feat(plugin): expose model generation options to session hooks

Co-authored-by: Brendonovich <14191578+Brendonovich@users.noreply.github.com>

.github/workflows/pr-standards.yml

@@ -135,7 +135,16 @@ jobs:
 
             const linkedIssues = result.repository.pullRequest.closingIssuesReferences.totalCount;
 
-            if (linkedIssues === 0) {
+            // GitHub only populates closingIssuesReferences when a PR targets the repository's
+            // default branch (dev). PRs targeting other branches like v2 always return totalCount 0.
+            // Fall back to checking the PR description for closing keywords (e.g. Closes #123).
+            const body = pr.body || '';
+            const issueMatch = body.match(/### Issue for this PR\s*\n([\s\S]*?)(?=###|$)/);
+            const issueContent = issueMatch ? issueMatch[1].trim() : body;
+            const hasBodyIssueRef = /(closes|fixes|resolves)\s+#\d+/i.test(issueContent) || /#\d+/.test(issueContent);
+            const hasLinkedIssue = linkedIssues > 0 || hasBodyIssueRef;
+
+            if (!hasLinkedIssue) {
               await addLabel('needs:issue');
               await comment('issue', `Thanks for your contribution!
 

.github/workflows/test.yml

@@ -22,6 +22,36 @@ env:
   FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: true
 
 jobs:
+  affected:
+    name: affected packages
+    runs-on: blacksmith-4vcpu-ubuntu-2404
+    outputs:
+      app: ${{ steps.packages.outputs.app }}
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+        with:
+          token: ${{ secrets.GITHUB_TOKEN }}
+          fetch-depth: 0
+
+      - name: Setup Bun
+        uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0
+        with:
+          bun-version-file: package.json
+
+      - name: Find affected packages
+        id: packages
+        env:
+          TURBO_SCM_BASE: ${{ github.event_name == 'pull_request' && format('{0}^1', github.sha) || github.event.before }}
+          TURBO_SCM_HEAD: ${{ github.sha }}
+        run: |
+          if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
+            echo "app=true" >> "$GITHUB_OUTPUT"
+            exit 0
+          fi
+          bun x turbo@2.10.2 ls --affected --filter=@opencode-ai/app --output=json > affected.json
+          bun -e 'const result = await Bun.file("affected.json").json(); console.log(`app=${result.packages.count > 0}`)' >> "$GITHUB_OUTPUT"
+
   unit:
     name: unit (${{ matrix.settings.name }})
     strategy:
@@ -41,6 +71,7 @@ jobs:
         uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
         with:
           token: ${{ secrets.GITHUB_TOKEN }}
+          fetch-depth: 0
 
       - name: Setup Node
         uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
@@ -80,9 +111,34 @@ jobs:
 
       - name: Run unit tests
         timeout-minutes: 20
-        run: GITHUB_ACTIONS=false bun turbo test
+        run: |
+          if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
+            GITHUB_ACTIONS=false bun turbo test
+            exit 0
+          fi
+          GITHUB_ACTIONS=false bun turbo test --affected
         env:
           OPENCODE_EXPERIMENTAL_DISABLE_FILEWATCHER: ${{ runner.os == 'Windows' && 'true' || 'false' }}
+          TURBO_SCM_BASE: ${{ github.event_name == 'pull_request' && format('{0}^1', github.sha) || github.event.before }}
+          TURBO_SCM_HEAD: ${{ github.sha }}
+
+      - name: Verify published codemode package
+        if: runner.os == 'Linux'
+        working-directory: packages/codemode
+        run: bun run script/publish.ts --dry-run
+
+      - name: Verify packed workerd SDK
+        if: runner.os == 'Linux'
+        timeout-minutes: 15
+        run: |
+          if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
+            bun turbo verify:package --filter=@opencode-ai/sdk
+            exit 0
+          fi
+          bun turbo verify:package --affected --filter=@opencode-ai/sdk
+        env:
+          TURBO_SCM_BASE: ${{ github.event_name == 'pull_request' && format('{0}^1', github.sha) || github.event.before }}
+          TURBO_SCM_HEAD: ${{ github.sha }}
 
       - name: Verify compiled service lifecycle
         if: always()
@@ -122,7 +178,7 @@ jobs:
 
   e2e:
     name: e2e (${{ matrix.settings.name }})
-    if: github.ref_name != 'v2' && github.head_ref != 'v2'
+    needs: affected
     strategy:
       fail-fast: false
       matrix:
@@ -133,32 +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:
@@ -166,23 +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 }}

bun.lock

@@ -71,6 +71,7 @@
         "@solid-primitives/event-bus": "1.1.2",
         "@solid-primitives/event-listener": "2.4.5",
         "@solid-primitives/i18n": "2.2.1",
+        "@solid-primitives/keyed": "1.5.3",
         "@solid-primitives/media": "2.3.3",
         "@solid-primitives/resize-observer": "2.1.5",
         "@solid-primitives/scheduled": "1.5.3",
@@ -124,6 +125,7 @@
         "@effect/platform-node": "catalog:",
         "@opencode-ai/client": "workspace:*",
         "@opencode-ai/plugin": "workspace:*",
+        "@opencode-ai/pty": "0.1.8",
         "@opencode-ai/schema": "workspace:*",
         "@opencode-ai/server": "workspace:*",
         "@opencode-ai/tui": "workspace:*",
@@ -345,9 +347,7 @@
         "@ai-sdk/amazon-bedrock": "4.0.112",
         "@ai-sdk/anthropic": "3.0.82",
         "@ai-sdk/azure": "3.0.88",
-        "@ai-sdk/cerebras": "2.0.41",
         "@ai-sdk/cohere": "3.0.27",
-        "@ai-sdk/deepinfra": "2.0.41",
         "@ai-sdk/gateway": "3.0.104",
         "@ai-sdk/google-vertex": "4.0.128",
         "@ai-sdk/groq": "3.0.31",
@@ -356,7 +356,6 @@
         "@ai-sdk/perplexity": "3.0.26",
         "@ai-sdk/provider": "3.0.8",
         "@ai-sdk/provider-utils": "4.0.23",
-        "@ai-sdk/togetherai": "2.0.41",
         "@ai-sdk/vercel": "2.0.39",
         "@aws-sdk/credential-providers": "3.1057.0",
         "@ff-labs/fff-bun": "0.10.5",
@@ -366,6 +365,7 @@
         "@opencode-ai/ai": "workspace:*",
         "@opencode-ai/codemode": "workspace:*",
         "@opencode-ai/plugin": "workspace:*",
+        "@opencode-ai/pty": "0.1.8",
         "@opencode-ai/schema": "workspace:*",
         "@opencode-ai/util": "workspace:*",
         "@parcel/watcher": "2.5.1",
@@ -429,6 +429,7 @@
       },
       "devDependencies": {
         "@actions/artifact": "4.0.0",
+        "@brendonovich/vite-plugin-opencode": "0.1.1",
         "@lydell/node-pty": "catalog:",
         "@opencode-ai/app": "workspace:*",
         "@opencode-ai/client": "workspace:*",
@@ -594,8 +595,8 @@
       },
       "peerDependencies": {
         "@opencode-ai/theme": "workspace:*",
-        "@opentui/core": ">=0.5.7",
-        "@opentui/solid": ">=0.5.7",
+        "@opentui/core": ">=0.5.8",
+        "@opentui/solid": ">=0.5.8",
         "solid-js": ">=1.9.0",
       },
       "optionalPeers": [
@@ -665,12 +666,14 @@
       "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:*",
         "effect": "catalog:",
       },
       "devDependencies": {
+        "@opencode-ai/ai": "workspace:*",
         "@opencode-ai/httpapi-codegen": "workspace:*",
         "@opencode-ai/protocol": "workspace:*",
         "@tsconfig/bun": "catalog:",
@@ -683,6 +686,7 @@
       "version": "1.18.4",
       "dependencies": {
         "@effect/platform-node": "catalog:",
+        "@effect/platform-node-shared": "catalog:",
         "@opencode-ai/core": "workspace:*",
         "@opencode-ai/protocol": "workspace:*",
         "@opencode-ai/schema": "workspace:*",
@@ -969,6 +973,7 @@
       "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",
@@ -1054,7 +1059,6 @@
   "patchedDependencies": {
     "@pierre/trees@1.0.0-beta.4": "patches/@pierre%2Ftrees@1.0.0-beta.4.patch",
     "@tanstack/virtual-core@3.17.8": "patches/@tanstack%2Fvirtual-core@3.17.8.patch",
-    "drizzle-orm@1.0.0-rc.2": "patches/drizzle-orm@1.0.0-rc.2.patch",
     "@modelcontextprotocol/sdk@1.29.0": "patches/@modelcontextprotocol%2Fsdk@1.29.0.patch",
     "@standard-community/standard-openapi@0.2.9": "patches/@standard-community%2Fstandard-openapi@0.2.9.patch",
     "@ai-sdk/mistral@3.0.51": "patches/@ai-sdk%2Fmistral@3.0.51.patch",
@@ -1089,9 +1093,9 @@
     "@npmcli/arborist": "9.4.0",
     "@octokit/rest": "22.0.0",
     "@openauthjs/openauth": "0.0.0-20250322224806",
-    "@opentui/core": "0.5.7",
-    "@opentui/keymap": "0.5.7",
-    "@opentui/solid": "0.5.7",
+    "@opentui/core": "0.5.8",
+    "@opentui/keymap": "0.5.8",
+    "@opentui/solid": "0.5.8",
     "@pierre/diffs": "1.2.10",
     "@playwright/test": "1.59.1",
     "@sentry/solid": "10.36.0",
@@ -1116,8 +1120,8 @@
     "cross-spawn": "7.0.6",
     "diff": "8.0.2",
     "dompurify": "3.3.1",
-    "drizzle-kit": "1.0.0-rc.2",
-    "drizzle-orm": "1.0.0-rc.2",
+    "drizzle-kit": "1.0.0-rc.5-ab785fc",
+    "drizzle-orm": "1.0.0-rc.5-169397b",
     "effect": "4.0.0-rc.111",
     "fuzzysort": "3.1.0",
     "get-east-asian-width": "1.6.0",
@@ -1175,8 +1179,6 @@
 
     "@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=="],
@@ -1203,8 +1205,6 @@
 
     "@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=="],
@@ -1609,6 +1609,8 @@
 
     "@braintree/sanitize-url": ["@braintree/sanitize-url@7.1.2", "", {}, "sha512-jigsZK+sMF/cuiB7sERuo9V7N9jx+dhmHHnQyDSVdpZwVutaBu7WvNYqMDLSgFgfB30n452TP3vjDAvFC973mA=="],
 
+    "@brendonovich/vite-plugin-opencode": ["@brendonovich/vite-plugin-opencode@0.1.1", "", { "dependencies": { "@babel/core": "^7.29.0", "@opencode-ai/client": "0.0.0-beta-18050" }, "peerDependencies": { "vite": "^6.0.0 || ^7.0.0 || ^8.0.0" } }, "sha512-aPG0ct8ctxAqndbNOx7NW0GhU6QY6sOUfi/DaKqH9c5WdxICSsUop6uSkJwPDHP9WpN9eg0dd2D2qwYpG6UdHw=="],
+
     "@bruits/satteri-darwin-arm64": ["@bruits/satteri-darwin-arm64@0.9.5", "", { "os": "darwin", "cpu": "arm64" }, "sha512-iw4nZgx9v30lWo/MTngQqi1pI78KI0DnkSm+lVJGYdmPLgAyDNJigVhpG42/Iq55A6c1Ll8q66ljyyRiQUxwow=="],
 
     "@bruits/satteri-darwin-x64": ["@bruits/satteri-darwin-x64@0.9.5", "", { "os": "darwin", "cpu": "x64" }, "sha512-6T26Z5Kf3cFW2PSlk9p7zT7yVxvuBSiJvYyz9u8KjYwMTqZyIDOj2wDyNpxKV4+6yUVG7rddq2QwvG/8LJA2+Q=="],
@@ -1683,7 +1685,7 @@
 
     "@dot/log": ["@dot/log@0.1.5", "", { "dependencies": { "chalk": "^4.1.2", "loglevelnext": "^6.0.0", "p-defer": "^3.0.0" } }, "sha512-ECraEVJWv2f2mWK93lYiefUkphStVlKD6yKDzisuoEmxuLKrxO9iGetHK2DoEAkj7sxjE886n0OUVVCUx0YPNg=="],
 
-    "@drizzle-team/brocli": ["@drizzle-team/brocli@0.11.0", "", {}, "sha512-hD3pekGiPg0WPCCGAZmusBBJsDqGUR66Y452YgQsZOnkdQ7ViEPKuyP4huUGEZQefp8g34RRodXYmJ2TbCH+tg=="],
+    "@drizzle-team/brocli": ["@drizzle-team/brocli@0.12.0", "", {}, "sha512-mlUE+rZ8CatQekLhnaiN91Iemdd+e2gFKooGlnRB3oPTL3VghLfX24dx7HrzMNeC1JrIB/0kpsfyty3f5HNfxQ=="],
 
     "@effect/opentelemetry": ["@effect/opentelemetry@4.0.0-rc.111", "", { "peerDependencies": { "@opentelemetry/api": ">=1.9.0 <2.0.0", "@opentelemetry/api-logs": ">=0.203.0 <0.300.0", "@opentelemetry/resources": ">=2.0.0 <3.0.0", "@opentelemetry/sdk-logs": ">=0.203.0 <0.300.0", "@opentelemetry/sdk-metrics": ">=2.0.0 <3.0.0", "@opentelemetry/sdk-trace-base": ">=2.0.0 <3.0.0", "@opentelemetry/sdk-trace-node": ">=2.0.0 <3.0.0", "@opentelemetry/sdk-trace-web": ">=2.0.0 <3.0.0", "@opentelemetry/semantic-conventions": ">=1.33.0 <2.0.0", "effect": "^4.0.0-rc.111" }, "optionalPeers": ["@opentelemetry/api", "@opentelemetry/api-logs", "@opentelemetry/resources", "@opentelemetry/sdk-logs", "@opentelemetry/sdk-metrics", "@opentelemetry/sdk-trace-base", "@opentelemetry/sdk-trace-node", "@opentelemetry/sdk-trace-web"] }, "sha512-bztAYpWoipn/a4tFjECMNvtWLkf2TS+Qn23PBI+f4q5DP9cBKfV00uhX8cwEVcfAtZd+lflSbGURHa4xlSewOA=="],
 
@@ -2155,6 +2157,20 @@
 
     "@opencode-ai/protocol": ["@opencode-ai/protocol@workspace:packages/protocol"],
 
+    "@opencode-ai/pty": ["@opencode-ai/pty@0.1.8", "", { "optionalDependencies": { "@opencode-ai/pty-darwin-arm64": "0.1.8", "@opencode-ai/pty-darwin-x64": "0.1.8", "@opencode-ai/pty-linux-arm64-gnu": "0.1.8", "@opencode-ai/pty-linux-arm64-musl": "0.1.8", "@opencode-ai/pty-linux-x64-gnu": "0.1.8", "@opencode-ai/pty-linux-x64-musl": "0.1.8" }, "bin": { "opencode-pty": "bin/opencode-pty.js" } }, "sha512-+OHAANB6xgtO4zVMK5Q3l50Iot2fYcHL003rNpvxxCXLkpdO90UnnRCL9jrll1SWkWoDZKaUDlR/Scw6TUTN4A=="],
+
+    "@opencode-ai/pty-darwin-arm64": ["@opencode-ai/pty-darwin-arm64@0.1.8", "", { "os": "darwin", "cpu": "arm64" }, "sha512-5DqVcCfa9P5zpM0QpvXD/QqZXFdXvtIoaBHVHjYnTt8oqlnmMWqG4OVowzPNQg07NudW6apwX1Jop1wuERgWGw=="],
+
+    "@opencode-ai/pty-darwin-x64": ["@opencode-ai/pty-darwin-x64@0.1.8", "", { "os": "darwin", "cpu": "x64" }, "sha512-keEDxfpy0mFJYmsTqea9IUbpLKpbe9yL4ISU8tYA0daWiKuR8ZlfLTlLGltaFNO7l7q6mURMeI4cZzwySOX/Kw=="],
+
+    "@opencode-ai/pty-linux-arm64-gnu": ["@opencode-ai/pty-linux-arm64-gnu@0.1.8", "", { "os": "linux", "cpu": "arm64" }, "sha512-NVJvlVtPbYRyjABwLkm1iboijpSZMXv4lfg4qda7UCzAgiDZ0cWauXeW5wtd9bbnUbHw4rdNVU6YweuwASgZXw=="],
+
+    "@opencode-ai/pty-linux-arm64-musl": ["@opencode-ai/pty-linux-arm64-musl@0.1.8", "", { "os": "linux", "cpu": "arm64" }, "sha512-GAhnCs7b7B8Hqa7AEN7k5ra4IqybFYePgbCSOHn3Y4mofp/5NjDlkOG/VFM7rHAy05iEmRi5WtOPRv0Mkca8zA=="],
+
+    "@opencode-ai/pty-linux-x64-gnu": ["@opencode-ai/pty-linux-x64-gnu@0.1.8", "", { "os": "linux", "cpu": "x64" }, "sha512-vSbty1L0BuOQlK7DmukuhdxNKv2vzqyxOjHmh+TULptMDbDZwE1EybASoXB7dYog9as85adFCz8ZTF7Boyfxjg=="],
+
+    "@opencode-ai/pty-linux-x64-musl": ["@opencode-ai/pty-linux-x64-musl@0.1.8", "", { "os": "linux", "cpu": "x64" }, "sha512-2GhmJLl+Ly/JjzMS2Eqvk7SgeuHIB0cQ4Wrmnj4WKhr2V7lS+pdUEApN6S0SG/FK0TYJlWeHeOIpNZgDqzEPPg=="],
+
     "@opencode-ai/schema": ["@opencode-ai/schema@workspace:packages/schema"],
 
     "@opencode-ai/script": ["@opencode-ai/script@workspace:packages/script"],
@@ -2217,27 +2233,27 @@
 
     "@opentelemetry/semantic-conventions": ["@opentelemetry/semantic-conventions@1.43.0", "", {}, "sha512-eSYWTm620tTk45EKSedaUL8MFYI8hW164hIXsgIHyxu3VobUB3fFCu5t0hQby6OoWRPsG1KkKUG2M5UadiLiVg=="],
 
-    "@opentui/core": ["@opentui/core@0.5.7", "", { "dependencies": { "bun-ffi-structs": "0.3.1", "diff": "9.0.0", "marked": "17.0.1", "string-width": "7.2.0", "strip-ansi": "7.1.2" }, "optionalDependencies": { "@opentui/core-darwin-arm64": "0.5.7", "@opentui/core-darwin-x64": "0.5.7", "@opentui/core-linux-arm64": "0.5.7", "@opentui/core-linux-arm64-musl": "0.5.7", "@opentui/core-linux-x64": "0.5.7", "@opentui/core-linux-x64-musl": "0.5.7", "@opentui/core-win32-arm64": "0.5.7", "@opentui/core-win32-x64": "0.5.7" }, "peerDependencies": { "web-tree-sitter": "0.25.10" } }, "sha512-/XDabTkfBs2Wy2FhlC4jvzbpphYAs4SlnCRKYEvi+metlXNTSAshSs43wqZ9O4IPd0E4EcQqdfeQ0sdc3yPpYw=="],
+    "@opentui/core": ["@opentui/core@0.5.8", "", { "dependencies": { "bun-ffi-structs": "0.3.1", "diff": "9.0.0", "marked": "17.0.1", "string-width": "7.2.0", "strip-ansi": "7.1.2" }, "optionalDependencies": { "@opentui/core-darwin-arm64": "0.5.8", "@opentui/core-darwin-x64": "0.5.8", "@opentui/core-linux-arm64": "0.5.8", "@opentui/core-linux-arm64-musl": "0.5.8", "@opentui/core-linux-x64": "0.5.8", "@opentui/core-linux-x64-musl": "0.5.8", "@opentui/core-win32-arm64": "0.5.8", "@opentui/core-win32-x64": "0.5.8" }, "peerDependencies": { "web-tree-sitter": "0.25.10" } }, "sha512-GbZ+nSLYZqxj2Z5TU19Mx6IsAvVsn2+7WEXz+6OlMGoootvt3TxP3vWwfGI0jWk9qp1ftRlML1JPNEkzy+9I8g=="],
 
-    "@opentui/core-darwin-arm64": ["@opentui/core-darwin-arm64@0.5.7", "", { "os": "darwin", "cpu": "arm64" }, "sha512-75TDJgFD6hDoCElIX35Yg3TIRF/jhrtVQO/9snhjGuTsZ7bd0W88jUlvSXSNhqB3CX431rwgi47B+asWPDI1lQ=="],
+    "@opentui/core-darwin-arm64": ["@opentui/core-darwin-arm64@0.5.8", "", { "os": "darwin", "cpu": "arm64" }, "sha512-c9Y1FBrSnA4sKUCMETsrLYOmsMTyJae8mU9cE6M4o9rXcr3ZLPA7o9AkPA3S0+kH+kZ0o1Fn0wLejgNxvEp5mg=="],
 
-    "@opentui/core-darwin-x64": ["@opentui/core-darwin-x64@0.5.7", "", { "os": "darwin", "cpu": "x64" }, "sha512-j+Dwu2yV8zahBFjnVIssYdPrHz/+pkd7xWxI+nHjV41V3c+scvH6vJY1U85VCVaoQs/zCH5xAJbmAAn7Sgeh8g=="],
+    "@opentui/core-darwin-x64": ["@opentui/core-darwin-x64@0.5.8", "", { "os": "darwin", "cpu": "x64" }, "sha512-oZ/6Iz1KN+4volMFKmmvziYJhMgyyJ99LfK1S+uPRyIRqzT3CESoJ58D4h04M1g0dHeHE4PvkU1p3uQKrXiP3g=="],
 
-    "@opentui/core-linux-arm64": ["@opentui/core-linux-arm64@0.5.7", "", { "os": "linux", "cpu": "arm64" }, "sha512-gWO9NWaivXRPc0XhEbxehj3ApyC1SW8Bf1cnHnME1zhw/EZNCNE3TQHgKwOnnNKtvlbhzLQjOI4OqyPYV8UcEQ=="],
+    "@opentui/core-linux-arm64": ["@opentui/core-linux-arm64@0.5.8", "", { "os": "linux", "cpu": "arm64" }, "sha512-N6i/ocrsTjIq9aUQyrfJkqUo+tc4P5ZS6xz38Cm1MhtDzwBE+CrNIJdHfoJfmRLhJkYHlw6Z0ToICn8ZNj4Bpw=="],
 
-    "@opentui/core-linux-arm64-musl": ["@opentui/core-linux-arm64-musl@0.5.7", "", { "os": "linux", "cpu": "arm64" }, "sha512-RumSHTasIAWU7jPBKiTuEZG/m+prfQKCHhL3JAQnBLgp7eAB20XZGfSLVWDBgxkyKq1rHCUgLtZktC5ZAyKfTA=="],
+    "@opentui/core-linux-arm64-musl": ["@opentui/core-linux-arm64-musl@0.5.8", "", { "os": "linux", "cpu": "arm64" }, "sha512-eFMB41AWODaYf8PsCx3vtTMX33tFgtSpR9tKNnUErCXlcnZ+WCRHVaJ9F6DrVNX3ir6HmgbwAmaWXvaVF+kLtg=="],
 
-    "@opentui/core-linux-x64": ["@opentui/core-linux-x64@0.5.7", "", { "os": "linux", "cpu": "x64" }, "sha512-fXDmrIlfp9xaoVkk3BNn3yUO/b7plBSOfUh2oWOezRA6E/g5z+hTO1alGFQCy+VV+1kfA6iuYadAww0xNtfACg=="],
+    "@opentui/core-linux-x64": ["@opentui/core-linux-x64@0.5.8", "", { "os": "linux", "cpu": "x64" }, "sha512-/2QM7/wMnML/sxchzbwgoU5tUu/7k836/kSOKMti8opjuecv1K+WWNKGXufhTNeRcXFZaba5rsCdYrf3VqnVsQ=="],
 
-    "@opentui/core-linux-x64-musl": ["@opentui/core-linux-x64-musl@0.5.7", "", { "os": "linux", "cpu": "x64" }, "sha512-1opDV+W7C1F4iWivCNKuYl5Hen+IeFXAuUPm3hVyN0UtdcP8LESpFJTr8QRwHpGSE9Jz1K2g2S2ipzI3u5mhgw=="],
+    "@opentui/core-linux-x64-musl": ["@opentui/core-linux-x64-musl@0.5.8", "", { "os": "linux", "cpu": "x64" }, "sha512-YXo+qUHYmep2uvv3ECvTeqr10aD7+lBsavYmsTLBzS5hHabbzlQ10oX/99nIRPC3Au1BMWD6d6zQP5czPx23eg=="],
 
-    "@opentui/core-win32-arm64": ["@opentui/core-win32-arm64@0.5.7", "", { "os": "win32", "cpu": "arm64" }, "sha512-gbiZUyttjq8s+bsnsaopygWN7LTS/RyzD6GOmpgOB2TVitvo0P6mxM+AebBGuu/7xri17Rqaxf3uqrgdKiJAeg=="],
+    "@opentui/core-win32-arm64": ["@opentui/core-win32-arm64@0.5.8", "", { "os": "win32", "cpu": "arm64" }, "sha512-7qBdhEAlh4tLFzW7nWLPlREtNiF6NZMDMi+4uDpUlAMhRavLr6wjcIcgfhNAF/puq06DjuZlPTwaMCgB3qOuwA=="],
 
-    "@opentui/core-win32-x64": ["@opentui/core-win32-x64@0.5.7", "", { "os": "win32", "cpu": "x64" }, "sha512-bDwon45lUxbV3rMcekTCg4mXcBu9uhdG6HLeLK5TVZf05/h+Ibkf2UeZ8A2jjQt+MK6jpYmxWwDCVCTzZ7EU7Q=="],
+    "@opentui/core-win32-x64": ["@opentui/core-win32-x64@0.5.8", "", { "os": "win32", "cpu": "x64" }, "sha512-Z76YaTKnmRDSHKdKa7iTBCXBQdInQLq7UG3qIE84nvyHCDtkhYtPWOaCzDisSgXC8Y6hJt6NiFbaNjQyxPkfpQ=="],
 
-    "@opentui/keymap": ["@opentui/keymap@0.5.7", "", { "dependencies": { "@opentui/core": "0.5.7" }, "peerDependencies": { "@opentui/react": "0.5.7", "@opentui/solid": "0.5.7", "react": ">=19.2.0", "solid-js": "1.9.12" }, "optionalPeers": ["@opentui/react", "@opentui/solid", "react", "solid-js"] }, "sha512-BqfSbjlLuctnew2aoPPdK3wpzJfEPp7ttLBkTu1wJTkps9AcfUgXqwloqvY4DUKjUBNFYYEpU/0CPxI4Blj4MA=="],
+    "@opentui/keymap": ["@opentui/keymap@0.5.8", "", { "dependencies": { "@opentui/core": "0.5.8" }, "peerDependencies": { "@opentui/react": "0.5.8", "@opentui/solid": "0.5.8", "react": ">=19.2.0", "solid-js": "1.9.12" }, "optionalPeers": ["@opentui/react", "@opentui/solid", "react", "solid-js"] }, "sha512-KQHKRnLZroSZIbHmGmSeDPsXi7Yykdu9x9ACmDzBFkgPGx0EIuXHqRcmrqmUaX485MQEZuCD9t6FdDsOMGyIFg=="],
 
-    "@opentui/solid": ["@opentui/solid@0.5.7", "", { "dependencies": { "@babel/core": "7.28.0", "@babel/preset-typescript": "7.27.1", "@opentui/core": "0.5.7", "babel-plugin-module-resolver": "5.0.2", "babel-preset-solid": "1.9.12", "entities": "7.0.1", "s-js": "^0.4.9" }, "peerDependencies": { "solid-js": "1.9.12" } }, "sha512-qrKAZd9xt4D67LXZUARg1Aw0dwEWi29GyOxpM6abqmyExGPMS7jih5Z9oMQJ9EXsouDKpW7Mejjq7WbX2AaecQ=="],
+    "@opentui/solid": ["@opentui/solid@0.5.8", "", { "dependencies": { "@babel/core": "7.28.0", "@babel/preset-typescript": "7.27.1", "@opentui/core": "0.5.8", "babel-plugin-module-resolver": "5.0.2", "babel-preset-solid": "1.9.12", "entities": "7.0.1", "s-js": "^0.4.9" }, "peerDependencies": { "solid-js": "1.9.12" } }, "sha512-L0NxuAU8XT+jlE5G90oA3kspqkof48b0hmzi5XLw+1gxnkxrkTb+YfKys+GzVK4UqhgwY9aW+TeDfrKbnfwCMw=="],
 
     "@oslojs/asn1": ["@oslojs/asn1@1.0.0", "", { "dependencies": { "@oslojs/binary": "1.0.0" } }, "sha512-zw/wn0sj0j0QKbIXfIlnEcTviaCzYOY3V5rAyjR6YtOByFtJiT574+8p9Wlach0lZH9fddD4yb9laEAIl4vXQA=="],
 
@@ -3217,12 +3233,6 @@
 
     "@valibot/to-json-schema": ["@valibot/to-json-schema@1.6.0", "", { "peerDependencies": { "valibot": "^1.3.0" } }, "sha512-d6rYyK5KVa2XdqamWgZ4/Nr+cXhxjy7lmpe6Iajw15J/jmU+gyxl2IEd1Otg1d7Rl3gOQL5reulnSypzBtYy1A=="],
 
-    "@vercel/cli-config": ["@vercel/cli-config@0.2.3", "", { "dependencies": { "xdg-app-paths": "5", "zod": "4.1.11" } }, "sha512-Ggh0Wmi92TUkUexmSUPkkDtvJmbjUr7IvF5T3FkSsWrXXs3GFzOujfxFpECdJZpux1JG4SWDv9BT4w++TDgD6A=="],
-
-    "@vercel/cli-exec": ["@vercel/cli-exec@1.0.1", "", { "dependencies": { "execa": "5.1.1" } }, "sha512-g9XerViJ/paZujufXYcu5XYI2vU2rtB4sgdpjUHde5RnOkdmpu0ngH46LCFGHoPXO/C+qDPSczIHIRN+8Q2YKQ=="],
-
-    "@vercel/functions": ["@vercel/functions@3.9.3", "", { "dependencies": { "@vercel/oidc": "3.8.4" }, "peerDependencies": { "@aws-sdk/credential-provider-web-identity": "*", "ws": ">=8" }, "optionalPeers": ["@aws-sdk/credential-provider-web-identity", "ws"] }, "sha512-cbzTdASCZDnufrABc8oO00e/FqlqFFSdld+iGZPhrWBDHP4Pu8ESKKYIzASqYvbYYUIWujBvEa5LLCcZzm7WEw=="],
-
     "@vercel/oidc": ["@vercel/oidc@3.2.0", "", {}, "sha512-UycprH3T6n3jH0k44NHMa7pnFHGu/N05MjojYr+Mc6I7obkoLIJujSWwin1pCvdy/eOxrI/l3uDLQsmcrOb4ug=="],
 
     "@vitejs/plugin-react": ["@vitejs/plugin-react@4.7.0", "", { "dependencies": { "@babel/core": "^7.28.0", "@babel/plugin-transform-react-jsx-self": "^7.27.1", "@babel/plugin-transform-react-jsx-source": "^7.27.1", "@rolldown/pluginutils": "1.0.0-beta.27", "@types/babel__core": "^7.20.5", "react-refresh": "^0.17.0" }, "peerDependencies": { "vite": "^4.2.0 || ^5.0.0 || ^6.0.0 || ^7.0.0" } }, "sha512-gUu9hwfWvvEDBBmgtAowQCojwZmJ5mcLn3aufeCsitijs3+f2NsrPtlAWIR6OPiqljl96GVCUbLe0HyqIpVaoA=="],
@@ -3835,9 +3845,9 @@
 
     "dotenv-expand": ["dotenv-expand@11.0.7", "", { "dependencies": { "dotenv": "^16.4.5" } }, "sha512-zIHwmZPRshsCdpMDyVsqGmgyP0yT8GAgXUnkdAoJisxvf33k7yO6OuoKmcTGuXPWSsm8Oh88nZicRLA9Y0rUeA=="],
 
-    "drizzle-kit": ["drizzle-kit@1.0.0-rc.2", "", { "dependencies": { "@drizzle-team/brocli": "^0.11.0", "@js-temporal/polyfill": "^0.5.1", "esbuild": "^0.25.10", "get-tsconfig": "^4.13.6", "jiti": "^2.6.1" }, "bin": { "drizzle-kit": "bin.cjs" } }, "sha512-TRxUmj1wDA2QCt3GvuhfamvIa66wJ7+MzSxBMKkpRtYScjHTumT9BE+x6daSzuEacSrPEuUH5/cW1uo5RkoPIg=="],
+    "drizzle-kit": ["drizzle-kit@1.0.0-rc.5-ab785fc", "", { "dependencies": { "@drizzle-team/brocli": "^0.12.0", "@js-temporal/polyfill": "^0.5.1", "esbuild": "^0.25.10", "get-tsconfig": "^4.13.6", "jiti": "^2.6.1" }, "bin": { "drizzle-kit": "bin.cjs" } }, "sha512-TSg7sK1finOxdo48O9mDfyMY3lhi/aHCVIil5TFhthyq/H/89H/3nwX5V9Gdp7HYMUGFHyWJD5iowSn98l0elw=="],
 
-    "drizzle-orm": ["drizzle-orm@1.0.0-rc.2", "", { "peerDependencies": { "@aws-sdk/client-rds-data": ">=3", "@cloudflare/workers-types": ">=4", "@effect/sql-pg": ">=4.0.0-beta.58 || >=4.0.0", "@electric-sql/pglite": ">=0.2.0", "@libsql/client": ">=0.10.0", "@libsql/client-wasm": ">=0.10.0", "@neondatabase/serverless": ">=0.10.0", "@op-engineering/op-sqlite": ">=2", "@opentelemetry/api": "^1.4.1", "@planetscale/database": ">=1.13", "@sinclair/typebox": ">=0.34.8", "@sqlitecloud/drivers": ">=1.0.653", "@tidbcloud/serverless": "*", "@tursodatabase/database": ">=0.2.1", "@tursodatabase/database-common": ">=0.2.1", "@tursodatabase/database-wasm": ">=0.2.1", "@types/better-sqlite3": "*", "@types/mssql": "^9.1.4", "@types/pg": "*", "@types/sql.js": "*", "@upstash/redis": ">=1.34.7", "@vercel/postgres": ">=0.8.0", "@xata.io/client": "*", "arktype": ">=2.0.0", "better-sqlite3": ">=9.3.0", "bun-types": "*", "effect": ">=4.0.0-beta.58 || >=4.0.0", "expo-sqlite": ">=14.0.0", "mssql": "^11.0.1", "mysql2": ">=2", "pg": ">=8", "postgres": ">=3", "sql.js": ">=1", "sqlite3": ">=5", "typebox": ">=1.0.0", "valibot": ">=1.0.0-beta.7", "zod": "^3.25.0 || ^4.0.0" }, "optionalPeers": ["@aws-sdk/client-rds-data", "@cloudflare/workers-types", "@effect/sql-pg", "@electric-sql/pglite", "@libsql/client", "@libsql/client-wasm", "@neondatabase/serverless", "@op-engineering/op-sqlite", "@opentelemetry/api", "@planetscale/database", "@sinclair/typebox", "@sqlitecloud/drivers", "@tidbcloud/serverless", "@tursodatabase/database", "@tursodatabase/database-common", "@tursodatabase/database-wasm", "@types/better-sqlite3", "@types/mssql", "@types/pg", "@types/sql.js", "@upstash/redis", "@vercel/postgres", "@xata.io/client", "arktype", "better-sqlite3", "bun-types", "effect", "expo-sqlite", "mssql", "mysql2", "pg", "postgres", "sql.js", "sqlite3", "typebox", "valibot", "zod"] }, "sha512-UXYDkbplF5wX0hwxll+80QhEwUvAJLBu+tAK/d4fna18kLE6VuliAzufF/ieDEIJeSnLRYgtmsXD6x1Xuy1kIg=="],
+    "drizzle-orm": ["drizzle-orm@1.0.0-rc.5-169397b", "", { "peerDependencies": { "@aws-sdk/client-rds-data": ">=3", "@cloudflare/workers-types": ">=4", "@drizzle-team/minipg": ">=0.3.2", "@effect/sql-d1": ">=4.0.0-beta.105 || >=4.0.0", "@effect/sql-libsql": ">=4.0.0-beta.105 || >=4.0.0", "@effect/sql-mysql2": ">=4.0.0-beta.105 || >=4.0.0", "@effect/sql-pg": ">=4.0.0-beta.105 || >=4.0.0", "@effect/sql-pglite": ">=4.0.0-beta.105 || >=4.0.0", "@effect/sql-sqlite-bun": ">=4.0.0-beta.105 || >=4.0.0", "@effect/sql-sqlite-do": ">=4.0.0-beta.105 || >=4.0.0", "@effect/sql-sqlite-node": ">=4.0.0-beta.105 || >=4.0.0", "@effect/sql-sqlite-wasm": ">=4.0.0-beta.105 || >=4.0.0", "@electric-sql/pglite": ">=0.2.0", "@libsql/client": ">=0.10.0", "@libsql/client-wasm": ">=0.10.0", "@neondatabase/serverless": ">=0.10.0", "@op-engineering/op-sqlite": ">=17", "@opentelemetry/api": "^1.4.1", "@planetscale/database": ">=1.13", "@sinclair/typebox": ">=0.34.8", "@sqlitecloud/drivers": ">=1.0.653", "@tidbcloud/serverless": "*", "@tursodatabase/database": ">=0.7.1", "@tursodatabase/database-common": ">=0.7.1", "@tursodatabase/database-wasm": ">=0.7.1", "@tursodatabase/serverless": ">=1.4.0", "@tursodatabase/sync": ">=0.7.1", "@types/better-sqlite3": "*", "@types/mssql": "^9.1.4", "@types/pg": "*", "@types/sql.js": "*", "@upstash/redis": ">=1.34.7", "@vercel/postgres": ">=0.8.0", "@xata.io/client": "*", "arktype": ">=2.0.0", "better-sqlite3": ">=9.3.0", "bun-types": "*", "effect": ">=4.0.0-beta.105 || >=4.0.0", "expo-sqlite": ">=14.0.0", "mssql": "^11.0.1", "mysql2": ">=2", "pg": ">=8", "postgres": ">=3", "sql.js": ">=1", "sqlite3": ">=5", "typebox": ">=1.2.0", "valibot": ">=1.0.0-beta.7", "zod": "^3.25.0 || ^4.0.0" }, "optionalPeers": ["@aws-sdk/client-rds-data", "@cloudflare/workers-types", "@drizzle-team/minipg", "@effect/sql-d1", "@effect/sql-libsql", "@effect/sql-mysql2", "@effect/sql-pg", "@effect/sql-pglite", "@effect/sql-sqlite-bun", "@effect/sql-sqlite-do", "@effect/sql-sqlite-node", "@effect/sql-sqlite-wasm", "@electric-sql/pglite", "@libsql/client", "@libsql/client-wasm", "@neondatabase/serverless", "@op-engineering/op-sqlite", "@opentelemetry/api", "@planetscale/database", "@sinclair/typebox", "@sqlitecloud/drivers", "@tidbcloud/serverless", "@tursodatabase/database", "@tursodatabase/database-common", "@tursodatabase/database-wasm", "@tursodatabase/serverless", "@tursodatabase/sync", "@types/better-sqlite3", "@types/mssql", "@types/pg", "@types/sql.js", "@upstash/redis", "@vercel/postgres", "@xata.io/client", "arktype", "better-sqlite3", "bun-types", "effect", "expo-sqlite", "mssql", "mysql2", "pg", "postgres", "sql.js", "sqlite3", "typebox", "valibot", "zod"] }, "sha512-xf5TudQhFplMMzjwgmMY5f64Sqbv4iwXxdTHBHqvLMt+SHyY8rAYhzNLQchEkDxiqMclTS/txPEOg5pdQy0Tcw=="],
 
