  await page.keyboard.press("Control+,")

-  const dialog = page.locator(".settings-dialog")
-  const autoAccept = dialog.locator('[data-action="settings-auto-accept-permissions"]')
+  const settings = page.getByTestId("settings-screen")
+  await expect(settings).toBeVisible()
+  await expect(page.getByRole("dialog")).toHaveCount(0)
+  await expect(settings.getByRole("tablist")).toHaveCSS("width", "328px")
+  await expect(sessionHeading).toBeAttached()
+  await expect(sessionHeading).toBeHidden()
+  const autoAccept = settings.locator('[data-action="settings-auto-accept-permissions"]')
  const input = autoAccept.getByRole("switch")
  await expect(autoAccept).toBeVisible()
  await expect(input).toBeEnabled()
@@ -55,9 +61,12 @@ test("session settings use the remote server context", async ({ page }) => {
      },
    ])

-  await dialog.getByRole("tab", { name: "Models" }).click()
-  await expect(dialog.getByRole("switch", { name: "Server B Model" })).toBeEnabled()
-  await expect(dialog.getByRole("switch", { name: "Server A Model" })).toHaveCount(0)
+  await settings.getByRole("tab", { name: "Models" }).click()
+  await expect(settings.getByRole("switch", { name: "Server B Model" })).toBeEnabled()
+  await expect(settings.getByRole("switch", { name: "Server A Model" })).toHaveCount(0)
+  await settings.getByRole("button", { name: "Back to app" }).click()
+  await expect(settings).toBeHidden()
+  await expect(sessionHeading).toBeVisible()
 })

 test("auto-accept responds for an unfocused server session", async ({ page }) => {
@@ -78,7 +87,7 @@ test("auto-accept responds for an unfocused server session", async ({ page }) =>
  await page.goto(`/server/${base64Encode(serverA)}/session/${sessionA.id}`)
  await expect(page.getByRole("heading", { name: sessionA.title, exact: true })).toBeVisible()
  await page.keyboard.press("Control+,")
-  const autoAccept = page.locator(".settings-dialog").locator('[data-action="settings-auto-accept-permissions"]')
+  const autoAccept = page.getByTestId("settings-screen").locator('[data-action="settings-auto-accept-permissions"]')
  await autoAccept.locator('[data-slot="switch-control"]').click()
  await expect(autoAccept.getByRole("switch")).toBeChecked()
  await expect
@@ -178,7 +187,7 @@ test("auto-accept sweeps again after a reconnect", async ({ page }) => {
  const first = await transport.waitForConnection()

  await page.keyboard.press("Control+,")
-  const autoAccept = page.locator(".settings-dialog").locator('[data-action="settings-auto-accept-permissions"]')
+  const autoAccept = page.getByTestId("settings-screen").locator('[data-action="settings-auto-accept-permissions"]')
  await autoAccept.locator('[data-slot="switch-control"]').click()
  await expect(autoAccept.getByRole("switch")).toBeChecked()
  await expect
@@ -234,7 +243,7 @@ test("auto-accept approves a request discovered by opening a session", async ({
  await expect(page.getByRole("heading", { name: sessionA.title, exact: true })).toBeVisible()

  await page.keyboard.press("Control+,")
-  const autoAccept = page.locator(".settings-dialog").locator('[data-action="settings-auto-accept-permissions"]')
+  const autoAccept = page.getByTestId("settings-screen").locator('[data-action="settings-auto-accept-permissions"]')
  await autoAccept.locator('[data-slot="switch-control"]').click()
  await expect(autoAccept.getByRole("switch")).toBeChecked()

@@ -2,7 +2,7 @@ import { expect, test, type Route } from "@playwright/test"

 const server = "http://127.0.0.1:4097"

-test("nested server dialog keeps focus inside the top layer", async ({ page }) => {
+test("server dialog keeps focus above fullscreen settings", async ({ page }) => {
  await page.addInitScript((server) => {
    localStorage.setItem("opencode.global.dat:server", JSON.stringify({ list: [server] }))
  }, server)
@@ -24,8 +24,9 @@ test("nested server dialog keeps focus inside the top layer", async ({ page }) =

  await page.goto("/")
  await page.keyboard.press("Control+,")
-  const settings = page.locator(".settings-dialog")
+  const settings = page.getByTestId("settings-screen")
  await expect(settings).toBeVisible()
+  await expect(page.getByRole("dialog")).toHaveCount(0)
  await settings.getByRole("tab", { name: "Servers" }).click()
  await settings.getByRole("button", { name: "Add server" }).click()

@@ -41,6 +42,9 @@ test("nested server dialog keeps focus inside the top layer", async ({ page }) =
  await expect(password).toBeFocused()
  await password.fill("secret")
  await expect(password).toHaveValue("secret")
+  await page.keyboard.press("Escape")
+  await expect(editor).toBeHidden()
+  await expect(settings).toBeVisible()
 })

 function json(route: Route, body: unknown, status = 200) {
@@ -9,7 +9,7 @@ test("space activates a focused timeline button instead of scrolling", async ({
    reducedMotion: true,
  })
  const scroller = page.locator(".scroll-view__viewport", { has: page.locator("[data-timeline-row]") })
-  const trigger = page.locator(`[data-timeline-part-id="${shellID}"] [data-slot="collapsible-trigger"]`)
+  const trigger = page.getByRole("button", { name: "Used Shell" })
  await trigger.focus()
  const before = await scroller.evaluate((element) => element.scrollTop)
  await trigger.press("Space")
@@ -40,7 +40,7 @@ test.describe("regression: session timeline context group resize", () => {
    expect(samples.at(-1)?.expanded).toBe("true")
  })

