refactor(tui): simplify plugin update controls

fix(tui): refine plugin update states
docs(cli): document plugin update controls
2026-08-27 20:16:17 +00:00 · 2026-08-25 22:02:54 -04:00 · 2026-08-25 22:02:54 -04:00 · 2026-08-25 22:02:54 -04:00 · 2026-08-25 22:02:54 -04:00 · 2026-08-25 22:02:54 -04:00
620 changed files with 10740 additions and 34087 deletions
@@ -1,5 +0,0 @@
---
-"@opencode-ai/core": patch
---
-
-Propagate fatal Git ignore-check errors during index refresh so best-effort snapshot capture returns unavailable instead of treating the check as having no matches.
@@ -1,5 +0,0 @@
---
-"@opencode-ai/core": patch
---
-
-Rename the write, patch, and question tool formatting helpers from `toModelOutput` to `toModelContent` to match the result field they populate. Direct imports of these helpers must use the new name; generated content and declared machine output are unchanged.
@@ -1,5 +0,0 @@
---
-"@opencode-ai/core": minor
---
-
-Rename ModelResolver.supported to ModelResolver.hasPackage. Consumers of the old export must update the name; the predicate remains Boolean(model.package), checking only whether a catalog model declares a provider package, not whether it can be loaded. Default-model selection behavior is unchanged.
@@ -1,5 +0,0 @@
---
-"@opencode-ai/core": patch
---
-
-Include both paths of renamed files in new snapshot change lists so undo restores the original file instead of only deleting the renamed file.
@@ -1,10 +0,0 @@
---
-"@opencode-ai/core": patch
---
-
-Make the experimental portable shell scanner authoritative, with no Tree-sitter
-fallback. Scan common Bash and PowerShell control flow, heredocs, functions,
-expressions, quoting, and substitutions natively. Preserve existing redirect and
-declaration permission matching, and make PowerShell saved approvals cover the
-original command spelling. Parser failures remain visible without changing the
-permission engine. The default Tree-sitter path is unchanged.
@@ -1,5 +0,0 @@
---
-"@opencode-ai/core": patch
---
-
-Use parent directory names for SKILL.md IDs so individually downloaded skills do not overwrite each other.
@@ -42,7 +42,7 @@ jobs:
      - 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_BASE: ${{ github.event.pull_request.base.sha || github.event.before }}
          TURBO_SCM_HEAD: ${{ github.sha }}
        run: |
          if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
@@ -119,7 +119,7 @@ jobs:
          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_BASE: ${{ github.event.pull_request.base.sha || github.event.before }}
          TURBO_SCM_HEAD: ${{ github.sha }}

      - name: Verify published codemode package
@@ -137,7 +137,7 @@ jobs:
          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_BASE: ${{ github.event.pull_request.base.sha || github.event.before }}
          TURBO_SCM_HEAD: ${{ github.sha }}

      - name: Verify compiled service lifecycle
@@ -237,19 +237,13 @@ jobs:
        working-directory: packages/app
        run: bunx playwright install chromium

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

-      - name: Verify service worker precaching and upgrades
-        if: env.E2E_ENABLED == 'true'
-        working-directory: packages/app
-        run: bunx playwright test --config e2e/service-worker/playwright.config.ts
-        timeout-minutes: 5
-
      - name: Upload Playwright artifacts
        if: always() && env.E2E_ENABLED == 'true'
        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
@@ -125,7 +125,6 @@
        "@effect/platform-node": "catalog:",
        "@opencode-ai/client": "workspace:*",
        "@opencode-ai/plugin": "workspace:*",
-        "@opencode-ai/pty": "0.1.13",
        "@opencode-ai/schema": "workspace:*",
        "@opencode-ai/server": "workspace:*",
        "@opencode-ai/tui": "workspace:*",
@@ -348,8 +347,10 @@
        "@ai-sdk/anthropic": "3.0.82",
        "@ai-sdk/azure": "3.0.88",
        "@ai-sdk/cohere": "3.0.27",
+        "@ai-sdk/deepinfra": "2.0.41",
        "@ai-sdk/gateway": "3.0.104",
        "@ai-sdk/google-vertex": "4.0.128",
+        "@ai-sdk/groq": "3.0.31",
        "@ai-sdk/mistral": "3.0.51",
        "@ai-sdk/openai-compatible": "2.0.41",
        "@ai-sdk/perplexity": "3.0.26",
@@ -364,7 +365,6 @@
        "@opencode-ai/ai": "workspace:*",
        "@opencode-ai/codemode": "workspace:*",
        "@opencode-ai/plugin": "workspace:*",
-        "@opencode-ai/pty": "0.1.13",
        "@opencode-ai/schema": "workspace:*",
        "@opencode-ai/util": "workspace:*",
        "@parcel/watcher": "2.5.1",
@@ -553,20 +553,6 @@
        "@typescript/native-preview": "catalog:",
      },
    },
-    "packages/latex": {
-      "name": "@opencode-ai/latex",
-      "version": "0.0.0",
-      "dependencies": {
-        "@opencode-ai/plugin": "workspace:*",
-        "@opentui/core": "catalog:",
-        "string-width": "catalog:",
-      },
-      "devDependencies": {
-        "@tsconfig/bun": "catalog:",
-        "@types/bun": "catalog:",
-        "@typescript/native-preview": "catalog:",
-      },
-    },
    "packages/merman": {
      "name": "@opencode-ai/merman",
      "version": "0.0.0",
@@ -745,7 +731,6 @@
      },
      "devDependencies": {
        "@happy-dom/global-registrator": "20.0.11",
-        "@playwright/test": "catalog:",
        "@tsconfig/node22": "catalog:",
        "@types/bun": "catalog:",
        "@types/luxon": "catalog:",
@@ -851,7 +836,6 @@
        "@opencode-ai/client": "workspace:*",
        "@opencode-ai/session-ui": "workspace:*",
        "@opencode-ai/ui": "workspace:*",
-        "@playwright/test": "catalog:",
        "@solidjs/meta": "catalog:",
        "@storybook/addon-a11y": "10.4.4",
        "@storybook/addon-docs": "10.4.4",
@@ -893,7 +877,6 @@
      "dependencies": {
        "@opencode-ai/client": "workspace:*",
        "@opencode-ai/core": "workspace:*",
-        "@opencode-ai/latex": "workspace:*",
        "@opencode-ai/merman": "workspace:*",
        "@opencode-ai/plugin": "workspace:*",
        "@opencode-ai/schema": "workspace:*",
@@ -1003,6 +986,7 @@
        "mime-types": "3.0.2",
        "minimatch": "10.2.5",
        "npm-package-arg": "13.0.2",
+        "pacote": "21.5.1",
        "resolve.exports": "catalog:",
      },
      "devDependencies": {
@@ -1012,6 +996,7 @@
        "@types/node": "catalog:",
        "@types/npm-package-arg": "6.1.4",
        "@types/npmcli__arborist": "6.3.3",
+        "@types/pacote": "11.1.8",
        "@typescript/native-preview": "catalog:",
      },
    },
@@ -1195,6 +1180,8 @@