     "dset": ["dset@3.1.4", "", {}, "sha512-2QF/g9/zTaPDc3BjNcVTGoBbXBgYfMTTceLaYcFJ/W9kggFUkhxD/hMEeuLKbugyef9SqAx8cpgwlIP/jinUTA=="],
 
@@ -3981,8 +3991,6 @@
 
     "eventsource-parser": ["eventsource-parser@3.1.1", "", {}, "sha512-EKN1vKAMcZ8MlYMpaNuxN6R9yakzH6uajHcHVTqWJzvu5pWw9DyhbP35HH8MVBQ+dZjAfDxk+A8NiR9KWaXiyQ=="],
 
-    "execa": ["execa@5.1.1", "", { "dependencies": { "cross-spawn": "^7.0.3", "get-stream": "^6.0.0", "human-signals": "^2.1.0", "is-stream": "^2.0.0", "merge-stream": "^2.0.0", "npm-run-path": "^4.0.1", "onetime": "^5.1.2", "signal-exit": "^3.0.3", "strip-final-newline": "^2.0.0" } }, "sha512-8uSpZZocAZRBAPIEINJj3Lo9HyGitllczc27Eh5YYojjMFMn8yHMDMaUHE2Jqfq05D/wucwI4JGURyXt1vchyg=="],
-
     "exit-hook": ["exit-hook@2.2.1", "", {}, "sha512-eNTPlAD67BmP31LDINZ3U7HSF8l57TxOY2PmBJ1shpCvpnxBF93mWCE8YHBnXs8qiUZJc9WDcWIeC3a2HIAMfw=="],
 
     "expect-type": ["expect-type@1.4.0", "", {}, "sha512-KfYbmpRm0VbLjEvVa9yGwCi9GI34xvi7A/HXYWQO65CSD2u3MczUJSuwXKFIxlGsgBQizV9q5J9NHj4VG0n+pA=="],
@@ -4257,8 +4265,6 @@
 
     "https-proxy-agent": ["https-proxy-agent@7.0.6", "", { "dependencies": { "agent-base": "^7.1.2", "debug": "4" } }, "sha512-vK9P5/iUfdl95AI+JVyUuIcVtd4ofvtrOr3HNtM2yxC9bnMbEdp3x01OhQNnjb8IJYi38VlTE3mBXwcfvywuSw=="],
 
-    "human-signals": ["human-signals@2.1.0", "", {}, "sha512-B4FFZ6q/T2jhhksgkbEW3HBvWIfDW85snkQgawt07S7J5QXTk6BkNV+0yAeZrM5QpMAdYlocGoljn0sJ/WQkFw=="],
-
     "humanize-ms": ["humanize-ms@1.2.1", "", { "dependencies": { "ms": "^2.0.0" } }, "sha512-Fl70vYtsAFb/C06PTS9dZBo7ihau+Tu/DNCk/OyHhea07S+aeMWpFFkUaXRa8fI+ScZbEI8dfSxwY7gxZ9SAVQ=="],
 
     "husky": ["husky@9.1.7", "", { "bin": { "husky": "bin.js" } }, "sha512-5gs5ytaNjBrh5Ow3zrvdUUY+0VxIuWVL4i9irt6friV+BqdCfmV11CQTWMiBYWHbXhco+J1kHfTOUkePhCDvMA=="],
@@ -4623,8 +4629,6 @@
 
     "merge-descriptors": ["merge-descriptors@2.0.0", "", {}, "sha512-Snk314V5ayFLhp3fkUREub6WtjBfPdCPY1Ln8/8munuLuiYhsABgBVWsozAG+MWMbVEvcdcpbi9R7ww22l9Q3g=="],
 
-    "merge-stream": ["merge-stream@2.0.0", "", {}, "sha512-abv/qOcuPfk3URPfDzmZU1LKmuw8kT+0nIHvKrKgFrwifol/doWcdA4ZqsWQ8ENrFKkd67Mfpo/LovbIUsbt3w=="],
-
     "merge2": ["merge2@1.4.1", "", {}, "sha512-8q7VEgMJW4J8tcfVPy8g09NcQwZdbwFEqhe/WZkoIzjn/3TGDwtOCYtXGxA3O8tPzpczCCDgv+P2P5y00ZJOOg=="],
 
     "mermaid": ["mermaid@11.16.1", "", { "dependencies": { "@braintree/sanitize-url": "^7.1.2", "@iconify/utils": "^3.0.2", "@mermaid-js/parser": "^1.2.0", "@types/d3": "^7.4.3", "@upsetjs/venn.js": "^2.0.0", "cytoscape": "^3.33.3", "cytoscape-cose-bilkent": "^4.1.0", "cytoscape-fcose": "^2.2.0", "d3": "^7.9.0", "d3-sankey": "^0.12.3", "dagre-d3-es": "7.0.14", "dayjs": "^1.11.20", "dompurify": "^3.3.3", "es-toolkit": "^1.45.1", "katex": "^0.16.45", "khroma": "^2.1.0", "marked": "^16.3.0", "roughjs": "^4.6.6", "stylis": "^4.3.6", "ts-dedent": "^2.2.0", "uuid": "^11.1.0 || ^12 || ^13 || ^14.0.0" } }, "sha512-TQsq6u22fAn3rek5VOubrhKPo1g5hwC3FXUN9hiyupTckcYiGuuKGkNQrKYwGJkXUxZdojwRG46gsSCFZMDp4g=="],
@@ -4709,8 +4713,6 @@
 
     "mime-types": ["mime-types@3.0.2", "", { "dependencies": { "mime-db": "^1.54.0" } }, "sha512-Lbgzdk0h4juoQ9fCKXW4by0UJqj+nOOrI9MJ1sSj4nI8aI2eo1qmvQEie4VD1glsS250n15LsWsYtCugiStS5A=="],
 
-    "mimic-fn": ["mimic-fn@2.1.0", "", {}, "sha512-OqbOk5oEQeAZ8WXWydlu9HJjz9WVdEIvamMCcXmuqUYjTknH/sqsWvhQ3vgwKFRR1HpjvNBKQ37nbJgYzGqGcg=="],
-
     "mimic-function": ["mimic-function@5.0.1", "", {}, "sha512-VP79XUPxV2CigYP3jWwAUFSku2aKqBH7uTAapFWCBqutsbmDo96KY5o8uh6U+/YSIn5OxJnXp73beVkpqMIGhA=="],
 
     "mimic-response": ["mimic-response@3.1.0", "", {}, "sha512-z0yWI+4FDrrweS8Zmt4Ej5HdJmky15+L2e6Wgn3+iK5fWzb6T3fhNFq2+MeTRb064c6Wr4N/wv0DzQTjNzHNGQ=="],
@@ -4827,8 +4829,6 @@
 
     "npm-registry-fetch": ["npm-registry-fetch@19.1.1", "", { "dependencies": { "@npmcli/redact": "^4.0.0", "jsonparse": "^1.3.1", "make-fetch-happen": "^15.0.0", "minipass": "^7.0.2", "minipass-fetch": "^5.0.0", "minizlib": "^3.0.1", "npm-package-arg": "^13.0.0", "proc-log": "^6.0.0" } }, "sha512-TakBap6OM1w0H73VZVDf44iFXsOS3h+L4wVMXmbWOQroZgFhMch0juN6XSzBNlD965yIKvWg2dfu7NSiaYLxtw=="],
 
-    "npm-run-path": ["npm-run-path@4.0.1", "", { "dependencies": { "path-key": "^3.0.0" } }, "sha512-S48WzZW777zhNIrn7gxOlISNAqi9ZC/uQFnRdbeIHhZhCA6UqpkOT8T1G7BvfdgP4Er8gF4sUbaS0i7QvIfCWw=="],
-
     "nth-check": ["nth-check@2.1.1", "", { "dependencies": { "boolbase": "^1.0.0" } }, "sha512-lqjrjmaOoAnWfMmBPL+XNnynZh2+swxiX3WUE0s4yEHI6m+AwrK2UZOimIRl3X/4QctVqS8AiZjFqyOGrMXb/w=="],
 
     "object-assign": ["object-assign@4.1.1", "", {}, "sha512-rJgTQnkUnH1sFw8yT6VSU3zD3sWmu6sZhIseY8VX+GRu3P6F7Fu+JNDoXfklElbLJSnc3FUQHVe4cU5hj+BcUg=="],
@@ -4853,8 +4853,6 @@
 
     "once": ["once@1.4.0", "", { "dependencies": { "wrappy": "1" } }, "sha512-lNaJgI+2Q5URQBkccEKHTQOPaXdUxnZZElQTZY0MFUAuaEqe1E+Nyvgdz/aIyNi6Z9MzO5dv1H8n58/GELp3+w=="],
 
-    "onetime": ["onetime@5.1.2", "", { "dependencies": { "mimic-fn": "^2.1.0" } }, "sha512-kbpaSSGJTWdAY5KPVeMOKXSrPtr8C8C7wodJbcsd51jRnmD+GZu8Y0VoU6Dm5Z4vWr0Ig/1NKuWRKf7j5aaYSg=="],
-
     "oniguruma-parser": ["oniguruma-parser@0.12.2", "", {}, "sha512-6HVa5oIrgMC6aA6WF6XyyqbhRPJrKR02L20+2+zpDtO5QAzGHAUGw5TKQvwi5vctNnRHkJYmjAhRVQF2EKdTQw=="],
 
     "oniguruma-to-es": ["oniguruma-to-es@4.3.6", "", { "dependencies": { "oniguruma-parser": "^0.12.2", "regex": "^6.1.0", "regex-recursion": "^6.0.2" } }, "sha512-csuQ9x3Yr0cEIs/Zgx/OEt9iBw9vqIunAPQkx19R/fiMq2oGVTgcMqO/V3Ybqefr1TBvosI6jU539ksaBULJyA=="],
@@ -4869,8 +4867,6 @@
 
     "opentui-spinner": ["opentui-spinner@0.0.7", "", { "dependencies": { "cli-spinners": "^3.3.0" }, "peerDependencies": { "@opentui/core": "^0.3.4", "@opentui/react": "^0.3.4", "@opentui/solid": "^0.3.4", "typescript": "^5" }, "optionalPeers": ["@opentui/react", "@opentui/solid"] }, "sha512-nPzwAvJG+y9rVEwwHLHqbsMzLnIk2zw+F9LqwA7aYJvpM5gsrKC2rrGi36A+tZpA+1RnWxXeWEgVZMchnaH18Q=="],
 
-    "os-paths": ["os-paths@4.4.0", "", {}, "sha512-wrAwOeXp1RRMFfQY8Sy7VaGVmPocaLwSFOYCGKSyo8qmJ+/yaafCl5BCA1IQZWqFSRBrKDYFeR9d/VyQzfH/jg=="],
-
     "own-keys": ["own-keys@1.0.2", "", { "dependencies": { "call-bound": "^1.0.4", "get-intrinsic": "^1.3.0", "object-keys": "^1.1.1", "safe-push-apply": "^1.0.0" } }, "sha512-19YVAg7T+WTrxggPukVq7DjTv6+PJ867TmhCvBsYwmbFCsZd344rq2Ld1p0wo8f8Qrrhgp82c6FJRqdXWtSEhg=="],
 
     "oxc-minify": ["oxc-minify@0.96.0", "", { "optionalDependencies": { "@oxc-minify/binding-android-arm64": "0.96.0", "@oxc-minify/binding-darwin-arm64": "0.96.0", "@oxc-minify/binding-darwin-x64": "0.96.0", "@oxc-minify/binding-freebsd-x64": "0.96.0", "@oxc-minify/binding-linux-arm-gnueabihf": "0.96.0", "@oxc-minify/binding-linux-arm-musleabihf": "0.96.0", "@oxc-minify/binding-linux-arm64-gnu": "0.96.0", "@oxc-minify/binding-linux-arm64-musl": "0.96.0", "@oxc-minify/binding-linux-riscv64-gnu": "0.96.0", "@oxc-minify/binding-linux-s390x-gnu": "0.96.0", "@oxc-minify/binding-linux-x64-gnu": "0.96.0", "@oxc-minify/binding-linux-x64-musl": "0.96.0", "@oxc-minify/binding-wasm32-wasi": "0.96.0", "@oxc-minify/binding-win32-arm64-msvc": "0.96.0", "@oxc-minify/binding-win32-x64-msvc": "0.96.0" } }, "sha512-dXeeGrfPJJ4rMdw+NrqiCRtbzVX2ogq//R0Xns08zql2HjV3Zi2SBJ65saqfDaJzd2bcHqvGWH+M44EQCHPAcA=="],
@@ -5453,8 +5449,6 @@
 
     "strip-comments": ["strip-comments@2.0.1", "", {}, "sha512-ZprKx+bBLXv067WTCALv8SSz5l2+XhpYCsVtSqlMnkAXMWDq+/ekVbl1ghqP9rUHTzv6sm/DwCOiYutU/yp1fw=="],
 
-    "strip-final-newline": ["strip-final-newline@2.0.0", "", {}, "sha512-BrpvfNAE3dcvq7ll3xVumzjKjZQ5tI1sEUIKr3Uoks0XUl45St3FlatVqef9prk4jRDzhW6WZg+3bk93y6pLjA=="],
-
     "strip-indent": ["strip-indent@3.0.0", "", { "dependencies": { "min-indent": "^1.0.0" } }, "sha512-laJTa3Jb+VQpaC6DseHhF7dXVqHTfJPCRDaEbid/drOhgitgYku/letMUqOXFoWV0zIIUbjpdH2t+tYj4bQMRQ=="],
 
     "stripe": ["stripe@18.0.0", "", { "dependencies": { "@types/node": ">=8.1.0", "qs": "^6.11.0" } }, "sha512-3Fs33IzKUby//9kCkCa1uRpinAoTvj6rJgQ2jrBEysoxEvfsclvXdna1amyEYbA2EKkjynuB4+L/kleCCaWTpA=="],
@@ -5847,10 +5841,6 @@
 
     "wsl-utils": ["wsl-utils@0.1.0", "", { "dependencies": { "is-wsl": "^3.1.0" } }, "sha512-h3Fbisa2nKGPxCpm89Hk33lBLsnaGBvctQopaBSOW/uIs6FTe1ATyAnKFJrzVs9vpGdsTe73WF3V4lIsk4Gacw=="],
 
-    "xdg-app-paths": ["xdg-app-paths@5.5.1", "", { "dependencies": { "os-paths": "^4.0.1", "xdg-portable": "^7.2.0" } }, "sha512-hI3flOB4PLZIy5prbtTpirobtPE2ZtZ52szO+2mM9Efp6ErM398La+C1lIpNWDfNoQk+6Lsi6nMcCwVB7pxeMQ=="],
-
-    "xdg-portable": ["xdg-portable@7.3.0", "", { "dependencies": { "os-paths": "^4.0.1" } }, "sha512-sqMMuL1rc0FmMBOzCpd0yuy9trqF2yTTVe+E9ogwCSWQCdDEtQUwrZPT6AxqtsFGRNxycgncbP/xmOOSPw5ZUw=="],
-
     "xml-naming": ["xml-naming@0.3.0", "", {}, "sha512-ghig2TBE/H11aOVgmahA3MhimvkBr6JIYknH/Dhdk10nXwdbIqBJsbfMxpvFPG8bAw77gN29aQWvKpmVoPlvPQ=="],
 
     "xml2js": ["xml2js@0.5.0", "", { "dependencies": { "sax": ">=0.6.0", "xmlbuilder": "~11.0.0" } }, "sha512-drPFnkQJik/O+uPKpqSgr22mpuFHqKdbS835iAQrUC73L2F5WkboIRd63ai/2Yg6I1jzifPFKH2NTK+cfglkIA=="],
@@ -5933,10 +5923,6 @@
 
     "@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=="],
@@ -5975,10 +5961,6 @@
 
     "@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=="],
@@ -6157,6 +6139,8 @@
 
     "@babel/preset-env/semver": ["semver@6.3.1", "", { "bin": { "semver": "bin/semver.js" } }, "sha512-BR7VvDCVHO+q2xBEWskxS6DJE1qRnb7DxzUrogb71CWoSficBxYsiAGd+Kl0mmq/MprG9yArRkyrQxTO6XjMzA=="],
 
+    "@brendonovich/vite-plugin-opencode/@opencode-ai/client": ["@opencode-ai/client@0.0.0-beta-18050", "", { "dependencies": { "@opencode-ai/protocol": "0.0.0-beta-18050", "@opencode-ai/schema": "0.0.0-beta-18050" }, "peerDependencies": { "effect": "4.0.0-rc.111", "solid-js": ">=1.9.0" }, "optionalPeers": ["effect", "solid-js"] }, "sha512-zWZv5X23iyx+/mxwiAi18YY/VMjQofTaH7RyKMBt7KL6FmaKAWf9Q05zbQhrLX8DYJD3MOtbjBeQCGLsPUDU8g=="],
+
     "@bruits/satteri-wasm32-wasi/@emnapi/core": ["@emnapi/core@1.11.1", "", { "dependencies": { "@emnapi/wasi-threads": "1.2.2", "tslib": "^2.4.0" } }, "sha512-RSvbQmHzdKzNsLYa/wHrbc3KN4sYLKAdPZxqiM2HATqv/SBk2/ENSHpvXGaLOMcsAyz0poEGqkmmKYG3OWiJEQ=="],
 
     "@bruits/satteri-wasm32-wasi/@emnapi/runtime": ["@emnapi/runtime@1.11.1", "", { "dependencies": { "tslib": "^2.4.0" } }, "sha512-vgj7R3y3Wgx24IQaGPA/R6YFXLHVMOZ0uVEyIQPaWs+rd1AzfEMXlAC22FYwO1XkKR6NPsq7mUandH8oIRdZFw=="],
@@ -6423,10 +6407,6 @@
 
     "@testing-library/dom/dom-accessibility-api": ["dom-accessibility-api@0.5.16", "", {}, "sha512-X7BJ2yElsnOJ30pZF4uIIDfBEVgF4XEBxL9Bxhy6dnrm5hkzqmsWHGTiHqRiITNhMyFLyAiWndIJP7Z1NTteDg=="],
 
-    "@vercel/cli-config/zod": ["zod@4.1.11", "", {}, "sha512-WPsqwxITS2tzx1bzhIKsEs19ABD5vmCVa4xBo2tq/SrV4RNZtfws1EnCWQXM6yh8bD08a1idvkB5MZSBiZsjwg=="],
-
-    "@vercel/functions/@vercel/oidc": ["@vercel/oidc@3.8.4", "", { "dependencies": { "@vercel/cli-config": "0.2.3", "@vercel/cli-exec": "1.0.1", "jose": "^5.9.6" } }, "sha512-FGNvVZ5pgX9FaBqkPt6VkYFZ6bWAMDzYi7nxW+1Xt+Z4fn5PuTULVwsxjKc+0uKhysyWBQmvsmM50Oh6C2/oMA=="],
-
     "@vitejs/plugin-react/@rolldown/pluginutils": ["@rolldown/pluginutils@1.0.0-beta.27", "", {}, "sha512-+d0F4MKMCbeVUJwG96uQ4SgAznZNSq93I3V+9NHA4OpvqG8mRCpGdKmK8l/dl02h2CCDHwW2FqilnTyDcAnqjA=="],
 
     "@vitejs/plugin-react/vite": ["vite@7.1.11", "", { "dependencies": { "esbuild": "^0.25.0", "fdir": "^6.5.0", "picomatch": "^4.0.3", "postcss": "^8.5.6", "rollup": "^4.43.0", "tinyglobby": "^0.2.15" }, "optionalDependencies": { "fsevents": "~2.3.3" }, "peerDependencies": { "@types/node": "^20.19.0 || >=22.12.0", "jiti": ">=1.21.0", "less": "^4.0.0", "lightningcss": "^1.21.0", "sass": "^1.70.0", "sass-embedded": "^1.70.0", "stylus": ">=0.54.8", "sugarss": "^5.0.0", "terser": "^5.16.0", "tsx": "^4.8.1", "yaml": "^2.4.2" }, "optionalPeers": ["@types/node", "jiti", "less", "lightningcss", "sass", "sass-embedded", "stylus", "sugarss", "terser", "tsx", "yaml"], "bin": { "vite": "bin/vite.js" } }, "sha512-uzcxnSDVjAopEUjljkWh8EIrg6tlzrjFUfMcR1EVsRDGwf/ccef0qQPRyOrROwhrTDaApueq+ja+KLPlzR/zdg=="],
@@ -6565,10 +6545,6 @@
 
     "esbuild-plugin-copy/chokidar": ["chokidar@3.6.0", "", { "dependencies": { "anymatch": "~3.1.2", "braces": "~3.0.2", "glob-parent": "~5.1.2", "is-binary-path": "~2.1.0", "is-glob": "~4.0.1", "normalize-path": "~3.0.0", "readdirp": "~3.6.0" }, "optionalDependencies": { "fsevents": "~2.3.2" } }, "sha512-7VT13fmjotKpGipCW9JEQAusEPE+Ei8nl6/g4FBAmIm0GOOLMua9NDDo/DWp0ZAxCr3cPq5ZpBqmPAQgDda2Pw=="],
 
-    "execa/get-stream": ["get-stream@6.0.1", "", {}, "sha512-ts6Wi+2j3jQjqi70w5AlN8DFnkSwC+MqmxEzdEALB2qXZYV3X/b1CTfgPLGJNMeAWxdPfU8FO1ms3NUfaHCPYg=="],
-
-    "execa/signal-exit": ["signal-exit@3.0.7", "", {}, "sha512-wnD2ZE+l+SPC/uoS0vXeE9L1+0wuaMqKlfz9AMUo38JsyLSBWSFcHR1Rri62LZc12vLr1gb3jl7iwQhgwpAbGQ=="],
-
     "express/cookie": ["cookie@0.7.2", "", {}, "sha512-yki5XnKuf750l50uGTllt6kKILY4nQ1eNIQatoXEByZ5dWgnKqbnqmTrBE5B4N7lrMJKQ2ytWMiTO2o0v6Ew/w=="],
 
     "fetch-blob/web-streams-polyfill": ["web-streams-polyfill@3.3.3", "", {}, "sha512-d2JWLCivmZYTSIoge9MsgFCZrt571BikcWGYkjC1khllbTeDlGqZ2D8vD8E/lJa8WGWbb7Plm8/XJYV7IJHZZw=="],
@@ -6989,6 +6965,10 @@
 
     "@babel/helper-compilation-targets/lru-cache/yallist": ["yallist@3.1.1", "", {}, "sha512-a4UGQaWPH59mOXUYnAG2ewncQS4i4F43Tv3JoAM+s2VDAmS9NsK8GpDMLrCHPksFT7h3K6TOoUNn2pb7RoXx4g=="],
 
+    "@brendonovich/vite-plugin-opencode/@opencode-ai/client/@opencode-ai/protocol": ["@opencode-ai/protocol@0.0.0-beta-18050", "", { "dependencies": { "@opencode-ai/schema": "0.0.0-beta-18050", "effect": "4.0.0-rc.111" } }, "sha512-HDQMnvGp8IU0MdBRbEuydX1WQm09BZ4HJm9iSMQwzweJuQ2HNscgzHJPIH6P02BsbbtfJ8J7sZGPItrz1tWSgw=="],
+
+    "@brendonovich/vite-plugin-opencode/@opencode-ai/client/@opencode-ai/schema": ["@opencode-ai/schema@0.0.0-beta-18050", "", { "dependencies": { "@standard-schema/spec": "1.1.0", "effect": "4.0.0-rc.111" } }, "sha512-/D6VXaWlytTXR3IOiMLIKuPcfp7FQNUzRPm9z3K7UBFd1Bw4q/WZksaf5RVcBGz+0YRxYMc1V4D7MFlceSgtyg=="],
+
     "@bruits/satteri-wasm32-wasi/@emnapi/core/@emnapi/wasi-threads": ["@emnapi/wasi-threads@1.2.2", "", { "dependencies": { "tslib": "^2.4.0" } }, "sha512-c95qOXkHdydNKhscBTebqEC1CVAZpyqOfVfBzQ1qgzyl3gfeldUjIggDbIZgDKsHLgnsM+igH7TJ/eAasaVuMA=="],
 
     "@electron/asar/minimatch/brace-expansion": ["brace-expansion@1.1.18", "", { "dependencies": { "balanced-match": "^1.0.0", "concat-map": "0.0.1" } }, "sha512-Edep/X9fGqVNmzKBVsDYIOtD+z1tuezV70LBjdCst9Tqu76lsnvRiZ6oTic1n+/BIwX6QDGAO94PN4N2SADvtw=="],
@@ -7325,8 +7305,6 @@
 
     "@tailwindcss/node/lightningcss/lightningcss-win32-x64-msvc": ["lightningcss-win32-x64-msvc@1.32.0", "", { "os": "win32", "cpu": "x64" }, "sha512-Amq9B/SoZYdDi1kFrojnoqPLxYhQ4Wo5XiL8EVJrVsB8ARoC1PWW6VGtT0WKCemjy8aC+louJnjS7U18x3b06Q=="],
 
-    "@vercel/functions/@vercel/oidc/jose": ["jose@5.9.6", "", {}, "sha512-AMlnetc9+CV9asI19zHmrgS/WYsWUwCn2R7RzlbJWD7F9eWYUTGyBmU9o6PxngtLGOiDGPRu+Uc4fhKzbpteZQ=="],
-
     "@vitest/expect/@vitest/utils/@vitest/pretty-format": ["@vitest/pretty-format@3.2.4", "", { "dependencies": { "tinyrainbow": "^2.0.0" } }, "sha512-IVNZik8IVRJRTr9fxlitMKeJeXFFFN0JaB9PHPGQ8NKQbGpfjlTx9zO4RefN8gp7eqjNy8nyK3NZmBzOPeIxtA=="],
 
     "ai-gateway-provider/@ai-sdk/openai/@ai-sdk/provider": ["@ai-sdk/provider@3.0.14", "", { "dependencies": { "json-schema": "^0.4.0" } }, "sha512-5X1k57JBJ4H7H1QjX7CnJYAB1I19r/trVZTMcSms7/kLNZ8RaU4Nt2agcwZzv82Hfx6Q7/TOLU7agAKeFfc8cA=="],

bunfig.toml

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

nix/hashes.json

@@ -1,8 +1,8 @@
 {
   "nodeModules": {
-    "x86_64-linux": "sha256-pnyqHqM8nt1Dh/d6xRm8DploZG9PiaO3HWGVIFyew54=",
-    "aarch64-linux": "sha256-cWPfoDku1bFN7zwCy6rsLP/WMmp6hEpO3Ous3FPGcfo=",
-    "aarch64-darwin": "sha256-2S3opWupuF3AncBbZLd5B7IsrUxYxA4S1mKWy58gMlo=",
-    "x86_64-darwin": "sha256-fu6WCBh9CWqsk8ACvLdeleP55uZ08i8MfEPFqgt7ob0="
+    "x86_64-linux": "sha256-6Fc26RulNnJZtMUxeryWYqCt/7xIs1E+cOQSddAC9LA=",
+    "aarch64-linux": "sha256-p7cyVH12IGAo6HyaoSgO1fBx5fNTCZGTpvbqzIpe+A8=",
+    "aarch64-darwin": "sha256-U90pWWqe7DRiUpY5KUUp4exQpa/ddb/bQ5Ldxe46LfQ=",
+    "x86_64-darwin": "sha256-Ak6omOq2BBRLQEMxPydiQY9ajaJGQR61P6az4+xxZVM="
   }
 }

package.json

@@ -49,9 +49,9 @@
       "@octokit/rest": "22.0.0",
       "@hono/standard-validator": "0.2.0",
       "@hono/zod-validator": "0.4.2",
-      "@opentui/core": "0.5.7",
-      "@opentui/keymap": "0.5.7",
-      "@opentui/solid": "0.5.7",
+      "@opentui/core": "0.5.8",
+      "@opentui/keymap": "0.5.8",
+      "@opentui/solid": "0.5.8",
       "@tanstack/solid-virtual": "3.13.37",
       "@shikijs/stream": "4.2.0",
       "@standard-schema/spec": "1.1.0",
@@ -71,8 +71,8 @@
       "@tailwindcss/vite": "4.1.11",
       "diff": "8.0.2",
       "dompurify": "3.3.1",
-      "drizzle-kit": "1.0.0-rc.2",
-      "drizzle-orm": "1.0.0-rc.2",
+      "drizzle-kit": "1.0.0-rc.5-ab785fc",
+      "drizzle-orm": "1.0.0-rc.5-169397b",
       "effect": "4.0.0-rc.111",
       "ai": "6.0.168",
       "cross-spawn": "7.0.6",
@@ -166,7 +166,6 @@
     "@npmcli/agent@4.0.2": "patches/@npmcli%2Fagent@4.0.2.patch",
     "@silvia-odwyer/photon-node@0.3.4": "patches/@silvia-odwyer%2Fphoton-node@0.3.4.patch",
     "@standard-community/standard-openapi@0.2.9": "patches/@standard-community%2Fstandard-openapi@0.2.9.patch",
-    "drizzle-orm@1.0.0-rc.2": "patches/drizzle-orm@1.0.0-rc.2.patch",
     "solid-js@1.9.10": "patches/solid-js@1.9.10.patch",
     "@ai-sdk/mistral@3.0.51": "patches/@ai-sdk%2Fmistral@3.0.51.patch",
     "gcp-metadata@8.1.2": "patches/gcp-metadata@8.1.2.patch",

packages/ai/script/setup-recording-env.ts

@@ -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)),
   },

packages/ai/src/cache-policy.ts

@@ -36,7 +36,7 @@ const resolve = (policy: CachePolicy | undefined): CachePolicyObject => {
 // Protocols whose wire format ignores inline cache markers (OpenAI's implicit
 // prefix caching, Gemini's implicit + out-of-band CachedContent). Skip the
 // whole policy pass for these — emitting hints would be harmless but pointless.
-const RESPECTS_INLINE_HINTS = new Set(["anthropic-messages", "bedrock-converse", "openrouter"])
+const RESPECTS_INLINE_HINTS = new Set(["anthropic-messages", "google-vertex-messages", "bedrock-converse", "openrouter"])
 
 const makeHint = (ttlSeconds: number | undefined): CacheHint =>
   ttlSeconds !== undefined ? new CacheHint({ type: "ephemeral", ttlSeconds }) : new CacheHint({ type: "ephemeral" })

packages/ai/src/image-client.ts

@@ -1,7 +1,8 @@
 import { Context, Effect, Layer } from "effect"
 import { RequestExecutor } from "./route/executor.js"
+import { mergeHttpOptions, type AIError } from "./schema/index.js"
+import { sanitizeSurrogates } from "./utils/sanitize.js"
 import type { ImageOptions, ImageRequest, ImageRequestFor, ImageResponse } from "./image.js"
-import type { AIError } from "./schema/index.js"
 
 export type Execute = RequestExecutor.Interface["execute"]
 
@@ -26,7 +27,18 @@ export const layer: Layer.Layer<Service, never, RequestExecutor.Service> = Layer
   Effect.gen(function* () {
     const executor = yield* RequestExecutor.Service
     return Service.of({
-      generate: (request) => request.model.route.generate(request, executor.execute),
+      generate: (request) =>
+        request.model.route.generate(
+          {
+            ...sanitizeSurrogates({
+              ...request,
+              model: undefined,
+              http: mergeHttpOptions(request.model.http, request.http),
+            }),
+            model: request.model,
+          },
+          executor.execute,
+        ),
     })
   }),
 )

packages/ai/src/protocols/anthropic-messages.ts

@@ -1,4 +1,5 @@
-import { Effect, Schema } from "effect"
+import { Buffer } from "node:buffer"
+import { Effect, Option, Schema } from "effect"
 import { Tool } from "@opencode-ai/schema/tool"
 import { Route } from "../route/client.js"
 import { Auth } from "../route/auth.js"
@@ -54,7 +55,7 @@ export type ThinkingInput =
   | {
       readonly type: "disabled"
     }
-  | ({ readonly type: "enabled" } & (
+  | ({ readonly type: "enabled"; readonly display?: "summarized" | "omitted" } & (
       | { readonly budgetTokens: number; readonly budget_tokens?: number }
       | { readonly budgetTokens?: number; readonly budget_tokens: number }
     ))
@@ -63,6 +64,19 @@ export interface OptionsInput {
   readonly [key: string]: unknown
   readonly thinking?: ThinkingInput
   readonly effort?: string
+  readonly service_tier?: "auto" | "standard_only"
+  readonly serviceTier?: "auto" | "standard_only"
+  // SDK Metadata:2649 {user_id?: string | null}
+  readonly metadata?: { readonly user_id?: string | null }
+  // SDK MessageCreateParamsContainer:2596 ContainerParams|string
+  readonly container?: string | { readonly id?: string | null; readonly skills?: ReadonlyArray<Record<string, unknown>> | null }
+  readonly inference_geo?: string | null
+  readonly inferenceGeo?: string | null
+  readonly cache_control?: { readonly type: "ephemeral"; readonly ttl?: "5m" | "1h" }
+  readonly cacheControl?: { readonly type: "ephemeral"; readonly ttl?: "5m" | "1h" }
+  // SDK OutputConfig:2684 {effort, format: JSONOutputFormat}
+  readonly output_config?: { readonly effort?: string | null; readonly format?: { readonly type: "json_schema"; readonly schema: Record<string, unknown> } | null }
+  readonly outputConfig?: { readonly effort?: string | null; readonly format?: { readonly type: "json_schema"; readonly schema: Record<string, unknown> } | null }
 }
 
 export type ProviderOptionsInput = OptionsInput
@@ -82,32 +96,68 @@ const AnthropicTextBlock = Schema.Struct({
 })
 type AnthropicTextBlock = Schema.Schema.Type<typeof AnthropicTextBlock>
 
+// SDK: Base64ImageSource:201 {type:"base64", media_type:"image/jpeg"|... , data}, URLImageSource:3817 {type:"url", url}, FileImageSource:2350 {type:"file", file_id}
+// SDK: ImageBlockParam:2356 {source: Base64|URL|File, cache_control, transformations:2381 {oversized_image?}}
+const AnthropicBase64ImageSource = Schema.Struct({
+  type: Schema.tag("base64"),
+  media_type: Schema.String,
+  data: Schema.String,
+})
+const AnthropicURLImageSource = Schema.Struct({ type: Schema.tag("url"), url: Schema.String })
+const AnthropicFileImageSource = Schema.Struct({ type: Schema.tag("file"), file_id: Schema.String })
+const AnthropicImageSource = Schema.Union([
+  AnthropicBase64ImageSource,
+  AnthropicURLImageSource,
+  AnthropicFileImageSource,
+])
+const AnthropicImageTransformations = Schema.Struct({
+  oversized_image: Schema.optional(Schema.Literals(["downsize", "error"])),
+})
+
 const AnthropicImageBlock = Schema.Struct({
   type: Schema.tag("image"),
-  source: Schema.Struct({
-    type: Schema.tag("base64"),
-    media_type: Schema.String,
-    data: Schema.String,
-  }),
+  source: AnthropicImageSource,
   cache_control: Schema.optional(AnthropicCacheControl),
+  transformations: Schema.optional(AnthropicImageTransformations),
 })
 type AnthropicImageBlock = Schema.Schema.Type<typeof AnthropicImageBlock>
 
+// SDK: Base64PDFSource:209 {type:"base64", media_type:"application/pdf", data}, PlainTextSource:2716 {type:"text", media_type:"text/plain", data},
+// SDK: URLPDFSource:3823 {type:"url", url}, FileDocumentSource:2344 {type:"file", file_id}, ContentBlockSource:2266 {type:"content", content}
+// SDK: DocumentBlockParam:2297 {source: 5-way union, cache_control, citations, context, title}
+const AnthropicBase64PDFSource = Schema.Struct({
+  type: Schema.tag("base64"),
+  media_type: Schema.Literal("application/pdf"),
+  data: Schema.String,
+})
+const AnthropicPlainTextSource = Schema.Struct({
+  type: Schema.tag("text"),
+  media_type: Schema.Literal("text/plain"),
+  data: Schema.String,
+})
+const AnthropicURLPDFSource = Schema.Struct({ type: Schema.tag("url"), url: Schema.String })
+const AnthropicFileDocumentSource = Schema.Struct({ type: Schema.tag("file"), file_id: Schema.String })
+const AnthropicDocumentSource = Schema.Union([
+  AnthropicBase64PDFSource,
+  AnthropicPlainTextSource,
+  AnthropicURLPDFSource,
+  AnthropicFileDocumentSource,
+])
+
 const AnthropicDocumentBlock = Schema.Struct({
   type: Schema.tag("document"),
-  source: Schema.Struct({
-    type: Schema.tag("base64"),
-    media_type: Schema.Literal("application/pdf"),
-    data: Schema.String,
-  }),
+  source: AnthropicDocumentSource,
   cache_control: Schema.optional(AnthropicCacheControl),
+  title: Schema.optional(Schema.String),
+  context: Schema.optional(Schema.String),
+  citations: Schema.optional(Schema.Struct({ enabled: Schema.Boolean })),
 })
 type AnthropicDocumentBlock = Schema.Schema.Type<typeof AnthropicDocumentBlock>
 
 const AnthropicThinkingBlock = Schema.Struct({
   type: Schema.tag("thinking"),
   thinking: Schema.String,
-  signature: Schema.optional(Schema.String),
+  signature: Schema.String,
   cache_control: Schema.optional(AnthropicCacheControl),
 })
 
@@ -205,14 +255,18 @@ const AnthropicTool = Schema.Struct({
 type AnthropicTool = Schema.Schema.Type<typeof AnthropicTool>
 
 const AnthropicToolChoice = Schema.Union([
-  Schema.Struct({ type: Schema.Literals(["auto", "any", "none"]) }),
-  Schema.Struct({ type: Schema.tag("tool"), name: Schema.String }),
+  Schema.Struct({
+    type: Schema.Literals(["auto", "any", "none"]),
+    disable_parallel_tool_use: Schema.optional(Schema.Boolean),
+  }),
+  Schema.Struct({ type: Schema.tag("tool"), name: Schema.String, disable_parallel_tool_use: Schema.optional(Schema.Boolean) }),
 ])
 
 const AnthropicThinking = Schema.Union([
   Schema.Struct({
     type: Schema.tag("enabled"),
     budget_tokens: Schema.Number,
+    display: Schema.optional(Schema.Literals(["summarized", "omitted"])),
   }),
   Schema.Struct({
     type: Schema.tag("adaptive"),
@@ -223,10 +277,28 @@ const AnthropicThinking = Schema.Union([
   }),
 ])
 
+// SDK OutputConfig:2684 {effort?: "low"|"medium"|"high"|"xhigh"|"max"|null, format?: JSONOutputFormat:2399}
+const AnthropicJsonOutputFormat = Schema.Struct({
+  type: Schema.Literal("json_schema"),
+  schema: JsonObject,
+})
 const AnthropicOutputConfig = Schema.Struct({
   effort: Schema.optional(Schema.String),
+  format: Schema.optional(Schema.NullOr(AnthropicJsonOutputFormat)),
 })
 