-  test("paints a stable exploring to explored transition", async ({ page }) => {
+  test("keeps a grouped tool summary stable as its calls complete", async ({ page }) => {
    const events: OpenCodeEvent[] = []
    await page.setViewportSize({ width: 1400, height: 900 })
    await mockServer(page, events, [
@@ -55,13 +55,12 @@ test.describe("regression: session timeline context group resize", () => {
    await devtools.send("Emulation.setCPUThrottlingRate", { rate: 4 })
    const context = page.locator(`[data-timeline-part-ids="${contextIDs.join(",")}"]`).first()
    await expectAppVisible(context)
-    await expect(context.locator('[data-component="tool-status-title"]')).toHaveAttribute("aria-label", "Exploring")
+    await expect(context.getByRole("button")).toHaveAccessibleName("Used Read, Glob, Grep, List")

    const contextSelector = `[data-timeline-part-ids="${contextIDs.join(",")}"]`
    const regions = defineVisualRegions({
      status: {
-        selector: `${contextSelector} [data-component="tool-status-title"]`,
-        opacitySelectors: ['[data-slot="tool-status-active"]', '[data-slot="tool-status-done"]'],
+        selector: `${contextSelector} [data-component="context-tool-group-trigger"]`,
      },
      context: { selector: contextSelector, closest: '[data-timeline-row="AssistantPart"]' },
      following: {
@@ -89,7 +88,7 @@ test.describe("regression: session timeline context group resize", () => {
      await page.waitForTimeout(delay)
    }

-    await expect(context.locator('[data-component="tool-status-title"]')).toHaveAttribute("aria-label", "Explored")
+    await expect(context.getByRole("button")).toHaveAccessibleName("Used Read, Glob, Grep, List")
    await page.waitForTimeout(700)
    const trace = await stopVisualProbe<keyof typeof regions>(page)
    const labels = trace.samples
@@ -108,7 +107,7 @@ test.describe("regression: session timeline context group resize", () => {
      ]),
    )

-    expect(labels).toEqual(["Exploring", "Explored"])
+    expect(labels).toEqual(["Used Read, Glob, Grep, List"])
    expect(issues, JSON.stringify(trace.samples, null, 2)).toEqual([])
  })
 })
@@ -209,13 +208,7 @@ function turn(index: number, target: boolean, status: "running" | "completed" =
  const content: SessionMessageAssistant["content"] = target
    ? [
        toolContent(
-          contextTool(
-            contextIDs[0]!,
-            assistantID,
-            "read",
-            { path: "src/recent-a.ts", offset: 0, limit: 120 },
-            status,
-          ),
+          contextTool(contextIDs[0]!, assistantID, "read", { path: "src/recent-a.ts", offset: 0, limit: 120 }, status),
        ),
        toolContent(contextTool(contextIDs[1]!, assistantID, "glob", { path: directory, pattern: "**/*.ts" }, status)),
        toolContent(
@@ -83,6 +83,7 @@ test("keeps an expanded file diff header at the same viewport position", async (
  const before = Array.from({ length: 80 }, (_, index) => `export const value${index} = ${index}\n`).join("")
  const after = before.replaceAll(" = ", " = compute(").replaceAll("\n", ")\n")
  await setupTimeline(page, {
+    settings: { editToolPartsExpanded: true },
    messages: [
      userMessage([userText("Preceding context ".repeat(120))]),
      assistantMessage([
@@ -26,7 +26,9 @@ for (const expanded of [false, true]) {
      messages: [userMessage(), assistantMessage([shell(id, "completed", lines(3))])],
      settings: { shellToolPartsExpanded: expanded },
    })
-    const trigger = page.locator(`[data-timeline-part-id="${id}"] [data-slot="collapsible-trigger"]`)
+    const trigger = expanded
+      ? page.locator(`[data-timeline-part-id="${id}"] [data-slot="collapsible-trigger"]`)
+      : page.getByRole("button", { name: "Used Shell" })
    await expect(trigger).toHaveAttribute("aria-expanded", String(expanded))
    await trigger.click()
    await expect(trigger).toHaveAttribute("aria-expanded", String(!expanded))
@@ -1,12 +1,9 @@
 import { expect, test } from "@playwright/test"
 import { assistantMessage, setupTimeline, toolPart, userMessage } from "../performance/timeline-stability/fixture"

-for (const profile of [
-  { locale: "de", label: "Erkundung abgeschlossen" },
-  { locale: "ar", label: "تم الاستكشاف" },
-] as const) {
-  test(`projects translated context status in ${profile.locale}`, async ({ page }) => {
-    const ids = [`prt_locale_${profile.locale}_01_read`, `prt_locale_${profile.locale}_02_glob`]
+for (const locale of ["de", "ar"] as const) {
+  test(`projects localized tool names with an English fallback in ${locale}`, async ({ page }) => {
+    const ids = [`prt_locale_${locale}_01_read`, `prt_locale_${locale}_02_glob`]
    await setupTimeline(page, {
      messages: [
        userMessage(),
@@ -15,11 +12,12 @@ for (const profile of [
          toolPart(ids[1]!, "glob", "completed", { path: ".", pattern: "**/*.ts" }),
        ]),
      ],
-      locale: profile.locale,
+      locale,
    })

    const group = page.locator(`[data-timeline-part-ids="${ids.join(",")}"]`)
-    await expect(group.locator('[data-component="tool-status-title"]')).toHaveAttribute("aria-label", profile.label)
-    await expect(page.locator("html")).toHaveAttribute("lang", profile.locale)
+    await expect(group.getByRole("button")).toHaveAccessibleName(/^Used /)
+    await expect(group.locator('[data-component="tag"]')).toHaveText("2")
+    await expect(page.locator("html")).toHaveAttribute("lang", locale)
  })
 }
@@ -53,9 +53,14 @@ test.describe("session timeline projection", () => {
    ]
    await setupTimeline(page, { messages: [userMessage(), assistantMessage(parts)] })

-    await expect(
-      page.locator('[data-timeline-part-ids="prt_01_read,prt_02_glob,prt_03_grep,prt_04_list"]'),
-    ).toBeVisible()
+    const first = page.locator(
+      '[data-timeline-part-ids="prt_01_read,prt_02_glob,prt_03_grep,prt_04_list,prt_webfetch,prt_websearch,prt_task,prt_bash,prt_edit,prt_write,prt_patch"]',
+    )
+    const second = page.locator('[data-timeline-part-ids="prt_skill,prt_custom"]')
+    await expect(first).toBeVisible()
+    await expect(second).toBeVisible()
+    await first.getByRole("button").click()
+    await second.getByRole("button").click()
    for (const id of [
      "prt_webfetch",
      "prt_websearch",
@@ -78,8 +83,7 @@ test.describe("session timeline projection", () => {
    await expect(patch.locator('[data-slot="message-part-title-filename"]')).toHaveCount(0)
    await expect(patch.locator('[data-slot="message-part-actions"]')).toHaveCount(0)
    const edit = page.locator('[data-timeline-part-id="prt_edit"]')
-    await expect(edit.locator('[data-component="apply-patch-tool"]')).toBeVisible()
-    await expect(edit.locator('[data-slot="basic-tool-tool-title"]')).toContainText("Edit")
+    await expect(edit).toContainText("Edit")
    await expect(page.locator('[data-timeline-part-id="prt_todo"]')).toHaveCount(0)
  })