+// SDK Metadata:2649 {user_id?: string|null}
+const AnthropicMetadata = Schema.Struct({ user_id: optionalNull(Schema.String) })
+
+// SDK MessageCreateParamsContainer:2596 ContainerParams|string; ContainerParams:2172 {id?, skills?}
+const AnthropicContainer = Schema.Union([
+  Schema.String,
+  Schema.Struct({
+    id: optionalNull(Schema.String),
+    skills: optionalNull(Schema.Array(JsonObject)),
+  }),
+])
+
 const AnthropicBodyFields = {
   model: Schema.String,
   system: optionalArray(AnthropicTextBlock),
@@ -241,6 +313,12 @@ const AnthropicBodyFields = {
   stop_sequences: optionalArray(Schema.String),
   thinking: Schema.optional(AnthropicThinking),
   output_config: Schema.optional(AnthropicOutputConfig),
+  // SDK top-level passthrough: cache_control:4638, container:4643, inference_geo:4649, metadata:4654, service_tier:4670
+  cache_control: Schema.optional(AnthropicCacheControl),
+  container: Schema.optional(Schema.NullOr(AnthropicContainer)),
+  inference_geo: Schema.optional(Schema.NullOr(Schema.String)),
+  metadata: Schema.optional(AnthropicMetadata),
+  service_tier: Schema.optional(Schema.Literals(["auto", "standard_only"])),
 }
 export const AnthropicMessagesBody = Schema.Struct(AnthropicBodyFields)
 export type AnthropicMessagesBody = Schema.Schema.Type<typeof AnthropicMessagesBody>
@@ -283,6 +361,8 @@ const AnthropicStreamBlock = Schema.Struct({
   tool_use_id: Schema.optional(Schema.String),
   content: Schema.optional(Schema.Unknown),
 })
+type AnthropicStreamBlock = Schema.Schema.Type<typeof AnthropicStreamBlock>
+const decodeAnthropicStreamBlock = Schema.decodeUnknownOption(AnthropicStreamBlock)
 
 const AnthropicStreamDelta = Schema.Struct({
   type: Schema.optional(Schema.String),
@@ -293,13 +373,15 @@ const AnthropicStreamDelta = Schema.Struct({
   stop_reason: optionalNull(Schema.String),
   stop_sequence: optionalNull(Schema.String),
 })
+type AnthropicStreamDelta = Schema.Schema.Type<typeof AnthropicStreamDelta>
+const decodeAnthropicStreamDelta = Schema.decodeUnknownOption(AnthropicStreamDelta)
 
 const AnthropicEvent = Schema.Struct({
   type: Schema.String,
   index: Schema.optional(Schema.Number),
   message: Schema.optional(Schema.Struct({ usage: Schema.optional(AnthropicUsage) })),
-  content_block: Schema.optional(AnthropicStreamBlock),
-  delta: Schema.optional(AnthropicStreamDelta),
+  content_block: Schema.optional(Schema.Unknown),
+  delta: Schema.optional(Schema.Unknown),
   usage: Schema.optional(AnthropicUsage),
   // `type` and `message` are both required per Anthropic's spec, but
   // OpenAI-compatible proxies and gateway translations occasionally drop one
@@ -369,10 +451,26 @@ const lowerTool = (breakpoints: Cache.Breakpoints, tool: ToolDefinition, inputSc
 
 const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>) =>
   ProviderShared.matchToolChoice("Anthropic Messages", toolChoice, {
-    auto: () => ({ type: "auto" as const }),
+    auto: () => ({
+      type: "auto" as const,
+      ...(toolChoice.disableParallelToolUse === undefined
+        ? {}
+        : { disable_parallel_tool_use: toolChoice.disableParallelToolUse }),
+    }),
     none: () => ({ type: "none" as const }),
-    required: () => ({ type: "any" as const }),
-    tool: (name) => ({ type: "tool" as const, name }),
+    required: () => ({
+      type: "any" as const,
+      ...(toolChoice.disableParallelToolUse === undefined
+        ? {}
+        : { disable_parallel_tool_use: toolChoice.disableParallelToolUse }),
+    }),
+    tool: (name) => ({
+      type: "tool" as const,
+      name,
+      ...(toolChoice.disableParallelToolUse === undefined
+        ? {}
+        : { disable_parallel_tool_use: toolChoice.disableParallelToolUse }),
+    }),
   })
 
 const scrubToolCallID = (id: string) => id.replace(/[^a-zA-Z0-9_-]/g, "_")
@@ -411,7 +509,152 @@ const lowerServerToolResult = Effect.fn("AnthropicMessages.lowerServerToolResult
   return { type: wireType, tool_use_id: scrubToolCallID(part.id), content: payload } satisfies AnthropicServerToolResultBlock
 })
 
-const lowerMedia = Effect.fn("AnthropicMessages.lowerMedia")(function* (part: MediaPart) {
+const fileIdFromMetadata = (metadata: MediaPart["metadata"]): string | undefined => {
+  if (!ProviderShared.isRecord(metadata)) return undefined
+  const anthropic = metadata.anthropic
+  if (ProviderShared.isRecord(anthropic)) {
+    if (typeof anthropic.file_id === "string") return anthropic.file_id
+    if (typeof anthropic.fileId === "string") return anthropic.fileId
+  }
+  if (typeof metadata.file_id === "string") return metadata.file_id
+  if (typeof metadata.fileId === "string") return metadata.fileId
+  return undefined
+}
+
+const transformationsFromMetadata = (
+  metadata: MediaPart["metadata"],
+): AnthropicImageBlock["transformations"] | undefined => {
+  if (!ProviderShared.isRecord(metadata)) return undefined
+  const anthropic = ProviderShared.isRecord(metadata.anthropic) ? metadata.anthropic : undefined
+  const raw = anthropic?.transformations ?? metadata.transformations
+  if (ProviderShared.isRecord(raw)) {
+    const value = raw.oversized_image
+    if (value === "downsize" || value === "error") return { oversized_image: value }
+  }
+  if (anthropic && (anthropic.oversized_image === "downsize" || anthropic.oversized_image === "error"))
+    return { oversized_image: anthropic.oversized_image }
+  return undefined
+}
+
+const documentTitleFromPart = (part: MediaPart): string | undefined => {
+  if (ProviderShared.isRecord(part.metadata)) {
+    const anthropic = part.metadata.anthropic
+    if (ProviderShared.isRecord(anthropic) && typeof anthropic.title === "string") return anthropic.title
+    if (typeof part.metadata.title === "string") return part.metadata.title
+  }
+  if (typeof part.filename === "string" && part.filename.length > 0) return part.filename
+  return undefined
+}
+
+const documentContextFromMetadata = (metadata: MediaPart["metadata"]): string | undefined => {
+  if (!ProviderShared.isRecord(metadata)) return undefined
+  const anthropic = ProviderShared.isRecord(metadata.anthropic) ? metadata.anthropic : undefined
+  if (anthropic && typeof anthropic.context === "string") return anthropic.context
+  if (typeof metadata.context === "string") return metadata.context
+  return undefined
+}
+
+const citationsFromMetadata = (
+  metadata: MediaPart["metadata"],
+): AnthropicDocumentBlock["citations"] | undefined => {
+  if (!ProviderShared.isRecord(metadata)) return undefined
+  const raw = ProviderShared.isRecord(metadata.anthropic)
+    ? (metadata.anthropic.citations ?? metadata.citations)
+    : metadata.citations
+  if (ProviderShared.isRecord(raw) && typeof raw.enabled === "boolean") return { enabled: raw.enabled }
+  return undefined
+}
+
+const isHttpUrl = (value: string) => /^https?:\/\//i.test(value.trim())
+
+const lowerMedia = Effect.fn("AnthropicMessages.lowerMedia")(function* (
+  part: MediaPart,
+  breakpoints?: Cache.Breakpoints,
+) {
+  const mime = part.mediaType.toLowerCase()
+  const cacheControlValue = breakpoints ? cacheControl(breakpoints, part.cache) : undefined
+  const fileId = fileIdFromMetadata(part.metadata)
+
+  // SDK file sources: FileImageSource:2350 / FileDocumentSource:2344 {type:"file", file_id}
+  if (fileId) {
+    if (mime.startsWith("image/"))
+      return {
+        type: "image" as const,
+        source: { type: "file" as const, file_id: fileId },
+        ...(cacheControlValue === undefined ? {} : { cache_control: cacheControlValue }),
+        ...(transformationsFromMetadata(part.metadata) === undefined
+          ? {}
+          : { transformations: transformationsFromMetadata(part.metadata)! }),
+      } satisfies AnthropicImageBlock
+    return {
+      type: "document" as const,
+      source: { type: "file" as const, file_id: fileId },
+      ...(cacheControlValue === undefined ? {} : { cache_control: cacheControlValue }),
+      ...(documentTitleFromPart(part) === undefined ? {} : { title: documentTitleFromPart(part)! }),
+      ...(documentContextFromMetadata(part.metadata) === undefined
+        ? {}
+        : { context: documentContextFromMetadata(part.metadata)! }),
+      ...(citationsFromMetadata(part.metadata) === undefined
+        ? {}
+        : { citations: citationsFromMetadata(part.metadata)! }),
+    } satisfies AnthropicDocumentBlock
+  }
+
+  const rawString = typeof part.data === "string" ? part.data.trim() : undefined
+  // SDK URL sources: URLImageSource:3817 / URLPDFSource:3823 {type:"url", url}
+  if (rawString && isHttpUrl(rawString) && !rawString.startsWith("data:")) {
+    if (mime.startsWith("image/"))
+      return {
+        type: "image" as const,
+        source: { type: "url" as const, url: rawString },
+        ...(cacheControlValue === undefined ? {} : { cache_control: cacheControlValue }),
+        ...(transformationsFromMetadata(part.metadata) === undefined
+          ? {}
+          : { transformations: transformationsFromMetadata(part.metadata)! }),
+      } satisfies AnthropicImageBlock
+    if (mime === "application/pdf")
+      return {
+        type: "document" as const,
+        source: { type: "url" as const, url: rawString },
+        ...(cacheControlValue === undefined ? {} : { cache_control: cacheControlValue }),
+        ...(documentTitleFromPart(part) === undefined ? {} : { title: documentTitleFromPart(part)! }),
+        ...(documentContextFromMetadata(part.metadata) === undefined
+          ? {}
+          : { context: documentContextFromMetadata(part.metadata)! }),
+        ...(citationsFromMetadata(part.metadata) === undefined
+          ? {}
+          : { citations: citationsFromMetadata(part.metadata)! }),
+      } satisfies AnthropicDocumentBlock
+  }
+
+  // SDK PlainTextSource:2716 {type:"text", media_type:"text/plain", data}
+  if (mime === "text/plain") {
+    const textData =
+      typeof part.data !== "string"
+        ? Buffer.from(part.data).toString("utf8")
+        : part.data.startsWith("data:")
+          ? (() => {
+              const comma = part.data.indexOf(",")
+              const payload = comma >= 0 ? part.data.slice(comma + 1) : part.data
+              return part.data.includes(";base64")
+                ? Buffer.from(payload, "base64").toString("utf8")
+                : decodeURIComponent(payload)
+            })()
+          : part.data
+    return {
+      type: "document" as const,
+      source: { type: "text" as const, media_type: "text/plain" as const, data: textData },
+      ...(cacheControlValue === undefined ? {} : { cache_control: cacheControlValue }),
+      ...(documentTitleFromPart(part) === undefined ? {} : { title: documentTitleFromPart(part)! }),
+      ...(documentContextFromMetadata(part.metadata) === undefined
+        ? {}
+        : { context: documentContextFromMetadata(part.metadata)! }),
+      ...(citationsFromMetadata(part.metadata) === undefined
+        ? {}
+        : { citations: citationsFromMetadata(part.metadata)! }),
+    } satisfies AnthropicDocumentBlock
+  }
+
   const media = ProviderShared.normalizeMedia(part)
   if (media.mime === "application/pdf")
     return {
@@ -421,6 +664,14 @@ const lowerMedia = Effect.fn("AnthropicMessages.lowerMedia")(function* (part: Me
         media_type: "application/pdf" as const,
         data: media.base64,
       },
+      ...(cacheControlValue === undefined ? {} : { cache_control: cacheControlValue }),
+      ...(documentTitleFromPart(part) === undefined ? {} : { title: documentTitleFromPart(part)! }),
+      ...(documentContextFromMetadata(part.metadata) === undefined
+        ? {}
+        : { context: documentContextFromMetadata(part.metadata)! }),
+      ...(citationsFromMetadata(part.metadata) === undefined
+        ? {}
+        : { citations: citationsFromMetadata(part.metadata)! }),
     } satisfies AnthropicDocumentBlock
   if (!media.mime.startsWith("image/"))
     return yield* invalid(`Anthropic Messages does not support media type ${part.mediaType}`)
@@ -431,6 +682,10 @@ const lowerMedia = Effect.fn("AnthropicMessages.lowerMedia")(function* (part: Me
       media_type: media.mime,
       data: media.base64,
     },
+    ...(cacheControlValue === undefined ? {} : { cache_control: cacheControlValue }),
+    ...(transformationsFromMetadata(part.metadata) === undefined
+      ? {}
+      : { transformations: transformationsFromMetadata(part.metadata)! }),
   } satisfies AnthropicImageBlock
 })
 
@@ -450,6 +705,26 @@ const lowerToolResultContent = Effect.fnUntraced(function* (part: ToolResultPart
   return yield* Effect.forEach(content, lowerToolResultContentItem)
 })
 
+const requireThinkingSignature = (request: LLMRequest) => {
+  if (request.model.compatibility?.requireSignature !== undefined)
+    return request.model.compatibility.requireSignature
+  const provider = request.model.provider.toLowerCase()
+  const model = request.model.id.toLowerCase()
+  const baseURL = (request.model.route.endpoint.baseURL ?? "").toLowerCase()
+  if (
+    provider === "kimi-for-coding" ||
+    provider === "moonshotai" ||
+    provider === "moonshotai-cn" ||
+    model.startsWith("kimi-") ||
+    baseURL.includes("api.kimi.com/coding") ||
+    baseURL.includes("api.moonshot.ai/anthropic") ||
+    baseURL.includes("api.moonshot.cn/anthropic")
+  )
+    return false
+  if (provider.includes("xiaomi") || model.includes("mimo") || baseURL.includes("xiaomimimo.com")) return false
+  return true
+}
+
 // Mid-conversation system messages became available with Opus 4.8 and version
 // 5 of the other supported Claude families. Treat later family versions as
 // compatible without assuming that every Anthropic Messages model is Claude.
@@ -469,9 +744,12 @@ const endsInServerToolUse = (message: LLMRequest["messages"][number]) => {
   return message.role === "assistant" && last?.type === "tool-call" && last.providerExecuted === true
 }
 
-const canUseNativeSystemUpdate = (messages: LLMRequest["messages"], index: number) => {
-  const previous = messages[index - 1]
-  const next = messages[index + 1]
+const canUseNativeSystemUpdate = (request: LLMRequest, index: number) => {
+  const previous = request.messages[index - 1]
+  const next = request.messages[index + 1]
+  // Vertex currently rejects/404s for a system message after local tool results,
+  // so fold it into the user tool-result turn across continuations and history.
+  if (request.model.route.id === "google-vertex-messages" && previous?.role === "tool") return false
   return (
     previous !== undefined &&
     previous.role !== "system" &&
@@ -518,7 +796,7 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
     if (message.role === "system") {
       if (splitsLocalToolResults(request.messages, index))
         return yield* invalid("Anthropic Messages system updates cannot split a local tool call from its tool result")
-      if (supportsNativeSystemUpdates(request) && canUseNativeSystemUpdate(request.messages, index)) {
+      if (supportsNativeSystemUpdates(request) && canUseNativeSystemUpdate(request, index)) {
         messages.push(yield* lowerNativeSystemUpdate(message, breakpoints))
         continue
       }
@@ -539,7 +817,7 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
           continue
         }
         if (part.type === "media") {
-          content.push(yield* lowerMedia(part))
+          content.push(yield* lowerMedia(part, breakpoints))
           continue
         }
         return yield* ProviderShared.unsupportedContent("Anthropic Messages", "user", ["text", "media"])
@@ -556,15 +834,30 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
           continue
         }
         if (part.type === "reasoning") {
-          // Mirrors Vercel's @ai-sdk/anthropic: a signature marks visible
-          // thinking; only signature-less parts carrying redactedData
-          // round-trip as opaque redacted_thinking blocks.
+          // A signature marks visible thinking; only signature-less parts carrying
+          // redactedData round-trip as opaque redacted_thinking blocks.
           const signature = part.encrypted ?? signatureFromMetadata(part.providerMetadata)
           const redactedData = redactedDataFromMetadata(part.providerMetadata)
           if (signature === undefined && redactedData !== undefined) {
             content.push({ type: "redacted_thinking", data: redactedData })
             continue
           }
+          if (typeof signature !== "string" || signature.trim().length === 0) {
+            if (part.text.trim().length === 0) continue
+            if (!requireThinkingSignature(request)) {
+              content.push({ type: "thinking", thinking: part.text, signature: "" })
+              continue
+            }
+            // Without a signature this cannot be a valid thinking block per
+            // the SDK ThinkingBlockParam:3217 — demote to text so the
+            // conversation remains sendable.
+            content.push({
+              type: "text",
+              text: part.text,
+              cache_control: cacheControl(breakpoints, part.cache),
+            })
+            continue
+          }
           content.push({ type: "thinking", thinking: part.text, signature })
           continue
         }
@@ -606,24 +899,74 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
 })
 
 const resolveOptions = Effect.fn("AnthropicMessages.resolveOptions")(function* (request: LLMRequest) {
-  const input = request.providerOptions
+  const input = request.providerOptions as Record<string, unknown> | undefined
+  const rawServiceTier = (input as Record<string, unknown> | undefined)?.service_tier ?? (input as Record<string, unknown> | undefined)?.serviceTier
+  const service_tier =
+    rawServiceTier === "auto" || rawServiceTier === "standard_only"
+      ? (rawServiceTier as "auto" | "standard_only")
+      : undefined
+  const rawMetadata = (input as Record<string, unknown> | undefined)?.metadata
+  const metadata =
+    ProviderShared.isRecord(rawMetadata) &&
+    (typeof rawMetadata.user_id === "string" || rawMetadata.user_id === null)
+      ? { user_id: rawMetadata.user_id as string | null }
+      : undefined
+  const container =
+    typeof (input as Record<string, unknown> | undefined)?.container === "string" ||
+    ProviderShared.isRecord((input as Record<string, unknown> | undefined)?.container)
+      ? ((input as Record<string, unknown>).container as string | { id?: string | null; skills?: ReadonlyArray<Record<string, unknown>> | null })
+      : undefined
+  const rawInferenceGeo =
+    (input as Record<string, unknown> | undefined)?.inference_geo ??
+    (input as Record<string, unknown> | undefined)?.inferenceGeo
+  const inference_geo = typeof rawInferenceGeo === "string" ? rawInferenceGeo : undefined
+  const rawCacheControl =
+    (input as Record<string, unknown> | undefined)?.cache_control ??
+    (input as Record<string, unknown> | undefined)?.cacheControl
+  const cache_control =
+    ProviderShared.isRecord(rawCacheControl) && rawCacheControl.type === "ephemeral"
+      ? (rawCacheControl as { type: "ephemeral"; ttl?: "5m" | "1h" })
+      : undefined
+  const rawOutputConfig =
+    (input as Record<string, unknown> | undefined)?.output_config ??
+    (input as Record<string, unknown> | undefined)?.outputConfig
+  const outputConfigEffort =
+    typeof (input as Record<string, unknown> | undefined)?.effort === "string"
+      ? ((input as Record<string, unknown>).effort as string)
+      : ProviderShared.isRecord(rawOutputConfig) && typeof rawOutputConfig.effort === "string"
+        ? (rawOutputConfig.effort as string)
+        : undefined
+  const outputConfigFormat =
+    ProviderShared.isRecord(rawOutputConfig) && ProviderShared.isRecord(rawOutputConfig.format)
+      ? (rawOutputConfig.format as { type: "json_schema"; schema: Record<string, unknown> })
+      : undefined
+  const output_config =
+    outputConfigEffort === undefined && outputConfigFormat === undefined
+      ? undefined
+      : {
+          ...(outputConfigEffort === undefined ? {} : { effort: outputConfigEffort }),
+          ...(outputConfigFormat === undefined ? {} : { format: outputConfigFormat }),
+        }
   return {
     thinking: yield* resolveThinking(input?.thinking),
-    effort: typeof input?.effort === "string" ? input.effort : undefined,
+    effort: outputConfigEffort,
+    output_config,
+    service_tier,
+    metadata,
+    container,
+    inference_geo,
+    cache_control,
   }
 })
 
 const resolveThinking = Effect.fn("AnthropicMessages.resolveThinking")(function* (input: unknown) {
   if (!ProviderShared.isRecord(input)) return undefined
-  if (input.type === "adaptive") {
-    const display =
-      input.display === "summarized"
-        ? ("summarized" as const)
-        : input.display === "omitted"
-          ? ("omitted" as const)
-          : undefined
+  const display =
+    input.display === "summarized" || input.display === "omitted"
+      ? (input.display as "summarized" | "omitted")
+      : undefined
+  if (input.type === "adaptive")
     return { type: "adaptive" as const, ...(display === undefined ? {} : { display }) }
-  }
   if (input.type === "disabled") return { type: "disabled" as const }
   if (input.type !== "enabled") return undefined
   const budget =
@@ -634,7 +977,7 @@ const resolveThinking = Effect.fn("AnthropicMessages.resolveThinking")(function*
         : undefined
   if (budget === undefined)
     return yield* ProviderShared.invalidRequest("Anthropic thinking provider option requires budgetTokens")
-  return { type: "enabled" as const, budget_tokens: budget }
+  return { type: "enabled" as const, budget_tokens: budget, ...(display === undefined ? {} : { display }) }
 })
 
 const fromRequest = Effect.fn("AnthropicMessages.fromRequest")(function* (request: LLMRequest) {
@@ -684,7 +1027,13 @@ const fromRequest = Effect.fn("AnthropicMessages.fromRequest")(function* (reques
     top_k: generation?.topK,
     stop_sequences: generation?.stop,
     thinking: options.thinking,
-    output_config: options.effort === undefined ? undefined : { effort: options.effort },
+    output_config: options.output_config,
+    // top-level passthrough per SDK MessageCreateParamsBase:4638,4643,4649,4654,4670
+    cache_control: options.cache_control,
+    container: options.container,
+    inference_geo: options.inference_geo,
+    metadata: options.metadata,
+    service_tier: options.service_tier,
   }
 })
 
@@ -764,7 +1113,7 @@ const SERVER_TOOL_RESULT_NAMES: Record<AnthropicServerToolResultType, string> =
 
 const isServerToolResultType = (type: string): type is AnthropicServerToolResultType => type in SERVER_TOOL_RESULT_NAMES
 
-const serverToolResultEvent = (block: NonNullable<AnthropicEvent["content_block"]>): LLMEvent | undefined => {
+const serverToolResultEvent = (block: AnthropicStreamBlock): LLMEvent | undefined => {
   if (!block.type || !isServerToolResultType(block.type)) return undefined
   const errorPayload =
     typeof block.content === "object" && block.content !== null && "type" in block.content
@@ -791,11 +1140,15 @@ const onMessageStart = (state: ParserState, event: AnthropicEvent): StepResult =
   return [usage ? { ...state, usage: mergeUsage(state.usage, usage) } : state, NO_EVENTS]
 }
 
-const onContentBlockStart = (state: ParserState, event: AnthropicEvent): StepResult => {
+const onContentBlockStart = (
+  state: ParserState,
+  event: AnthropicEvent & { readonly content_block: AnthropicStreamBlock },
+): StepResult => {
   const block = event.content_block
   if (!block) return [state, NO_EVENTS]
 
-  if ((block.type === "tool_use" || block.type === "server_tool_use") && event.index !== undefined) {
+  if (block.type === "tool_use" || block.type === "server_tool_use") {
+    if (event.index === undefined || !block.id) return [state, NO_EVENTS]
     const events: LLMEvent[] = []
     const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
     return [
@@ -803,7 +1156,7 @@ const onContentBlockStart = (state: ParserState, event: AnthropicEvent): StepRes
         ...state,
         lifecycle,
         tools: ToolStream.start(state.tools, event.index, {
-          id: block.id ?? String(event.index),
+          id: block.id,
           name: block.name ?? "",
           input:
             block.input !== undefined && (!ProviderShared.isRecord(block.input) || Object.keys(block.input).length > 0)
@@ -815,7 +1168,7 @@ const onContentBlockStart = (state: ParserState, event: AnthropicEvent): StepRes
       [
         ...events,
         LLMEvent.toolInputStart({
-          id: block.id ?? String(event.index),
+          id: block.id,
           name: block.name ?? "",
           providerExecuted: block.type === "server_tool_use" ? true : undefined,
         }),
@@ -881,11 +1234,12 @@ const onContentBlockStart = (state: ParserState, event: AnthropicEvent): StepRes
 
 const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(function* (
   state: ParserState,
-  event: AnthropicEvent,
+  event: AnthropicEvent & { readonly delta: AnthropicStreamDelta },
 ) {
   const delta = event.delta
 
   if (delta?.type === "text_delta" && delta.text) {
+    if (!state.lifecycle.text.has(`text-${event.index ?? 0}`)) return [state, NO_EVENTS] satisfies StepResult
     const events: LLMEvent[] = []
     return [
       { ...state, lifecycle: Lifecycle.textDelta(state.lifecycle, events, `text-${event.index ?? 0}`, delta.text) },
@@ -894,6 +1248,7 @@ const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(f
   }
 
   if (delta?.type === "thinking_delta" && delta.thinking) {
+    if (!state.lifecycle.reasoning.has(`reasoning-${event.index ?? 0}`)) return [state, NO_EVENTS] satisfies StepResult
     const events: LLMEvent[] = []
     return [
       {
@@ -906,6 +1261,7 @@ const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(f
 
   if (delta?.type === "signature_delta" && delta.signature) {
     const index = event.index ?? 0
+    if (!state.lifecycle.reasoning.has(`reasoning-${index}`)) return [state, NO_EVENTS] satisfies StepResult
     return [
       {
         ...state,
@@ -958,7 +1314,10 @@ const onContentBlockStop = Effect.fn("AnthropicMessages.onContentBlockStop")(fun
   return [{ ...state, lifecycle, tools: result.tools, reasoningSignatures }, events] satisfies StepResult
 })
 
-const onMessageDelta = (state: ParserState, event: AnthropicEvent): StepResult => {
+const onMessageDelta = (
+  state: ParserState,
+  event: AnthropicEvent & { readonly delta?: AnthropicStreamDelta },
+): StepResult => {
   const usage = mergeUsage(state.usage, mapUsage(event.usage))
   return [
     {
@@ -1013,12 +1372,72 @@ const onError = (event: AnthropicEvent) =>
     }),
   )
 
+const isKnownStreamBlockType = (type: string) =>
+  type === "text" ||
+  type === "thinking" ||
+  type === "redacted_thinking" ||
+  type === "tool_use" ||
+  type === "server_tool_use" ||
+  isServerToolResultType(type)
+
+const isKnownStreamDeltaType = (type: string) =>
+  type === "text_delta" || type === "thinking_delta" || type === "signature_delta" || type === "input_json_delta"
+
+const invalidStreamEvent = (event: AnthropicEvent) =>
+  Effect.fail(
+    ProviderShared.eventError(
+      ADAPTER,
+      "Invalid anthropic/anthropic-messages stream event",
+      ProviderShared.encodeJson(event),
+    ),
+  )
+
 const step = (state: ParserState, event: AnthropicEvent) => {
+  if (!SSE_EVENTS.has(event.type)) return Effect.succeed<StepResult>([state, NO_EVENTS])
+  if (
+    event.type !== "content_block_start" &&
+    event.content_block !== undefined &&
+    Option.isNone(decodeAnthropicStreamBlock(event.content_block))
+  )
+    return invalidStreamEvent(event)
+  if (
+    event.type !== "content_block_delta" &&
+    event.delta !== undefined &&
+    Option.isNone(decodeAnthropicStreamDelta(event.delta))
+  )
+    return invalidStreamEvent(event)
   if (event.type === "message_start") return Effect.succeed(onMessageStart(state, event))
-  if (event.type === "content_block_start") return Effect.succeed(onContentBlockStart(state, event))
-  if (event.type === "content_block_delta") return onContentBlockDelta(state, event)
+  if (event.type === "content_block_start") {
+    if (!ProviderShared.isRecord(event.content_block) || typeof event.content_block.type !== "string")
+      return invalidStreamEvent(event)
+    if (!isKnownStreamBlockType(event.content_block.type)) return Effect.succeed<StepResult>([state, NO_EVENTS])
+    const decoded = decodeAnthropicStreamBlock(event.content_block)
+    if (Option.isNone(decoded)) return invalidStreamEvent(event)
+    const block = decoded.value
+    if (block.type === "tool_use" || block.type === "server_tool_use") {
+      if (event.index === undefined)
+        return Effect.fail(ProviderShared.eventError(ADAPTER, `Anthropic ${block.type} missing index`))
+      if (!block.id)
+        return Effect.fail(
+          ProviderShared.eventError(ADAPTER, `Anthropic tool_use missing id at index ${event.index}`),
+        )
+    }
+    return Effect.succeed(onContentBlockStart(state, { ...event, content_block: block }))
+  }
+  if (event.type === "content_block_delta") {
+    if (!ProviderShared.isRecord(event.delta)) return invalidStreamEvent(event)
+    if (typeof event.delta.type === "string" && !isKnownStreamDeltaType(event.delta.type))
+      return Effect.succeed<StepResult>([state, NO_EVENTS])
+    const decoded = decodeAnthropicStreamDelta(event.delta)
+    if (Option.isNone(decoded)) return invalidStreamEvent(event)
+    return onContentBlockDelta(state, { ...event, delta: decoded.value })
+  }
   if (event.type === "content_block_stop") return onContentBlockStop(state, event)
-  if (event.type === "message_delta") return Effect.succeed(onMessageDelta(state, event))
+  if (event.type === "message_delta") {
+    const decoded = decodeAnthropicStreamDelta(event.delta)
+    if (Option.isNone(decoded)) return invalidStreamEvent(event)
+    return Effect.succeed(onMessageDelta(state, { ...event, delta: decoded.value }))
+  }
   if (event.type === "message_stop") return onMessageStop(state)
   if (event.type === "error") return onError(event)
   return Effect.succeed<StepResult>([state, NO_EVENTS])
@@ -1054,7 +1473,10 @@ export const route = Route.make({
   provider: "anthropic",
   providerMetadataKey: "anthropic",
   protocol,
-  endpoint: Endpoint.path(PATH, { baseURL: DEFAULT_BASE_URL }),
+  endpoint: Endpoint.path(
+    (input) => (input.request.model.provider === "anthropic" ? `${PATH}?beta=true` : PATH),
+    { baseURL: DEFAULT_BASE_URL },
+  ),
   auth: Auth.none,
   framing,
   headers: () => ({ "anthropic-version": "2023-06-01" }),

packages/ai/src/protocols/bedrock-converse.ts

@@ -212,11 +212,7 @@ const BedrockEvent = Schema.Struct({
       metrics: Schema.optional(Schema.Unknown),
     }),
   ),
-  internalServerException: Schema.optional(BedrockStreamException),
-  modelStreamErrorException: Schema.optional(BedrockStreamException),
-  validationException: Schema.optional(BedrockStreamException),
-  throttlingException: Schema.optional(BedrockStreamException),
-  serviceUnavailableException: Schema.optional(BedrockStreamException),
+  exception: Schema.optional(Schema.Struct({ type: Schema.String, details: BedrockStreamException })),
 })
 type BedrockEvent = Schema.Schema.Type<typeof BedrockEvent>
 
@@ -650,22 +646,14 @@ const step = (state: ParserState, event: BedrockEvent) =>
       ] as const
     }
 
-    const exception = (
-      [
-        ["internalServerException", event.internalServerException],
-        ["modelStreamErrorException", event.modelStreamErrorException],
-        ["serviceUnavailableException", event.serviceUnavailableException],
-        ["throttlingException", event.throttlingException],
-        ["validationException", event.validationException],
-      ] as const
-    ).find((entry) => entry[1] !== undefined)
-    if (exception) {
+    if (event.exception) {
       return yield* new AIError({
         module: ADAPTER,
         method: "stream",
         reason: classifyProviderFailure({
-          message: exception[1]?.message ?? exception[1]?.originalMessage ?? "Bedrock Converse stream error",
-          code: exception[0],
+          message:
+            event.exception.details.message ?? event.exception.details.originalMessage ?? "Bedrock Converse stream error",
+          code: event.exception.type,
         }),
       })
     }
@@ -716,7 +704,7 @@ export const protocol = Protocol.make({
       reasoningSignatures: {},
     }),
     step,
-    onHalt,
+    onHalt: (state) => Effect.succeed(onHalt(state)),
   },
 })
 

packages/ai/src/protocols/bedrock-event-stream.ts

@@ -82,7 +82,7 @@ const consumeFrames = (route: string) => (state: FrameBufferState, chunk: Uint8A
         "Failed to parse Bedrock Converse event-stream payload",
       )) as Record<string, unknown>
       delete parsed.p
-      out.push({ [eventType]: parsed })
+      out.push(messageType === "exception" ? { exception: { type: eventType, details: parsed } } : { [eventType]: parsed })
     }
     return [cursor, out] as const
   })

packages/ai/src/protocols/gemini.ts

@@ -1,4 +1,4 @@
-import { Effect, Schema } from "effect"
+import { Effect, Option, Schema } from "effect"
 import { Tool } from "@opencode-ai/schema/tool"
 import { Route } from "../route/client.js"
 import { Auth } from "../route/auth.js"
@@ -125,6 +125,7 @@ const GeminiContentPart = Schema.Union([
   GeminiFunctionCallPart,
   GeminiFunctionResponsePart,
 ])
+const decodeGeminiContentPart = Schema.decodeUnknownOption(GeminiContentPart)
 
 const GeminiContent = Schema.Struct({
   role: optionalNull(Schema.Literals(["user", "model"])),
@@ -132,6 +133,11 @@ const GeminiContent = Schema.Struct({
 })
 type GeminiContent = Schema.Schema.Type<typeof GeminiContent>
 
+const GeminiResponseContent = Schema.Struct({
+  role: optionalNull(Schema.Literals(["user", "model"])),
+  parts: optionalNull(Schema.Array(Schema.Unknown)),
+})
+
 const GeminiSystemInstruction = Schema.Struct({
   parts: Schema.Array(Schema.Struct({ text: Schema.String })),
 })
@@ -200,7 +206,7 @@ const GeminiUsage = Schema.Struct({
 type GeminiUsage = Schema.Schema.Type<typeof GeminiUsage>
 
 const GeminiCandidate = Schema.Struct({
-  content: optionalNull(GeminiContent),
+  content: optionalNull(GeminiResponseContent),
   finishReason: optionalNull(Schema.String),
 })
 
@@ -222,6 +228,7 @@ const GeminiEvent = Schema.Struct({
 type GeminiEvent = Schema.Schema.Type<typeof GeminiEvent>
 
 interface ParserState {
+  readonly route: string
   readonly finishReason?: string
   readonly hasToolCalls: boolean
   readonly promptFeedback?: GeminiPromptFeedback
@@ -598,7 +605,21 @@ const step = (state: ParserState, event: GeminiEvent) => {
   // Supplier ids must be tracked across chunks of the same response, not just within one event's parts.
   const seenCallIds = new Set(nextState.seenCallIds)
 
-  for (const part of candidate.content.parts ?? []) {
+  for (const input of candidate.content.parts ?? []) {
+    if (
+      ProviderShared.isRecord(input) &&
+      !("text" in input) &&
+      !("inlineData" in input) &&
+      !("functionCall" in input) &&
+      !("functionResponse" in input)
+    )
+      continue
+    const decoded = decodeGeminiContentPart(input)
+    if (Option.isNone(decoded))
+      return Effect.fail(
+        ProviderShared.eventError(ADAPTER, `Invalid ${state.route} stream event`, ProviderShared.encodeJson(event)),
+      )
+    const part = decoded.value
     const signature = "thoughtSignature" in part && part.thoughtSignature ? part.thoughtSignature : undefined
     // Gemini attaches replay signatures to thought parts, visible text, or function calls;
     // each block kind must retain the signature attached to its own parts.
@@ -691,9 +712,13 @@ export const protocol = Protocol.make({
   },
   stream: {
     event: Protocol.jsonEvent(GeminiEvent),
-    initial: () => ({ hasToolCalls: false, lifecycle: Lifecycle.initial() }),
+    initial: (request) => ({
+      route: `${request.model.provider}/${request.model.route.id}`,
+      hasToolCalls: false,
+      lifecycle: Lifecycle.initial(),
+    }),
     step,
-    onHalt: finish,
+    onHalt: (state) => Effect.succeed(finish(state)),
   },
 })
 

packages/ai/src/protocols/open-responses-channel.ts

@@ -112,6 +112,8 @@ const driver = (options: Options, body: string): WebSocketChannelDriver => {
           responseID = created
           return { type: "frame", frame }
         }
+        // Keepalives carry no response state and may arrive before response.created.
+        if (event.type === "keepalive") return { type: "frame", frame }
         if (!responseID)
           return yield* ProviderShared.eventError(
             options.id,

packages/ai/src/protocols/open-responses-continuation.ts

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

packages/ai/src/protocols/open-responses.ts

@@ -79,10 +79,60 @@ const OpenResponsesReasoningItem = Schema.Struct({
   encrypted_content: optionalNull(Schema.String),
 })
 
-const OpenResponsesItemReference = Schema.Struct({
-  type: Schema.tag("item_reference"),
-  id: Schema.String,
-})
+const OpenResponsesWebSearchCall = Schema.StructWithRest(
+  Schema.Struct({
+    type: Schema.tag("web_search_call"),
+    id: Schema.String,
+    status: Schema.optional(Schema.String),
+    action: optionalNull(JsonObject),
+  }),
+  [JsonObject],
+)
+
+const OpenResponsesFileSearchCall = Schema.StructWithRest(
+  Schema.Struct({
+    type: Schema.tag("file_search_call"),
+    id: Schema.String,
+    status: Schema.optional(Schema.String),
+    queries: Schema.optional(Schema.Array(Schema.String)),
+    results: optionalNull(Schema.Array(JsonObject)),
+  }),
+  [JsonObject],
+)
+
+const OpenResponsesCodeInterpreterCall = Schema.StructWithRest(
+  Schema.Struct({
+    type: Schema.tag("code_interpreter_call"),
+    id: Schema.String,
+    status: Schema.optional(Schema.String),
+    code: optionalNull(Schema.String),
+    container_id: optionalNull(Schema.String),
+    outputs: optionalNull(Schema.Array(JsonObject)),
+  }),
+  [JsonObject],
+)
+
+const OpenResponsesMCPCall = Schema.StructWithRest(
+  Schema.Struct({
+    type: Schema.tag("mcp_call"),
+    id: Schema.String,
+    status: Schema.optional(Schema.String),
+    server_label: Schema.optional(Schema.String),
+    name: Schema.optional(Schema.String),
+    arguments: Schema.optional(Schema.String),
+    output: optionalNull(Schema.String),
+    error: Schema.optional(Schema.Unknown),
+  }),
+  [JsonObject],
+)
+
+export const HostedToolItem = Schema.Union([
+  OpenResponsesWebSearchCall,
+  OpenResponsesFileSearchCall,
+  OpenResponsesCodeInterpreterCall,
+  OpenResponsesMCPCall,
+])
+export type HostedToolItem = Schema.Schema.Type<typeof HostedToolItem>
 
 // `function_call_output.output` accepts either a plain string or an ordered
 // array of content items so tools can return images and files in addition to text.
@@ -111,7 +161,6 @@ export const InputItem = Schema.Union([
     phase: Schema.optionalKey(MessagePhase),
   }),
   OpenResponsesReasoningItem,
-  OpenResponsesItemReference,
   Schema.Struct({
     type: Schema.tag("function_call"),
     id: Schema.optionalKey(Schema.String),
@@ -124,10 +173,17 @@ export const InputItem = Schema.Union([
     call_id: Schema.String,
     output: OpenResponsesFunctionCallOutput,
   }),
+  HostedToolItem,
 ])
 type OpenResponsesInputItem = Schema.Schema.Type<typeof InputItem>
+export type ExtendedHostedToolItem = {
+  readonly type: string
+  readonly id: string
+  readonly [key: string]: unknown
+}
 type LoweredInputItem =
   | OpenResponsesInputItem
+  | ExtendedHostedToolItem
   | {
       readonly type: "message"
       readonly id?: string
@@ -140,7 +196,7 @@ type LoweredInputItem =
 // multiple streamed summary parts into the same item before flushing.
 type OpenResponsesReasoningInput = {
   type: "reasoning"
-  id: string
+  id?: string
   summary: Array<{ type: "summary_text"; text: string }>
   encrypted_content?: string | null
 }
@@ -285,8 +341,10 @@ export const Event = Schema.StructWithRest(
   Schema.Struct({
     type: Schema.String,
     delta: Schema.optional(Schema.String),
+    arguments: Schema.optional(Schema.String),
     text: Schema.optional(Schema.String),
     item_id: Schema.optional(Schema.String),
+    output_index: Schema.optional(Schema.Number),
     summary_index: Schema.optional(Schema.Number),
     item: Schema.optional(StreamItem),
     response: Schema.optional(
@@ -295,6 +353,7 @@ export const Event = Schema.StructWithRest(
           id: Schema.optional(Schema.String),
           service_tier: optionalNull(Schema.String),
           incomplete_details: optionalNull(Schema.Struct({ reason: Schema.optional(Schema.String) })),
+          output: Schema.optional(Schema.Array(StreamItem)),
           usage: optionalNull(OpenResponsesUsage),
           error: optionalNull(OpenResponsesErrorPayload),
         }),
@@ -321,6 +380,7 @@ export interface Extension {
     readonly media: ProviderShared.NormalizedMedia
     readonly request: LLMRequest
   }) => MediaInput | undefined
+  readonly lowerHostedToolItem?: (item: unknown) => ExtendedHostedToolItem | undefined
 }
 
 const BASE: Extension = { id: ADAPTER, name: NAME }
@@ -332,10 +392,10 @@ export interface ParserState {
   readonly tools: ToolStream.State<string>
   readonly hasFunctionCall: boolean
   readonly lifecycle: Lifecycle.State
+  readonly outputItems: Readonly<Record<number, string>>
   readonly messageItems: ReadonlySet<string>
   readonly messagePhases: Readonly<Record<string, MessagePhase | null>>
   readonly reasoningItems: Readonly<Record<string, ReasoningStreamItem>>
-  readonly store: boolean | undefined
 }
 
 type ReasoningSummaryStatus = "active" | "can-conclude" | "concluded"
@@ -346,6 +406,10 @@ interface ReasoningStreamItem {
   // strings, but typing the map as `Record<number, ...>` documents intent
   // and matches the wire field.
   readonly summaryParts: Readonly<Record<number, ReasoningSummaryStatus>>
+  // Summary indexes that received at least one streamed delta. The `:0` block
+  // is started eagerly when the item opens, so block existence cannot tell
+  // whether a `.done` final would duplicate streamed text.
+  readonly deltaIndexes: ReadonlySet<number>
 }
 
 // =============================================================================
@@ -362,7 +426,7 @@ export const lowerTool = Effect.fn("OpenResponses.lowerTool")(function* (
     type: "function" as const,
     name: tool.name,
     description: tool.description,
-    parameters: ToolSchemaProjection.responses(inputSchema),
+    parameters: inputSchema,
     // The common tool definition does not currently express Responses strict-schema policy.
     strict: false,
   }
@@ -376,18 +440,20 @@ export const lowerToolChoice = (protocolName: string, toolChoice: NonNullable<LL
     tool: (toolName) => ({ type: "function" as const, name: toolName }),
   })
 
+// Server-issued item ids need a nonempty prefix and suffix, but the prefix is
+// provider-defined and does not necessarily identify the item's semantic type.
 const itemID = (providerMetadata: ProviderMetadata | undefined, providerMetadataKey: string) => {
   const metadata = providerMetadata?.[providerMetadataKey]
-  return ProviderShared.isRecord(metadata) && typeof metadata.itemId === "string" && metadata.itemId.length > 0
-    ? 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 lowerToolCall = (part: ToolCallPart, providerMetadataKey: string): OpenResponsesInputItem => {
   const id = itemID(part.providerMetadata, providerMetadataKey)
   return {
     type: "function_call",
-    ...(id ? { id } : {}),
+    ...(id === undefined ? {} : { id }),
     call_id: part.id,
     name: part.name,
     arguments: ProviderShared.encodeJson(part.input),
@@ -396,15 +462,15 @@ const lowerToolCall = (part: ToolCallPart, providerMetadataKey: string): OpenRes
 
 const lowerReasoning = (part: ReasoningPart, providerMetadataKey: string): OpenResponsesReasoningInput | undefined => {
   const metadata = part.providerMetadata?.[providerMetadataKey]
+  if (!ProviderShared.isRecord(metadata)) return undefined
   const id = itemID(part.providerMetadata, providerMetadataKey)
-  if (!ProviderShared.isRecord(metadata) || !id) return undefined
   const encryptedContent =
     typeof metadata.reasoningEncryptedContent === "string" || metadata.reasoningEncryptedContent === null
       ? metadata.reasoningEncryptedContent
       : undefined
   return {
     type: "reasoning",
-    id,
+    ...(id === undefined ? {} : { id }),
     summary: part.text.length > 0 ? [{ type: "summary_text", text: part.text }] : [],
     encrypted_content: encryptedContent,
   }
@@ -495,10 +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) {
@@ -521,8 +584,7 @@ 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<
@@ -556,19 +618,14 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
           flushText()
           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
         }
@@ -581,17 +638,29 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
         if (part.type === "tool-result" && part.providerExecuted === true) {
           flushText()
           const id = itemID(part.providerMetadata, providerMetadataKey)
-          if (store !== false && id && !hostedToolReferences.has(id)) input.push({ type: "item_reference", id })
-          if (store === false && part.result.type === "content") {
-            const content: ReadonlyArray<Content> = part.result.value
-            input.push({
-              role: "user",
-              content: yield* Effect.forEach(content, (item) =>
-                lowerHostedToolResultContentItem(item, request, extension),
-              ),
-            })
+          const hosted =
+            part.result.type !== "json"
+              ? undefined
+              : Schema.is(HostedToolItem)(part.result.value)
+                ? part.result.value
+                : extension.lowerHostedToolItem?.(part.result.value)
+          if (id !== undefined && hosted?.id === id) {
+            if (!hostedToolItems.has(id)) {
+              input.push(hosted)
+              hostedToolItems.add(id)
+            }
+            continue
           }
-          if (id) hostedToolReferences.add(id)
+          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", [
@@ -616,20 +685,16 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
     }
   }
 
-  // With store:false, Responses APIs only accept previous reasoning items when the
-  // complete item has encrypted state. Summary blocks for one item may carry
-  // that state only on the last block, so filter after they have been joined.
-  return store === false
-    ? input.filter(
-        (item) => !("type" in item) || item.type !== "reasoning" || typeof item.encrypted_content === "string",
-      )
-    : input
+  return input
 })
 
 const lowerOptions = (request: LLMRequest) => {
   const options = OpenResponsesOptions.resolve(request)
+  const 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 } : {}),
@@ -637,7 +702,7 @@ const lowerOptions = (request: LLMRequest) => {
       ? { stream_options: { include_obfuscation: options.streamOptions.includeObfuscation } }
       : {}),
     ...(options.topLogprobs !== undefined ? { top_logprobs: options.topLogprobs } : {}),
-    ...(request.promptCacheKey ? { prompt_cache_key: request.promptCacheKey } : {}),
+    ...(cacheKey ? { prompt_cache_key: cacheKey } : {}),
     ...(options.include ? { include: options.include } : {}),
     ...(options.reasoningEffort || options.reasoningSummary
       ? { reasoning: { effort: options.reasoningEffort, summary: options.reasoningSummary } }
@@ -645,11 +710,18 @@ const lowerOptions = (request: LLMRequest) => {
     ...(options.textVerbosity ? { text: { verbosity: options.textVerbosity } } : {}),
     ...(options.serviceTier ? { service_tier: options.serviceTier } : {}),
     ...(options.maxToolCalls !== undefined ? { max_tool_calls: options.maxToolCalls } : {}),
-    ...(options.parallelToolCalls !== undefined ? { parallel_tool_calls: options.parallelToolCalls } : {}),
+    ...(parallelToolCalls !== undefined ? { parallel_tool_calls: parallelToolCalls } : {}),
     ...(options.truncation ? { truncation: options.truncation } : {}),
   }
 }
 
+export const resolveParallelToolCalls = (request: LLMRequest) => {
+  const configured = OpenResponsesOptions.resolve(request).parallelToolCalls
+  if (configured !== undefined) return configured
+  const disabled = request.toolChoice?.disableParallelToolUse
+  return disabled === undefined ? undefined : !disabled
+}
+
 const allowedToolChoice = (request: LLMRequest) => {
   const allowed = OpenResponsesOptions.resolve(request).allowedTools
   if (!allowed) return undefined
@@ -753,7 +825,7 @@ const TERMINAL_TYPES = new Set(["error", "response.completed", "response.incompl
 export const terminal = (event: Event) => TERMINAL_TYPES.has(event.type)
 
 const onOutputTextDelta = (state: ParserState, event: Event, id: string): StepResult => {
-  if (!event.delta) return [state, NO_EVENTS]
+  if (!event.delta || !state.messageItems.has(id)) return [state, NO_EVENTS]
   const events: LLMEvent[] = []
   const phase = state.messagePhases[id]
   const metadata = providerMetadata(state, { itemId: id, ...(phase === undefined ? {} : { phase }) })
@@ -770,21 +842,37 @@ const onOutputTextDone = (state: ParserState, event: Event, id: string): StepRes
   return [{ ...state, lifecycle: Lifecycle.textEnd(state.lifecycle, events, id) }, events]
 }
 
+export const outputItemID = (state: ParserState, event: Event) =>
+  event.output_index === undefined ? event.item_id : (state.outputItems[event.output_index] ?? event.item_id)
+
 export const onReasoningDelta = (state: ParserState, event: Event, itemID: string): StepResult => {
-  if (!event.delta) return [state, NO_EVENTS]
+  const item = state.reasoningItems[itemID]
+  if (!event.delta || !item) return [state, NO_EVENTS]
+  const index = event.summary_index ?? 0
   const events: LLMEvent[] = []
-  const id =
-    event.summary_index !== undefined || state.reasoningItems[itemID] ? `${itemID}:${event.summary_index ?? 0}` : itemID
   return [
     {
       ...state,
-      lifecycle: Lifecycle.reasoningDelta(state.lifecycle, events, id, event.delta),
+      lifecycle: Lifecycle.reasoningDelta(state.lifecycle, events, `${itemID}:${index}`, event.delta),
+      reasoningItems: {
+        ...state.reasoningItems,
+        [itemID]: { ...item, deltaIndexes: new Set([...item.deltaIndexes, index]) },
+      },
     },
     events,
   ]
 }
 
-export const onReasoningDone = (state: ParserState, _event: Event): StepResult => [state, NO_EVENTS]
+// Some compatible gateways emit a reasoning final without streaming any
+// deltas, mirroring `response.output_text.done`. Reconcile the complete text
+// as a single delta unless that summary index already streamed one.
+export const onReasoningDone = (state: ParserState, event: Event, itemID: string): StepResult => {
+  const item = state.reasoningItems[itemID]
+  if (!item || typeof event.text !== "string") return [state, NO_EVENTS]
+  const index = event.summary_index ?? 0
+  if (item.deltaIndexes.has(index)) return [state, NO_EVENTS]
+  return onReasoningDelta(state, { ...event, delta: event.text }, itemID)
+}
 
 const reasoningMetadata = (state: ParserState, item: StreamItem & { id: string }) =>
   providerMetadata(state, { itemId: item.id, reasoningEncryptedContent: item.encrypted_content ?? null })
@@ -822,57 +910,41 @@ const onOutputItemAdded = (state: ParserState, event: Event): StepResult => {
         lifecycle: Lifecycle.reasoningStart(state.lifecycle, events, `${item.id}:0`, reasoningMetadata(state, item)),
         reasoningItems: {
           ...state.reasoningItems,
-          [item.id]: { encryptedContent: item.encrypted_content, summaryParts: { 0: "active" } },
+          [item.id]: {
+            encryptedContent: item.encrypted_content,
+            summaryParts: { 0: "active" },
+            deltaIndexes: new Set(),
+          },
         },
       },
       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 })],
   ]
 }
 
 const onReasoningSummaryPartAdded = (state: ParserState, event: Event): StepResult => {
   if (!event.item_id || event.summary_index === undefined) return [state, NO_EVENTS]
-  const item = state.reasoningItems[event.item_id] ?? { encryptedContent: undefined, summaryParts: {} }
-  if (event.summary_index === 0) {
-    if (state.reasoningItems[event.item_id]) return [state, NO_EVENTS]
-    const events: LLMEvent[] = []
-    return [
-      {
-        ...state,
-        lifecycle: Lifecycle.reasoningStart(
-          state.lifecycle,
-          events,
-          `${event.item_id}:0`,
-          providerMetadata(state, { itemId: event.item_id, reasoningEncryptedContent: null }),
-        ),
-        reasoningItems: {
-          ...state.reasoningItems,
-          [event.item_id]: { ...item, summaryParts: { 0: "active" } },
-        },
-      },
-      events,
-    ]
-  }
+  const item = state.reasoningItems[event.item_id]
+  if (!item) return [state, NO_EVENTS]
+  if (event.summary_index === 0) return [state, NO_EVENTS]
 
   const events: LLMEvent[] = []
   const closed = Object.entries(item.summaryParts)
@@ -919,31 +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,
   ]
 }
 
@@ -951,12 +1013,24 @@ const onFunctionCallArgumentsDelta = Effect.fn("OpenResponses.onFunctionCallArgu
   state: ParserState,
   event: Event,
 ) {
-  if (!event.item_id || !event.delta) return [state, NO_EVENTS] satisfies StepResult
+  if (!event.item_id) return [state, NO_EVENTS] satisfies StepResult
+  const tool = state.tools[event.item_id]
+  if (!tool) return [state, NO_EVENTS] satisfies StepResult
+  const final = event.type === "response.function_call_arguments.done" ? event.arguments : undefined
+  if (event.type === "response.function_call_arguments.done" && final === undefined)
+    return [state, NO_EVENTS] satisfies StepResult
+  if (final !== undefined && !final.startsWith(tool.input))
+    return [
+      { ...state, tools: ToolStream.start(state.tools, event.item_id, { ...tool, input: final }) },
+      NO_EVENTS,
+    ] satisfies StepResult
+  const delta = final === undefined ? event.delta : final.slice(tool.input.length)
+  if (!delta) return [state, NO_EVENTS] satisfies StepResult
   const result = ToolStream.appendExisting(
     state.id,
     state.tools,
     event.item_id,
-    event.delta,
+    delta,
     `${state.name} tool argument delta is missing its tool call`,
   )
   if (ToolStream.isError(result)) return yield* result
@@ -994,18 +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
@@ -1053,30 +1128,50 @@ const onOutputItemDone = Effect.fn("OpenResponses.onOutputItemDone")(function* (
 })
 
 const onResponseFinish = Effect.fn("OpenResponses.onResponseFinish")(function* (state: ParserState, event: Event) {
+  const reconciled =
+    event.type === "response.completed"
+      ? yield* Effect.reduce(
+          event.response?.output ?? [],
+          () => [state, NO_EVENTS] satisfies StepResult,
+          ([current, events], item) => {
+            const id = item.id ?? (item.type === "function_call" ? item.call_id : undefined)
+            if (
+              !id ||
+              ((item.type !== "function_call" || !current.tools[id]) &&
+                (item.type !== "reasoning" || !current.reasoningItems[id]))
+            )
+              return Effect.succeed([current, events] satisfies StepResult)
+            return onOutputItemDone(current, { type: "response.output_item.done", item }).pipe(
+              Effect.map(([next, emitted]) => [next, [...events, ...emitted]] satisfies StepResult),
+            )
+          },
+        )
+      : ([state, NO_EVENTS] satisfies StepResult)
+  const current = reconciled[0]
   // Some compatible providers omit output_item.done even after completing the response.
   const pending =
     event.type === "response.completed"
-      ? yield* ToolStream.finishAll(state.id, state.tools)
-      : { tools: state.tools, events: NO_EVENTS }
-  const events: LLMEvent[] = [...pending.events]
+      ? yield* ToolStream.finishAll(current.id, current.tools)
+      : { tools: current.tools, events: NO_EVENTS }
+  const events: LLMEvent[] = [...reconciled[1], ...pending.events]
   const hasFunctionCall =
     pending.events.some((event) => LLMEvent.is.toolCall(event) || LLMEvent.is.toolInputError(event)) ||
-    state.hasFunctionCall
-  const lifecycle = Lifecycle.finish(state.lifecycle, events, {
+    current.hasFunctionCall
+  const lifecycle = Lifecycle.finish(current.lifecycle, events, {
     reason: {
       normalized: mapFinishReason(event, hasFunctionCall),
       raw: event.response?.incomplete_details?.reason,
     },
-    usage: mapUsage(event.response?.usage, state.providerMetadataKey),
+    usage: mapUsage(event.response?.usage, current.providerMetadataKey),
     providerMetadata:
       event.response?.id || event.response?.service_tier
-        ? providerMetadata(state, {
+        ? providerMetadata(current, {
             responseId: event.response.id,
             serviceTier: event.response.service_tier,
           })
         : undefined,
   })
-  return [{ ...state, lifecycle, hasFunctionCall, tools: pending.tools }, events] satisfies StepResult
+  return [{ ...current, lifecycle, hasFunctionCall, tools: pending.tools }, events] satisfies StepResult
 })
 
 // Build the prettiest summary available from whatever the provider supplied.
@@ -1123,7 +1218,11 @@ export const providerFailure = (id: string, event: Event, fallback: string) => {
 
 const providerError = (state: ParserState, event: Event, fallback: string) => providerFailure(state.id, event, fallback)
 
-export const step = (state: ParserState, event: Event) => {
+export const step = (state: ParserState, input: Event) => {
+  const event =
+    input.item_id && outputItemID(state, input) !== input.item_id
+      ? { ...input, item_id: outputItemID(state, input) }
+      : input
   if (event.type === "response.output_text.delta" || event.type === "response.output_text.done") {
     if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
     return Effect.succeed(
@@ -1146,6 +1245,14 @@ export const step = (state: ParserState, event: Event) => {
     if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
     return Effect.succeed(onReasoningDelta(state, event, event.item_id))
   }
+  if (
+    event.type === "response.reasoning.done" ||
+    event.type === "response.reasoning_summary_text.done" ||
+    event.type === "response.reasoning_text.done"
+  ) {
+    if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
+    return Effect.succeed(onReasoningDone(state, event, event.item_id))
+  }
   if (event.type === "response.reasoning_summary_part.added")
     return event.item_id
       ? Effect.succeed(onReasoningSummaryPartAdded(state, event))
@@ -1157,9 +1264,20 @@ export const step = (state: ParserState, event: Event) => {
   if (event.type === "response.output_item.added") {
     if (event.item?.type === "message" && !event.item.id)
       return ProviderShared.eventError(state.id, `${event.type} message is missing id`)
-    return Effect.succeed(onOutputItemAdded(state, event))
+    const id = event.item?.id ?? (event.item?.type === "function_call" ? event.item.call_id : undefined)
+    return Effect.succeed(
+      onOutputItemAdded(
+        event.output_index !== undefined && id
+          ? { ...state, outputItems: { ...state.outputItems, [event.output_index]: id } }
+          : state,
+        event,
+      ),
+    )
   }
-  if (event.type === "response.function_call_arguments.delta") return onFunctionCallArgumentsDelta(state, event)
+  if (event.type === "response.function_call_arguments.delta" || event.type === "response.function_call_arguments.done")
+    return event.item_id
+      ? onFunctionCallArgumentsDelta(state, event)
+      : ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
   if (event.type === "response.output_item.done") {
     if (event.item?.type === "message" && !event.item.id)
       return ProviderShared.eventError(state.id, `${event.type} message is missing id`)
@@ -1189,10 +1307,10 @@ export const initial = (request: LLMRequest, extension: Extension = BASE): Parse
   hasFunctionCall: false,
   tools: ToolStream.empty<string>(),
   lifecycle: Lifecycle.initial(),
+  outputItems: {},
   messageItems: new Set<string>(),
   messagePhases: {},
   reasoningItems: {},
-  store: OpenResponsesOptions.resolve(request).store,
 })
 
 export const protocol = Protocol.make({

packages/ai/src/protocols/openai-chat.ts

@@ -7,7 +7,10 @@ import { HttpTransport } from "../route/transport/index.js"
 import { Protocol } from "../route/protocol.js"
 import {
   AIError,
+  InvalidProviderOutputReason,
   LLMEvent,
+  ProviderInternalReason,
+  UnknownProviderReason,
   Usage,
   type FinishReason,
   type FinishReasonDetails,
@@ -51,7 +54,12 @@ const OpenAIChatFunction = Schema.Struct({
 
 const OpenAIChatTool = Schema.Struct({
   type: Schema.tag("function"),
-  function: OpenAIChatFunction,
+  function: Schema.Struct({
+    name: Schema.String,
+    description: Schema.String,
+    parameters: JsonObject,
+    strict: Schema.optional(Schema.Boolean),
+  }),
   cache_control: Schema.optional(OpenAIChatCacheControl),
 })
 type OpenAIChatTool = Schema.Schema.Type<typeof OpenAIChatTool>
@@ -133,6 +141,7 @@ export const bodyFields = {
   store: Schema.optional(Schema.Boolean),
   prompt_cache_key: Schema.optional(Schema.String),
   reasoning_effort: Schema.optional(OpenAIOptions.OpenAIReasoningEffort),
+  tool_stream: Schema.optional(Schema.Boolean),
   max_completion_tokens: Schema.optional(Schema.Number),
   max_tokens: Schema.optional(Schema.Number),
   temperature: Schema.optional(Schema.Number),
@@ -218,16 +227,22 @@ const OpenAIChatChoice = Schema.StructWithRest(
   [Schema.Record(Schema.String, Schema.Unknown)],
 )
 
-const OpenAIChatError = Schema.Struct({
-  code: optionalNull(Schema.Union([Schema.String, Schema.Number])),
-  message: Schema.String,
-})
+const OpenAIChatError = Schema.StructWithRest(
+  Schema.Struct({
+    code: optionalNull(Schema.Union([Schema.String, Schema.Number])),
+    message: Schema.String,
+  }),
+  [Schema.Record(Schema.String, Schema.Unknown)],
+)
 
-export const OpenAIChatEvent = Schema.Struct({
-  choices: optionalNull(Schema.Array(OpenAIChatChoice)),
-  usage: optionalNull(OpenAIChatUsage),
-  error: optionalNull(OpenAIChatError),
-})
+export const OpenAIChatEvent = Schema.StructWithRest(
+  Schema.Struct({
+    choices: optionalNull(Schema.Array(OpenAIChatChoice)),
+    usage: optionalNull(OpenAIChatUsage),
+    error: optionalNull(OpenAIChatError),
+  }),
+  [Schema.Record(Schema.String, Schema.Unknown)],
+)
 export type OpenAIChatEvent = Schema.Schema.Type<typeof OpenAIChatEvent>
 type OpenAIChatRequestMessage = LLMRequest["messages"][number]
 
@@ -250,6 +265,7 @@ export interface ParserState {
   readonly reasoningEmitted: boolean
   readonly latestToolIndex?: number
   readonly nextToolIndex: number
+  readonly requireFinishReason: boolean
 }
 
 // =============================================================================
@@ -262,14 +278,21 @@ interface LoweringOptions {
   readonly cacheControl?: (
     cache: CacheHint | undefined,
   ) => Schema.Schema.Type<typeof OpenAIChatCacheControl> | undefined
+  readonly toolCallID?: (id: string) => string
 }
 
-const lowerTool = (tool: ToolDefinition, inputSchema: JsonSchema, options: LoweringOptions): OpenAIChatTool => ({
+const lowerTool = (
+  tool: ToolDefinition,
+  inputSchema: JsonSchema,
+  options: LoweringOptions,
+  supportsStrictMode: boolean,
+): OpenAIChatTool => ({
   type: "function",
   function: {
     name: tool.name,
     description: tool.description,
-    parameters: ToolSchemaProjection.openAI(inputSchema),
+    parameters: inputSchema,
+    ...(supportsStrictMode ? { strict: false } : {}),
   },
   cache_control: options.cacheControl?.(tool.cache),
 })
@@ -282,8 +305,8 @@ const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>) =>
     tool: (name) => ({ type: "function" as const, function: { name } }),
   })
 
-const lowerToolCall = (part: ToolCallPart): OpenAIChatAssistantToolCall => ({
-  id: part.id,
+const lowerToolCall = (part: ToolCallPart, options: LoweringOptions): OpenAIChatAssistantToolCall => ({
+  id: options.toolCallID?.(part.id) ?? part.id,
   type: "function",
   function: {
     name: part.name,
@@ -341,8 +364,9 @@ const lowerUserMessage = Effect.fn("OpenAIChat.lowerUserMessage")(function* (
 
 const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(function* (
   message: OpenAIChatRequestMessage,
-  configuredField?: string,
-  options: LoweringOptions = {},
+  configuredField: string | undefined,
+  requireReasoning: boolean,
+  options: LoweringOptions,
 ) {
   const content: TextPart[] = []
   const reasoning: ReasoningPart[] = []
@@ -359,7 +383,7 @@ const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(func
       continue
     }
     if (part.type === "tool-call") {
-      toolCalls.push(lowerToolCall(part))
+      toolCalls.push(lowerToolCall(part, options))
       continue
     }
   }
@@ -369,15 +393,17 @@ const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(func
   const nativeReasoning = openAICompatibleReasoningContent(message.native?.openaiCompatible)
   const fullyStructured = reasoning.every((part) => Array.isArray(part.providerMetadata?.openai?.reasoningDetails))
   const field = (() => {
-    if (configuredField !== undefined) return configuredField
-    if (reasoning.length === 0) return undefined
+    if (configuredField !== undefined && (requireReasoning || reasoning.length > 0 || nativeReasoning !== undefined))
+      return configuredField
+    if (reasoning.length === 0) return requireReasoning ? "reasoning_content" : undefined
     if (observedField !== undefined) return observedField
     if (nativeReasoning !== undefined) return "reasoning_content"
-    if (!fullyStructured) return "reasoning_content"
+    if (!fullyStructured || requireReasoning) return "reasoning_content"
   })()
   const reasoningText = (() => {
-    if (configuredField !== undefined) return reasoning.length === 0 ? (nativeReasoning ?? "") : text
-    if (reasoning.length === 0) return nativeReasoning
+    if (configuredField !== undefined)
+      return reasoning.length === 0 ? (nativeReasoning ?? (requireReasoning ? "" : undefined)) : text
+    if (reasoning.length === 0) return nativeReasoning ?? (requireReasoning ? "" : undefined)
     return text
   })()
   const cached = message.content.findLast((part) => "cache" in part && part.cache !== undefined)
@@ -405,7 +431,7 @@ const lowerToolMessages = Effect.fn("OpenAIChat.lowerToolMessages")(function* (
     if (part.result.type !== "content") {
       messages.push({
         role: "tool",
-        tool_call_id: part.id,
+        tool_call_id: options.toolCallID?.(part.id) ?? part.id,
         content: ProviderShared.toolResultText(part),
         cache_control: options.cacheControl?.(part.cache),
       })
@@ -415,7 +441,7 @@ const lowerToolMessages = Effect.fn("OpenAIChat.lowerToolMessages")(function* (
     const text = content.filter((item) => item.type === "text").map((item) => item.text)
     messages.push({
       role: "tool",
-      tool_call_id: part.id,
+      tool_call_id: options.toolCallID?.(part.id) ?? part.id,
       content: text.join("\n"),
       cache_control: options.cacheControl?.(part.cache),
     })
@@ -431,11 +457,13 @@ const lowerToolMessages = Effect.fn("OpenAIChat.lowerToolMessages")(function* (
 
 const lowerMessage = Effect.fn("OpenAIChat.lowerMessage")(function* (
   message: OpenAIChatRequestMessage,
-  reasoningField?: string,
-  options: LoweringOptions = {},
+  reasoningField: string | undefined,
+  requireReasoning: boolean,
+  options: LoweringOptions,
 ) {
   if (message.role === "user") return [yield* lowerUserMessage(message, options)]
-  if (message.role === "assistant") return [yield* lowerAssistantMessage(message, reasoningField, options)]
+  if (message.role === "assistant")
+    return [yield* lowerAssistantMessage(message, reasoningField, requireReasoning, options)]
   return (yield* lowerToolMessages(message, options)).messages
 })
 
@@ -456,12 +484,37 @@ const lowerMessages = Effect.fn("OpenAIChat.lowerMessages")(function* (request:
           ]
         : [{ role: "system", content: ProviderShared.joinText(request.system) }]
   const messages = [...system]
+  const modelID = request.model.id.toLowerCase()
+  const requireReasoning =
+    request.model.compatibility?.requireReasoning ??
+    (request.model.compatibility?.reasoningField !== undefined ||
+      request.model.provider === "deepseek" ||
+      request.model.route.endpoint.baseURL?.toLowerCase().includes("deepseek.com") ||
+      modelID.includes("deepseek"))
+  const reasoningField = request.model.compatibility?.reasoningField
+  const mistral = ["mistral", "devstral", "codestral", "pixtral", "mixtral"].some((family) => modelID.includes(family))
+  const lowering = {
+    ...options,
+    toolCallID: (id: string) => {
+      if (mistral) return id.replace(/[^a-zA-Z0-9]/g, "").slice(0, 9).padEnd(9, "0")
+      if (modelID.includes("claude")) return id.replace(/[^a-zA-Z0-9_-]/g, "_")
+      if (request.model.provider === "openai" || request.model.provider === "azure" || modelID.startsWith("openai/"))
+        return id.slice(0, 40)
+      return id
+    },
+  }
+  const requireAssistantAfterTool = request.model.compatibility?.requireAssistantAfterTool ?? mistral
+  const bridgeTools = () => {
+    if (requireAssistantAfterTool && messages.at(-1)?.role === "tool") messages.push({ role: "assistant", content: "Done." })
+  }
   const pendingImages: Array<Schema.Schema.Type<typeof OpenAIChatUserContent>> = []
   const flushImages = () => {
     if (pendingImages.length === 0) return
+    bridgeTools()
     messages.push({ role: "user", content: pendingImages.splice(0) })
   }
   for (const message of request.messages) {
+    if (message.role === "user") bridgeTools()
     if (message.role === "system") {
       const part = yield* ProviderShared.wrappedSystemUpdate("OpenAI Chat", message)
       if (pendingImages.length > 0) {
@@ -504,14 +557,16 @@ const lowerMessages = Effect.fn("OpenAIChat.lowerMessages")(function* (request:
         )
       continue
     }
+    if (message.role === "assistant" && message.content.every((part) => part.type === "text" && part.text.trim() === ""))
+      continue
     if (message.role === "tool") {
-      const lowered = yield* lowerToolMessages(message, options)
+      const lowered = yield* lowerToolMessages(message, lowering)
       messages.push(...lowered.messages)
       pendingImages.push(...lowered.images)
       continue
     }
     flushImages()
-    messages.push(...(yield* lowerMessage(message, request.model.compatibility?.reasoningField, options)))
+    messages.push(...(yield* lowerMessage(message, reasoningField, requireReasoning, lowering)))
   }
   flushImages()
   return messages
@@ -528,11 +583,123 @@ const hasToolHistory = (messages: ReadonlyArray<LLMRequest["messages"][number]>)
   return false
 }
 
-const lowerOptions = (request: LLMRequest) => {
+// Derive `max_tokens` vs `max_completion_tokens` from provider/baseURL when
+// explicit `compatibility.maxTokensField` is not set. Aligned with
+// models.dev provider naming: DeepSeek, Moonshot AI, Together AI, ZAI
+// (Zhipu + Coding Plan variants), Nvidia, Cerebras, Chutes, etc. still
+// require `max_tokens`.
+const detectMaxTokensField = (provider: string, baseURL: string | undefined): "max_tokens" | "max_completion_tokens" => {
+  const p = provider.toLowerCase()
+  const url = (baseURL ?? "").toLowerCase()
+  if (
+    p === "deepseek" ||
+    url.includes("deepseek.com") ||
+    p === "moonshotai" ||
+    url.includes("api.moonshot.ai") ||
+    p === "togetherai" ||
+    url.includes("api.together.") ||
+    p === "zai" ||
+    p === "zai-coding-plan" ||
+    p === "zhipuai" ||
+    p === "zhipuai-coding-plan" ||
+    url.includes("api.z.ai") ||
+    url.includes("open.bigmodel.cn") ||
+    p === "nvidia" ||
+    url.includes("integrate.api.nvidia.com") ||
+    p === "cerebras" ||
+    url.includes("cerebras.ai") ||
+    url.includes("llm.chutes.ai") ||
+    p === "chutes" ||
+    p === "cloudflare-ai-gateway" ||
+    url.includes("gateway.ai.cloudflare.com") ||
+    p === "cloudflare-workers-ai" ||
+    url.includes("api.cloudflare.com")
+  )
+    return "max_tokens"
+  return "max_completion_tokens"
+}
+
+const detectSupportsStore = (provider: string, baseURL: string | undefined): boolean => {
+  const p = provider.toLowerCase()
+  const url = (baseURL ?? "").toLowerCase()
+  const isNvidia = p === "nvidia" || url.includes("integrate.api.nvidia.com")
+  const isMoonshot = p === "moonshotai" || p === "moonshotai-cn" || url.includes("api.moonshot.")
+  const isTogether = p === "togetherai" || p === "together" || url.includes("api.together.")
+  const isZai =
+    p === "zai" ||
+    p === "zai-coding-plan" ||
+    p === "zhipuai" ||
+    p === "zhipuai-coding-plan" ||
+    url.includes("api.z.ai") ||
+    url.includes("open.bigmodel.cn")
+  const isDeepSeek = p === "deepseek" || url.includes("deepseek.com")
+  const isCerebras = p === "cerebras" || url.includes("cerebras.ai")
+  const isXai = p === "xai" || url.includes("api.x.ai")
+  const isChutes = p === "chutes" || url.includes("chutes.ai")
+  const isCloudflareWorkersAI = p === "cloudflare-workers-ai" || url.includes("api.cloudflare.com")
+  const isCloudflareAiGateway = p === "cloudflare-ai-gateway" || url.includes("gateway.ai.cloudflare.com")
+  const isVercelAiGateway = p === "vercel-ai-gateway" || url.includes("ai-gateway.vercel.sh") || url.includes("vercel.sh")
+  const isAntLing = p === "ant-ling" || url.includes("api.ant-ling.com")
+  const isOpencode = p === "opencode" || url.includes("opencode.ai")
+  const isNonStandard =
+    isNvidia ||
+    isCerebras ||
+    isXai ||
+    isTogether ||
+    isChutes ||
+    isDeepSeek ||
+    isZai ||
+    isMoonshot ||
+    isOpencode ||
+    isCloudflareWorkersAI ||
+    isCloudflareAiGateway ||
+    isVercelAiGateway ||
+    isAntLing
+  return !isNonStandard
+}
+
+const detectSupportsUsageInStreaming = (): boolean => true
+
+const detectSupportsStrictMode = (provider: string, baseURL: string | undefined): boolean => {
+  const p = provider.toLowerCase()
+  const url = (baseURL ?? "").toLowerCase()
+  const isMoonshot = p === "moonshotai" || p === "moonshotai-cn" || url.includes("api.moonshot.")
+  const isTogether = p === "togetherai" || p === "together" || url.includes("api.together.")
+  const isCloudflareAiGateway = p === "cloudflare-ai-gateway" || url.includes("gateway.ai.cloudflare.com")
+  const isNvidia = p === "nvidia" || url.includes("integrate.api.nvidia.com")
+  return !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia
+}
+
+const detectZaiToolStream = (
+  provider: string,
+  baseURL: string | undefined,
+  modelID: string,
+): boolean => {
+  const p = provider.toLowerCase()
+  const url = (baseURL ?? "").toLowerCase()
+  const isZai =
+    p === "zai" ||
+    p === "zai-coding-plan" ||
+    p === "zhipuai" ||
+    p === "zhipuai-coding-plan" ||
+    url.includes("api.z.ai") ||
+    url.includes("open.bigmodel.cn")
+  if (!isZai) return false
+  const id = modelID.toLowerCase()
+  if (id === "glm-4.5" || id === "glm-4.5-air" || id === "glm-4.5-flash" || id === "glm-4.5v") return false
+  return true
+}
+
+const lowerOptions = (request: LLMRequest, supportsStore: boolean) => {
   const options = OpenAIOptions.resolve(request)
+  const cacheKey = ProviderShared.promptCacheKey(request)
   return {
-    ...(options.store !== undefined ? { store: options.store } : {}),
-    ...(request.promptCacheKey ? { prompt_cache_key: request.promptCacheKey } : {}),
+    ...(supportsStore && options.store !== undefined ? { store: options.store } : {}),
+    // For providers that support `store`, ensure stateless `store:false` is sent
+    // even when no explicit `providerOptions.store` was supplied, mirroring the
+    // native OpenAI Chat default. Non-standard providers omit `store` entirely.
+    ...(supportsStore && options.store === undefined ? { store: false } : {}),
+    ...(cacheKey ? { prompt_cache_key: cacheKey } : {}),
     ...(options.reasoningEffort ? { reasoning_effort: options.reasoningEffort } : {}),
   }
 }
@@ -550,8 +717,19 @@ export const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (
     )
   const generation = request.generation
   const toolSchemaCompatibility = request.model.compatibility?.toolSchema
-  const maxTokensField = request.model.compatibility?.maxTokensField ?? "max_tokens"
+  const provider = String(request.model.provider)
+  const baseURL = request.model.route.endpoint.baseURL
+  const detectedMaxTokensField = detectMaxTokensField(provider, baseURL)
+  const maxTokensField = request.model.compatibility?.maxTokensField ?? detectedMaxTokensField
+  const supportsStore = request.model.compatibility?.supportsStore ?? detectSupportsStore(provider, baseURL)
+  const supportsUsageInStreaming =
+    request.model.compatibility?.supportsUsageInStreaming ?? detectSupportsUsageInStreaming()
+  const supportsStrictMode = request.model.compatibility?.supportsStrictMode ?? detectSupportsStrictMode(provider, baseURL)
+  const zaiToolStream =
+    request.model.compatibility?.zaiToolStream ??
+    detectZaiToolStream(provider, baseURL, request.model.id)
   const hasHistory = hasToolHistory(request.messages)
+  const hasActiveTools = request.tools.length > 0
   return {
     model: request.model.id,
     messages: yield* lowerMessages(request, options),
@@ -565,11 +743,13 @@ export const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (
               tool,
               ToolSchemaProjection.modelCompatibility(tool.inputSchema, toolSchemaCompatibility),
               options,
+              supportsStrictMode,
             ),
           ),
     tool_choice: request.toolChoice ? yield* lowerToolChoice(request.toolChoice) : undefined,
     stream: true as const,
-    stream_options: { include_usage: true },
+    ...(supportsUsageInStreaming ? { stream_options: { include_usage: true } } : {}),
+    ...(zaiToolStream && hasActiveTools ? { tool_stream: true } : {}),
     ...(maxTokensField === "max_completion_tokens"
       ? { max_completion_tokens: generation?.maxTokens }
       : { max_tokens: generation?.maxTokens }),
@@ -579,7 +759,7 @@ export const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (
     presence_penalty: generation?.presencePenalty,
     seed: generation?.seed,
     stop: generation?.stop,
-    ...lowerOptions(request),
+    ...lowerOptions(request, supportsStore),
   }
 })
 
@@ -589,14 +769,40 @@ export const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (
 // Streaming parsers are small state machines: every event returns a new state
 // plus the common `LLMEvent`s produced by that event. Tool calls are accumulated
 // because OpenAI streams JSON arguments across multiple deltas.
-const mapFinishReason = (reason: string | null | undefined): FinishReason => {
-  if (reason === "stop") return "stop"
-  if (reason === "length") return "length"
-  if (reason === "content_filter") return "content-filter"
-  if (reason === "function_call" || reason === "tool_calls") return "tool-calls"
-  if (reason === "error") return "error"
-  return "unknown"
-}
+const finishReasonError = (event: OpenAIChatEvent, reason: AIError["reason"]) =>
+  new AIError({
+    module: ADAPTER,
+    method: "stream",
+    body: ProviderShared.encodeJson(event),
+    reason,
+  })
+
+const mapFinishReason = Effect.fn("OpenAIChat.mapFinishReason")(function* (event: OpenAIChatEvent, reason: string) {
+  switch (reason) {
+    case "error":
+      return yield* finishReasonError(
+        event,
+        new UnknownProviderReason({ message: "Provider reported an error (finish_reason: error)" }),
+      )
+    case "network_error":
+      return yield* finishReasonError(
+        event,
+        new ProviderInternalReason({ message: "Provider reported a network error (finish_reason: network_error)" }),
+      )
+    case "stop":
+    case "end":
+      return "stop" as const
+    case "length":
+      return "length" as const
+    case "content_filter":
+      return "content-filter" as const
+    case "function_call":
+    case "tool_calls":
+      return "tool-calls" as const
+    default:
+      return "unknown" as const
+  }
+})
 