@@ -87,6 +91,7 @@ test.describe("session timeline projection", () => {
    const first = "prt_patch_first"
    const second = "prt_patch_second"
    const timeline = await setupTimeline(page, {
+      settings: { editToolPartsExpanded: true },
      messages: [
        userMessage(),
        assistantMessage([
@@ -13,7 +13,7 @@ import {
  userMessage,
 } from "../performance/timeline-stability/fixture"

-test("groups singleton and separated context operations at correct boundaries", async ({ page }) => {
+test("groups every collapsed tool until visible text separates the stack", async ({ page }) => {
  const parts = [
    toolPart("prt_boundary_01_read", "read", "completed", { path: "src/a.ts" }),
    textPart("prt_boundary_02_text", "Boundary text"),
@@ -25,9 +25,112 @@ test("groups singleton and separated context operations at correct boundaries",
  await setupTimeline(page, { messages: [userMessage(), assistantMessage(parts)] })

  await expect(page.locator('[data-timeline-part-ids="prt_boundary_01_read"]')).toBeVisible()
-  await expect(page.locator('[data-timeline-part-ids="prt_boundary_03_glob,prt_boundary_04_grep"]')).toBeVisible()
-  await expect(page.locator('[data-timeline-part-ids="prt_boundary_06_list"]')).toBeVisible()
-  await expect(page.locator('[data-timeline-row="AssistantPart"]')).toHaveCount(5)
+  const group = page.locator(
+    '[data-timeline-part-ids="prt_boundary_03_glob,prt_boundary_04_grep,prt_boundary_05_shell,prt_boundary_06_list"]',
+  )
+  await expect(group).toBeVisible()
+  await expect(group.getByRole("button")).toHaveAccessibleName("Used Glob, Grep, Shell, List")
+  await expect(group.locator('[data-component="tag"]')).toHaveText("4")
+  await expect(page.locator('[data-timeline-row="AssistantPart"]')).toHaveCount(3)
+  await expect(page.locator('[data-timeline-spacing="content"]')).toHaveCount(2)
+  await expect(page.locator('[data-timeline-spacing="content"]').nth(0)).toHaveCSS("padding-top", "16px")
+})
+
+test("expands a mixed collapsed tool stack without expanding its individual calls", async ({ page }) => {
+  const parts = [
+    shell("prt_stack_shell_1", "completed", "first"),
+    toolPart("prt_stack_explore", "subagent", "completed", {
+      agent: "explore",
+      description: "Inspect the project",
+      prompt: "Explore the project",
+    }),
+    toolPart("prt_stack_patch", "patch", "completed", { patchText: "Update src/value.ts" }),
+    shell("prt_stack_shell_2", "completed", "second"),
+  ]
+  await setupTimeline(page, { messages: [userMessage(), assistantMessage(parts)] })
+
+  const group = page.locator(
+    '[data-timeline-part-ids="prt_stack_shell_1,prt_stack_explore,prt_stack_patch,prt_stack_shell_2"]',
+  )
+  const summary = group.getByRole("button", { name: "Used Shell, Explore, Patch" })
+  await expect(summary).toHaveAttribute("aria-expanded", "false")
+  await expect(summary).toHaveCSS("height", "28px")
+  await expect(group.locator('[data-component="tag"]')).toHaveText("4")
+  await summary.click()
+  await expect(summary).toHaveAttribute("aria-expanded", "true")
+  await expect(group.locator('[data-slot="context-tool-group-item"]')).toHaveCount(4)
+  await expect(group.locator('[data-timeline-part-id="prt_stack_shell_1"]')).toBeVisible()
+  await expect(group.locator('[data-timeline-part-id="prt_stack_patch"]')).toBeVisible()
+  await expect(group.locator('[data-component="context-tool-group-list"]')).toHaveCSS("row-gap", "8px")
+  const content = group.locator(':scope > [data-component="collapsible"] > [data-slot="collapsible-content"]')
+  await expect(content).toHaveCSS("margin-left", "0px")
+  await expect(content).toHaveCSS("padding-left", "12px")
+  await expect.poll(() => content.evaluate((element) => getComputedStyle(element, "::before").content)).toBe("none")
+})
+
+test("leaves tools expanded by settings outside the collapsed stack", async ({ page }) => {
+  const parts = [
+    shell("prt_expanded_shell", "completed", "expanded"),
+    toolPart("prt_collapsed_patch", "patch", "completed", { patchText: "Update src/value.ts" }),
+    toolPart("prt_collapsed_read", "read", "completed", { path: "src/value.ts" }),
+  ]
+  await setupTimeline(page, {
+    messages: [userMessage(), assistantMessage(parts)],
+    settings: { shellToolPartsExpanded: true },
+  })
+
+  await expect(page.locator('[data-timeline-part-id="prt_expanded_shell"]')).toBeVisible()
+  const group = page.locator('[data-timeline-part-ids="prt_collapsed_patch,prt_collapsed_read"]')
+  await expect(group.getByRole("button", { name: "Used Patch, Read" })).toBeVisible()
+  await expect(group.locator('[data-component="tag"]')).toHaveText("2")
+  await expect(page.locator('[data-timeline-spacing="tool"]')).toHaveCSS("padding-top", "8px")
+})
+
+test("keeps failed search calls and their error cards inside the collapsed stack", async ({ page }) => {
+  const parts = [
+    toolPart(
+      "prt_error_glob",
+      "glob",
+      "error",
+      { path: "C:/Users", pattern: "*.ts" },
+      {
+        error: "Invalid tool input",
+      },
+    ),
+    toolPart(
+      "prt_error_grep",
+      "grep",
+      "error",
+      { path: "C:/Users", pattern: "value" },
+      {
+        error: "Search timed out after 30 seconds",
+      },
+    ),
+  ]
+  await setupTimeline(page, { messages: [userMessage(), assistantMessage(parts)] })
+
+  const group = page.locator('[data-timeline-part-ids="prt_error_glob,prt_error_grep"]')
+  const summary = group.getByRole("button", { name: "Used Glob, Grep" })
+  await expect(group.locator('[data-component="tag"]')).toHaveText("2")
+  await summary.click()
+  await expect(group.locator('[data-kind="tool-error-card"]')).toHaveCount(2)
+  const glob = group.locator('[data-timeline-part-id="prt_error_glob"]')
+  await expect(glob).toContainText("Invalid tool input")
+  await expect(glob.locator('[data-component="tool-error-card-icon"]')).toBeVisible()
+  await expect(glob.locator('[data-component="tool-error-card-icon"] use')).toHaveAttribute(
+    "href",
+    "#opencode-v2-icon-circle-exclamation",
+  )
+  await expect
+    .poll(() =>
+      glob
+        .locator('[data-kind="tool-error-card"]')
+        .evaluate((element) => getComputedStyle(element, "::before").display),
+    )
+    .toBe("none")
+  await expect(group.locator('[data-timeline-part-id="prt_error_grep"]')).toContainText(
+    "Search timed out after 30 seconds",
+  )
 })