 // OpenAI Chat reports `prompt_tokens` (inclusive total) with a
 // cached-read and cache-write subsets, and `completion_tokens` (inclusive
@@ -709,16 +915,20 @@ const reasoningMetadata = (field: ParserState["reasoningField"], details?: Reado
 
 const step = (state: ParserState, event: OpenAIChatEvent) =>
   Effect.gen(function* () {
-    if (event.error)
+    if (event.error) {
+      const body = ProviderShared.encodeJson(event)
       return yield* new AIError({
         module: ADAPTER,
         method: "stream",
+        body,
         reason: classifyProviderFailure({
           message: event.error.message,
           code: event.error.code === undefined || event.error.code === null ? undefined : String(event.error.code),
           status: typeof event.error.code === "number" ? event.error.code : undefined,
+          rawBody: body,
         }),
       })
+    }
     const events: LLMEvent[] = []
     const choice = event.choices?.[0]
     // Moonshot (and a few other OpenAI-compatible providers) attach usage to
@@ -727,8 +937,11 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
     const usage = mapUsage(event.usage) ?? (choiceUsage ? mapUsage(choiceUsage) : undefined) ?? state.usage
     const rawFinishReason = choice?.finish_reason
     const finishReason =
-      rawFinishReason !== undefined && rawFinishReason !== null
-        ? { normalized: mapFinishReason(rawFinishReason), raw: choice?.native_finish_reason ?? rawFinishReason }
+      rawFinishReason
+        ? {
+            normalized: yield* mapFinishReason(event, rawFinishReason),
+            raw: choice?.native_finish_reason ?? rawFinishReason,
+          }
         : state.finishReason
     const delta = choice?.delta
     const toolDeltas = delta?.tool_calls ?? []
@@ -748,7 +961,11 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
       toolDeltas.some((tool) => Boolean(tool.id) || Boolean(tool.function?.name) || Boolean(tool.function?.arguments))
     if (state.finishReason !== undefined) {
       if (hasLateContent)
-        return yield* ProviderShared.eventError(ADAPTER, "OpenAI Chat received content after the finish reason")
+        return yield* ProviderShared.eventError(
+          ADAPTER,
+          "OpenAI Chat received content after the finish reason",
+          ProviderShared.encodeJson(event),
+        )
       return [{ ...state, usage }, events] as const
     }
 
@@ -820,14 +1037,19 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
         { id: id || undefined, name: name || undefined, text },
         "OpenAI Chat tool call delta is missing id or name",
       )
-      if (ToolStream.isError(result)) return yield* result
+      if (ToolStream.isError(result))
+        return yield* ProviderShared.eventError(ADAPTER, result.reason.message, ProviderShared.encodeJson(event))
       tools = result.tools
       if (result.events.length) lifecycle = Lifecycle.stepStart(lifecycle, events)
       events.push(...result.events)
     }
 
     if (finishReason !== undefined && state.finishReason === undefined && Object.keys(pendingTools).length > 0)
-      return yield* ProviderShared.eventError(ADAPTER, "OpenAI Chat tool call delta is missing id or name")
+      return yield* ProviderShared.eventError(
+        ADAPTER,
+        "OpenAI Chat tool call delta is missing id or name",
+        ProviderShared.encodeJson(event),
+      )
 
     // Finalize accumulated tool inputs eagerly when finish_reason arrives so
     // valid calls and malformed local calls settle independently.
@@ -850,16 +1072,27 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
         reasoningEmitted,
         latestToolIndex,
         nextToolIndex,
+        requireFinishReason: state.requireFinishReason,
       },
       events,
     ] as const
   })
 
-const finishEvents = (state: ParserState): ReadonlyArray<LLMEvent> => {
+const finishEvents = Effect.fn("OpenAIChat.finishEvents")(function* (state: ParserState) {
+  if (state.finishReason === undefined && state.requireFinishReason)
+    return yield* new AIError({
+      module: ADAPTER,
+      method: "stream",
+      reason: new InvalidProviderOutputReason({
+        classification: "incomplete-stream",
+        message: "OpenAI Chat stream ended without finish_reason",
+        route: ADAPTER,
+      }),
+    })
   const events: LLMEvent[] = []
   const toolCallEvents =
     state.finishReason === undefined && Object.keys(state.tools).length > 0
-      ? Effect.runSync(ToolStream.finishAll(ADAPTER, state.tools)).events
+      ? (yield* ToolStream.finishAll(ADAPTER, state.tools)).events
       : state.toolCallEvents
   const hasToolCalls = toolCallEvents.length > 0
   const reason = state.finishReason
@@ -868,7 +1101,7 @@ const finishEvents = (state: ParserState): ReadonlyArray<LLMEvent> => {
         normalized:
           state.finishReason.normalized === "stop" && hasToolCalls ? "tool-calls" : state.finishReason.normalized,
       }
-    : { normalized: hasToolCalls ? ("tool-calls" as const) : ("unknown" as const) }
+    : { normalized: hasToolCalls ? ("tool-calls" as const) : ("stop" as const) }
   const metadata = reasoningMetadata(
     state.reasoningField,
     state.reasoningDetailsObserved ? state.reasoningDetails : undefined,
@@ -882,7 +1115,7 @@ const finishEvents = (state: ParserState): ReadonlyArray<LLMEvent> => {
   events.push(...toolCallEvents)
   Lifecycle.finish(lifecycle, events, { reason, usage: state.usage })
   return events
-}
+})
 
 // =============================================================================
 // Protocol And OpenAI Route
@@ -911,6 +1144,7 @@ export const protocol = Protocol.make({
       reasoningDetailsObserved: false,
       reasoningEmitted: false,
       nextToolIndex: 0,
+      requireFinishReason: request.model.compatibility?.requireFinishReason ?? true,
     }),
     step,
     onHalt: finishEvents,

packages/ai/src/protocols/openai-compatible-responses.ts

@@ -17,6 +17,7 @@ export const route = Route.make({
   protocol: OpenResponses.protocol,
   endpoint: Endpoint.path(OpenResponses.PATH),
   transport: OpenResponses.httpTransport,
+  defaults: { providerOptions: { store: false, include: ["reasoning.encrypted_content"] } },
 })
 
 export * as OpenAICompatibleResponses from "./openai-compatible-responses.js"

packages/ai/src/protocols/openai-responses.ts

@@ -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),
 }
@@ -54,6 +89,7 @@ export type OpenAIResponsesBody = Schema.Schema.Type<typeof OpenAIResponsesBody>
 const extension = {
   id: ADAPTER,
   name: NAME,
+  lowerHostedToolItem: (item: unknown) => (Schema.is(OpenAIResponsesHostedToolItem)(item) ? item : undefined),
 } satisfies OpenResponses.Extension
 
 const nativeImageToolInput = (tool: ToolDefinition) => {
@@ -86,14 +122,18 @@ const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>, tool
         : { type: "function" as const, name },
   })
 
+const decodeBody = ProviderShared.validateWith(Schema.decodeUnknownEffect(OpenAIResponsesBody))
+
 const fromRequest = Effect.fn("OpenAIResponses.fromRequest")(function* (request: LLMRequest) {
   const body = yield* OpenResponses.fromRequestWithExtension(
     LLMRequest.update(request, { tools: [], toolChoice: undefined }),
     extension,
   )
   const toolSchemaCompatibility = request.model.compatibility?.toolSchema
-  return {
+  const parallelToolCalls = OpenResponses.resolveParallelToolCalls(request)
+  return yield* decodeBody({
     ...body,
+    ...(parallelToolCalls === undefined ? {} : { parallel_tool_calls: parallelToolCalls }),
     tools:
       request.tools.length === 0
         ? undefined
@@ -102,7 +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) {
@@ -143,13 +183,11 @@ const HOSTED_TOOLS = {
 } as const satisfies ResponsesHostedTools.Definitions
 
 const step = (state: OpenResponses.ParserState, event: OpenResponses.Event) => {
-  if (event.type === "response.reasoning_text.delta" || event.type === "response.reasoning_summary.delta")
+  if (event.type === "response.reasoning_text.delta")
     return event.item_id
-      ? Effect.succeed(OpenResponses.onReasoningDelta(state, event, event.item_id))
-      : ProviderShared.eventError(ADAPTER, `${event.type} is missing item_id`)
-  if (event.type === "response.reasoning_text.done" || event.type === "response.reasoning_summary.done")
-    return event.item_id
-      ? Effect.succeed(OpenResponses.onReasoningDone(state, event))
+      ? Effect.succeed(
+          OpenResponses.onReasoningDelta(state, event, OpenResponses.outputItemID(state, event) ?? event.item_id),
+        )
       : ProviderShared.eventError(ADAPTER, `${event.type} is missing item_id`)
   if (event.type === "response.output_item.done" && event.item && ResponsesHostedTools.isItem(event.item, HOSTED_TOOLS))
     return ResponsesHostedTools.onDone(state, event.item, HOSTED_TOOLS)
@@ -190,7 +228,7 @@ export const route = Route.make({
   endpoint,
   auth,
   transport,
-  defaults: { providerOptions: { store: false } },
+  defaults: { providerOptions: { store: false, include: ["reasoning.encrypted_content"] } },
 })
 
 export * as OpenAIResponses from "./openai-responses.js"

packages/ai/src/protocols/shared.ts

@@ -24,6 +24,17 @@ export const JsonObject = Schema.Record(Schema.String, Schema.Unknown)
 export const optionalArray = <const S extends Schema.Top>(schema: S) => Schema.optional(Schema.Array(schema))
 export const optionalNull = <const S extends Schema.Top>(schema: S) => Schema.optional(Schema.NullOr(schema))
 
+export const OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH = 64
+
+// OpenAI limits `prompt_cache_key` to 64 chars; DeepSeek and Zai inherit the same
+// limit via their OpenAI-compatible APIs. Clamp with unicode-aware slicing.
+export const promptCacheKey = (request: LLMRequest): string | undefined => {
+  if (request.cache === "none" || request.promptCacheKey === undefined) return undefined
+  const chars = Array.from(request.promptCacheKey)
+  if (chars.length <= OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH) return request.promptCacheKey
+  return chars.slice(0, OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH).join("")
+}
+
 /**
  * Streaming tool-call accumulator. Adapters that build a tool call across
  * multiple `tool-input-delta` chunks store the partial JSON input string here
@@ -199,10 +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>,
@@ -210,9 +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)) &&

packages/ai/src/protocols/utils/open-responses-options.ts

@@ -56,7 +56,6 @@ export const StreamOptions = Schema.Struct({
 })
 
 export const Options = Schema.Struct({
-  instructions: Schema.optional(Schema.String),
   store: Schema.optional(Schema.Boolean),
   metadata: Schema.optional(Schema.Record(Schema.String, Schema.String)),
   safetyIdentifier: Schema.optional(Schema.String),

packages/ai/src/protocols/utils/partial-json-options.ts

@@ -0,0 +1,64 @@
+/*
+ * Adapted from partial-json by the Promplate Dev Team:
+ * https://github.com/promplate/partial-json-parser-js/blob/main/src/options.ts
+ * Licensed under the MIT License; see partial-json.ts for the complete notice.
+ */
+
+/**
+ * allow partial strings like `"hello \u12` to be parsed as `"hello `
+ */
+export const STR = 0b000000001
+
+/**
+ * allow partial numbers like `123.` to be parsed as `123`
+ */
+export const NUM = 0b000000010
+
+/**
+ * allow partial arrays like `[1, 2,` to be parsed as `[1, 2]`
+ */
+export const ARR = 0b000000100
+
+/**
+ * allow partial objects like `{"a": 1, "b":` to be parsed as `{"a": 1}`
+ */
+export const OBJ = 0b000001000
+
+/**
+ * allow `nu` to be parsed as `null`
+ */
+export const NULL = 0b000010000
+
+/**
+ * allow `tr` to be parsed as `true`, and `fa` to be parsed as `false`
+ */
+export const BOOL = 0b000100000
+
+/**
+ * allow `Na` to be parsed as `NaN`
+ */
+export const NAN = 0b001000000
+
+/**
+ * allow `Inf` to be parsed as `Infinity`
+ */
+export const INFINITY = 0b010000000
+
+/**
+ * allow `-Inf` to be parsed as `-Infinity`
+ */
+export const _INFINITY = 0b100000000
+
+export const INF = INFINITY | _INFINITY
+export const SPECIAL = NULL | BOOL | INF | NAN
+export const ATOM = STR | NUM | SPECIAL
+export const COLLECTION = ARR | OBJ
+export const ALL = ATOM | COLLECTION
+
+/**
+ * Control what types you allow to be partially parsed.
+ * The default is to allow all types to be partially parsed, which in most cases is the best option.
+ */
+export const Allow = { STR, NUM, ARR, OBJ, NULL, BOOL, NAN, INFINITY, _INFINITY, INF, SPECIAL, ATOM, COLLECTION, ALL }
+
+export default Allow

packages/ai/src/protocols/utils/partial-json.ts

@@ -0,0 +1,282 @@
+/*
+ * Adapted from partial-json by the Promplate Dev Team:
+ * https://github.com/promplate/partial-json-parser-js
+ *
+ * MIT License
+ *
+ * Copyright (c) 2023 Promplate Dev Team
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+import { Schema } from "effect"
+import { Allow } from "./partial-json-options.js"
+export * from "./partial-json-options.js"
+
+export class PartialJSON extends Error {}
+export class MalformedJSON extends Error {}
+
+const decodeJson = Schema.decodeUnknownSync(Schema.fromJsonString(Schema.Unknown))
+
+/** Parse complete or incomplete JSON, restricted by the supplied partial-value flags. */
+export function parseJSON(jsonString: string, allowPartial = Allow.ALL): unknown {
+  if (typeof jsonString !== "string") throw new TypeError(`expecting str, got ${typeof jsonString}`)
+  const input = jsonString.trim()
+  if (!input) throw new Error(`${jsonString} is empty`)
+  try {
+    return decodeJson(input)
+  } catch {}
+
+  const repaired = repairJSON(input)
+  if (repaired !== input) {
+    try {
+      return decodeJson(repaired)
+    } catch {}
+  }
+
+  try {
+    return _parseJSON(input, allowPartial)
+  } catch (error) {
+    if (repaired !== input) return _parseJSON(repaired, allowPartial)
+    throw error
+  }
+}
+
+const repairJSON = (input: string) => {
+  let repaired = ""
+  let quoted = false
+
+  for (let index = 0; index < input.length; index++) {
+    const character = input[index]
+    if (!quoted) {
+      repaired += character
+      if (character === '"') quoted = true
+      continue
+    }
+
+    if (character === '"') {
+      repaired += character
+      quoted = false
+      continue
+    }
+
+    if (character === "\\") {
+      const next = input[index + 1]
+      if (next === "u" && /^[0-9a-fA-F]{4}$/.test(input.slice(index + 2, index + 6))) {
+        repaired += input.slice(index, index + 6)
+        index += 5
+        continue
+      }
+      if (next !== undefined && '"\\/bfnrtu'.includes(next)) {
+        repaired += `\\${next}`
+        index++
+        continue
+      }
+      repaired += "\\\\"
+      continue
+    }
+
+    const code = character.charCodeAt(0)
+    repaired += code <= 0x1f ? `\\u${code.toString(16).padStart(4, "0")}` : character
+  }
+
+  return repaired
+}
+
+const _parseJSON = (jsonString: string, allow: number) => {
+  const length = jsonString.length
+  let index = 0
+
+  const markPartialJSON = (message: string): never => {
+    throw new PartialJSON(`${message} at position ${index}`)
+  }
+
+  const throwMalformedError = (message: string): never => {
+    throw new MalformedJSON(`${message} at position ${index}`)
+  }
+
+  const parseAny = (): unknown => {
+    skipBlank()
+    if (index >= length) markPartialJSON("Unexpected end of input")
+    if (jsonString[index] === '"') return parseStr()
+    if (jsonString[index] === "{") return parseObj()
+    if (jsonString[index] === "[") return parseArr()
+    if (
+      jsonString.substring(index, index + 4) === "null" ||
+      (Allow.NULL & allow && length - index < 4 && "null".startsWith(jsonString.substring(index)))
+    ) {
+      index += 4
+      return null
+    }
+    if (
+      jsonString.substring(index, index + 4) === "true" ||
+      (Allow.BOOL & allow && length - index < 4 && "true".startsWith(jsonString.substring(index)))
+    ) {
+      index += 4
+      return true
+    }
+    if (
+      jsonString.substring(index, index + 5) === "false" ||
+      (Allow.BOOL & allow && length - index < 5 && "false".startsWith(jsonString.substring(index)))
+    ) {
+      index += 5
+      return false
+    }
+    if (
+      jsonString.substring(index, index + 8) === "Infinity" ||
+      (Allow.INFINITY & allow && length - index < 8 && "Infinity".startsWith(jsonString.substring(index)))
+    ) {
+      index += 8
+      return Infinity
+    }
+    if (
+      jsonString.substring(index, index + 9) === "-Infinity" ||
+      (Allow._INFINITY & allow &&
+        1 < length - index &&
+        length - index < 9 &&
+        "-Infinity".startsWith(jsonString.substring(index)))
+    ) {
+      index += 9
+      return -Infinity
+    }
+    if (
+      jsonString.substring(index, index + 3) === "NaN" ||
+      (Allow.NAN & allow && length - index < 3 && "NaN".startsWith(jsonString.substring(index)))
+    ) {
+      index += 3
+      return NaN
+    }
+    return parseNum()
+  }
+
+  const parseStr = (): string => {
+    const start = index
+    let escape = false
+    index++
+    while (index < length && (jsonString[index] !== '"' || (escape && jsonString[index - 1] === "\\"))) {
+      escape = jsonString[index] === "\\" ? !escape : false
+      index++
+    }
+    if (jsonString.charAt(index) === '"') {
+      try {
+        return decodeJson(jsonString.substring(start, ++index - Number(escape))) as string
+      } catch (error) {
+        throwMalformedError(String(error))
+      }
+    }
+    if (Allow.STR & allow) {
+      try {
+        return decodeJson(`${jsonString.substring(start, index - Number(escape))}"`) as string
+      } catch {
+        return decodeJson(`${jsonString.substring(start, jsonString.lastIndexOf("\\"))}"`) as string
+      }
+    }
+    return markPartialJSON("Unterminated string literal")
+  }
+
+  const parseObj = (): Record<string, unknown> => {
+    index++
+    skipBlank()
+    const object: Record<string, unknown> = {}
+    try {
+      while (jsonString[index] !== "}") {
+        skipBlank()
+        if (index >= length && Allow.OBJ & allow) return object
+        const key = parseStr()
+        skipBlank()
+        index++
+        try {
+          Object.defineProperty(object, key, {
+            value: parseAny(),
+            enumerable: true,
+            configurable: true,
+            writable: true,
+          })
+        } catch (error) {
+          if (Allow.OBJ & allow) return object
+          throw error
+        }
+        skipBlank()
+        if (jsonString[index] === ",") index++
+      }
+    } catch {
+      if (Allow.OBJ & allow) return object
+      return markPartialJSON("Expected '}' at end of object")
+    }
+    index++
+    return object
+  }
+
+  const parseArr = (): unknown[] => {
+    index++
+    const array: unknown[] = []
+    try {
+      while (jsonString[index] !== "]") {
+        array.push(parseAny())
+        skipBlank()
+        if (jsonString[index] === ",") index++
+      }
+    } catch {
+      if (Allow.ARR & allow) return array
+      return markPartialJSON("Expected ']' at end of array")
+    }
+    index++
+    return array
+  }
+
+  const parseNum = (): unknown => {
+    if (index === 0) {
+      if (jsonString === "-") throwMalformedError("Not sure what '-' is")
+      try {
+        return decodeJson(jsonString)
+      } catch (error) {
+        if (Allow.NUM & allow) {
+          try {
+            return decodeJson(jsonString.substring(0, jsonString.lastIndexOf("e")))
+          } catch {}
+        }
+        throwMalformedError(String(error))
+      }
+    }
+
+    const start = index
+    if (jsonString[index] === "-") index++
+    while (jsonString[index] && !",]}".includes(jsonString[index])) index++
+    if (index === length && !(Allow.NUM & allow)) markPartialJSON("Unterminated number literal")
+
+    try {
+      return decodeJson(jsonString.substring(start, index))
+    } catch (error) {
+      if (jsonString.substring(start, index) === "-") markPartialJSON("Not sure what '-' is")
+      try {
+        return decodeJson(jsonString.substring(start, jsonString.lastIndexOf("e")))
+      } catch {
+        throwMalformedError(String(error))
+      }
+    }
+  }
+
+  const skipBlank = () => {
+    while (index < length && " \n\r\t".includes(jsonString[index])) index++
+  }
+
+  return parseAny()
+}
+
+export const parse = parseJSON

packages/ai/src/protocols/utils/tool-schema.ts

@@ -2,20 +2,6 @@ import type { JsonSchema, LanguageModelToolSchemaCompatibility } from "../../sch
 import { isRecord } from "../../utils/record.js"
 import { GeminiToolSchema } from "./gemini-tool-schema.js"
 
-const removeNullSchemas = (value: unknown): unknown => {
-  if (Array.isArray(value)) return value.map(removeNullSchemas)
-  if (!isRecord(value)) return value
-  const fields = Object.fromEntries(
-    Object.entries(value)
-      .filter(([key]) => key !== "anyOf")
-      .map(([key, field]) => [key, removeNullSchemas(field)]),
-  )
-  if (!Array.isArray(value.anyOf)) return fields
-  const variants = value.anyOf.filter((variant) => !isRecord(variant) || variant.type !== "null").map(removeNullSchemas)
-  if (variants.length === 1 && isRecord(variants[0])) return { ...fields, ...variants[0] }
-  return { ...fields, anyOf: variants }
-}
-
 const tupleItemsSchema = (items: ReadonlyArray<unknown>) => {
   const projected = items.map(moonshotNode)
   if (projected.length === 0) return {}
@@ -45,24 +31,7 @@ const moonshot = (schema: JsonSchema): JsonSchema => {
   return isRecord(projected) ? projected : {}
 }
 
-const openAI = (schema: JsonSchema): JsonSchema => {
-  const variants = Array.isArray(schema.anyOf) ? schema.anyOf.filter(isRecord) : []
-  const flattened =
-    variants.length === 0
-      ? { ...schema, type: "object" }
-      : {
-          ...Object.fromEntries(Object.entries(schema).filter(([key]) => key !== "anyOf")),
-          type: "object",
-          properties: variants.reduce(
-            (properties, variant) => ({ ...(isRecord(variant.properties) ? variant.properties : {}), ...properties }),
-            {},
-          ),
-          additionalProperties: false,
-        }
-  const normalized = removeNullSchemas(flattened)
-  return isRecord(normalized) ? normalized : { type: "object" }
-}
-
+const openAI = (schema: JsonSchema): JsonSchema => schema
 const responses = openAI
 
 const gemini = (schema: JsonSchema): JsonSchema => GeminiToolSchema.convert(schema) ?? {}

packages/ai/src/protocols/utils/tool-stream.ts

@@ -1,8 +1,10 @@
-import { Effect } from "effect"
-import { AIError, LLMEvent, type ProviderMetadata, type ToolCall, type ToolInputError } from "../../schema/index.js"
+import { Effect, Option } from "effect"
+import { AIError, LLMEvent, type ProviderMetadata, type ToolCall } from "../../schema/index.js"
 import { eventError, parseToolInput, type ToolAccumulator } from "../shared.js"
+import { parse } from "./partial-json.js"
 
 type StreamKey = string | number
+const parsePartialInput = Option.liftThrowable(parse)
 
 /**
  * One pending streamed tool call. Providers emit the tool identity and JSON
@@ -62,38 +64,39 @@ const inputDelta = (tool: PendingTool, text: string) =>
     id: tool.id,
     name: tool.name,
     text,
+    input: Option.getOrElse(parsePartialInput(tool.input), () => ({})),
   })
 
 const toolCall = (route: string, tool: PendingTool, inputOverride?: string) => {
   const raw = inputOverride ?? tool.input
   return parseToolInput(route, tool.name, raw).pipe(
-    Effect.map((input): ToolCall | ToolInputError =>
-      LLMEvent.toolCall({
-        id: tool.id,
-        name: tool.name,
-        input,
-        providerExecuted: tool.providerExecuted ? true : undefined,
-        providerMetadata: tool.providerMetadata,
-      }),
-    ),
     Effect.catch((error) =>
       tool.providerExecuted
         ? Effect.fail(error)
         : Effect.succeed(
-            LLMEvent.toolInputError({
-              id: tool.id,
-              name: tool.name,
-              raw,
-            }),
+            Option.getOrElse(
+              Option.map(parsePartialInput(raw), (input) => input ?? {}),
+              () => ({}),
+            ),
           ),
     ),
+    Effect.map(
+      (input): ToolCall =>
+        LLMEvent.toolCall({
+          id: tool.id,
+          name: tool.name,
+          input,
+          providerExecuted: tool.providerExecuted ? true : undefined,
+          providerMetadata: tool.providerMetadata,
+        }),
+    ),
   )
 }
 
-const finishEvents = (tool: PendingTool, event: ToolCall | ToolInputError): ReadonlyArray<LLMEvent> =>
-  event.type === "tool-input-error"
-    ? [event]
-    : [LLMEvent.toolInputEnd({ id: tool.id, name: tool.name, providerMetadata: tool.providerMetadata }), event]
+const finishEvents = (tool: PendingTool, event: ToolCall): ReadonlyArray<LLMEvent> => [
+  LLMEvent.toolInputEnd({ id: tool.id, name: tool.name, providerMetadata: tool.providerMetadata }),
+  event,
+]
 
 /** Store the updated tool and produce the optional public delta event. */
 const appendTool = <K extends StreamKey>(
@@ -176,7 +179,7 @@ export const appendExisting = <K extends StreamKey>(
 
 /**
  * Finalize one pending tool call: parse the accumulated raw JSON, remove it
- * from state, and return either a call or a non-executable local input error.
+ * from state, and recover incomplete local arguments when needed.
  * Missing keys are a no-op because some providers emit stop events for
  * non-tool content blocks.
  */

packages/ai/src/protocols/xai-responses.ts

@@ -1,17 +1,52 @@
-import { Effect } from "effect"
+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 { ProviderShared } from "./shared.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 ?? {} },
@@ -27,15 +62,9 @@ const HOSTED_TOOLS = {
   },
 } as const satisfies ResponsesHostedTools.Definitions
 
+// Grok speaks the standard Responses reasoning dialect (`reasoning_summary_text.*`,
+// handled by the baseline); only its hosted tool vocabulary differs.
 const step = (state: OpenResponses.ParserState, event: OpenResponses.Event) => {
-  if (event.type === "response.reasoning_text.delta" || event.type === "response.reasoning_summary.delta")
-    return event.item_id
-      ? Effect.succeed(OpenResponses.onReasoningDelta(state, event, event.item_id))
-      : ProviderShared.eventError(ADAPTER, `${event.type} is missing item_id`)
-  if (event.type === "response.reasoning_text.done" || event.type === "response.reasoning_summary.done")
-    return event.item_id
-      ? Effect.succeed(OpenResponses.onReasoningDone(state, event))
-      : ProviderShared.eventError(ADAPTER, `${event.type} is missing item_id`)
   if (event.type === "response.output_item.done" && event.item && ResponsesHostedTools.isItem(event.item, HOSTED_TOOLS))
     return ResponsesHostedTools.onDone(state, event.item, HOSTED_TOOLS)
   return OpenResponses.step(state, event)
@@ -43,7 +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),

packages/ai/src/provider-error.ts

@@ -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" })

packages/ai/src/providers/amazon-bedrock-mantle.ts

@@ -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({

packages/ai/src/providers/cerebras.ts

@@ -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)

packages/ai/src/providers/deepinfra.ts

@@ -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)

packages/ai/src/providers/index.ts

@@ -3,8 +3,10 @@ export * as AnthropicCompatible from "./anthropic-compatible.js"
 export * as AmazonBedrock from "./amazon-bedrock.js"
 export * as AmazonBedrockMantle from "./amazon-bedrock-mantle.js"
 export * as Azure from "./azure.js"
+export * as Cerebras from "./cerebras.js"
 export * as Cloudflare from "./cloudflare.js"
 export { CloudflareAIGateway, CloudflareWorkersAI } from "./cloudflare.js"
+export * as DeepInfra from "./deepinfra.js"
 export * as Google from "./google.js"
 export * as GoogleVertex from "./google-vertex.js"
 export * as GoogleVertexChat from "./google-vertex-chat.js"
@@ -14,5 +16,6 @@ export * as OpenAI from "./openai.js"
 export * as OpenAICompatible from "./openai-compatible.js"
 export * as OpenAICompatibleResponses from "./openai-compatible-responses.js"
 export * as OpenRouter from "./openrouter.js"
+export * as TogetherAI from "./togetherai.js"
 export * as XAI from "./xai.js"
 export * as ZAI from "./zai.js"

packages/ai/src/providers/openrouter.ts

@@ -119,7 +119,6 @@ export const protocol = Protocol.make({
             ...body,
             messages,
             ...bodyOptions(request.providerOptions),
-            ...(request.promptCacheKey ? { prompt_cache_key: request.promptCacheKey } : {}),
           } as OpenRouterBody
         }),
       ),

packages/ai/src/providers/togetherai.ts

@@ -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)

packages/ai/src/providers/xai.ts

@@ -42,7 +42,7 @@ const responsesRoute = Route.make({
     name: "xAI Responses",
     rotateAfterMs: RESPONSES_WEBSOCKET_ROTATE_AFTER_MS,
   }),
-  defaults: { providerOptions: { store: false } },
+  defaults: { providerOptions: { store: false, include: ["reasoning.encrypted_content"] } },
 })
 
 const chatRoute = Route.make({

packages/ai/src/route/client.ts

@@ -7,6 +7,7 @@ import { HttpTransport } from "./transport/index.js"
 import type { HttpMiddleware, Transport, TransportRuntime, WebSocketChannelExecutor } from "./transport/index.js"
 import type { Protocol } from "./protocol.js"
 import { applyCachePolicy } from "../cache-policy.js"
+import { sanitizeSurrogates } from "../utils/sanitize.js"
 import * as ProviderShared from "../protocols/shared.js"
 import type { ProtocolID, ProviderOptions } from "../schema/index.js"
 import {
@@ -321,12 +322,30 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
                 Stream.mapEffect(decodeEvent(route)),
                 protocol.stream.terminal ? Stream.takeUntil(protocol.stream.terminal) : (stream) => stream,
               )
-              const stream = events.pipe(
-                Stream.mapAccumEffect(
-                  () => protocol.stream.initial(request),
-                  protocol.stream.step,
-                  protocol.stream.onHalt ? { onHalt: protocol.stream.onHalt } : undefined,
-                ),
+              const stream = Stream.suspend(() => {
+                let state = protocol.stream.initial(request)
+                const parsed = events.pipe(
+                  Stream.mapEffect((event) =>
+                    protocol.stream.step(state, event).pipe(
+                      Effect.map(([next, output]) => {
+                        state = next
+                        return output
+                      }),
+                    ),
+                  ),
+                  Stream.flatMap(Stream.fromIterable),
+                )
+                const onHalt = protocol.stream.onHalt
+                return onHalt
+                  ? parsed.pipe(
+                      Stream.concat(
+                        Stream.suspend(() =>
+                          Stream.unwrap(onHalt(state).pipe(Effect.map(Stream.fromIterable))),
+                        ),
+                      ),
+                    )
+                  : parsed
+              }).pipe(
                 Stream.catchCause((cause) => Stream.fail(streamError(route, `Failed to read ${route} stream`, cause))),
                 requireTerminalEvent(route),
               )
@@ -382,7 +401,8 @@ export function make<Body, Prepared, Frame, Event, State>(
 }
 
 const compile = Effect.fn("LLM.compile")(function* (request: LLMRequest, options?: StreamOptions) {
-  const resolved = applyCachePolicy(resolveRequestOptions(request))
+  const original = applyCachePolicy(resolveRequestOptions(request))
+  const resolved = LLMRequest.update(original, sanitizeSurrogates({ ...LLMRequest.input(original), model: undefined }))
   const route = resolved.model.route
 
   const body = yield* route.body

packages/ai/src/route/protocol.ts

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

packages/ai/src/schema/events.ts

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

packages/ai/src/schema/messages.ts

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

packages/ai/src/schema/options.ts

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

packages/ai/src/utils/sanitize.ts

@@ -0,0 +1,12 @@
+import { isRecord } from "./record.js"
+
+export const sanitizeSurrogates = <T>(value: T): T => {
+  if (typeof value === "string") return value.toWellFormed() as T
+  if (Array.isArray(value)) return value.map(sanitizeSurrogates) as T
+  if (value instanceof Uint8Array || value instanceof Error) return value
+  if (isRecord(value))
+    return Object.fromEntries(
+      Object.entries(value).map(([key, entry]) => [key.toWellFormed(), sanitizeSurrogates(entry)]),
+    ) as T
+  return value
+}

packages/ai/test/cache-policy.test.ts

@@ -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(

packages/ai/test/compile.test.ts

@@ -1,7 +1,7 @@
 import { describe, expect, test } from "bun:test"
 import { Effect, Ref, Schema } from "effect"
 import { HttpClientRequest, HttpClientResponse } from "effect/unstable/http"
-import { LLM, mergeProviderOptions } from "../src/index.js"
+import { LLM, Message, ToolCallPart, mergeProviderOptions } from "../src/index.js"
 import { AnthropicMessages, OpenAIChat } from "../src/protocols.js"
 import { Auth, LLMClient } from "../src/route.js"
 import { compileRequest } from "../src/route/client.js"
@@ -66,7 +66,7 @@ describe("request option precedence", () => {
       expect(prepared.body).toMatchObject({
         model: "gpt-4o-mini",
         stream: true,
-        max_tokens: 30,
+        max_completion_tokens: 30,
         temperature: 0.5,
         top_p: 0.9,
         frequency_penalty: 0.25,
@@ -247,6 +247,73 @@ describe("request option precedence", () => {
     }),
   )
 
+  it.effect("sanitizes outbound JSON without an HTTP overlay", () =>
+    LLMClient.generate(
+      LLM.request({
+        model: OpenAIChat.route
+          .with({ endpoint: { baseURL: "https://api.openai.test/v1/" }, auth: Auth.bearer("test") })
+          .model({ id: "gpt-4o-mini" }),
+        prompt: "hello \uD800 \u{1F600}",
+      }),
+    ).pipe(
+      Effect.provide(
+        dynamicResponse((input) =>
+          Effect.gen(function* () {
+            expect(decodeJson(input.text)).toMatchObject({
+              messages: [{ role: "user", content: "hello \uFFFD \u{1F600}" }],
+            })
+            return input.respond(sseEvents(deltaChunk({}, "stop")), {
+              headers: { "content-type": "text/event-stream" },
+            })
+          }),
+        ),
+      ),
+    ),
+  )
+
+  it.effect("sanitizes unpaired surrogates throughout outbound JSON", () =>
+    LLMClient.generate(
+      LLM.request({
+        model: OpenAIChat.route
+          .with({ endpoint: { baseURL: "https://api.openai.test/v1/" }, auth: Auth.bearer("test") })
+          .model({ id: "gpt-4o-mini" }),
+        system: "system \uD800 \u{1F600}",
+        messages: [
+          Message.user("user \uDC00"),
+          Message.assistant([
+            Message.text("assistant \uD800"),
+            ToolCallPart.make({ id: "call_1", name: "lookup", input: { query: "input \uDC00" } }),
+          ]),
+          Message.tool({ id: "call_1", name: "lookup", result: { output: "result \uD800" } }),
+        ],
+        http: { body: { metadata: { "key\uD800": ["overlay \uDC00", "valid \u{1F600}"] } } },
+      }),
+    ).pipe(
+      Effect.provide(
+        dynamicResponse((input) =>
+          Effect.gen(function* () {
+            expect(decodeJson(input.text)).toMatchObject({
+              messages: [
+                { role: "system", content: "system \uFFFD \u{1F600}" },
+                { role: "user", content: "user \uFFFD" },
+                {
+                  role: "assistant",
+                  content: "assistant \uFFFD",
+                  tool_calls: [{ function: { arguments: '{"query":"input \uFFFD"}' } }],
+                },
+                { role: "tool", content: '{"output":"result \uFFFD"}' },
+              ],
+              metadata: { "key\uFFFD": ["overlay \uFFFD", "valid \u{1F600}"] },
+            })
+            return input.respond(sseEvents(deltaChunk({}, "stop")), {
+              headers: { "content-type": "text/event-stream" },
+            })
+          }),
+        ),
+      ),
+    ),
+  )
+
   it.effect("applies raw body overlays after protocol lowering", () =>
     LLMClient.generate(
       LLM.request({

packages/ai/test/executor.test.ts

@@ -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

packages/ai/test/fixtures/recordings/cerebras-chat/cerebras-gpt-oss-120b-text.json

@@ -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"
+      }
+    }
+  ]
+}

packages/ai/test/fixtures/recordings/cerebras-chat/cerebras-gpt-oss-120b-tool-call.json

@@ -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"
+      }
+    }
+  ]
+}

packages/ai/test/fixtures/recordings/cloudflare-workers-ai/cloudflare-workers-ai-gpt-oss-20b-tools-tool-call.json

@@ -7,7 +7,13 @@
     "route": "cloudflare-workers-ai",
     "transport": "http",
     "model": "@cf/openai/gpt-oss-20b",
-    "tags": ["prefix:cloudflare-workers-ai", "provider:cloudflare-workers-ai", "tool", "tool-call", "golden"]
+    "tags": [
+      "prefix:cloudflare-workers-ai",
+      "provider:cloudflare-workers-ai",
+      "tool",
+      "tool-call",
+      "golden"
+    ]
   },
   "interactions": [
     {
@@ -18,7 +24,7 @@
         "headers": {
           "content-type": "application/json"
         },
-        "body": "{\"model\":\"@cf/openai/gpt-oss-20b\",\"messages\":[{\"role\":\"system\",\"content\":\"Call tools exactly as requested.\"},{\"role\":\"user\",\"content\":\"Call get_weather with city exactly Paris.\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false}}}],\"tool_choice\":{\"type\":\"function\",\"function\":{\"name\":\"get_weather\"}},\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":120,\"temperature\":0}"
+        "body": "{\"model\": \"@cf/openai/gpt-oss-20b\", \"messages\": [{\"role\": \"system\", \"content\": \"Call tools exactly as requested.\"}, {\"role\": \"user\", \"content\": \"Call get_weather with city exactly Paris.\"}], \"tools\": [{\"type\": \"function\", \"function\": {\"name\": \"get_weather\", \"description\": \"Get current weather for a city.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"city\": {\"type\": \"string\"}}, \"required\": [\"city\"], \"additionalProperties\": false}, \"strict\": false}}], \"tool_choice\": {\"type\": \"function\", \"function\": {\"name\": \"get_weather\"}}, \"stream\": true, \"stream_options\": {\"include_usage\": true}, \"max_tokens\": 120, \"temperature\": 0}"
       },
       "response": {
         "status": 200,
@@ -29,4 +35,4 @@
       }
     }
   ]
-}
+}

packages/ai/test/fixtures/recordings/deepinfra-chat/deepinfra-llama-3-3-70b-text.json

@@ -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"
+      }
+    }
+  ]
+}

packages/ai/test/fixtures/recordings/deepinfra-chat/deepinfra-llama-3-3-70b-tool-call.json