 test("reducer-hardening: converges when idle arrives before final part and message completion", async ({ page }) => {
@@ -21,6 +21,9 @@ test("renders every tool error outcome without leaking hidden tools", async ({ p
  )
  await setupTimeline(page, { messages: [userMessage(), assistantMessage(parts)] })

+  const group = page.locator(`[data-timeline-part-ids="${ordinary.map((_, index) => `prt_error_${index}`).join(",")}"]`)
+  await expect(group.locator('[data-component="tag"]')).toHaveText(String(ordinary.length))
+  await group.getByRole("button").click()
  await expect(page.locator('[data-kind="tool-error-card"]')).toHaveCount(ordinary.length + 1)
  await expect(page.getByText(/dismissed/i)).toBeVisible()
  await expect(page.locator('[data-timeline-part-id="prt_todo_error"]')).toHaveCount(0)
@@ -33,6 +36,7 @@ test("transitions shell and question through running error outcomes", async ({ p
  const shellID = "prt_transition_error_shell"
  const questionID = "prt_transition_error_question"
  const timeline = await setupTimeline(page, {
+    settings: { shellToolPartsExpanded: true },
    messages: [
      userMessage(),
      assistantMessage(
@@ -44,7 +48,6 @@ test("transitions shell and question through running error outcomes", async ({ p
      ),
    ],
  })
-  await timeline.waitForPart(shellID)
  await expect(page.locator(`[data-timeline-part-id="${questionID}"]`)).toHaveCount(0)
  await timeline.send(partUpdated(toolPart(shellID, "shell", "running", { command: "exit 1" })), 120)
  await timeline.send(partUpdated(toolPart(questionID, "question", "running", questionInput())), 180)
@@ -68,6 +71,7 @@ test("preserves surviving grouped patch state when its first patch fails", async
  const failed = "prt_grouped_patch_failed"
  const surviving = "prt_grouped_patch_surviving"
  const timeline = await setupTimeline(page, {
+    settings: { editToolPartsExpanded: true },
    messages: [
      userMessage(),
      assistantMessage(
@@ -147,10 +151,12 @@ test("labels all web search provider variants", async ({ page }) => {
    toolPart("prt_search_generic", "websearch", "completed", { query: "generic" }),
  ]
  await setupTimeline(page, { messages: [userMessage(), assistantMessage(parts)] })
+  await page.getByRole("button", { name: "Used Parallel Web Search, Exa Web Search, Web Search" }).click()

-  await expect(page.getByRole("button", { name: /Parallel Web Search/ })).toBeVisible()
-  await expect(page.getByRole("button", { name: /Exa Web Search/ })).toBeVisible()
-  await expect(page.getByRole("button", { name: /^Web Search/ })).toBeVisible()
+  const tools = page.locator('[data-component="context-tool-group-list"]')
+  await expect(tools.getByRole("button", { name: /Parallel Web Search/ })).toBeVisible()
+  await expect(tools.getByRole("button", { name: /Exa Web Search/ })).toBeVisible()
+  await expect(tools.getByRole("button", { name: /^Web Search/ })).toBeVisible()
 })

 test("labels completed searches with result counts", async ({ page }) => {
@@ -188,6 +194,33 @@ test("labels read tools from their path input", async ({ page }) => {
  ).toContainText("a.ts")
 })

+test("groups instruction files loaded by the same read", async ({ page }) => {
+  const id = "prt_read_instructions"
+  await setupTimeline(page, {
+    messages: [
+      userMessage(),
+      assistantMessage([
+        toolPart(
+          id,
+          "read",
+          "completed",
+          { path: "src/a.ts" },
+          { metadata: { loaded: ["AGENTS.md", "packages/app/AGENTS.md", "packages/ui/AGENTS.md"] } },
+        ),
+      ]),
+    ],
+  })
+
+  const tool = page.locator(`[data-timeline-part-id="${id}"]`)
+  const loaded = tool.locator('[data-component="tool-loaded-item"]')
+  await expect(loaded).toHaveCount(1)
+  await expect(loaded).toHaveAttribute("aria-label", "Loaded AGENTS.md, packages/app/AGENTS.md, packages/ui/AGENTS.md")
+  await expect(loaded.locator('[data-slot="tool-loaded-value"]')).toHaveText(
+    "AGENTS.md, packages/app/AGENTS.md, packages/ui/AGENTS.md",
+  )
+  await expect(loaded.locator('[data-slot="tool-loaded-kind"]')).toHaveCount(0)
+})
+
 test("labels skill tools from IDs and result metadata", async ({ page }) => {
  const pending = "prt_skill_id"
  const completed = "prt_skill_name"
@@ -201,18 +234,40 @@ test("labels skill tools from IDs and result metadata", async ({ page }) => {
    ],
  })

-  for (const [id, name] of [
-    [pending, "frontend-design"],
-    [completed, "OpenCode"],
-  ] as const) {
-    const skill = page.locator(`[data-timeline-part-id="${id}"]`)
-    const loaded = skill.locator('[data-component="tool-loaded-item"]')
-    await expect(loaded).toHaveAttribute("aria-label", `Loaded ${name} skill`)
-    await expect(loaded).toHaveCSS("line-height", "16px")
-    await expect(loaded.locator('[data-slot="tool-loaded-label"]')).toHaveText("Loaded")
-    await expect(loaded.locator('[data-slot="tool-loaded-kind"]')).toHaveText("skill")
-    await expect(loaded.locator('[data-component="text-shimmer"]')).toHaveAttribute("aria-label", name)
-  }
+  const group = page.locator(`[data-timeline-part-ids="${pending},${completed}"]`)
+  await expect(group.getByRole("button")).toHaveAccessibleName("Used Skill")
+  await expect(group.locator('[data-component="tag"]')).toHaveText("2")
+  await group.getByRole("button").click()
+
+  const loaded = group.locator('[data-component="tool-loaded-item"]')
+  await expect(loaded).toHaveCount(1)
+  await expect(loaded).toHaveAttribute("aria-label", "Loaded frontend-design, OpenCode skills")
+  await expect(loaded).toHaveCSS("line-height", "16px")
+  await expect(loaded.locator('[data-slot="tool-loaded-label"]')).toHaveText("Loaded")
+  await expect(loaded.locator('[data-slot="tool-loaded-kind"]')).toHaveText("skills")
+  const names = loaded.locator('[data-component="text-shimmer"]')
+  await expect(names).toHaveCount(2)
+  await expect(names.nth(0)).toHaveAttribute("aria-label", "frontend-design")
+  await expect(names.nth(1)).toHaveAttribute("aria-label", "OpenCode")
+})
+
+test("groups only consecutive successful skill tools", async ({ page }) => {
+  const parts = [
+    toolPart("prt_skill_first", "skill", "completed", { id: "ocpr" }),
+    toolPart("prt_skill_second", "skill", "completed", { id: "effect" }),
+    toolPart("prt_skill_third", "skill", "completed", { id: "ui-pr-screenshots" }),
+    toolPart("prt_skill_break", "read", "completed", { path: "src/a.ts" }),
+    toolPart("prt_skill_last", "skill", "completed", { id: "opencode" }),
+  ]
+  await setupTimeline(page, { messages: [userMessage(), assistantMessage(parts)] })
+
+  const group = page.locator(`[data-timeline-part-ids="${parts.map((part) => part.id).join(",")}"]`)
+  await group.getByRole("button").click()
+
+  const loaded = group.locator('[data-component="tool-loaded-item"]')
+  await expect(loaded).toHaveCount(2)
+  await expect(loaded.nth(0)).toHaveAttribute("aria-label", "Loaded ocpr, effect, ui-pr-screenshots skills")
+  await expect(loaded.nth(1)).toHaveAttribute("aria-label", "Loaded opencode skill")
 })

 function questionInput() {
@@ -42,8 +42,7 @@ test("shows parent lineage while the child timeline loads", async ({ page }) =>
  const release = Promise.withResolvers<void>()
  await page.route(
    (url) =>
-      url.pathname === `/api/session/${childID}/message` &&
-      url.port === (process.env.PLAYWRIGHT_SERVER_PORT ?? "4096"),
+      url.pathname === `/api/session/${childID}/message` && url.port === (process.env.PLAYWRIGHT_SERVER_PORT ?? "4096"),
    async (route) => {
      requested.resolve()
      await release.promise
@@ -53,6 +52,7 @@ test("shows parent lineage while the child timeline loads", async ({ page }) =>

  await page.goto(sessionHref(parentID))
  await expectSessionTitle(page, parentTitle)
+  await page.getByRole("button", { name: "Used Explore" }).click()
  await page.locator(`a[href="${sessionHref(childID)}"]`).click()
  await Promise.all([requested.promise, expect(page).toHaveURL(sessionHref(childID))])
  await Promise.all([
@@ -77,6 +77,7 @@ test("keeps the parent visible while the child session resolves", async ({ page
  await page.goto(sessionHref(parentID))
  await expectSessionTitle(page, parentTitle)

+  await page.getByRole("button", { name: "Used Explore" }).click()
  await page.locator(`a[href="${sessionHref(childID)}"]`).click()
  await requested.promise
  await Promise.all([expect(page).toHaveURL(sessionHref(parentID)), expectSessionTitle(page, parentTitle)]).finally(
@@ -194,6 +195,7 @@ async function setup(page: Page, events?: () => OpenCodeEvent[]) {
 async function openChildFromParent(page: Page) {
  await page.goto(sessionHref(parentID))
  await expectSessionTitle(page, parentTitle)
+  await page.getByRole("button", { name: "Used Explore" }).click()

  const card = page.locator(`a[href="${sessionHref(childID)}"]`)
  await expect(card).toBeVisible()
@@ -103,6 +103,127 @@ test("cramped tabs only show the close button for the active tab", async ({ page
  await expect(tabB.locator('[data-slot="tab-close"]')).toBeVisible()
 })