@@ -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"
+      }
+    }
+  ]
+}

packages/ai/test/fixtures/recordings/deepinfra-chat/deepinfra-llama-3-3-70b-tool-loop.json

@@ -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"
+      }
+    }
+  ]
+}

packages/ai/test/fixtures/recordings/openai-responses-websocket/continues-a-tool-call-over-one-socket.json

@@ -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",

packages/ai/test/fixtures/recordings/openai-responses-websocket/reconstructs-full-context-after-reconnect.json

@@ -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",

packages/ai/test/fixtures/recordings/openai-responses-websocket/recovers-from-explicit-continuation-rejection.json

@@ -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",

packages/ai/test/fixtures/recordings/openai-responses/openai-responses-gpt-5-5-reasoning-continuation.json

@@ -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,

packages/ai/test/fixtures/recordings/openai-responses/openai-responses-gpt-5-5-reasoning.json

@@ -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,

packages/ai/test/fixtures/recordings/openai-responses/openai-responses-gpt-5-5-tool-loop.json

@@ -25,7 +25,7 @@
         "headers": {
           "content-type": "application/json"
         },
-        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"system\",\"content\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"What is the weather in Paris?\"}]}],\"tools\":[{\"type\":\"function\",\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":80,\"stream\":true}"
+        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"What is the weather in Paris?\"}]}],\"tools\":[{\"type\":\"function\",\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":80,\"stream\":true,\"instructions\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"}"
       },
       "response": {
         "status": 200,
@@ -43,7 +43,7 @@
         "headers": {
           "content-type": "application/json"
         },
-        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"system\",\"content\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"What is the weather in Paris?\"}]},{\"type\":\"reasoning\",\"id\":\"rs_0ad67c31d9ddad95016a869efbd02487d1a51eed850e6f87f5\",\"summary\":[],\"encrypted_content\":\"gAAAAABqhp79U9UPKmTdmo9tmdil0C2KXpkFqUc4MNkYHT53Lzos9omncFPg76QzUmmSOdBcajisWBEo-xiTCvhp135uACUq8TJcdw4DluieYq6dWszijy28PFFfeO-6MmHwi7zeln1Z202zErJUEyuf1bML68VAeam5PqlMLG-a4-pmnWiH2ExWKibTUX37QoMQoArrkccJOCmxwDflV_kWDPMFxQVDfeMg9fd1gVv2u-x1Mjk0b9mJDOq0Fe5Gh-IkpWzfXgZTdptFmCM75cksvs61Rqsx6P33czal-LSixEF0WMizCvbMQmqKGs7MKGMeoa6j6vWOnB3ICIbv6FShnSaTpZWJFwejvOurkfuxa-2q6xVDZsBoQCgMWPHsqLxwAo1JKdfBk0pMvSuvpw2BRxykUZ1ULCYJ-BypST65292-EuSZFIuXPMPir-_raSCTsgsZNMscDG6ll3qksDTDS6_o5NutD7Ra-WZzaUe_HQlSLKLACTc4qv2EK1QoC4aYv4goxkTSx17WhS2D86lILgkUd-TIHjJ6iR3uxSNx7YeBNxiJgddIAEjAaSrdF-WDouSNT9k3efd5HhT3zahIOMKgb3XIQzFOYWfWgea5-SbaIdKwne9hU0QyhcBQs6yoifSg-fJZtahbPb-GCDYnOLlH-bV94vldoccb-2P1JdB3jaLj5tJUecfr2H4qiu8MgkPj0TkwYNbJynYmJo9H5Lm-XJ9gfzIXzJh0arKwsS4gwDLf4J3LOEF3WEW3mknOjjb9PrLmHRYXQQh9tTiX9ILPZpbufkyCurTUMQgWiSCitXBC6FoLXRHilSmb-6_avBnlUMziMfey-FkKvRfiPox6BaJrnOq6SGlOv11y7EKvzrn29la7HKPygYenDAkyq2mq0Zk2nLWNmJcv9sQTBrkBdFMmJYPi2J2im8XD5MmAjEL8R4FCBHoPIIZ6pENQykvH8PhpWKuzF5gJlY3Vwz4iJ0Qb9TrNI0hzBoI1U0LeB5FJ2HgjZQwCFF5x3ubh72xrUsFpuyYyYPa8GDT0Bo-LW_IlJ_mN4EwI5Nk9n-8Bt015yxsfpa5YaDeCeQFcdj8SD0UAd7QWtGACpzKcIj1-vJJU7OiwscV_v1dLvoiEe1ehI9jcvPn28TgHlo_dippe0iMN4FAm1Bf8vtWVMFDvfV1rPv1pAFFnSa9XqFszD5Exo_xzcQEoKXvQv3OnUtoiM4Db4uadClazLjoep2TQgHcJBVbTbLySTVPmok4ROFQZsU_mq4vu2M__d8HOjADfIIYz5VLVQKNpo0Hv_QkT2bn56Q==\"},{\"type\":\"function_call\",\"id\":\"fc_0ad67c31d9ddad95016a869efd126887d1b7e2f17f155cb5dc\",\"call_id\":\"call_qrzOfKDfzaq8fqbSNNVHlNsV\",\"name\":\"get_weather\",\"arguments\":\"{\\\"city\\\":\\\"Paris\\\"}\"},{\"type\":\"function_call_output\",\"call_id\":\"call_qrzOfKDfzaq8fqbSNNVHlNsV\",\"output\":\"{\\\"temperature\\\":22,\\\"condition\\\":\\\"sunny\\\"}\"}],\"tools\":[{\"type\":\"function\",\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":80,\"stream\":true}"
+        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"What is the weather in Paris?\"}]},{\"type\":\"reasoning\",\"id\":\"rs_0ad67c31d9ddad95016a869efbd02487d1a51eed850e6f87f5\",\"summary\":[],\"encrypted_content\":\"gAAAAABqhp79U9UPKmTdmo9tmdil0C2KXpkFqUc4MNkYHT53Lzos9omncFPg76QzUmmSOdBcajisWBEo-xiTCvhp135uACUq8TJcdw4DluieYq6dWszijy28PFFfeO-6MmHwi7zeln1Z202zErJUEyuf1bML68VAeam5PqlMLG-a4-pmnWiH2ExWKibTUX37QoMQoArrkccJOCmxwDflV_kWDPMFxQVDfeMg9fd1gVv2u-x1Mjk0b9mJDOq0Fe5Gh-IkpWzfXgZTdptFmCM75cksvs61Rqsx6P33czal-LSixEF0WMizCvbMQmqKGs7MKGMeoa6j6vWOnB3ICIbv6FShnSaTpZWJFwejvOurkfuxa-2q6xVDZsBoQCgMWPHsqLxwAo1JKdfBk0pMvSuvpw2BRxykUZ1ULCYJ-BypST65292-EuSZFIuXPMPir-_raSCTsgsZNMscDG6ll3qksDTDS6_o5NutD7Ra-WZzaUe_HQlSLKLACTc4qv2EK1QoC4aYv4goxkTSx17WhS2D86lILgkUd-TIHjJ6iR3uxSNx7YeBNxiJgddIAEjAaSrdF-WDouSNT9k3efd5HhT3zahIOMKgb3XIQzFOYWfWgea5-SbaIdKwne9hU0QyhcBQs6yoifSg-fJZtahbPb-GCDYnOLlH-bV94vldoccb-2P1JdB3jaLj5tJUecfr2H4qiu8MgkPj0TkwYNbJynYmJo9H5Lm-XJ9gfzIXzJh0arKwsS4gwDLf4J3LOEF3WEW3mknOjjb9PrLmHRYXQQh9tTiX9ILPZpbufkyCurTUMQgWiSCitXBC6FoLXRHilSmb-6_avBnlUMziMfey-FkKvRfiPox6BaJrnOq6SGlOv11y7EKvzrn29la7HKPygYenDAkyq2mq0Zk2nLWNmJcv9sQTBrkBdFMmJYPi2J2im8XD5MmAjEL8R4FCBHoPIIZ6pENQykvH8PhpWKuzF5gJlY3Vwz4iJ0Qb9TrNI0hzBoI1U0LeB5FJ2HgjZQwCFF5x3ubh72xrUsFpuyYyYPa8GDT0Bo-LW_IlJ_mN4EwI5Nk9n-8Bt015yxsfpa5YaDeCeQFcdj8SD0UAd7QWtGACpzKcIj1-vJJU7OiwscV_v1dLvoiEe1ehI9jcvPn28TgHlo_dippe0iMN4FAm1Bf8vtWVMFDvfV1rPv1pAFFnSa9XqFszD5Exo_xzcQEoKXvQv3OnUtoiM4Db4uadClazLjoep2TQgHcJBVbTbLySTVPmok4ROFQZsU_mq4vu2M__d8HOjADfIIYz5VLVQKNpo0Hv_QkT2bn56Q==\"},{\"type\":\"function_call\",\"id\":\"fc_0ad67c31d9ddad95016a869efd126887d1b7e2f17f155cb5dc\",\"call_id\":\"call_qrzOfKDfzaq8fqbSNNVHlNsV\",\"name\":\"get_weather\",\"arguments\":\"{\\\"city\\\":\\\"Paris\\\"}\"},{\"type\":\"function_call_output\",\"call_id\":\"call_qrzOfKDfzaq8fqbSNNVHlNsV\",\"output\":\"{\\\"temperature\\\":22,\\\"condition\\\":\\\"sunny\\\"}\"}],\"tools\":[{\"type\":\"function\",\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":80,\"stream\":true,\"instructions\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"}"
       },
       "response": {
         "status": 200,

packages/ai/test/partial-json.test.ts

@@ -0,0 +1,58 @@
+import { describe, expect, test } from "bun:test"
+import { Allow, MalformedJSON, PartialJSON, parse } from "../src/protocols/utils/partial-json.js"
+
+describe("partial JSON", () => {
+  test("parses complete JSON", () => {
+    expect(parse('{"key":"value","items":[1,true,null]}')).toEqual({
+      key: "value",
+      items: [1, true, null],
+    })
+
+    const object = parse('{"__proto__":{"safe":true}}') as Record<string, unknown>
+    expect(Object.hasOwn(object, "__proto__")).toBe(true)
+  })
+
+  test("parses partial strings", () => {
+    expect(parse('"hello')).toBe("hello")
+    expect(parse('"hello \\u12')).toBe("hello ")
+    expect(() => parse('"hello', ~Allow.STR)).toThrow(PartialJSON)
+  })
+
+  test("repairs invalid escapes and raw control characters", () => {
+    expect(parse('{"path":"A\\H","text":"first\tsecond"}')).toEqual({
+      path: "A\\H",
+      text: "first\tsecond",
+    })
+  })
+
+  test("preserves prototype keys in partial objects", () => {
+    const object = parse('{"__proto__":{"safe":true}') as Record<string, unknown>
+
+    expect(Object.hasOwn(object, "__proto__")).toBe(true)
+    expect(Object.getPrototypeOf(object)).toBe(Object.prototype)
+    expect(object.__proto__).toEqual({ safe: true })
+  })
+
+  test("controls partial collection values independently", () => {
+    expect(parse('["', Allow.ARR)).toEqual([])
+    expect(parse('["', Allow.ARR | Allow.STR)).toEqual([""])
+    expect(parse('{"key":"', Allow.OBJ)).toEqual({})
+    expect(parse('{"key":"', Allow.OBJ | Allow.STR)).toEqual({ key: "" })
+  })
+
+  test("parses partial literals and numbers", () => {
+    expect(parse("nu", Allow.NULL)).toBeNull()
+    expect(parse("tr", Allow.BOOL)).toBe(true)
+    expect(parse("fa", Allow.BOOL)).toBe(false)
+    expect(parse("1e", Allow.NUM)).toBe(1)
+  })
+
+  test("distinguishes disallowed partial values from malformed values", () => {
+    expect(() => parse("[", Allow.STR)).toThrow(PartialJSON)
+    expect(() => parse("n", ~Allow.NULL)).toThrow(MalformedJSON)
+  })
+
+  test("rejects empty input", () => {
+    expect(() => parse("  ")).toThrow("is empty")
+  })
+})

packages/ai/test/provider-error.test.ts

@@ -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", () => {

packages/ai/test/provider-package.test.ts

@@ -26,6 +26,9 @@ describe("provider package entrypoints", () => {
       import("@opencode-ai/ai/providers/amazon-bedrock/mantle"),
       import("@opencode-ai/ai/providers/amazon-bedrock/mantle/chat"),
       import("@opencode-ai/ai/providers/amazon-bedrock/mantle/responses"),
+      import("@opencode-ai/ai/providers/togetherai"),
+      import("@opencode-ai/ai/providers/cerebras"),
+      import("@opencode-ai/ai/providers/deepinfra"),
     ])
 
     for (const module of modules) expect(module.model).toBeFunction()
@@ -35,6 +38,24 @@ describe("provider package entrypoints", () => {
     expect(modules[19].model).toBe(modules[20].model)
   })
 
+  test("maps DeepInfra package settings onto its native executable model", async () => {
+    const DeepInfra = await import("@opencode-ai/ai/providers/deepinfra")
+    const settings = {
+      apiKey: "fixture",
+      baseURL: "https://provider.example.test/v1/",
+      headers: { "x-application": "opencode" },
+      body: { service_tier: "priority" },
+      providerOptions: { reasoningEffort: "high" as const },
+    }
+    const deepinfra = DeepInfra.model("google/gemma-3-27b-it", settings)
+
+    expect(deepinfra.route.id).toBe("deepinfra-chat")
+    expect(deepinfra.route.endpoint.baseURL).toBe("https://provider.example.test/v1/openai")
+    expect(deepinfra.route.defaults.providerOptions).toEqual(settings.providerOptions)
+    expect(deepinfra.route.defaults.headers).toEqual(settings.headers)
+    expect(deepinfra.route.defaults.http?.body).toEqual(settings.body)
+  })
+
   test("maps OpenRouter and xAI package settings onto executable models", async () => {
     const OpenRouter = await import("@opencode-ai/ai/providers/openrouter")
     const XAI = await import("@opencode-ai/ai/providers/xai")
@@ -95,7 +116,11 @@ describe("provider package entrypoints", () => {
     })
     expect(selected.route.defaults.headers).toEqual({ "x-application": "opencode" })
     expect(selected.route.defaults.http?.body).toEqual({ service_tier: "priority" })
-    expect(selected.route.defaults.providerOptions).toEqual({ reasoningEffort: "low", store: true })
+    expect(selected.route.defaults.providerOptions).toEqual({
+      reasoningEffort: "low",
+      store: true,
+      include: ["reasoning.encrypted_content"],
+    })
   })
 
   test("maps Anthropic-compatible settings onto the executable model", async () => {
@@ -285,7 +310,10 @@ describe("provider package entrypoints", () => {
       baseURL: "https://aiplatform.googleapis.com/v1/projects/vertex-project/locations/global/endpoints/openapi",
       path: "/responses",
     })
-    expect(responses.route.defaults.providerOptions).toEqual({ store: false })
+    expect(responses.route.defaults.providerOptions).toEqual({
+      store: false,
+      include: ["reasoning.encrypted_content"],
+    })
   })
 
   test("rejects conflicting Vertex auth settings at runtime", async () => {

packages/ai/test/provider/anthropic-messages.test.ts

@@ -5,6 +5,7 @@ import { CacheHint, LLM, AIError, LLMRequest, Message, ToolCallPart, ToolDefinit
 import { Auth, LLMClient } from "../../src/route.js"
 import { compileRequest } from "../../src/route/client.js"
 import * as AnthropicMessages from "../../src/protocols/anthropic-messages.js"
+import { GoogleVertexMessages } from "../../src/providers.js"
 import { continuationRequest, nativeAnthropicMessagesContinuation } from "../continuation-scenarios.js"
 import { it } from "../lib/effect.js"
 import { dynamicResponse, fixedResponse } from "../lib/http.js"
@@ -18,6 +19,21 @@ const opus48 = AnthropicMessages.route
   .with({ endpoint: { baseURL: "https://api.anthropic.test/v1/" }, auth: Auth.header("x-api-key", "test") })
   .model({ id: "claude-opus-4-8" })
 
+const compileUnsignedReasoning = (model: LLMRequest["model"]) =>
+  compileRequest(
+    LLM.request({
+      model,
+      messages: [Message.assistant([{ type: "reasoning", text: "unsigned reasoning" }])],
+      cache: "none",
+    }),
+  )
+
+const vertexOpus48 = GoogleVertexMessages.configure({
+  accessToken: "test",
+  location: "global",
+  project: "test",
+}).model("claude-opus-4-8")
+
 const request = LLM.request({
   id: "req_1",
   model,
@@ -277,6 +293,149 @@ describe("Anthropic Messages route", () => {
     }),
   )
 
+  it.effect("keeps a terminal Vertex system update in the tool-result turn", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: vertexOpus48,
+          messages: [
+            Message.assistant([ToolCallPart.make({ id: "call_1", name: "lookup", input: {} })]),
+            Message.tool({ id: "call_1", name: "lookup", result: "Done." }),
+            Message.system("Operator update."),
+          ],
+          cache: "none",
+        }),
+      )
+
+      expect(prepared.body.messages).toEqual([
+        {
+          role: "assistant",
+          content: [{ type: "tool_use", id: "call_1", name: "lookup", input: {} }],
+        },
+        {
+          role: "user",
+          content: [
+            {
+              type: "tool_result",
+              tool_use_id: "call_1",
+              content: '"Done."',
+              is_error: undefined,
+              cache_control: undefined,
+            },
+            {
+              type: "text",
+              text: "<system-update>\nOperator update.\n</system-update>",
+              cache_control: undefined,
+            },
+          ],
+        },
+      ])
+    }),
+  )
+
+  it.effect("preserves folded tool-result system updates across multi-turn Vertex history", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: vertexOpus48,
+          messages: [
+            Message.assistant([ToolCallPart.make({ id: "call_1", name: "lookup", input: {} })]),
+            Message.tool({ id: "call_1", name: "lookup", result: "Done." }),
+            Message.system("Operator update."),
+            Message.assistant("Acknowledged."),
+            Message.user("Next step."),
+          ],
+          cache: "none",
+        }),
+      )
+
+      expect(prepared.body.messages).toEqual([
+        {
+          role: "assistant",
+          content: [{ type: "tool_use", id: "call_1", name: "lookup", input: {} }],
+        },
+        {
+          role: "user",
+          content: [
+            {
+              type: "tool_result",
+              tool_use_id: "call_1",
+              content: '"Done."',
+              is_error: undefined,
+              cache_control: undefined,
+            },
+            {
+              type: "text",
+              text: "<system-update>\nOperator update.\n</system-update>",
+              cache_control: undefined,
+            },
+          ],
+        },
+        { role: "assistant", content: [{ type: "text", text: "Acknowledged." }] },
+        { role: "user", content: [{ type: "text", text: "Next step." }] },
+      ])
+    }),
+  )
+
+  it.effect("keeps a terminal direct Anthropic system update native", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: opus48,
+          messages: [
+            Message.assistant([ToolCallPart.make({ id: "call_1", name: "lookup", input: {} })]),
+            Message.tool({ id: "call_1", name: "lookup", result: "Done." }),
+            Message.system("Operator update."),
+          ],
+          cache: "none",
+        }),
+      )
+
+      expect(prepared.body.messages).toEqual([
+        {
+          role: "assistant",
+          content: [{ type: "tool_use", id: "call_1", name: "lookup", input: {} }],
+        },
+        {
+          role: "user",
+          content: [
+            {
+              type: "tool_result",
+              tool_use_id: "call_1",
+              content: '"Done."',
+              is_error: undefined,
+              cache_control: undefined,
+            },
+          ],
+        },
+        {
+          role: "system",
+          content: [{ type: "text", text: "Operator update.", cache_control: undefined }],
+        },
+      ])
+    }),
+  )
+
+  it.effect("keeps an ordinary terminal Vertex system update native", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: vertexOpus48,
+          messages: [Message.user("Before."), Message.system("Operator update.")],
+          cache: "none",
+        }),
+      )
+
+      expect(prepared.body.messages).toEqual([
+        { role: "user", content: [{ type: "text", text: "Before." }] },
+        {
+          role: "system",
+          content: [{ type: "text", text: "Operator update.", cache_control: undefined }],
+        },
+      ])
+    }),
+  )
+
   it.effect("rejects a system update between a local tool call and its result", () =>
     Effect.gen(function* () {
       const error = yield* compileRequest(
@@ -564,6 +723,65 @@ describe("Anthropic Messages route", () => {
     }),
   )
 
+  it.effect("demotes unsigned reasoning when signatures are required", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileUnsignedReasoning(model)
+
+      expect(prepared.body.messages).toEqual([
+        { role: "assistant", content: [{ type: "text", text: "unsigned reasoning" }] },
+      ])
+    }),
+  )
+
+  it.effect("infers empty-signature compatibility across Kimi providers", () =>
+    Effect.gen(function* () {
+      const coding = AnthropicMessages.route.with({
+        provider: "kimi-for-coding",
+        endpoint: { baseURL: "https://compatible.test/v1/" },
+        auth: Auth.header("x-api-key", "test"),
+      })
+      const moonshot = AnthropicMessages.route
+        .with({
+          provider: "moonshotai",
+          endpoint: { baseURL: "https://api.moonshot.ai/anthropic" },
+          auth: Auth.bearer("test"),
+        })
+        .model({ id: "kimi-k2.6" })
+      const codingPrepared = yield* compileUnsignedReasoning(coding.model({ id: "k3" }))
+      const moonshotPrepared = yield* compileUnsignedReasoning(moonshot)
+
+      expect(codingPrepared.body.messages).toEqual([
+        {
+          role: "assistant",
+          content: [{ type: "thinking", thinking: "unsigned reasoning", signature: "" }],
+        },
+      ])
+      expect(moonshotPrepared.body.messages).toEqual([
+        {
+          role: "assistant",
+          content: [{ type: "thinking", thinking: "unsigned reasoning", signature: "" }],
+        },
+      ])
+    }),
+  )
+
+  it.effect("lets an explicit signature requirement override inference", () =>
+    Effect.gen(function* () {
+      const compatible = AnthropicMessages.route
+        .with({
+          provider: "kimi-for-coding",
+          endpoint: { baseURL: "https://api.kimi.com/coding/v1/" },
+          auth: Auth.header("x-api-key", "test"),
+        })
+        .model({ id: "k3", compatibility: { requireSignature: true } })
+      const prepared = yield* compileUnsignedReasoning(compatible)
+
+      expect(prepared.body.messages).toEqual([
+        { role: "assistant", content: [{ type: "text", text: "unsigned reasoning" }] },
+      ])
+    }),
+  )
+
   it.effect("round-trips redacted thinking as redacted_thinking blocks", () =>
     Effect.gen(function* () {
       const prepared = yield* compileRequest(
@@ -702,6 +920,108 @@ describe("Anthropic Messages route", () => {
     }),
   )
 
+  it.effect("ignores unknown content block and delta variants", () =>
+    Effect.gen(function* () {
+      const response = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              { type: "message_start", message: { usage: { input_tokens: 5 } } },
+              { type: "future_event", content_block: 42, delta: 42 },
+              { type: "content_block_start", index: 0, content_block: { type: "future_block", text: 42 } },
+              { type: "content_block_delta", index: 0, delta: { text: "ignored" } },
+              { type: "content_block_delta", index: 0, delta: { type: "future_delta", text: 42 } },
+              { type: "content_block_delta", index: 0, delta: { type: "text_delta", text: "hidden" } },
+              { type: "content_block_delta", index: 0, delta: { type: "thinking_delta", thinking: "hidden" } },
+              { type: "content_block_delta", index: 0, delta: { type: "signature_delta", signature: "hidden" } },
+              { type: "content_block_stop", index: 0 },
+              { type: "content_block_start", index: 1, content_block: { type: "text", text: "" } },
+              { type: "content_block_delta", index: 1, delta: { type: "text_delta", text: "Hello" } },
+              { type: "content_block_stop", index: 1 },
+              { type: "message_delta", delta: { stop_reason: "end_turn" }, usage: { output_tokens: 1 } },
+              { type: "message_stop" },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.message.content).toEqual([{ type: "text", text: "Hello" }])
+      expect(response.finishReason).toEqual({ normalized: "stop", raw: "end_turn" })
+    }),
+  )
+
+  it.effect("rejects malformed recognized content block variants", () =>
+    Effect.gen(function* () {
+      const error = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              { type: "message_start", message: { usage: { input_tokens: 5 } } },
+              { type: "content_block_start", index: 0, content_block: { type: "text", text: 42 } },
+            ),
+          ),
+        ),
+        Effect.flip,
+      )
+
+      expect(error.reason).toMatchObject({
+        _tag: "InvalidProviderOutput",
+        message: "Invalid anthropic/anthropic-messages stream event",
+      })
+    }),
+  )
+
+  it.effect("rejects malformed recognized content delta variants", () =>
+    Effect.gen(function* () {
+      const error = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              { type: "message_start", message: { usage: { input_tokens: 5 } } },
+              { type: "content_block_start", index: 0, content_block: { type: "text", text: "" } },
+              { type: "content_block_delta", index: 0, delta: { type: "text_delta", text: 42 } },
+            ),
+          ),
+        ),
+        Effect.flip,
+      )
+
+      expect(error.reason).toMatchObject({
+        _tag: "InvalidProviderOutput",
+        message: "Invalid anthropic/anthropic-messages stream event",
+      })
+    }),
+  )
+
+  it.effect("rejects malformed payloads on unrelated stream events", () =>
+    Effect.gen(function* () {
+      const events = [
+        { type: "message_start", message: { usage: { input_tokens: 1 } }, delta: 42 },
+        { type: "content_block_start", index: 0 },
+        { type: "content_block_delta", index: 0 },
+        { type: "content_block_stop", index: 0, content_block: { type: "text", text: 42 } },
+        { type: "message_delta" },
+        { type: "message_delta", delta: { stop_reason: 42 } },
+        { type: "message_stop", delta: { text: 42 } },
+        { type: "error", error: { type: "overloaded_error", message: "busy" }, content_block: 42 },
+      ]
+
+      yield* Effect.forEach(events, (event) =>
+        Effect.gen(function* () {
+          const error = yield* LLMClient.generate(request).pipe(
+            Effect.provide(fixedResponse(sseEvents(event))),
+            Effect.flip,
+          )
+
+          expect(error.reason).toMatchObject({
+            _tag: "InvalidProviderOutput",
+            message: "Invalid anthropic/anthropic-messages stream event",
+          })
+        }),
+      )
+    }),
+  )
+
   it.effect("rejects malformed recognized SSE events", () =>
     Effect.gen(function* () {
       const error = yield* LLMClient.generate(request).pipe(
@@ -1060,8 +1380,14 @@ describe("Anthropic Messages route", () => {
       expect(response.events).toEqual([
         { type: "step-start", index: 0 },
         { type: "tool-input-start", id: "call_1", name: "lookup" },
-        { type: "tool-input-delta", id: "call_1", name: "lookup", text: '{"query"' },
-        { type: "tool-input-delta", id: "call_1", name: "lookup", text: ':"weather"}' },
+        { type: "tool-input-delta", id: "call_1", name: "lookup", text: '{"query"', input: {} },
+        {
+          type: "tool-input-delta",
+          id: "call_1",
+          name: "lookup",
+          text: ':"weather"}',
+          input: { query: "weather" },
+        },
         { type: "tool-input-end", id: "call_1", name: "lookup", providerMetadata: undefined },
         {
           type: "tool-call",

packages/ai/test/provider/bedrock-converse.test.ts

@@ -475,8 +475,14 @@ describe("Bedrock Converse route", () => {
       ])
       const events = response.events.filter((event) => event.type === "tool-input-delta")
       expect(events).toEqual([
-        { type: "tool-input-delta", id: "tool_1", name: "lookup", text: '{"query"' },
-        { type: "tool-input-delta", id: "tool_1", name: "lookup", text: ':"weather"}' },
+        { type: "tool-input-delta", id: "tool_1", name: "lookup", text: '{"query"', input: {} },
+        {
+          type: "tool-input-delta",
+          id: "tool_1",
+          name: "lookup",
+          text: ':"weather"}',
+          input: { query: "weather" },
+        },
       ])
       expect(response.events.at(-1)).toMatchObject({
         type: "finish",
@@ -485,7 +491,7 @@ describe("Bedrock Converse route", () => {
     }),
   )
 
-  it.effect("emits malformed tool input as an unexecuted tool error", () =>
+  it.effect("recovers incomplete tool input at finalization", () =>
     Effect.gen(function* () {
       const body = eventStreamBody(
         ["messageStart", { role: "assistant" }],
@@ -502,10 +508,10 @@ describe("Bedrock Converse route", () => {
       )
       const response = yield* LLMClient.generate(baseRequest).pipe(Effect.provide(fixedBytes(body)))
 
-      expect(response.events.find((event) => event.type === "tool-input-error")).toMatchObject({
+      expect(response.events.find((event) => event.type === "tool-call")).toMatchObject({
         id: "tool_1",
         name: "lookup",
-        raw: '{"query":"partial',
+        input: { query: "partial" },
       })
       expect(response.finishReason).toEqual({ normalized: "tool-calls", raw: "end_turn" })
     }),
@@ -710,6 +716,32 @@ describe("Bedrock Converse route", () => {
     }),
   )
 
+  it.effect("ignores unknown normal stream events", () =>
+    Effect.gen(function* () {
+      const body = concat([
+        eventFrame("messageStart", { role: "assistant" }),
+        eventFrame("futureEvent", { message: "Ignore this" }),
+        eventFrame("messageStop", { stopReason: "end_turn" }),
+      ])
+      const response = yield* LLMClient.generate(baseRequest).pipe(Effect.provide(fixedBytes(body)))
+
+      expect(response.finishReason).toEqual({ normalized: "stop", raw: "end_turn" })
+    }),
+  )
+
+  it.effect("fails unknown stream exceptions after message stop", () =>
+    Effect.gen(function* () {
+      const body = concat([
+        eventFrame("messageStart", { role: "assistant" }),
+        eventFrame("messageStop", { stopReason: "end_turn" }),
+        exceptionFrame("futureException", { message: "A future provider failure" }),
+      ])
+      const error = yield* LLMClient.generate(baseRequest).pipe(Effect.provide(fixedBytes(body)), Effect.flip)
+
+      expect(error.reason).toMatchObject({ _tag: "UnknownProvider", message: "A future provider failure" })
+    }),
+  )
+
   it.effect("classifies throttlingException as a rate limit", () =>
     Effect.gen(function* () {
       const body = concat([

packages/ai/test/provider/bedrock-mantle.test.ts

@@ -1,11 +1,13 @@
 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"
 
 const credentials = {
@@ -18,6 +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" }),
@@ -71,7 +74,9 @@ describe("Amazon Bedrock Mantle provider", () => {
             Effect.gen(function* () {
               const request = yield* HttpClientRequest.toWeb(input.request)
               seen.push({ url: request.url, authorization: request.headers.get("authorization") ?? undefined })
-              return input.respond("", { headers: { "content-type": "text/event-stream" } })
+              return input.respond(sseEvents({ choices: [{ delta: {}, finish_reason: "stop" }] }), {
+                headers: { "content-type": "text/event-stream" },
+              })
             }),
           ),
         ),
@@ -80,6 +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({

packages/ai/test/provider/gemini.test.ts

@@ -906,6 +906,54 @@ describe("Gemini route", () => {
     }),
   )
 
+  it.effect("ignores unknown response parts", () =>
+    Effect.gen(function* () {
+      const response = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents({
+              candidates: [
+                {
+                  content: {
+                    role: "model",
+                    parts: [
+                      { text: "Hello " },
+                      { executableCode: { language: "PYTHON", code: "print('ignored')" } },
+                      { text: "world" },
+                    ],
+                  },
+                  finishReason: "STOP",
+                },
+              ],
+            }),
+          ),
+        ),
+      )
+
+      expect(response.text).toBe("Hello world")
+      expect(response.finishReason).toEqual({ normalized: "stop", raw: "STOP" })
+    }),
+  )
+
+  it.effect("rejects malformed recognized response parts", () =>
+    Effect.gen(function* () {
+      const error = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents({
+              candidates: [{ content: { role: "model", parts: [{ text: 42 }] } }],
+            }),
+          ),
+        ),
+        Effect.flip,
+      )
+
+      expect(error).toBeInstanceOf(AIError)
+      expect(error.reason).toMatchObject({ _tag: "InvalidProviderOutput" })
+      expect(error.message).toContain("Invalid google/gemini stream event")
+    }),
+  )
+
   it.effect("preserves thoughtSignature for reasoning and tool-call continuation", () =>
     Effect.gen(function* () {
       const body = sseEvents({

packages/ai/test/provider/golden.recorded.test.ts

@@ -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",

packages/ai/test/provider/google-vertex.test.ts

@@ -237,6 +237,7 @@ describe("Google Vertex providers", () => {
               })
               return input.respond(
                 sseEvents(
+                  { type: "response.output_item.added", item: { type: "message", id: "msg_1" } },
                   { type: "response.output_text.delta", item_id: "msg_1", delta: "Hello." },
                   { type: "response.completed", response: { id: "resp_1" } },
                 ),

packages/ai/test/provider/native-providers.test.ts

@@ -0,0 +1,184 @@
+import { describe, expect } from "bun:test"
+import { ConfigProvider, Effect } from "effect"
+import { HttpClientRequest } from "effect/unstable/http"
+import { LLM, Message, ToolDefinition } from "../../src/index.js"
+import { Cerebras, DeepInfra, TogetherAI } from "../../src/providers/index.js"
+import { compileRequest } from "../../src/route/client.js"
+import { it } from "../lib/effect.js"
+import { dynamicResponse } from "../lib/http.js"
+import { sseEvents } from "../lib/sse.js"
+
+describe("native OpenAI-compatible providers", () => {
+  it.effect("preserves native Together AI and Cerebras provider and route identities", () =>
+    Effect.gen(function* () {
+      const together = TogetherAI.configure({ apiKey: "fixture" }).model("meta-llama/Llama-3.3-70B")
+      const cerebras = Cerebras.configure({ apiKey: "fixture" }).model("qwen-3-235b-a22b")
+
+      expect(together).toMatchObject({
+        provider: "togetherai",
+        compatibility: { maxTokensField: "max_tokens", supportsStore: false, supportsStrictMode: false },
+        route: { id: "togetherai-chat", protocol: "openai-chat" },
+      })
+      expect(together.route.endpoint.baseURL).toBe("https://api.together.xyz/v1")
+      expect(cerebras).toMatchObject({
+        provider: "cerebras",
+        compatibility: { maxTokensField: "max_tokens", reasoningField: "reasoning", supportsStore: false },
+        route: { id: "cerebras-chat", protocol: "openai-chat" },
+      })
+      expect(cerebras.route.endpoint.baseURL).toBe("https://api.cerebras.ai/v1")
+    }),
+  )
+
+  it.effect("preserves native DeepInfra provider and route identity", () =>
+    Effect.gen(function* () {
+      const deepinfra = DeepInfra.configure({ apiKey: "fixture" }).model("google/gemma-3-27b-it")
+      expect(deepinfra).toMatchObject({
+        provider: "deepinfra",
+        compatibility: { maxTokensField: "max_tokens", reasoningField: "reasoning_content", supportsStore: false },
+        route: { id: "deepinfra-chat", protocol: "openai-chat" },
+      })
+      expect(deepinfra.route.endpoint.baseURL).toBe("https://api.deepinfra.com/v1/openai")
+    }),
+  )
+
+  it.effect("applies native provider request defaults even with a custom gateway URL", () =>
+    Effect.gen(function* () {
+      const together = yield* compileRequest(
+        LLM.request({
+          model: TogetherAI.configure({ apiKey: "fixture", baseURL: "https://gateway.example/v1" }).model("llama"),
+          prompt: "Use a tool.",
+          generation: { maxTokens: 32 },
+          tools: [
+            ToolDefinition.make({ name: "lookup", description: "Look up data", inputSchema: { type: "object" } }),
+          ],
+          providerOptions: { store: true },
+        }),
+      )
+
+      expect(together.body).toMatchObject({
+        max_tokens: 32,
+        stream_options: { include_usage: true },
+        tools: [{ function: { name: "lookup" } }],
+      })
+      expect(together.body).not.toHaveProperty("max_completion_tokens")
+      expect(together.body).not.toHaveProperty("store")
+      expect(together.body.tools?.[0]?.function).not.toHaveProperty("strict")
+
+      const cerebras = yield* compileRequest(
+        LLM.request({
+          model: Cerebras.configure({ apiKey: "fixture", baseURL: "https://gateway.example/v1" }).model("qwen"),
+          generation: { maxTokens: 48 },
+          messages: [
+            Message.user("Think first."),
+            Message.assistant([
+              { type: "reasoning", text: "A deliberate thought." },
+              { type: "text", text: "An answer." },
+            ]),
+            Message.user("Continue."),
+          ],
+          providerOptions: { store: true },
+        }),
+      )
+
+      expect(cerebras.body).toMatchObject({
+        max_tokens: 48,
+        messages: [
+          { role: "user", content: "Think first." },
+          { role: "assistant", content: "An answer.", reasoning: "A deliberate thought." },
+          { role: "user", content: "Continue." },
+        ],
+      })
+      expect(cerebras.body).not.toHaveProperty("max_completion_tokens")
+      expect(cerebras.body).not.toHaveProperty("store")
+      expect(cerebras.body.messages[1]).not.toHaveProperty("reasoning_content")
+    }),
+  )
+
+  it.effect("normalizes DeepInfra API roots without duplicating the OpenAI path", () =>
+    Effect.gen(function* () {
+      for (const baseURL of [
+        "https://gateway.example/v1",
+        "https://gateway.example/v1/",
+        "https://gateway.example/v1/openai",
+        "https://gateway.example/v1/openai/",
+      ]) {
+        expect(DeepInfra.configure({ apiKey: "fixture", baseURL }).model("gemma").route.endpoint.baseURL).toBe(
+          "https://gateway.example/v1/openai",
+        )
+      }
+    }),
+  )
+
+  it.effect("maps package settings onto native executable models", () =>
+    Effect.gen(function* () {
+      for (const native of [TogetherAI, Cerebras]) {
+        const selected = native.model("provider-model", {
+          apiKey: "fixture",
+          baseURL: "https://gateway.example/v1",
+          headers: { "x-application": "opencode" },
+          body: { service_tier: "priority" },
+          providerOptions: { reasoningEffort: "high" },
+        })
+
+        expect(selected.route.endpoint.baseURL).toBe("https://gateway.example/v1")
+        expect(selected.route.defaults.headers).toEqual({ "x-application": "opencode" })
+        expect(selected.route.defaults.http?.body).toEqual({ service_tier: "priority" })
+        expect(selected.route.defaults.providerOptions).toEqual({ reasoningEffort: "high" })
+      }
+    }),
+  )
+
+  it.effect("resolves provider environment credentials and preserves deprecated Together credentials", () =>
+    Effect.gen(function* () {
+      const scenarios = [
+        {
+          model: TogetherAI.configure().model("llama"),
+          env: { TOGETHER_API_KEY: "together-primary", TOGETHER_AI_API_KEY: "together-legacy" },
+          token: "together-primary",
+          url: "https://api.together.xyz/v1/chat/completions",
+        },
+        {
+          model: TogetherAI.configure().model("llama"),
+          env: { TOGETHER_AI_API_KEY: "together-legacy" },
+          token: "together-legacy",
+          url: "https://api.together.xyz/v1/chat/completions",
+        },
+        {
+          model: Cerebras.configure().model("qwen"),
+          env: { CEREBRAS_API_KEY: "cerebras-secret" },
+          token: "cerebras-secret",
+          url: "https://api.cerebras.ai/v1/chat/completions",
+        },
+        {
+          model: DeepInfra.configure().model("gemma"),
+          env: { DEEPINFRA_API_KEY: "deepinfra-secret" },
+          token: "deepinfra-secret",
+          url: "https://api.deepinfra.com/v1/openai/chat/completions",
+        },
+      ]
+
+      yield* Effect.forEach(scenarios, (scenario) =>
+        LLM.generate(LLM.request({ model: scenario.model, prompt: "Say hello." })).pipe(
+          Effect.provide(
+            dynamicResponse((input) =>
+              Effect.gen(function* () {
+                const request = yield* HttpClientRequest.toWeb(input.request).pipe(Effect.orDie)
+                expect(request.url).toBe(scenario.url)
+                expect(request.headers.get("authorization")).toBe(`Bearer ${scenario.token}`)
+                return input.respond(
+                  sseEvents(
+                    { id: "chatcmpl_fixture", choices: [{ delta: { content: "Hello" }, finish_reason: null }] },
+                    { id: "chatcmpl_fixture", choices: [{ delta: {}, finish_reason: "stop" }] },
+                  ),
+                  { headers: { "content-type": "text/event-stream" } },
+                )
+              }),
+            ),
+          ),
+          Effect.provide(ConfigProvider.layer(ConfigProvider.fromEnv({ env: scenario.env }))),
+          Effect.tap((response) => Effect.sync(() => expect(response.text).toBe("Hello"))),
+        ),
+      )
+    }),
+  )
+})

packages/ai/test/provider/openai-chat.test.ts

@@ -47,7 +47,7 @@ describe("OpenAI Chat route", () => {
     Effect.gen(function* () {
       const prepared = yield* compileRequest(request)
 