+test("vertical tabs show project details, resize, and navigate", async ({ page }) => {
+  await mockServer(page)
+  await page.addInitScript(
+    ({ server, sessionA, sessionB }) => {
+      localStorage.setItem("settings.v3", JSON.stringify({ appearance: { tabLayout: "vertical" } }))
+      localStorage.setItem(
+        "opencode.window.browser.dat:tabs",
+        JSON.stringify([
+          { type: "session", server, sessionId: sessionA },
+          { type: "session", server, sessionId: sessionB },
+        ]),
+      )
+    },
+    { server, sessionA: sessionA.id, sessionB: sessionB.id },
+  )
+
+  const hrefA = `/server/${base64Encode(server)}/session/${sessionA.id}`
+  const hrefB = `/server/${base64Encode(server)}/session/${sessionB.id}`
+  await page.goto(hrefA)
+
+  const sidebar = page.locator('[data-slot="vertical-tabs-sidebar"]')
+  const tabA = sidebar.locator(`[data-titlebar-tab-link][href="${hrefA}"]`)
+  const tabB = sidebar.locator(`[data-titlebar-tab-link][href="${hrefB}"]`)
+  await expect(sidebar).toHaveCSS("width", "260px")
+  await expect(tabA).toContainText(sessionA.title)
+  await expect(tabB).toContainText(sessionB.title)
+  await expect(tabB.locator('[data-slot="tab-project"]')).toHaveText("tab-project")
+  await expect(sidebar.getByRole("button", { name: "New session" })).toBeVisible()
+  await expect(page.locator('[data-slot="titlebar-tabs"]')).toHaveCount(0)
+
+  const handle = sidebar.locator('[data-component="resize-handle"]')
+  await expect(handle).toHaveCSS("cursor", "col-resize")
+  const box = await handle.boundingBox()
+  if (!box) throw new Error("vertical tab resize handle has no bounding box")
+  await page.mouse.move(box.x + box.width / 2, box.y + box.height / 2)
+  await page.mouse.down()
+  await page.mouse.move(box.x + box.width / 2 - 80, box.y + box.height / 2)
+  await page.mouse.up()
+  await expect(sidebar).toHaveCSS("width", "180px")
+  await expect(tabB.locator('[data-slot="tab-project"]')).toHaveText("tab-project")
+
+  const resized = await handle.boundingBox()
+  if (!resized) throw new Error("resized vertical tab handle has no bounding box")
+  await page.mouse.move(resized.x + resized.width / 2, resized.y + resized.height / 2)
+  await page.mouse.down()
+  await page.mouse.move(resized.x - 200, resized.y + resized.height / 2)
+  await page.mouse.up()
+  await expect(sidebar).toHaveCSS("width", "130px")
+
+  await tabB.click()
+  await expect(page).toHaveURL(new RegExp(`${hrefB.replace(/[.*+?^${}()|[\]\\]/g, "\\$&")}$`))
+  await expect(tabB).toBeVisible()
+})
+
+test("appearance experimental setting switches tab orientation", async ({ page }) => {
+  await mockServer(page)
+  await page.addInitScript(
+    ({ server, sessionA }) => {
+      localStorage.setItem(
+        "opencode.window.browser.dat:tabs",
+        JSON.stringify([{ type: "session", server, sessionId: sessionA }]),
+      )
+    },
+    { server, sessionA: sessionA.id },
+  )
+
+  await page.goto("/")
+  await expect(page.locator('[data-slot="titlebar-tabs"] [data-titlebar-tab-link]')).toBeVisible()
+  await page.keyboard.press("Control+,")
+
+  const settings = page.getByTestId("settings-screen")
+  await expect(settings).toBeVisible()
+  await settings.getByRole("tab", { name: "Appearance" }).click()
+  await expect(settings.getByRole("heading", { name: "Experimental" })).toBeVisible()
+
+  const layout = settings.locator('[data-action="settings-tab-layout"]')
+  await expect(layout).toContainText("Horizontal")
+  await layout.click()
+  await page.getByRole("option", { name: "Vertical" }).click()
+
+  await expect(layout).toContainText("Vertical")
+  await expect(page.locator('[data-slot="vertical-tabs-sidebar"]')).toBeVisible()
+  await expect(page.locator('[data-slot="titlebar-tabs"]')).toHaveCount(0)
+  await expect(settings.getByRole("tablist")).toHaveCSS("width", "240px")
+
+  await page.setViewportSize({ width: 920, height: 720 })
+  await expect(page.locator('[data-slot="vertical-tabs-sidebar"]')).toHaveCSS("width", "260px")
+  await expect(settings.getByRole("tablist")).toHaveCSS("width", "160px")
+
+  await page.setViewportSize({ width: 800, height: 720 })
+  await expect(settings.getByRole("tablist")).toHaveCSS("width", "160px")
+})
+
+test("vertical tab preference falls back to horizontal on mobile", async ({ page }) => {
+  await page.setViewportSize({ width: 390, height: 720 })
+  await mockServer(page)
+  await page.addInitScript(
+    ({ server, sessionA }) => {
+      localStorage.setItem("settings.v3", JSON.stringify({ appearance: { tabLayout: "vertical" } }))
+      localStorage.setItem(
+        "opencode.window.browser.dat:tabs",
+        JSON.stringify([{ type: "session", server, sessionId: sessionA }]),
+      )
+    },
+    { server, sessionA: sessionA.id },
+  )
+
+  const href = `/server/${base64Encode(server)}/session/${sessionA.id}`
+  await page.goto(href)
+
+  const tabs = page.locator('[data-slot="titlebar-tabs"]')
+  await expect(tabs.locator(`[data-titlebar-tab-link][href="${href}"]`)).toContainText(sessionA.title)
+  await expect(page.locator('[data-slot="vertical-tabs-sidebar"]')).toHaveCount(0)
+
+  await page.setViewportSize({ width: 1280, height: 720 })
+  await expect(
+    page.locator('[data-slot="vertical-tabs-sidebar"]').locator(`[data-titlebar-tab-link][href="${href}"]`),
+  ).toBeVisible()
+  await expect(page.locator('[data-slot="titlebar-tabs"]')).toHaveCount(0)
+})
+
 function session(id: string, title: string) {
  return {
    id,
@@ -58,7 +58,7 @@ export function AppBaseProviders(
  props: ParentProps<{
    locale?: Locale
    onNativeTranslations?: Parameters<typeof LanguageProvider>[0]["onNativeTranslations"]
-    onThemeApplied?: () => void
+    onThemeApplied?: (mode: "light" | "dark", scheme: "system" | "light" | "dark") => void
  }>,
 ) {
  return (
@@ -67,7 +67,7 @@ export function AppBaseProviders(
      <ThemeProvider
        onThemeApplied={(_, mode, scheme) => {
          void window.api?.setTitlebar?.({ mode, scheme })
-          props.onThemeApplied?.()
+          props.onThemeApplied?.(mode, scheme)
        }}
      >
        <LanguageProvider locale={props.locale} onNativeTranslations={props.onNativeTranslations}>
@@ -133,7 +133,7 @@ export function createNewSessionComposerAdapter(props: {

  return {
    adapter,
-    project: createComposerProjectControls({ draftId: props.draftID }),
+    project: createComposerProjectControls({ draftId: props.draftID, worktree: props.worktree }),
    model,
    ready: prompt.ready,
  }
@@ -4,10 +4,11 @@ import { useGlobal, useServerCtx } from "@/runtime/server/runtime"
 import { useServerSDK } from "@/runtime/server/client"
 import { serverName, ServerConnection, useServers } from "@/runtime/server/registry"
 import { useWorkspaceLocation } from "@/workspaces/location"
+import { workspaceSelectionDestination } from "@/workspaces/paths"
 import { useTabs } from "@/shell/tabs/tabs"
 import type { PromptProjectControls } from "./selector"

-export function createComposerProjectControls(props: { draftId: string }) {
+export function createComposerProjectControls(props: { draftId: string; worktree: () => string }) {
  const servers = useServers()
  const serverSDK = useServerSDK()
  const location = useWorkspaceLocation()
@@ -38,7 +39,7 @@ export function createComposerProjectControls(props: { draftId: string }) {
    tabs.updateDraft(props.draftId, {
      server: ServerConnection.key(connection),
      directory: worktree,
-      worktree: undefined,
+      worktree: workspaceSelectionDestination(props.worktree(), location().directory),
      branch: undefined,
    })
  }
@@ -5,6 +5,8 @@ import { useWorkspaceLocation } from "@/workspaces/location"
 import { useServerSDK } from "@/runtime/server/client"
 import { useData } from "@/runtime/server/current"
 import { useSettings } from "@/settings/model"
+import { useTabs } from "@/shell/tabs/tabs"
+import { ServerConnection } from "@/runtime/server/registry"
 import { normalizeProjectInfo } from "@/runtime/server/global-sync/utils"
 import {
  isWorkspaceDirectory,
@@ -12,6 +14,7 @@ import {
  sameDirectory,
  workspaceDefaultSelection,
  workspaceDirectories,
+  workspaceSelectionDestination,
 } from "@/workspaces/paths"

 export function resolveNewSessionWorktree(input: {
@@ -57,6 +60,7 @@ export function createNewSessionWorkspaceController(input: {
  const serverSDK = useServerSDK()
  const data = useData()
  const settings = useSettings()
+  const tabs = useTabs()
  const [state, setState] = createStore({ search: "" })
  const searchBranches = debounce((search: string) => setState("search", search.trim()), 100)
  const currentProject = createMemo(() => {
@@ -121,7 +125,8 @@ export function createNewSessionWorkspaceController(input: {
  const remember = (worktree = value()) => {
    const project = currentProject()
    if (!project) return
-    const local = worktree === "main" || sameDirectory(worktree, project.worktree)
+    tabs.initializeDraftWorktrees(ServerConnection.key(serverSDK.server), sdk().directory, fallback())
+    const local = workspaceSelectionDestination(worktree, project.worktree) === "main"
    settings.workspaces.setLastUsed(serverSDK.scope, project.id, local ? "local" : "workspace")
  }

@@ -141,6 +146,7 @@ export function createNewSessionWorkspaceController(input: {
      set: (worktree: string) => {
        input.setSelectedBranch(undefined)
        input.setSelectedWorktree(normalizeNewSessionWorktree(worktree, sdk().directory, currentProject()?.worktree))
+        remember(worktree)
      },
      create: (branch: string) => {
        input.setSelectedBranch(branch)
@@ -160,7 +166,10 @@ export function createNewSessionWorkspaceController(input: {
        const loaded = branches.latest
        const list = loaded?.directory === projectRoot() ? loaded.data : []
        return [
-          ...new Set([...list, ...(current && current.toLowerCase().includes(state.search.toLowerCase()) ? [current] : [])]),
+          ...new Set([
+            ...list,
+            ...(current && current.toLowerCase().includes(state.search.toLowerCase()) ? [current] : []),
+          ]),
        ].slice(0, 50)
      },
      searchBranches,
@@ -884,6 +884,7 @@ export const dict = {