-      expect(prepared.body).toEqual({
+      expect(prepared.body).toMatchObject({
         model: "gpt-4o-mini",
         messages: [
           { role: "system", content: "You are concise." },
@@ -55,7 +55,8 @@ describe("OpenAI Chat route", () => {
         ],
         stream: true,
         stream_options: { include_usage: true },
-        max_tokens: 20,
+        store: false,
+        max_completion_tokens: 20,
         temperature: 0,
       })
     }),
@@ -84,6 +85,28 @@ describe("OpenAI Chat route", () => {
     }),
   )
 
+  it.effect("omits empty and whitespace-only assistant messages", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model,
+          messages: [
+            Message.user("Before."),
+            Message.assistant([]),
+            Message.assistant(""),
+            Message.assistant(" \n\t "),
+            Message.assistant("After."),
+          ],
+        }),
+      )
+
+      expect(prepared.body.messages).toEqual([
+        { role: "user", content: "Before." },
+        { role: "assistant", content: "After." },
+      ])
+    }),
+  )
+
   it.effect("replays canonical reasoning as OpenAI-compatible reasoning_content", () =>
     Effect.gen(function* () {
       const prepared = yield* compileRequest(
@@ -146,6 +169,56 @@ describe("OpenAI Chat route", () => {
     }),
   )
 
+  it.effect("preserves observed reasoning fields when reasoning is required", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: LanguageModel.update(model, { compatibility: { requireReasoning: true } }),
+          messages: [
+            Message.assistant([
+              {
+                type: "reasoning",
+                text: "thinking",
+                providerMetadata: { openai: { reasoningField: "reasoning_text" } },
+              },
+              { type: "text", text: "Hello" },
+            ]),
+            Message.assistant("Done"),
+          ],
+        }),
+      )
+
+      expect(prepared.body.messages).toEqual([
+        { role: "assistant", content: "Hello", reasoning_text: "thinking" },
+        { role: "assistant", content: "Done", reasoning_content: "" },
+      ])
+    }),
+  )
+
+  it.effect("omits empty configured reasoning fields when reasoning is explicitly optional", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: LanguageModel.update(model, {
+            compatibility: { reasoningField: "reasoning_text", requireReasoning: false },
+          }),
+          messages: [
+            Message.assistant([
+              { type: "reasoning", text: "thinking" },
+              { type: "text", text: "Hello" },
+            ]),
+            Message.assistant("Done"),
+          ],
+        }),
+      )
+
+      expect(prepared.body.messages).toEqual([
+        { role: "assistant", content: "Hello", reasoning_text: "thinking" },
+        { role: "assistant", content: "Done" },
+      ])
+    }),
+  )
+
   it.effect("rejects reasoning fields that conflict with assistant message fields", () =>
     Effect.gen(function* () {
       const error = yield* compileRequest(
@@ -191,6 +264,21 @@ describe("OpenAI Chat route", () => {
     }),
   )
 
+  it.effect("omits the prompt cache key when caching is disabled", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model,
+          prompt: "Hello",
+          promptCacheKey: "session_123",
+          cache: "none",
+        }),
+      )
+
+      expect(prepared.body).not.toHaveProperty("prompt_cache_key")
+    }),
+  )
+
   it.effect("maps the xAI Chat prompt cache key to conversation affinity", () =>
     LLMClient.generate(
       LLM.request({
@@ -325,7 +413,7 @@ describe("OpenAI Chat route", () => {
         }),
       )
 
-      expect(prepared.body).toEqual({
+      expect(prepared.body).toMatchObject({
         model: "gpt-4o-mini",
         messages: [
           { role: "user", content: "What is the weather?" },
@@ -345,10 +433,40 @@ describe("OpenAI Chat route", () => {
         tools: [],
         stream: true,
         stream_options: { include_usage: true },
+        store: false,
       })
     }),
   )
 
+  it.effect("limits OpenAI and Azure Chat tool call IDs to 40 characters", () =>
+    Effect.gen(function* () {
+      const id = `call_${"a".repeat(48)}`
+      const models = [
+        model,
+        Azure.configure({ baseURL: "https://opencode-test.openai.azure.com/openai/", apiKey: "test" }).chat("gpt-4o"),
+      ]
+
+      yield* Effect.forEach(models, (selected) =>
+        Effect.gen(function* () {
+          const prepared = yield* compileRequest(
+            LLM.request({
+              model: selected,
+              messages: [
+                Message.assistant([ToolCallPart.make({ id, name: "lookup", input: {} })]),
+                Message.tool({ id, name: "lookup", result: "Sunny" }),
+              ],
+            }),
+          )
+
+          expect(prepared.body.messages).toMatchObject([
+            { role: "assistant", tool_calls: [{ id: id.slice(0, 40) }] },
+            { role: "tool", tool_call_id: id.slice(0, 40) },
+          ])
+        }),
+      )
+    }),
+  )
+
   it.effect("preserves structured tool errors for the model", () =>
     Effect.gen(function* () {
       const error = { error: { type: "unknown", message: "Tool execution interrupted" } }
@@ -414,6 +532,30 @@ describe("OpenAI Chat route", () => {
     }),
   )
 
+  it.effect("bridges image tool results before their synthetic user message when required", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: LanguageModel.update(model, { compatibility: { requireAssistantAfterTool: true } }),
+          messages: [
+            Message.assistant([ToolCallPart.make({ id: "call_image", name: "read", input: {} })]),
+            Message.tool({
+              id: "call_image",
+              name: "read",
+              result: {
+                type: "content",
+                value: [{ type: "file", uri: "data:image/png;base64,AAECAw==", mime: "image/png", name: "pixel.png" }],
+              },
+            }),
+          ],
+        }),
+      )
+
+      expect(prepared.body.messages.map((message) => message.role)).toEqual(["assistant", "tool", "assistant", "user"])
+      expect(prepared.body.messages[2]).toEqual({ role: "assistant", content: "Done." })
+    }),
+  )
+
   it.effect("orders parallel tool responses before one aggregated vision message", () =>
     Effect.gen(function* () {
       const prepared = yield* compileRequest(
@@ -1162,8 +1304,14 @@ describe("OpenAI Chat route", () => {
       expect(response.events).toEqual([
         { type: "step-start", index: 0 },
         { type: "tool-input-start", id: "call_1", name: "lookup", providerMetadata: undefined },
-        { type: "tool-input-delta", id: "call_1", name: "lookup", text: '{"query"' },
-        { type: "tool-input-delta", id: "call_1", name: "lookup", text: ':"weather"}' },
+        { type: "tool-input-delta", id: "call_1", name: "lookup", text: '{"query"', input: {} },
+        {
+          type: "tool-input-delta",
+          id: "call_1",
+          name: "lookup",
+          text: ':"weather"}',
+          input: { query: "weather" },
+        },
         { type: "tool-input-end", id: "call_1", name: "lookup", providerMetadata: undefined },
         {
           type: "tool-call",
@@ -1243,6 +1391,11 @@ describe("OpenAI Chat route", () => {
       ).pipe(Effect.provide(fixedResponse(body)), Effect.flip)
 
       expect(error.message).toContain("OpenAI Chat tool call delta is missing id or name")
+      expect(error.reason._tag).toBe("InvalidProviderOutput")
+      if (error.reason._tag !== "InvalidProviderOutput") return
+      expect(decodeJson(error.reason.raw ?? "")).toMatchObject({
+        choices: [{ finish_reason: "tool_calls" }],
+      })
     }),
   )
 
@@ -1256,6 +1409,7 @@ describe("OpenAI Chat route", () => {
         deltaChunk({ tool_calls: [{ index: 0, function: { arguments: ':"weather"}' } }] }),
       )
       const input = LLMRequest.update(request, {
+        model: LanguageModel.update(model, { compatibility: { requireFinishReason: false } }),
         tools: [ToolDefinition.make({ name: "lookup", description: "Lookup data", inputSchema: { type: "object" } })],
       })
       const response = yield* LLMClient.generate(input).pipe(Effect.provide(fixedResponse(body)))
@@ -1263,8 +1417,14 @@ describe("OpenAI Chat route", () => {
       expect(response.events).toEqual([
         { type: "step-start", index: 0 },
         { type: "tool-input-start", id: "call_1", name: "lookup", providerMetadata: undefined },
-        { type: "tool-input-delta", id: "call_1", name: "lookup", text: '{"query"' },
-        { type: "tool-input-delta", id: "call_1", name: "lookup", text: ':"weather"}' },
+        { type: "tool-input-delta", id: "call_1", name: "lookup", text: '{"query"', input: {} },
+        {
+          type: "tool-input-delta",
+          id: "call_1",
+          name: "lookup",
+          text: ':"weather"}',
+          input: { query: "weather" },
+        },
         { type: "tool-input-end", id: "call_1", name: "lookup", providerMetadata: undefined },
         {
           type: "tool-call",

packages/ai/test/provider/openai-compatible-chat.test.ts

@@ -70,7 +70,7 @@ describe("OpenAI-compatible Chat route", () => {
         baseURL: "https://api.deepseek.test/v1/",
         query: { "api-version": "2026-01-01" },
       })
-      expect(prepared.body).toEqual({
+      expect(prepared.body).toMatchObject({
         model: "deepseek-chat",
         messages: [
           { role: "system", content: "You are concise." },
@@ -79,7 +79,7 @@ describe("OpenAI-compatible Chat route", () => {
         tools: [
           {
             type: "function",
-            function: { name: "lookup", description: "Lookup data", parameters: { type: "object" } },
+            function: { name: "lookup", description: "Lookup data", parameters: { type: "object" }, strict: false },
           },
         ],
         tool_choice: "required",
@@ -130,7 +130,7 @@ describe("OpenAI-compatible Chat route", () => {
     Effect.gen(function* () {
       const prepared = yield* compileRequest(request)
 
-      expect(prepared.body).toEqual({
+      expect(prepared.body).toMatchObject({
         model: "deepseek-chat",
         messages: [
           { role: "system", content: "You are concise." },
@@ -158,6 +158,29 @@ describe("OpenAI-compatible Chat route", () => {
     }),
   )
 
+  it.effect("enables ZAI tool streaming except for GLM 4.5 models", () =>
+    Effect.gen(function* () {
+      const prepare = (provider: string, baseURL: string, id: string) =>
+        compileRequest(
+          LLM.request({
+            model: OpenAICompatibleChat.route.with({ provider, endpoint: { baseURL } }).model({ id }),
+            prompt: "Use a tool.",
+            tools: [ToolDefinition.make({ name: "lookup", description: "Lookup data", inputSchema: {} })],
+          }),
+        )
+
+      const current = yield* prepare("zai", "https://api.z.ai/api/paas/v4", "glm-4.7")
+      expect(current.body).toMatchObject({ tool_stream: true })
+
+      const legacy = yield* Effect.all(
+        ["glm-4.5", "glm-4.5-air", "glm-4.5-flash", "glm-4.5v"].map((id) =>
+          prepare("zhipuai", "https://open.bigmodel.cn/api/paas/v4", id),
+        ),
+      )
+      legacy.forEach((item) => expect(item.body).not.toHaveProperty("tool_stream"))
+    }),
+  )
+
   it.effect("matches AI SDK compatible tool request body fixture", () =>
     Effect.gen(function* () {
       const prepared = yield* compileRequest(
@@ -180,7 +203,7 @@ describe("OpenAI-compatible Chat route", () => {
         }),
       )
 
-      expect(prepared.body).toEqual({
+      expect(prepared.body).toMatchObject({
         model: "deepseek-chat",
         messages: [
           { role: "user", content: "What is the weather?" },
@@ -204,6 +227,7 @@ describe("OpenAI-compatible Chat route", () => {
               name: "lookup",
               description: "Lookup data",
               parameters: { type: "object", properties: { query: { type: "string" } }, required: ["query"] },
+              strict: false,
             },
           },
         ],
@@ -214,6 +238,135 @@ describe("OpenAI-compatible Chat route", () => {
     }),
   )
 
+  it.effect("normalizes tool call IDs for the selected model family", () =>
+    Effect.gen(function* () {
+      const longID = `call_${"a".repeat(48)}`
+      const cases = [
+        { provider: "custom", model: "mistral-small", id: "toolu_01CBhTTz95qkd9LJMdC9sf8t", expected: "toolu01CB" },
+        { provider: "custom", model: "devstral-small", id: "abc", expected: "abc000000" },
+        { provider: "custom", model: "codestral-latest", id: "toolu_01CBhTTz95", expected: "toolu01CB" },
+        { provider: "custom", model: "pixtral-large", id: "toolu_01CBhTTz95", expected: "toolu01CB" },
+        { provider: "custom", model: "open-mixtral-8x22b", id: "toolu_01CBhTTz95", expected: "toolu01CB" },
+        { provider: "gateway", model: "anthropic/claude-sonnet-4", id: "call|item/+", expected: "call_item__" },
+        { provider: "gateway", model: "openai/gpt-4o", id: longID, expected: longID.slice(0, 40) },
+        { provider: "custom", model: "ordinary-model", id: "call|item/+", expected: "call|item/+" },
+        { provider: "mistral", model: "zai-glm-5-2", id: "call_long_identifier", expected: "call_long_identifier" },
+      ]
+
+      yield* Effect.forEach(cases, (item) =>
+        Effect.gen(function* () {
+          const prepared = yield* compileRequest(
+            LLM.request({
+              model: OpenAICompatibleChat.route
+                .with({ provider: item.provider, endpoint: { baseURL: "https://api.custom.test/v1" } })
+                .model({ id: item.model }),
+              messages: [
+                Message.assistant([ToolCallPart.make({ id: item.id, name: "lookup", input: {} })]),
+                Message.tool({ id: item.id, name: "lookup", result: { type: "content", value: [] } }),
+              ],
+            }),
+          )
+
+          expect(prepared.body.messages).toMatchObject([
+            { role: "assistant", tool_calls: [{ id: item.expected }] },
+            { role: "tool", tool_call_id: item.expected },
+          ])
+        }),
+      )
+    }),
+  )
+
+  it.effect("bridges tool results for Mistral-family models and honors compatibility overrides", () =>
+    Effect.gen(function* () {
+      const cases = [
+        { id: "mistral-small", bridge: true },
+        { id: "devstral-small", bridge: true },
+        { id: "codestral-latest", bridge: true },
+        { id: "pixtral-large", bridge: true },
+        { id: "open-mixtral-8x22b", bridge: true },
+        { id: "ordinary-model", bridge: false },
+        { id: "ordinary-model", override: true, bridge: true },
+        { id: "mistral-small", override: false, bridge: false },
+      ] as const
+
+      yield* Effect.forEach(cases, (item) =>
+        Effect.gen(function* () {
+          const selected = OpenAICompatibleChat.route
+            .with({ provider: "custom", endpoint: { baseURL: "https://api.custom.test/v1" } })
+            .model({
+              id: item.id,
+              compatibility: "override" in item ? { requireAssistantAfterTool: item.override } : undefined,
+            })
+          const prepared = yield* compileRequest(
+            LLM.request({
+              model: selected,
+              messages: [
+                Message.assistant([ToolCallPart.make({ id: "call_1", name: "lookup", input: {} })]),
+                Message.tool({ id: "call_1", name: "lookup", result: "Sunny" }),
+                Message.user("What next?"),
+              ],
+            }),
+          )
+
+          expect(prepared.body.messages.map((message) => message.role)).toEqual(
+            item.bridge ? ["assistant", "tool", "assistant", "user"] : ["assistant", "tool", "user"],
+          )
+          if (item.bridge) expect(prepared.body.messages[2]).toEqual({ role: "assistant", content: "Done." })
+        }),
+      )
+    }),
+  )
+
+  it.effect("requires reasoning for DeepSeek models, providers, and endpoints unless explicitly overridden", () =>
+    Effect.gen(function* () {
+      const cases = [
+        { id: "DeepSeek-V3", provider: "custom", baseURL: "https://api.custom.test/v1", required: true },
+        { id: "custom-model", provider: "deepseek", baseURL: "https://api.custom.test/v1", required: true },
+        { id: "custom-model", provider: "custom", baseURL: "https://API.DeepSeek.COM/v1", required: true },
+        { id: "ordinary-model", provider: "custom", baseURL: "https://api.custom.test/v1", required: false },
+        {
+          id: "ordinary-model",
+          provider: "custom",
+          baseURL: "https://api.custom.test/v1",
+          compatibility: { requireReasoning: true, reasoningField: "reasoning" },
+          required: true,
+          field: "reasoning",
+        },
+        {
+          id: "deepseek-chat",
+          provider: "deepseek",
+          baseURL: "https://api.deepseek.com/v1",
+          compatibility: { requireReasoning: false },
+          required: false,
+        },
+      ] as const
+
+      yield* Effect.forEach(cases, (item) =>
+        Effect.gen(function* () {
+          const selected = OpenAICompatibleChat.route
+            .with({ provider: item.provider, endpoint: { baseURL: item.baseURL } })
+            .model({ id: item.id, compatibility: "compatibility" in item ? item.compatibility : undefined })
+          const prepared = yield* compileRequest(
+            LLM.request({
+              model: selected,
+              messages: [
+                Message.assistant("Hello"),
+                Message.assistant([ToolCallPart.make({ id: "call_1", name: "lookup", input: {} })]),
+                Message.tool({ id: "call_1", name: "lookup", result: "Sunny" }),
+              ],
+            }),
+          )
+          const field = "field" in item ? item.field : "reasoning_content"
+
+          for (const message of prepared.body.messages.filter((message) => message.role === "assistant")) {
+            if (item.required) expect(message).toHaveProperty(field, "")
+            else expect(message).not.toHaveProperty(field)
+          }
+        }),
+      )
+    }),
+  )
+
   it.effect("posts to the configured compatible endpoint and parses text usage", () =>
     Effect.gen(function* () {
       const response = yield* LLMClient.generate(request).pipe(
@@ -329,13 +482,106 @@ describe("OpenAI-compatible Chat route", () => {
     }),
   )
 
-  it.effect("treats an empty finish reason as terminal", () =>
+  it.effect("rejects a stream without a required finish reason", () =>
     Effect.gen(function* () {
-      const response = yield* LLMClient.generate(request).pipe(
+      const error = yield* LLMClient.generate(request).pipe(
+        Effect.provide(fixedResponse(sseEvents(deltaChunk({ content: "Hello" }), deltaChunk({}, "")))),
+        Effect.flip,
+      )
+
+      expect(error.reason).toMatchObject({
+        _tag: "InvalidProviderOutput",
+        classification: "incomplete-stream",
+        message: "OpenAI Chat stream ended without finish_reason",
+      })
+    }),
+  )
+
+  it.effect("infers stop when finish reasons are optional", () =>
+    Effect.gen(function* () {
+      const compatible = OpenAICompatibleChat.route
+        .with({ provider: "custom", endpoint: { baseURL: "https://api.custom.test/v1" } })
+        .model({ id: "custom-model", compatibility: { requireFinishReason: false } })
+      const response = yield* LLMClient.generate(LLMRequest.update(request, { model: compatible })).pipe(
         Effect.provide(fixedResponse(sseEvents(deltaChunk({ content: "Hello" }), deltaChunk({}, "")))),
       )
 
-      expect(response.finishReason).toEqual({ normalized: "unknown", raw: "" })
+      expect(response.finishReason).toEqual({ normalized: "stop" })
+    }),
+  )
+
+  it.effect("normalizes the end finish reason to stop", () =>
+    Effect.gen(function* () {
+      const response = yield* LLMClient.generate(request).pipe(
+        Effect.provide(fixedResponse(sseEvents(deltaChunk({ content: "Hello" }), deltaChunk({}, "end")))),
+      )
+
+      expect(response.finishReason).toEqual({ normalized: "stop", raw: "end" })
+    }),
+  )
+
+  it.effect("classifies provider error finish reasons", () =>
+    Effect.gen(function* () {
+      const error = yield* LLMClient.generate(request).pipe(
+        Effect.provide(fixedResponse(sseEvents(deltaChunk({}, "network_error")))),
+        Effect.flip,
+      )
+
+      expect(error.reason).toMatchObject({
+        _tag: "ProviderInternal",
+        message: "Provider reported a network error (finish_reason: network_error)",
+      })
+      expect(decodeJson(error.body ?? "")).toMatchObject({
+        id: "chatcmpl_fixture",
+        choices: [{ finish_reason: "network_error" }],
+      })
+
+      const generic = yield* LLMClient.generate(request).pipe(
+        Effect.provide(fixedResponse(sseEvents(deltaChunk({}, "error")))),
+        Effect.flip,
+      )
+      expect(generic.reason).toMatchObject({
+        _tag: "UnknownProvider",
+        message: "Provider reported an error (finish_reason: error)",
+      })
+    }),
+  )
+
+  it.effect("preserves explicit provider error events", () =>
+    Effect.gen(function* () {
+      const error = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents({
+              id: "chatcmpl_error",
+              error: { code: 502, message: "Provider disconnected", details: { upstream: "vendor" } },
+              trace_id: "trace_1",
+            }),
+          ),
+        ),
+        Effect.flip,
+      )
+
+      expect(error.reason).toMatchObject({ _tag: "ProviderInternal", message: "Provider disconnected", status: 502 })
+      expect(decodeJson(error.body ?? "")).toMatchObject({
+        id: "chatcmpl_error",
+        error: { code: 502, message: "Provider disconnected", details: { upstream: "vendor" } },
+        trace_id: "trace_1",
+      })
+    }),
+  )
+
+  it.effect("preserves provider finish outcomes in the common reason algebra", () =>
+    Effect.gen(function* () {
+      const filtered = yield* LLMClient.generate(request).pipe(
+        Effect.provide(fixedResponse(sseEvents(deltaChunk({}, "content_filter")))),
+      )
+      const future = yield* LLMClient.generate(request).pipe(
+        Effect.provide(fixedResponse(sseEvents(deltaChunk({}, "future_reason")))),
+      )
+
+      expect(filtered.finishReason).toEqual({ normalized: "content-filter", raw: "content_filter" })
+      expect(future.finishReason).toEqual({ normalized: "unknown", raw: "future_reason" })
     }),
   )
 
@@ -355,6 +601,11 @@ describe("OpenAI-compatible Chat route", () => {
       )
 
       expect(error.message).toContain("OpenAI Chat received content after the finish reason")
+      expect(error.reason._tag).toBe("InvalidProviderOutput")
+      if (error.reason._tag !== "InvalidProviderOutput") return
+      expect(decodeJson(error.reason.raw ?? "")).toMatchObject({
+        choices: [{ delta: { tool_calls: [{ id: "call_1" }] } }],
+      })
     }),
   )
 })

packages/ai/test/provider/openai-compatible-responses.test.ts

@@ -47,15 +47,31 @@ describe("Open Responses-compatible route", () => {
       })
       expect(prepared.body).toEqual({
         model: "example-model",
-        input: [
-          { role: "system", content: "You are concise." },
-          { role: "user", content: [{ type: "input_text", text: "Say hello." }] },
-        ],
+        input: [{ role: "user", content: [{ type: "input_text", text: "Say hello." }] }],
+        instructions: "You are concise.",
         stream: true,
+        store: false,
+        include: ["reasoning.encrypted_content"],
       })
     }),
   )
 
+  it.effect("allows callers to override stateless encrypted reasoning defaults", () =>
+    Effect.gen(function* () {
+      const model = configure({
+        apiKey: "test-key",
+        baseURL: "https://responses.example.test/v1",
+        provider: "example",
+      }).model("example-model")
+      const prepared = yield* compileRequest(
+        LLM.request({ model, prompt: "Say hello.", providerOptions: { store: true, include: [] } }),
+      )
+
+      expect(prepared.body.store).toBe(true)
+      expect(prepared.body.include).toBeUndefined()
+    }),
+  )
+
   it.effect("lowers chronological system updates as standard developer messages", () =>
     Effect.gen(function* () {
       const model = configure({
@@ -66,10 +82,12 @@ describe("Open Responses-compatible route", () => {
       const prepared = yield* compileRequest(
         LLM.request({
           model,
+          system: "Initial instructions.",
           messages: [Message.user("Before."), Message.system("Operator update."), Message.assistant("After.")],
         }),
       )
 
+      expect(prepared.body.instructions).toBe("Initial instructions.")
       expect(prepared.body.input).toEqual([
         { role: "user", content: [{ type: "input_text", text: "Before." }] },
         { role: "developer", content: "Operator update." },
@@ -132,6 +150,305 @@ describe("Open Responses-compatible route", () => {
     }),
   )
 
+  it.effect("lowers canonical parallel tool control", () =>
+    Effect.gen(function* () {
+      const model = configure({
+        apiKey: "test-key",
+        baseURL: "https://responses.example.test/v1",
+      }).model("example-model")
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model,
+          prompt: "Read the file.",
+          tools: [
+            ToolDefinition.make({
+              name: "read",
+              description: "Read a file.",
+              inputSchema: { type: "object" },
+            }),
+          ],
+          toolChoice: { type: "auto", disableParallelToolUse: true },
+        }),
+      )
+
+      expect(prepared.body.parallel_tool_calls).toBe(false)
+      expect(prepared.body.tools).toEqual([
+        {
+          type: "function",
+          name: "read",
+          description: "Read a file.",
+          parameters: { type: "object" },
+          strict: false,
+        },
+      ])
+    }),
+  )
+
+  it.effect("keeps foreign item id grammars but drops malformed ids", () =>
+    Effect.gen(function* () {
+      const model = configure({
+        apiKey: "test-key",
+        baseURL: "https://responses.example.test/v1",
+      }).model("example-model")
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model,
+          messages: [
+            Message.assistant([
+              { type: "text", text: "Kept.", providerMetadata: { openresponses: { itemId: "history_1" } } },
+              {
+                type: "text",
+                text: "Long.",
+                providerMetadata: { openresponses: { itemId: `history_${"a".repeat(64)}` } },
+              },
+              {
+                type: "text",
+                text: "Opaque.",
+                providerMetadata: { openresponses: { itemId: "provider_value/with+symbols" } },
+              },
+              { type: "text", text: "No suffix.", providerMetadata: { openresponses: { itemId: "msg_" } } },
+              { type: "text", text: "No prefix.", providerMetadata: { openresponses: { itemId: "_item" } } },
+            ]),
+          ],
+        }),
+      )
+
+      expect(prepared.body.input).toEqual([
+        {
+          type: "message",
+          id: "history_1",
+          role: "assistant",
+          content: [{ type: "output_text", text: "Kept." }],
+        },
+        {
+          type: "message",
+          id: `history_${"a".repeat(64)}`,
+          role: "assistant",
+          content: [{ type: "output_text", text: "Long." }],
+        },
+        {
+          type: "message",
+          id: "provider_value/with+symbols",
+          role: "assistant",
+          content: [{ type: "output_text", text: "Opaque." }],
+        },
+        {
+          type: "message",
+          role: "assistant",
+          content: [
+            { type: "output_text", text: "No suffix." },
+            { type: "output_text", text: "No prefix." },
+          ],
+        },
+      ])
+    }),
+  )
+
+  it.effect("replays only shared hosted tool items", () =>
+    Effect.gen(function* () {
+      const model = configure({
+        apiKey: "test-key",
+        baseURL: "https://responses.example.test/v1",
+        provider: "example",
+      }).model("example-model")
+      const items = [
+        { type: "web_search_call", id: "ws_1", status: "completed" },
+        { type: "x_search_call", id: "x_search_1", status: "completed" },
+        { type: "future_call", id: "future_1", status: "completed" },
+        { type: "file_search_call", id: "fs_1", queries: "not-an-array" },
+      ]
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model,
+          messages: items.map((item) =>
+            Message.assistant({
+              type: "tool-result",
+              id: item.id,
+              name: item.type,
+              result: { type: "json", value: item },
+              providerExecuted: true,
+              providerMetadata: { openresponses: { itemId: item.id } },
+            }),
+          ),
+        }),
+      )
+
+      expect(prepared.body.input).toEqual([
+        items[0],
+        { role: "user", content: [{ type: "input_text", text: JSON.stringify(items[1]) }] },
+        { role: "user", content: [{ type: "input_text", text: JSON.stringify(items[2]) }] },
+        { role: "user", content: [{ type: "input_text", text: JSON.stringify(items[3]) }] },
+      ])
+    }),
+  )
+
+  it.effect("routes response deltas by output index", () =>
+    Effect.gen(function* () {
+      const model = configure({
+        apiKey: "test-key",
+        baseURL: "https://responses.example.test/v1",
+      }).model("example-model")
+      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Say hello." })).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              { type: "response.output_item.added", output_index: 2, item: { type: "message", id: "msg_1" } },
+              { type: "response.output_text.delta", output_index: 2, item_id: "wrong_message", delta: "Indexed" },
+              { type: "response.output_item.done", output_index: 2, item: { type: "message", id: "msg_1" } },
+              { type: "response.completed", response: { id: "resp_1" } },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.message.content).toEqual([
+        { type: "text", text: "Indexed", providerMetadata: { openresponses: { itemId: "msg_1" } } },
+      ])
+    }),
+  )
+
+  it.effect("streams function calls without optional item ids through the shared baseline", () =>
+    Effect.gen(function* () {
+      const model = configure({
+        apiKey: "test-key",
+        baseURL: "https://responses.example.test/v1",
+        provider: "example",
+      }).model("example-model")
+      const item = { type: "function_call", call_id: "call_1", name: "lookup", arguments: "" }
+      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Look it up." })).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              { type: "response.output_item.added", output_index: 1, item },
+              {
+                type: "response.function_call_arguments.delta",
+                output_index: 1,
+                item_id: "opaque_item",
+                delta: '{"query":"shared"}',
+              },
+              {
+                type: "response.output_item.done",
+                output_index: 1,
+                item: { ...item, arguments: '{"query":"complete"}' },
+              },
+              { type: "response.completed", response: { id: "resp_1" } },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.events.filter(LLMEvent.is.toolCall)).toEqual([
+        expect.objectContaining({ id: "call_1", name: "lookup", input: { query: "complete" } }),
+      ])
+      expect(response.events.find(LLMEvent.is.toolCall)?.providerMetadata).toBeUndefined()
+    }),
+  )
+
+  it.effect("finalizes pending function calls from completed response output", () =>
+    Effect.gen(function* () {
+      const model = configure({
+        apiKey: "test-key",
+        baseURL: "https://responses.example.test/v1",
+        provider: "example",
+      }).model("example-model")
+      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Look it up." })).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              {
+                type: "response.output_item.added",
+                item: { type: "function_call", id: "item_1", call_id: "call_1", name: "lookup", arguments: "" },
+              },
+              { type: "response.function_call_arguments.delta", item_id: "item_1", delta: '{"query":"par' },
+              {
+                type: "response.completed",
+                response: {
+                  output: [
+                    {
+                      type: "function_call",
+                      id: "item_1",
+                      call_id: "call_1",
+                      name: "lookup",
+                      arguments: '{"query":"complete"}',
+                    },
+                  ],
+                },
+              },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.events.find(LLMEvent.is.toolCall)).toMatchObject({
+        input: { query: "complete" },
+        providerMetadata: { openresponses: { itemId: "item_1" } },
+      })
+    }),
+  )
+
+  it.effect("preserves terminal reasoning metadata when item completion is missing", () =>
+    Effect.gen(function* () {
+      const model = configure({
+        apiKey: "test-key",
+        baseURL: "https://responses.example.test/v1",
+      }).model("example-model")
+      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Think it through." })).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              {
+                type: "response.output_item.added",
+                item: { type: "reasoning", id: "rs_raw", encrypted_content: null },
+              },
+              { type: "response.reasoning_summary_text.delta", item_id: "rs_raw", delta: "Thinking" },
+              {
+                type: "response.completed",
+                response: {
+                  output: [{ type: "reasoning", id: "rs_raw", encrypted_content: "raw-state" }],
+                },
+              },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.events.find((event) => event.type === "reasoning-end")).toMatchObject({
+        providerMetadata: { openresponses: { itemId: "rs_raw", reasoningEncryptedContent: "raw-state" } },
+      })
+    }),
+  )
+
+  it.effect("reconciles raw reasoning finals without streamed deltas", () =>
+    Effect.gen(function* () {
+      const model = configure({
+        apiKey: "test-key",
+        baseURL: "https://responses.example.test/v1",
+      }).model("example-model")
+      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Think it through." })).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              {
+                type: "response.output_item.added",
+                item: { type: "reasoning", id: "rs_raw", encrypted_content: null },
+              },
+              // Raw reasoning finals carry no summary index; they reconcile
+              // into the item's first block.
+              { type: "response.reasoning.done", item_id: "rs_raw", text: "Raw chain of thought." },
+              {
+                type: "response.output_item.done",
+                item: { type: "reasoning", id: "rs_raw", encrypted_content: "raw-state" },
+              },
+              { type: "response.completed", response: { id: "resp_1" } },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.reasoning).toBe("Raw chain of thought.")
+    }),
+  )
+
   it.effect("preserves nullable phases in the forgiving Open Responses baseline", () =>
     Effect.gen(function* () {
       const model = configure({

packages/ai/test/provider/openai-responses-websocket.recorded.test.ts

@@ -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." }] },

packages/ai/test/provider/openrouter.test.ts

@@ -190,6 +190,21 @@ describe("OpenRouter", () => {
     }),
   )
 
+  it.effect("omits the prompt cache key when caching is disabled", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: OpenRouter.configure({ apiKey: "test-key" }).model("openai/gpt-4o-mini"),
+          prompt: "Hello",
+          promptCacheKey: "session_123",
+          cache: "none",
+        }),
+      )
+
+      expect(prepared.body).not.toHaveProperty("prompt_cache_key")
+    }),
+  )
+
   it.effect("filters invalid known OpenRouter options while preserving extensions", () =>
     Effect.gen(function* () {
       const invalid: Record<string, unknown> = {

packages/ai/test/provider/xai-responses.test.ts

@@ -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"
@@ -14,24 +14,44 @@ import { sseEvents } from "../lib/sse.js"
 const model = XAI.configure({ apiKey: "test", baseURL: "https://api.x.ai/v1" }).responses("grok-4.6")
 
 describe("xAI Responses route", () => {
-  it.effect("extends the Open Responses baseline directly", () =>
+  it.effect("composes the Open Responses baseline with xAI extensions", () =>
     Effect.gen(function* () {
-      expect(XAIResponses.protocol.body).toBe(OpenResponses.protocol.body)
+      expect(XAIResponses.protocol.body).not.toBe(OpenResponses.protocol.body)
       expect(XAIResponses.protocol.body).not.toBe(OpenAIResponses.protocol.body)
 
       const prepared = yield* compileRequest(LLM.request({ model, prompt: "Hello" }))
       expect(prepared.protocol).toBe("xai-responses")
+      expect(prepared.body.store).toBe(false)
+      expect(prepared.body.include).toEqual(["reasoning.encrypted_content"])
     }),
   )
 
-  it.effect("parses xAI reasoning text events", () =>
+  it.effect("allows callers to opt out of encrypted reasoning", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(LLM.request({ model, prompt: "Hello", providerOptions: { include: [] } }))
+
+      expect(prepared.body.store).toBe(false)
+      expect(prepared.body.include).toBeUndefined()
+    }),
+  )
+
+  it.effect("parses xAI reasoning summaries", () =>
     Effect.gen(function* () {
       const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Think" })).pipe(
         Effect.provide(
           fixedResponse(
             sseEvents(
-              { type: "response.reasoning_text.delta", item_id: "reasoning_1", delta: "Considering." },
-              { type: "response.reasoning_text.done", item_id: "reasoning_1" },
+              {
+                type: "response.output_item.added",
+                item: { type: "reasoning", id: "reasoning_1" },
+              },
+              // Grok streams reasoning with the standard summary event name.
+              {
+                type: "response.reasoning_summary_text.delta",
+                item_id: "reasoning_1",
+                summary_index: 0,
+                delta: "Considering.",
+              },
               {
                 type: "response.output_item.done",
                 item: { type: "reasoning", id: "reasoning_1", encrypted_content: "opaque" },
@@ -50,16 +70,106 @@ describe("xAI Responses route", () => {
     }),
   )
 
-  it.effect("parses xAI hosted tool items", () =>
+  it.effect("routes xAI reasoning summaries by output index", () =>
     Effect.gen(function* () {
-      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Search X" })).pipe(
+      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Think" })).pipe(
         Effect.provide(
           fixedResponse(
             sseEvents(
               {
-                type: "response.output_item.done",
-                item: { type: "x_search_call", id: "x_search_1", status: "completed", action: { query: "news" } },
+                type: "response.output_item.added",
+                output_index: 3,
+                item: { type: "reasoning", id: "reasoning_1" },
               },
+              {
+                type: "response.reasoning_summary_text.delta",
+                output_index: 3,
+                item_id: "wrong_reasoning",
+                summary_index: 0,
+                delta: "Considering.",
+              },
+              {
+                type: "response.output_item.done",
+                output_index: 3,
+                item: { type: "reasoning", id: "reasoning_1", encrypted_content: "opaque" },
+              },
+              { type: "response.completed", response: { id: "response_1" } },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.reasoning).toBe("Considering.")
+      expect(response.message.content.find((part) => part.type === "reasoning")).toMatchObject({
+        providerMetadata: { xai: { itemId: "reasoning_1", reasoningEncryptedContent: "opaque" } },
+      })
+    }),
+  )
+
+  it.effect("replays xAI hosted tool items when continuing with the same provider", () =>
+    Effect.gen(function* () {
+      const item = { type: "x_search_call", id: "x_search_1", status: "completed", action: { query: "news" } }
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model,
+          messages: [
+            Message.assistant([
+              {
+                type: "tool-result",
+                id: "x_search_1",
+                name: "x_search",
+                result: { type: "json", value: item },
+                providerExecuted: true,
+                providerMetadata: { xai: { itemId: "x_search_1" } },
+              },
+            ]),
+          ],
+        }),
+      )
+
+      expect(prepared.body.input).toEqual([item])
+    }),
+  )
+
+  it.effect("replays shared and xAI hosted tool items but rejects OpenAI extensions", () =>
+    Effect.gen(function* () {
+      const items = [
+        { type: "web_search_call", id: "ws_1", status: "completed" },
+        { type: "image_generation_call", id: "ig_1", status: "completed", result: "AQID" },
+        { type: "computer_call", id: "computer_1", status: "completed" },
+      ]
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model,
+          messages: items.map((item) =>
+            Message.assistant({
+              type: "tool-result",
+              id: item.id,
+              name: item.type,
+              result: { type: "json", value: item },
+              providerExecuted: true,
+              providerMetadata: { xai: { itemId: item.id } },
+            }),
+          ),
+        }),
+      )
+
+      expect(prepared.body.input).toEqual([
+        items[0],
+        items[1],
+        { role: "user", content: [{ type: "input_text", text: JSON.stringify(items[2]) }] },
+      ])
+    }),
+  )
+
+  it.effect("parses xAI hosted tool items", () =>
+    Effect.gen(function* () {
+      const item = { type: "x_search_call", id: "x_search_1", status: "completed", action: { query: "news" } }
+      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Search X" })).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              { type: "response.output_item.done", item },
               { type: "response.completed", response: { id: "response_1" } },
             ),
           ),
@@ -71,6 +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" } },
       })
     }),
   )

packages/ai/test/provider/zai-images.test.ts

@@ -78,6 +78,31 @@ describe("Z.ai Images", () => {
     ),
   )
 