  "settings.section.desktop": "Desktop",
  "settings.section.server": "Server",
+  "settings.backToApp": "Back to app",
  "settings.tab.general": "General",
  "settings.tab.preferences": "Preferences",
  "settings.tab.shortcuts": "Shortcuts",
@@ -892,6 +893,11 @@ export const dict = {
  "settings.tab.extensions": "Extensions",
  "settings.preferences.description": "Customize preferences and theme and default behavior",
  "settings.appearance.description": "Customize theme and fonts",
+  "settings.appearance.section.experimental": "Experimental",
+  "settings.appearance.row.tabs.title": "Tabs",
+  "settings.appearance.row.tabs.description": "Choose how session tabs are arranged",
+  "settings.appearance.row.tabs.horizontal": "Horizontal",
+  "settings.appearance.row.tabs.vertical": "Vertical",
  "settings.notifications.description": "Choose when to receive notifications and hear sounds",
  "settings.shortcuts.description": "Customize shortcuts for common actions",
  "settings.servers.description": "Manage server connections",
@@ -0,0 +1,54 @@
+import { describe, expect, test } from "bun:test"
+import type { SessionInboxInfo } from "@opencode-ai/client/promise"
+import { queuedPromptRows } from "./queue"
+
+const queued = [
+  {
+    id: "msg_original",
+    sessionID: "ses_1",
+    timeCreated: 1,
+    type: "user",
+    delivery: "queue",
+    payload: { text: "original" },
+  },
+  {
+    id: "msg_replacement",
+    sessionID: "ses_1",
+    timeCreated: 2,
+    type: "user",
+    delivery: "queue",
+    payload: { text: "edited" },
+  },
+] satisfies SessionInboxInfo[]
+
+describe("queuedPromptRows", () => {
+  test("keeps the edited prompt to one row while its replacement is admitted", () => {
+    expect(queuedPromptRows(queued, { original: "msg_original", replacement: "msg_replacement" })).toEqual([
+      { id: "msg_replacement", text: "edited", attachments: false },
+    ])
+  })
+
+  test("keeps the original visible until its replacement appears", () => {
+    expect(queuedPromptRows([queued[0]], { original: "msg_original", replacement: "msg_replacement" })).toEqual([
+      { id: "msg_original", text: "original", attachments: false },
+    ])
+  })
+
+  test("retains unrelated queue entries", () => {
+    expect(queuedPromptRows(queued)).toEqual([
+      { id: "msg_original", text: "original", attachments: false },
+      { id: "msg_replacement", text: "edited", attachments: false },
+    ])
+  })
+
+  test("keeps other prompts visible while a mutation replaces the edited prompt", () => {
+    const other = { ...queued[0], id: "msg_other", payload: { text: "other" } }
+
+    expect(
+      queuedPromptRows([queued[0], other, queued[1]], { original: "msg_original", replacement: "msg_replacement" }),
+    ).toEqual([
+      { id: "msg_other", text: "other", attachments: false },
+      { id: "msg_replacement", text: "edited", attachments: false },
+    ])
+  })
+})
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
opencode-agent[bot]	b731bc19e2	chore: update nix node_modules hashes	2026-08-26 16:44:38 +00:00
Kit Langton	fedf017e25	fix(tui): clarify code mode tool call rendering	2026-08-26 11:27:17 -04:00
Aiden Cline	f4a9b93013	feat(ai): add native Groq provider with provider-specific options (#45288 )	2026-08-26 10:12:52 -05:00
Dax Raad	cbef698861	fix(core): suppress state updates during location teardown	2026-08-26 10:48:44 -04:00
James Long	ab2d251155	fix(server): unblock persistent terminal input in Bun builds (#45287 )	2026-08-26 10:47:16 -04:00
Kit Langton	3d7ba38965	feat(tui): investigate errors in a new session	2026-08-26 10:43:40 -04:00
Aiden Cline	37a6ba893e	chore(ai): format package with prettier (#45280 )	2026-08-26 09:24:13 -05:00
Brendan Allan	6c6871fd2a	fix(app): include files in mention search (#45281 )	2026-08-26 22:18:34 +08:00
opencode-agent[bot]	1e864dd8c6	chore: update nix node_modules hashes	2026-08-26 13:49:52 +00:00
James Long	21980a4448	fix(tui): restore terminal settings and move pane toggle to sessions (#45271 )	2026-08-26 09:46:33 -04:00
James Long	9cca8dd6e0	feat(tui): add persistent session terminals (#44971 )	2026-08-26 09:30:03 -04:00
Dax	91028a690b	feat(tui): show LLM token throughput	2026-08-26 09:23:53 -04:00
Shoubhit Dash	e82aa92e64	fix(session): support assistant message content updates (#45015 )	2026-08-26 18:31:59 +05:30
opencode-agent[bot]andBrendonovich	874538d702	fix(desktop): restrict macOS app entitlements (#45258 ) Co-authored-by: Brendonovich <14191578+Brendonovich@users.noreply.github.com>	2026-08-26 12:53:00 +00:00
opencode-agent[bot]	afc26c72c0	chore: update nix node_modules hashes	2026-08-26 12:11:06 +00:00
James Long	d572a5756c	feat(cli): embed persistent PTY binaries from npm packages (#45248 )	2026-08-26 07:55:30 -04:00
opencode-agent[bot]andBrendonovich	ab95695b24	fix(app): combine loaded skill and instruction file entries (#45217 ) Co-authored-by: Brendonovich <14191578+Brendonovich@users.noreply.github.com>	2026-08-26 10:08:33 +00:00
Luke Parker	5d7c2ccfc0	feat(app): add experimental vertical session tabs (#45210 )	2026-08-26 19:40:09 +10:00
Luke Parker	24ac05868c	fix(desktop): align and theme Windows caption controls (#45208 )	2026-08-26 09:02:10 +00:00
Shoubhit Dash	4eaf533cd0	feat(plugin): add declarative vcs repository markers (#45192 )	2026-08-26 13:49:36 +05:30
Shoubhit Dash	0f7a76eff0	fix(ai): keep bedrock mantle responses on http (#45197 )	2026-08-26 13:38:27 +05:30
Luke Parker	16d731bd67	feat(app): render settings as a fullscreen surface (#45190 )	2026-08-26 07:57:36 +00:00
Luke Parker	ea582fc133	feat(app): stack collapsed tool calls (#45176 )	2026-08-26 07:56:53 +00:00
opencode-agent[bot]andBrendonovich	7e27e81bc7	fix(app): preserve timeline measurements when moving sessions (#45145 ) Co-authored-by: Brendonovich <14191578+Brendonovich@users.noreply.github.com>	2026-08-26 14:39:01 +08:00
opencode-agent[bot]andBrendonovich	667722897e	fix(app): identify session worktrees from project inventory (#45153 ) Co-authored-by: Brendonovich <14191578+Brendonovich@users.noreply.github.com>	2026-08-26 14:38:08 +08:00
opencode-agent[bot]andBrendonovich	d12dbd12a9	fix(app): prevent duplicate rows when editing queued prompts (#45169 )l Co-authored-by: Brendonovich <14191578+Brendonovich@users.noreply.github.com>	2026-08-26 06:21:44 +00:00
opencode-agent[bot]andBrendonovich	b0bd0bc394	fix(app): align project settings server sections (#45163 ) Co-authored-by: Brendonovich <14191578+Brendonovich@users.noreply.github.com>	2026-08-26 04:42:26 +00:00
Dax	437df1164c	feat(plugin): expose session move	2026-08-26 00:14:12 -04:00
opencode-agent[bot]andBrendonovich	9a91e21a76	fix(app): preserve workspace choices across draft tabs (#45149 ) Co-authored-by: Brendonovich <14191578+Brendonovich@users.noreply.github.com>	2026-08-26 03:51:14 +00:00
opencode-agent[bot]	d01ac2069a	chore: update nix node_modules hashes	2026-08-26 03:32:00 +00:00
Aiden Cline	a78d3c5438	fix(ai): recover Anthropic request_too_large as overflow (#45144 )	2026-08-25 22:30:14 -05:00
David Hill	474c3588c1	fix(tui): use unicode ellipses in interface text (#45143 )	2026-08-25 22:26:15 -05:00
Kit Langton	9361117504	feat(core): add directory projects (#45107 )	2026-08-25 23:22:23 -04:00
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