+  it.effect("sanitizes unpaired surrogates in outbound image requests", () =>
+    Image.generate({
+      model: ZAI.configure({ apiKey: "test", http: { body: { metadata: { source: "default\uDC00" } } } }).image("model"),
+      prompt: "A red circle \uD800 on a white background \u{1F600}",
+    }).pipe(
+      Effect.provide(
+        ImageClient.layer.pipe(
+          Layer.provide(
+            dynamicResponse((input) => {
+              expect(JSON.parse(input.text)).toMatchObject({
+                prompt: "A red circle \uFFFD on a white background \u{1F600}",
+                metadata: { source: "default\uFFFD" },
+              })
+              return Effect.succeed(
+                input.respond(JSON.stringify({ data: [{ url: "https://example.test/image.jpg" }] }), {
+                  headers: { "content-type": "application/json" },
+                }),
+              )
+            }),
+          ),
+        ),
+      ),
+    ),
+  )
+
   it.effect("lets raw native options override aliases", () =>
     Image.generate({
       model: ZAI.configure({ apiKey: "test" }).image("model"),

packages/ai/test/recorded-golden.ts

@@ -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,
             })

packages/ai/test/recorded-scenarios.ts

@@ -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,

packages/ai/test/schema.test.ts

@@ -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)

packages/ai/test/tool-runtime.test.ts

@@ -101,7 +101,7 @@ describe("LLMClient tools", () => {
       const messages = Reflect.get(second, "messages")
       const tools = Reflect.get(second, "tools")
 
-      expect(Reflect.get(second, "max_tokens")).toBe(50)
+      expect(Reflect.get(second, "max_completion_tokens")).toBe(50)
       expect(Reflect.get(second, "tool_choice")).toBe("auto")
       expect(tools).toHaveLength(1)
       expect(

packages/ai/test/tool-schema-projection.test.ts

@@ -50,32 +50,7 @@ describe("tool schema projections", () => {
     })
   })
 
-  test("openai keeps one flat object top-level schema", () => {
-    expect(
-      ToolSchemaProjection.openAI({
-        anyOf: [
-          {
-            type: "object",
-            properties: {
-              path: { type: "string" },
-              maybe: { anyOf: [{ type: "string" }, { type: "null" }] },
-            },
-          },
-          { type: "object", properties: { resource: { type: "string" } } },
-        ],
-      }),
-    ).toEqual({
-      type: "object",
-      properties: {
-        path: { type: "string" },
-        maybe: { type: "string" },
-        resource: { type: "string" },
-      },
-      additionalProperties: false,
-    })
-  })
-
-  it.effect("applies model compatibility before protocol projection", () =>
+  it.effect("applies model compatibility without changing schema semantics", () =>
     Effect.gen(function* () {
       const model = OpenAIChat.route
         .with({ endpoint: { baseURL: "https://api.openai.test/v1/" }, auth: Auth.bearer("test") })
@@ -107,11 +82,15 @@ describe("tool schema projections", () => {
 
       expect(prepared.body.tools?.[0]?.function.parameters).toEqual({
         type: "object",
-        properties: {
-          tuple: { type: "array", items: { anyOf: [{ type: "string" }, { type: "number" }] } },
-          linked: { $ref: "#/$defs/Linked" },
-        },
-        additionalProperties: false,
+        anyOf: [
+          {
+            type: "object",
+            properties: {
+              tuple: { type: "array", items: { anyOf: [{ type: "string" }, { type: "number" }] } },
+              linked: { $ref: "#/$defs/Linked" },
+            },
+          },
+        ],
       })
     }),
   )

packages/ai/test/tool-stream.test.ts

@@ -23,9 +23,11 @@ describe("ToolStream", () => {
 
       expect(first.events).toEqual([
         { type: "tool-input-start", id: "call_1", name: "lookup" },
-        { type: "tool-input-delta", id: "call_1", name: "lookup", text: '{"query"' },
+        { type: "tool-input-delta", id: "call_1", name: "lookup", text: '{"query"', input: {} },
+      ])
+      expect(second.events).toEqual([
+        { type: "tool-input-delta", id: "call_1", name: "lookup", text: ':"weather"}', input: { query: "weather" } },
       ])
-      expect(second.events).toEqual([{ type: "tool-input-delta", id: "call_1", name: "lookup", text: ':"weather"}' }])
       expect(finished).toEqual({
         tools: {},
         events: [
@@ -36,6 +38,45 @@ describe("ToolStream", () => {
     }),
   )
 
+  it.effect("exposes cumulative partial string values", () =>
+    Effect.gen(function* () {
+      const result = ToolStream.appendOrStart(
+        ADAPTER,
+        ToolStream.empty<number>(),
+        0,
+        { id: "call_1", name: "lookup", text: '{"query":"wea' },
+        "missing tool",
+      )
+      if (ToolStream.isError(result)) return yield* result
+
+      expect(result.events.at(-1)).toEqual({
+        type: "tool-input-delta",
+        id: "call_1",
+        name: "lookup",
+        text: '{"query":"wea',
+        input: { query: "wea" },
+      })
+    }),
+  )
+
+  it.effect("defaults partial input to an empty object when the accumulated value cannot be parsed", () =>
+    Effect.gen(function* () {
+      const result = ToolStream.appendOrStart(
+        ADAPTER,
+        ToolStream.empty<number>(),
+        0,
+        { id: "call_1", name: "lookup", text: "x" },
+        "missing tool",
+      )
+      if (ToolStream.isError(result)) return yield* result
+
+      expect(result.events).toEqual([
+        { type: "tool-input-start", id: "call_1", name: "lookup" },
+        { type: "tool-input-delta", id: "call_1", name: "lookup", text: "x", input: {} },
+      ])
+    }),
+  )
+
   it.effect("keeps accumulated identity when later deltas contain empty strings", () =>
     Effect.gen(function* () {
       const first = ToolStream.appendOrStart(
@@ -91,7 +132,7 @@ describe("ToolStream", () => {
     }),
   )
 
-  it.effect("finalizes malformed local input as a non-executable tool error", () =>
+  it.effect("finalizes incomplete local input using the partial JSON parser", () =>
     Effect.gen(function* () {
       const tools = ToolStream.start(ToolStream.empty<string>(), "item_1", {
         id: "call_1",
@@ -103,18 +144,46 @@ describe("ToolStream", () => {
       expect(finished).toEqual({
         tools: {},
         events: [
-          {
-            type: "tool-input-error",
-            id: "call_1",
-            name: "lookup",
-            raw: '{"query":"partial',
-          },
+          { type: "tool-input-end", id: "call_1", name: "lookup" },
+          { type: "tool-call", id: "call_1", name: "lookup", input: { query: "partial" } },
         ],
       })
     }),
   )
 
-  it.effect("preserves valid siblings when one parallel input is malformed", () =>
+  it.effect("repairs malformed string escapes in final local input", () =>
+    Effect.gen(function* () {
+      const tools = ToolStream.start(ToolStream.empty<string>(), "item_1", {
+        id: "call_1",
+        name: "lookup",
+        input: '{"path":"A\\H","text":"first\tsecond"}',
+      })
+      const finished = yield* ToolStream.finish(ADAPTER, tools, "item_1")
+
+      expect(finished.events).toEqual([
+        { type: "tool-input-end", id: "call_1", name: "lookup" },
+        { type: "tool-call", id: "call_1", name: "lookup", input: { path: "A\\H", text: "first\tsecond" } },
+      ])
+    }),
+  )
+
+  it.effect("defaults unrecoverable local input to an empty object", () =>
+    Effect.gen(function* () {
+      const tools = ToolStream.start(ToolStream.empty<string>(), "item_1", {
+        id: "call_1",
+        name: "lookup",
+        input: "invalid",
+      })
+      const finished = yield* ToolStream.finish(ADAPTER, tools, "item_1")
+
+      expect(finished.events).toEqual([
+        { type: "tool-input-end", id: "call_1", name: "lookup" },
+        { type: "tool-call", id: "call_1", name: "lookup", input: {} },
+      ])
+    }),
+  )
+
+  it.effect("recovers incomplete input alongside valid parallel tool calls", () =>
     Effect.gen(function* () {
       const valid = ToolStream.start(ToolStream.empty<number>(), 0, {
         id: "call_valid",
@@ -133,12 +202,8 @@ describe("ToolStream", () => {
         events: [
           { type: "tool-input-end", id: "call_valid", name: "lookup" },
           { type: "tool-call", id: "call_valid", name: "lookup", input: { query: "weather" } },
-          {
-            type: "tool-input-error",
-            id: "call_invalid",
-            name: "lookup",
-            raw: '{"query":"partial',
-          },
+          { type: "tool-input-end", id: "call_invalid", name: "lookup" },
+          { type: "tool-call", id: "call_invalid", name: "lookup", input: { query: "partial" } },
         ],
       })
     }),

packages/app/e2e/performance/timeline-stability/fixture.ts

@@ -899,8 +899,26 @@ function provider() {
         name: "OpenCode",
         models: { "claude-opus-4-6": { id: "claude-opus-4-6", name: "Claude Opus 4.6", limit: { context: 200_000 } } },
       },
+      {
+        id: "company-gateway",
+        name: "Company Gateway",
+        models: {
+          "fast-nano": {
+            id: "fast-nano",
+            api: { id: "openai/gpt-5.4-nano" },
+            name: "GPT-5.4 nano",
+            limit: { context: 128_000 },
+          },
+          "long-context": {
+            id: "long-context",
+            api: { id: "company/long-context" },
+            name: "Company Gateway Extra Long Context Model for Narrow Timeline Layouts",
+            limit: { context: 128_000 },
+          },
+        },
+      },
     ],
-    connected: ["opencode"],
+    connected: ["opencode", "company-gateway"],
     default: { providerID: "opencode", modelID: "claude-opus-4-6" },
   }
 }

packages/app/e2e/regression/home-session-actions.spec.ts

@@ -0,0 +1,106 @@
+import { expect, test } from "@playwright/test"
+import { fixture, pageMessages } from "../smoke/session-timeline.fixture"
+import { mockOpenCodeServer } from "../utils/mock-server"
+
+test("renames, exports, and deletes a home session from its context menu", async ({ page }) => {
+  const sessions = fixture.sessions.map((session) => ({ ...session }))
+  await mockOpenCodeServer(page, {
+    sessions,
+    provider: fixture.provider,
+    directory: fixture.directory,
+    project: fixture.project,
+    pageMessages,
+  })
+  await page.route("**/api/session/*/rename", async (route) => {
+    const sessionID = new URL(route.request().url()).pathname.split("/").at(-2)
+    const session = sessions.find((item) => item.id === sessionID)
+    const payload: unknown = route.request().postDataJSON()
+    if (!payload || typeof payload !== "object" || !("title" in payload) || typeof payload.title !== "string")
+      throw new Error("Invalid rename payload")
+    if (session) session.title = payload.title
+    await route.fulfill({ status: 204, headers: { "access-control-allow-origin": "*" } })
+  })
+  await page.addInitScript((directory) => {
+    localStorage.setItem(
+      "opencode.global.dat:server",
+      JSON.stringify({
+        projects: { local: [{ worktree: directory, expanded: true }] },
+        lastProject: { local: directory },
+      }),
+    )
+  }, fixture.directory)
+
+  await page.goto("/")
+  const row = page.locator('[data-component="home-session-row"]').filter({ hasText: fixture.expected.targetTitle })
+  await expect(row).toBeVisible()
+  const container = page.locator(`[data-component="home-session-row-container"][data-session-id="${fixture.targetID}"]`)
+  const titleBox = await container.locator('[data-component="home-session-title"]').boundingBox()
+  const avatarBox = await container.locator('[data-component="project-avatar-v2"]').boundingBox()
+  await expect(container.getByRole("button", { name: "More options" })).toHaveCount(0)
+
+  await row.focus()
+  await row.press("Shift+F10")
+  await expect(page.getByRole("menuitem", { name: "Rename" })).toBeVisible()
+  await page.keyboard.press("Escape")
+  await expect(page.getByRole("menuitem", { name: "Rename" })).toBeHidden()
+  await expect(row).toBeFocused()
+
+  const rowBox = await row.boundingBox()
+  await row.click({ button: "right", position: { x: 48, y: 12 } })
+  await expect(page).toHaveURL("/")
+  await expect(page.getByRole("menuitem", { name: "Rename" })).toBeVisible()
+  await expect(page.getByRole("menuitem", { name: "Export..." })).toBeVisible()
+  await expect(page.getByRole("menuitem", { name: "Delete..." })).toBeVisible()
+  const menuBox = await page.locator('[data-component="menu-v2-content"]').boundingBox()
+  expect(Math.abs((menuBox?.x ?? 0) - (rowBox?.x ?? 0) - 48)).toBeLessThan(4)
+
+  await page.getByRole("menuitem", { name: "Rename" }).click()
+  const title = page.locator('[data-component="home-session-rename"]')
+  await expect(title).toBeFocused()
+  await expect(title).toHaveValue(fixture.expected.targetTitle)
+  const editorBox = await title.boundingBox()
+  const editingAvatarBox = await container.locator('[data-component="project-avatar-v2"]').boundingBox()
+  expect(editorBox?.x).toBe(titleBox?.x)
+  expect(editingAvatarBox).toEqual(avatarBox)
+  expect(
+    await title.evaluate((element) => ({
+      outline: getComputedStyle(element).outlineStyle,
+      shadow: getComputedStyle(element).boxShadow,
+    })),
+  ).toEqual({ outline: "none", shadow: "none" })
+  expect(await container.evaluate((element) => getComputedStyle(element).outlineStyle)).toBe("none")
+  await title.fill("Renamed from Home")
+  const renamed = page.waitForRequest(
+    (request) => request.method() === "POST" && new URL(request.url()).pathname.endsWith("/rename"),
+  )
+  await title.press("Enter")
+  expect((await renamed).postDataJSON()).toEqual({ title: "Renamed from Home" })
+  let renamedRow = page.locator('[data-component="home-session-row"]').filter({ hasText: "Renamed from Home" })
+  await expect(renamedRow).toBeVisible()
+
+  await renamedRow.click()
+  await expect(page).toHaveURL(new RegExp(`/session/${fixture.targetID}$`))
+  await expect(page.locator('[data-slot="titlebar-tabs"] a').filter({ hasText: "Renamed from Home" })).toBeVisible()
+  await page.getByRole("button", { name: "Home" }).click()
+  await expect(page).toHaveURL("/")
+  renamedRow = page.locator('[data-component="home-session-row"]').filter({ hasText: "Renamed from Home" })
+  await expect(renamedRow).toBeVisible()
+
+  await renamedRow.click({ button: "right" })
+  const download = page.waitForEvent("download")
+  const exportItem = page.getByRole("menuitem", { name: "Export..." })
+  await exportItem.click()
+  expect((await download).suggestedFilename()).toBe("renamed-from-home.json")
+  await expect(exportItem).toBeHidden()
+
+  await renamedRow.click({ button: "right" })
+  await page.getByRole("menuitem", { name: "Delete..." }).click()
+  const dialog = page.getByRole("dialog")
+  await expect(dialog).toContainText('Delete session "Renamed from Home"?')
+  const removed = page.waitForRequest(
+    (request) => request.method() === "DELETE" && new URL(request.url()).pathname.endsWith(`/${fixture.targetID}`),
+  )
+  await dialog.getByRole("button", { name: "Delete session" }).click()
+  await removed
+  await expect(renamedRow).toBeHidden()
+})

packages/app/e2e/regression/new-session-workspace-branch.spec.ts

@@ -0,0 +1,53 @@
+import { expect, test } from "@playwright/test"
+import { mockOpenCodeServer } from "../utils/mock-server"
+import { expectAppVisible } from "../utils/waits"
+
+const draftID = "draft_new_session_workspace_branch"
+const directory = "C:/OpenCode/WorkspaceBranch"
+const server = `http://${process.env.PLAYWRIGHT_SERVER_HOST ?? "127.0.0.1"}:${process.env.PLAYWRIGHT_SERVER_PORT ?? "4096"}`
+
+test("selects a base branch for a new workspace", async ({ page }) => {
+  await mockOpenCodeServer(page, {
+    directory,
+    project: {
+      id: "proj_new_session_workspace_branch",
+      worktree: directory,
+      vcs: "git",
+      name: "workspace-branch",
+      time: { created: 1700000000000, updated: 1700000000000 },
+      sandboxes: [],
+    },
+    provider: { all: [], connected: [], default: {} },
+    sessions: [],
+    pageMessages: () => ({ items: [] }),
+    vcsBranches: ["feature/api", "main", "origin/release"],
+  })
+  await page.addInitScript(
+    ({ directory, draftID, server }) => {
+      localStorage.setItem(
+        "opencode.global.dat:server",
+        JSON.stringify({
+          projects: { local: [{ worktree: directory, expanded: true }] },
+          lastProject: { local: directory },
+        }),
+      )
+      localStorage.setItem(
+        "opencode.window.browser.dat:tabs",
+        JSON.stringify([{ type: "draft", draftID, server, directory }]),
+      )
+    },
+    { directory, draftID, server },
+  )
+
+  await page.goto(`/new-session?draftId=${draftID}`)
+  await expectAppVisible(page.locator('[data-component="composer-editor"]'))
+  await page.getByRole("button", { name: "Local", exact: true }).click()
+  await page.getByRole("menuitem", { name: "New workspace", exact: true }).click()
+  await page.getByRole("button", { name: "from main", exact: true }).click()
+  await page.getByRole("menuitemradio", { name: "feature/api", exact: true }).click()
+
+  const selected = page.getByRole("button", { name: "from feature/api", exact: true })
+  await expect(selected).toBeVisible()
+  await selected.click()
+  await expect(page.getByRole("menuitemradio", { name: "feature/api", exact: true })).toBeChecked()
+})

packages/app/e2e/regression/remote-session-settings.spec.ts

@@ -13,17 +13,28 @@ const sessionB = session("ses_server_b", directoryB, "Server B session")
 
 test("session settings use the remote server context", async ({ page }) => {
   const permissionRequests: string[] = []
+  const permissionResponses: PermissionResponse[] = []
   await installSseTransport(page, { server: serverA })
   await installSseTransport(page, { server: serverB })
-  await mockServers(page, permissionRequests)
+  // Server A has no tab and is never visited: a pending request there proves
+  // one toggle sweeps every connected server, not just the focused one.
+  await mockServers(page, permissionRequests, permissionResponses, {
+    pending: { [serverA]: [pendingPermission("permission-pending-a", sessionA.id)] },
+  })
   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()
@@ -38,11 +49,24 @@ test("session settings use the remote server context", async ({ page }) => {
       }),
     )
     .toBe(true)
-  expect(permissionRequests.every((request) => new URL(request).origin === serverB)).toBe(true)
+  await expect
+    .poll(() => permissionResponses)
+    .toEqual([
+      {
+        origin: serverA,
+        directory: undefined,
+        sessionID: sessionA.id,
+        permissionID: "permission-pending-a",
+        body: { reply: "once" },
+      },
+    ])
 
-  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 }) => {
@@ -63,7 +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
@@ -143,6 +167,99 @@ test("auto-accept responds for an unfocused server session", async ({ page }) =>
     ])
 })
 
+test("auto-accept sweeps again after a reconnect", async ({ page }) => {
+  const permissionRequests: string[] = []
+  const permissionResponses: PermissionResponse[] = []
+  const pendingA: MockPermission[] = []
+  const listFailures: Record<string, number> = {}
+  const sessionGets: string[] = []
+  await installSseTransport(page, { server: serverB })
+  const transport = await installSseTransport(page, { server: serverA, retry: 20 })
+  await mockServers(page, permissionRequests, permissionResponses, {
+    pending: { [serverA]: pendingA },
+    listFailures,
+    sessionGets,
+  })
+  await configureServers(page, [{ type: "session", server: serverA, sessionId: sessionA.id }])
+
+  await page.goto(`/server/${base64Encode(serverA)}/session/${sessionA.id}`)
+  await expect(page.getByRole("heading", { name: sessionA.title, exact: true })).toBeVisible()
+  const first = await transport.waitForConnection()
+
+  await page.keyboard.press("Control+,")
+  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
+    .poll(() =>
+      permissionRequests.some((request) => {
+        const url = new URL(request)
+        return url.origin === serverA && url.searchParams.get("location[directory]") === directoryA
+      }),
+    )
+    .toBe(true)
+  await page.keyboard.press("Escape")
+
+  // This request is asked while the client is disconnected, so it is never
+  // delivered as an event and only a reconnect sweep can find it. The first
+  // listing after the reconnect fails, so only the bounded sweep retry can
+  // deliver the reply.
+  pendingA.push(pendingPermission("permission-offline-a", sessionA.id))
+  listFailures[serverA] = 1
+  const syncsBeforeReconnect = sessionGets.length
+  await transport.disconnect()
+  await transport.waitForConnection({ after: first.id })
+
+  await expect
+    .poll(() => permissionResponses)
+    .toEqual([
+      {
+        origin: serverA,
+        directory: undefined,
+        sessionID: sessionA.id,
+        permissionID: "permission-offline-a",
+        body: { reply: "once" },
+      },
+    ])
+  // The reconnect sweep must resync active sessions instead of trusting
+  // cached locations, since another client may have moved them meanwhile.
+  expect(sessionGets.slice(syncsBeforeReconnect)).toContain(sessionA.id)
+})
+
+test("auto-accept approves a request discovered by opening a session", async ({ page }) => {
+  const permissionRequests: string[] = []
+  const permissionResponses: PermissionResponse[] = []
+  await installSseTransport(page, { server: serverA })
+  await installSseTransport(page, { server: serverB })
+  // The request is only served from the per-session permission list, so it
+  // reaches the client through the store sync when the session view opens,
+  // never through a location sweep or an event.
+  await mockServers(page, permissionRequests, permissionResponses, {
+    sessionPending: { [sessionA.id]: [pendingPermission("permission-synced-a", sessionA.id)] },
+  })
+  await configureServers(page, [{ type: "session", server: serverA, sessionId: sessionA.id }])
+
+  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.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
+    .poll(() => permissionResponses)
+    .toEqual([
+      {
+        origin: serverA,
+        directory: undefined,
+        sessionID: sessionA.id,
+        permissionID: "permission-synced-a",
+        body: { reply: "once" },
+      },
+    ])
+})
+
 type PermissionResponse = {
   origin: string
   directory?: string
@@ -151,6 +268,19 @@ type PermissionResponse = {
   body: unknown
 }
 
+type MockPermission = {
+  id: string
+  sessionID: string
+  action: string
+  resources: string[]
+  metadata: Record<string, unknown>
+  save: unknown[]
+}
+
+function pendingPermission(id: string, sessionID: string): MockPermission {
+  return { id, sessionID, action: "shell", resources: ["git status"], metadata: {}, save: [] }
+}
+
 async function configureServers(page: Page, tabs: { type: "session"; server: string; sessionId: string }[] = []) {
   await page.addInitScript(
     ({ serverB, tabs }) => {
@@ -161,7 +291,23 @@ async function configureServers(page: Page, tabs: { type: "session"; server: str
   )
 }
 
-async function mockServers(page: Page, permissionRequests: string[], permissionResponses: PermissionResponse[] = []) {
+type MockServerOptions = {
+  // Pending requests served from /api/permission/request, keyed by origin.
+  pending?: Record<string, MockPermission[]>
+  // Pending requests served from /api/session/:id/permission, keyed by session ID.
+  sessionPending?: Record<string, MockPermission[]>
+  // Counts of /api/permission/request calls to fail with a 500, keyed by origin.
+  listFailures?: Record<string, number>
+  // Records /api/session/:id GETs so tests can assert session resyncs.
+  sessionGets?: string[]
+}
+
+async function mockServers(
+  page: Page,
+  permissionRequests: string[],
+  permissionResponses: PermissionResponse[] = [],
+  options: MockServerOptions = {},
+) {
   await page.route("**/*", async (route) => {
     const url = new URL(route.request().url())
     if (url.origin !== serverA && url.origin !== serverB) return route.fallback()
@@ -178,8 +324,12 @@ async function mockServers(page: Page, permissionRequests: string[], permissionR
         permissionID: response[2]!,
         body: route.request().postDataJSON(),
       })
-      return json(route, true)
+      // The generated client requires exactly 204 for a successful reply.
+      return route.fulfill({ status: 204, headers: { "access-control-allow-origin": "*" } })
     }
+    const sessionPermission = url.pathname.match(/^\/api\/session\/([^/]+)\/permission$/)
+    if (route.request().method() === "GET" && sessionPermission)
+      return json(route, { data: options.sessionPending?.[sessionPermission[1]!] ?? [] })
     if (requestDirectory && requestDirectory !== directory) return json(route, { name: "InvalidDirectory" }, 500)
     if (url.pathname === "/api/provider")
       return json(route, {
@@ -197,7 +347,12 @@ async function mockServers(page: Page, permissionRequests: string[], permissionR
     if (url.pathname === "/api/agent") return json(route, { location: { directory }, data: [] })
     if (url.pathname === "/api/permission/request") {
       permissionRequests.push(url.toString())
-      return json(route, { location: { directory }, data: [] })
+      const failures = options.listFailures?.[url.origin] ?? 0
+      if (failures > 0) {
+        options.listFailures![url.origin] = failures - 1
+        return json(route, { name: "Internal" }, 500)
+      }
+      return json(route, { location: { directory }, data: options.pending?.[url.origin] ?? [] })
     }
     if (["/api/command", "/api/reference", "/api/question/request"].includes(url.pathname))
       return json(route, { location: { directory }, data: [] })
@@ -219,9 +374,13 @@ async function mockServers(page: Page, permissionRequests: string[], permissionR
       return json(route, { id: remote ? sessionB.projectID : "project-server-a", directory, canonical: directory })
     if (url.pathname === "/api/session")
       return json(route, { data: sessions.map((session) => currentSession(session)), cursor: {} })
-    if (url.pathname === "/api/session/active") return json(route, { data: {} })
+    if (url.pathname === "/api/session/active")
+      return json(route, { data: Object.fromEntries(sessions.map((session) => [session.id, { type: "running" }])) })
     const currentSessionInfo = sessions.find((session) => url.pathname === `/api/session/${session.id}`)
-    if (currentSessionInfo) return json(route, { data: currentSession(currentSessionInfo) })
+    if (currentSessionInfo) {
+      options.sessionGets?.push(currentSessionInfo.id)
+      return json(route, { data: currentSession(currentSessionInfo) })
+    }
     if (sessions.some((session) => url.pathname === `/api/session/${session.id}/message`))
       return json(route, { data: [], cursor: {} })
     if (sessions.some((session) => url.pathname === `/api/session/${session.id}/inbox`))

packages/app/e2e/regression/server-dialog-focus.spec.ts

@@ -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) {

packages/app/e2e/regression/session-message-revert.spec.ts

@@ -20,45 +20,47 @@ const messages = [
   },
   { id: "msg_second", type: "user", text: "Second prompt", time: { created: 4 } },
 ] satisfies SessionMessageInfo[]
+const session = {
+  id: sessionID,
+  slug: "session-message-revert",
+  projectID,
+  directory,
+  title: "Session message revert",
+  agent: "build",
+  model: { id: "test", providerID: "opencode" },
+  version: "dev",
+  time: { created: 1, updated: 4 },
+}
+const fixture = {
+  directory,
+  project: {
+    id: projectID,
+    worktree: directory,
+    canonical: directory,
+    vcs: "git",
+    name: "session-message-revert",
+    time: { created: 1, updated: 1 },
+    sandboxes: [],
+  },
+  provider: {
+    all: [
+      {
+        id: "opencode",
+        name: "OpenCode",
+        models: { test: { id: "test", name: "Test", variants: {}, limit: { context: 200_000 } } },
+      },
+    ],
+    connected: ["opencode"],
+    default: { providerID: "opencode", modelID: "test" },
+  },
+  pageMessages: () => ({ items: messages }),
+}
 
 test("reverts directly to the selected user message", async ({ page }) => {
   const staged: { sessionID: string; messageID: string }[] = []
   await mockOpenCodeServer(page, {
-    directory,
-    project: {
-      id: projectID,
-      worktree: directory,
-      canonical: directory,
-      vcs: "git",
-      name: "session-message-revert",
-      time: { created: 1, updated: 1 },
-      sandboxes: [],
-    },
-    provider: {
-      all: [
-        {
-          id: "opencode",
-          name: "OpenCode",
-          models: { test: { id: "test", name: "Test", variants: {}, limit: { context: 200_000 } } },
-        },
-      ],
-      connected: ["opencode"],
-      default: { providerID: "opencode", modelID: "test" },
-    },
-    sessions: [
-      {
-        id: sessionID,
-        slug: "session-message-revert",
-        projectID,
-        directory,
-        title: "Session message revert",
-        agent: "build",
-        model: { id: "test", providerID: "opencode" },
-        version: "dev",
-        time: { created: 1, updated: 4 },
-      },
-    ],
-    pageMessages: () => ({ items: messages }),
+    ...fixture,
+    sessions: [session],
     onRevertStage: (input) => staged.push(input),
   })
   await page.goto(`/server/${base64Encode(server)}/session/${sessionID}`)
@@ -77,3 +79,19 @@ test("reverts directly to the selected user message", async ({ page }) => {
   await expect(page.getByRole("textbox", { name: "Prompt" })).toHaveText("Second prompt")
   expect(staged).toEqual([{ sessionID, messageID: "msg_second" }])
 })
+
+test("hides revert actions in a child session", async ({ page }) => {
+  await mockOpenCodeServer(page, {
+    ...fixture,
+    sessions: [
+      { ...session, id: "ses_parent", slug: "parent", title: "Parent session" },
+      { ...session, parentID: "ses_parent" },
+    ],
+  })
+  await page.goto(`/server/${base64Encode(server)}/session/${sessionID}`)
+  await expectSessionTitle(page, "Session message revert")
+
+  const message = page.locator('[data-message-id="msg_second"]')
+  await message.hover()
+  await expect(message.getByRole("button", { name: "Revert message" })).toHaveCount(0)
+})

packages/app/e2e/regression/session-queue.spec.ts

@@ -0,0 +1,226 @@
+import { expect, test, type Page } from "@playwright/test"
+import type { OpenCodeEvent } from "@opencode-ai/client/promise"
+import { base64Encode } from "@opencode-ai/util/encode"
+import { mockOpenCodeServer } from "../utils/mock-server"
+import { expectAppVisible } from "../utils/waits"
+
+const directory = "C:/OpenCode/SessionQueueRegression"
+const projectID = "proj_session_queue_regression"
+const sessionID = "ses_session_queue_regression"
+const server = `http://${process.env.PLAYWRIGHT_SERVER_HOST ?? "127.0.0.1"}:${process.env.PLAYWRIGHT_SERVER_PORT ?? "4096"}`
+
+type InboxRow = {
+  id: string
+  sessionID: string
+  timeCreated: number
+  type: "user"
+  payload: { text: string; metadata?: Record<string, unknown> }
+  delivery: "steer" | "queue"
+}
+
+function createQueueMock(seed: string[]) {
+  const rows: InboxRow[] = seed.map((text, index) => ({
+    id: `inb_seed_${index + 1}`,
+    sessionID,
+    timeCreated: 1700000000000 + index,
+    type: "user",
+    payload: { text },
+    delivery: "queue",
+  }))
+  const events: OpenCodeEvent[] = []
+  const prompts: Record<string, unknown>[] = []
+  const changes: { inboxID: string; action: "cancel" | "steer" }[] = []
+  const log: string[] = []
+  let sequence = 0
+  const emit = (type: OpenCodeEvent["type"], data: OpenCodeEvent["data"]) => {
+    sequence += 1
+    events.push({
+      id: `evt_queue_${sequence}`,
+      type,
+      created: Date.now(),
+      durable: { aggregateID: sessionID, seq: sequence, version: 1 },
+      data,
+    } as OpenCodeEvent)
+  }
+  return {
+    rows,
+    prompts,
+    changes,
+    log,
+    events: () => events.splice(0),
+    onPrompt: (input: { sessionID: string; body: Record<string, unknown> }) => {
+      prompts.push(input.body)
+      log.push(`prompt:${String(input.body.delivery ?? "steer")}`)
+      const row: InboxRow = {
+        id: typeof input.body.id === "string" ? input.body.id : `inb_mock_${sequence}`,
+        sessionID: input.sessionID,
+        timeCreated: Date.now(),
+        type: "user",
+        payload: {
+          text: typeof input.body.text === "string" ? input.body.text : "",
+          ...(input.body.metadata === undefined ? {} : { metadata: input.body.metadata as Record<string, unknown> }),
+        },
+        delivery: input.body.delivery === "queue" ? "queue" : "steer",
+      }
+      rows.push(row)
+      emit("session.inbox.enqueued", {
+        sessionID: input.sessionID,
+        inboxID: row.id,
+        item: { type: "user", payload: row.payload, delivery: row.delivery },
+      })
+    },
+    onInboxChange: (input: { sessionID: string; inboxID: string; action: "cancel" | "steer" }) => {
+      changes.push({ inboxID: input.inboxID, action: input.action })
+      log.push(`${input.action}:${input.inboxID}`)
+      const index = rows.findIndex((row) => row.id === input.inboxID)
+      const row = rows[index]
+      if (!row) return
+      if (input.action === "cancel") {
+        rows.splice(index, 1)
+        emit("session.inbox.cancelled", { sessionID: input.sessionID, inboxID: input.inboxID })
+        return
+      }
+      row.delivery = "steer"
+      emit("session.inbox.delivery.changed", {
+        sessionID: input.sessionID,
+        inboxID: input.inboxID,
+        delivery: "steer",
+      })
+    },
+  }
+}
+
+async function openSession(page: Page, mock: ReturnType<typeof createQueueMock>, followUpBehavior?: "queue" | "steer") {
+  if (followUpBehavior) {
+    await page.addInitScript(
+      (behavior) => localStorage.setItem("settings.v3", JSON.stringify({ general: { followUpBehavior: behavior } })),
+      followUpBehavior,
+    )
+  }
+  await mockOpenCodeServer(page, {
+    directory,
+    project: {
+      id: projectID,
+      worktree: directory,
+      vcs: "git",
+      name: "session-queue-regression",
+      time: { created: 1700000000000, updated: 1700000000000 },
+      sandboxes: [],
+    },
+    provider: {
+      all: [
+        {
+          id: "opencode",
+          name: "OpenCode",
+          models: { "queue-model": { id: "queue-model", name: "Queue Model", limit: { context: 200_000 } } },
+        },
+      ],
+      connected: ["opencode"],
+      default: { providerID: "opencode", modelID: "queue-model" },
+    },
+    sessions: [
+      {
+        id: sessionID,
+        slug: "session-queue-regression",
+        projectID,
+        directory,
+        title: "Session queue regression",
+        version: "dev",
+        time: { created: 1700000000000, updated: 1700000000000 },
+      },
+    ],
+    pageMessages: () => ({ items: [] }),
+    sessionStatus: () => ({ [sessionID]: { type: "running" } }),
+    inbox: () => mock.rows.map((row) => ({ ...row, payload: { ...row.payload } })),
+    onPrompt: mock.onPrompt,
+    onInboxChange: mock.onInboxChange,
+    events: mock.events,
+  })
+  await page.goto(`/server/${base64Encode(server)}/session/${sessionID}`)
+  const composer = page.locator('[data-component="composer"]')
+  await expectAppVisible(composer)
+  return {
+    composer,
+    input: composer.locator('[data-component="composer-editor"]'),
+    rows: page.locator('[data-component="session-queue-row"]'),
+  }
+}
+
+test("follow-up preference controls Enter while Mod+Enter uses the alternate delivery", async ({ page }) => {
+  const mock = createQueueMock([])
+  const view = await openSession(page, mock, "queue")
+
+  await view.input.fill("queue this follow-up")
+  await expect(view.composer.locator('[data-action="composer-alternate-delivery"]')).toContainText("Steer")
+  await view.input.press("Enter")
+  await expect(view.rows.getByText("queue this follow-up", { exact: true })).toBeVisible()
+
+  await view.input.fill("steer this correction")
+  await view.input.press("ControlOrMeta+Enter")
+  await expect.poll(() => mock.prompts.map((prompt) => prompt.delivery)).toEqual(["queue", "steer"])
+  await expect(view.input).toHaveText("")
+})
+
+test("dragging reorders queued prompts", async ({ page }) => {
+  const mock = createQueueMock(["first queued prompt", "second queued prompt", "third queued prompt"])
+  const view = await openSession(page, mock)
+  await expect(view.rows).toHaveCount(3)
+
+  const first = view.rows.filter({ hasText: "first queued prompt" })
+  const third = view.rows.filter({ hasText: "third queued prompt" })
+  await first.getByRole("button", { name: "Reorder queued prompt" }).hover()
+  await page.mouse.down()
+  const target = await third.boundingBox()
+  if (!target) throw new Error("The target queue row is not visible")
+  await page.mouse.move(target.x + target.width / 2, target.y + target.height / 2, { steps: 10 })
+  await page.mouse.up()
+
+  await expect(view.rows.locator('[data-action="session-queue-edit"]')).toHaveText([
+    "second queued prompt",
+    "third queued prompt",
+    "first queued prompt",
+  ])
+  expect(mock.prompts.map((prompt) => prompt.text)).toEqual([
+    "second queued prompt",
+    "third queued prompt",
+    "first queued prompt",
+  ])
+  expect(mock.changes).toEqual([
+    { inboxID: "inb_seed_1", action: "cancel" },
+    { inboxID: "inb_seed_2", action: "cancel" },
+    { inboxID: "inb_seed_3", action: "cancel" },
+  ])
+})
+
+test("editing restores the existing draft and replaces only the original queue position", async ({ page }) => {
+  const mock = createQueueMock(["first queued prompt", "tighten the error copy", "third queued prompt"])
+  const view = await openSession(page, mock)
+  const original = view.rows.getByText("tighten the error copy", { exact: true })
+  await expect(original).toBeVisible()
+
+  await view.input.fill("my in-progress draft")
+  await original.click()
+  await expect(view.input).toHaveText("tighten the error copy")
+  await view.input.press("Escape")
+  await expect(view.input).toHaveText("my in-progress draft")
+
+  await original.click()
+  await expect(view.input).toHaveText("tighten the error copy")
+  await view.input.fill("tighten the error copy and add a retry hint")
+  await view.input.press("Enter")
+
+  await expect(view.rows.locator('[data-action="session-queue-edit"]')).toHaveText([
+    "first queued prompt",
+    "tighten the error copy and add a retry hint",
+    "third queued prompt",
+  ])
+  await expect(view.input).toHaveText("my in-progress draft")
+  expect(mock.prompts.map((prompt) => prompt.text)).toEqual([
+    "tighten the error copy and add a retry hint",
+    "tighten the error copy and add a retry hint",
+    "third queued prompt",
+  ])
+  expect(mock.prompts.every((prompt) => prompt.delivery === "queue" && prompt.resume === false)).toBe(true)
+  expect(mock.changes.map((change) => change.action)).toEqual(["cancel", "cancel", "cancel"])
+  expect(mock.log[0]).toBe("prompt:queue")
+})

packages/app/e2e/regression/session-timeline-accessibility.spec.ts

@@ -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")

packages/app/e2e/regression/session-timeline-collapse-state.spec.ts

@@ -139,7 +139,7 @@ test.describe("regression: session timeline local row state", () => {
     expect(siblingProbe).toEqual({
       fileMarker: "before",
       frameMarker: "before",
-      rowKey: `assistant-part:part:${assistantMessageID}:${editPartID}`,
+      rowKey: `assistant-part:file:part:${assistantMessageID}:${editPartID}`,
       rowMarker: "before",
       shadowRoots: 0,
       toolMarker: "before",

packages/app/e2e/regression/session-timeline-context-resize.spec.ts

@@ -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(