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

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

    "@ai-sdk/elevenlabs": ["@ai-sdk/elevenlabs@2.0.52", "", { "dependencies": { "@ai-sdk/provider": "3.0.15", "@ai-sdk/provider-utils": "4.0.46" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-ZgkausouWvO9U4ZtowNJ093bSNYOvH8zqls3uLC3+oxzWvbbTZO8SOdmFk0+gGafsXFJvq2yUX9+rEeJPwOJLw=="],
@@ -2165,8 +2152,6 @@

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

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

    "@opencode-ai/plugin": ["@opencode-ai/plugin@workspace:packages/plugin"],
@@ -2175,20 +2160,6 @@

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

-    "@opencode-ai/pty": ["@opencode-ai/pty@0.1.13", "", { "optionalDependencies": { "@opencode-ai/pty-darwin-arm64": "0.1.13", "@opencode-ai/pty-darwin-x64": "0.1.13", "@opencode-ai/pty-linux-arm64-gnu": "0.1.13", "@opencode-ai/pty-linux-arm64-musl": "0.1.13", "@opencode-ai/pty-linux-x64-gnu": "0.1.13", "@opencode-ai/pty-linux-x64-musl": "0.1.13" }, "bin": { "opencode-pty": "bin/opencode-pty.js" } }, "sha512-WPCN8h8HaZhhUcrMG0zu+4D9vco0EZiEg/gCF1K3JPRN6UsHMiXq1HVIy5IlyfcoyjfViRmQmXYE4AuU3laBjA=="],
-
-    "@opencode-ai/pty-darwin-arm64": ["@opencode-ai/pty-darwin-arm64@0.1.13", "", { "os": "darwin", "cpu": "arm64" }, "sha512-fVtQZqVLBuJx/aB+5ojfmQifS1KMc9gxlxpFQ6bxEFU8tn8xHQTiFPaNroZgOtaw7I4ceGyx/eXieK1wp68yAA=="],
-
-    "@opencode-ai/pty-darwin-x64": ["@opencode-ai/pty-darwin-x64@0.1.13", "", { "os": "darwin", "cpu": "x64" }, "sha512-b/tAEm0hCMXraPM9cxR8Rg7X1UBZInRTaxWAS4Ht9eH1nWj1rANOLvHWiWX/vVh5TB0Ubg8bWPu4B0nZkEHROQ=="],
-
-    "@opencode-ai/pty-linux-arm64-gnu": ["@opencode-ai/pty-linux-arm64-gnu@0.1.13", "", { "os": "linux", "cpu": "arm64" }, "sha512-I124aSYBBjpGZnYExHfIajkvVK1FiK+//OJBGdqqFp5pas2Oruq4O8tv+pMoxomZIYh2ce/QhOOYLHRwXsthTg=="],
-
-    "@opencode-ai/pty-linux-arm64-musl": ["@opencode-ai/pty-linux-arm64-musl@0.1.13", "", { "os": "linux", "cpu": "arm64" }, "sha512-feWsfKpaDytGJzutoK43GqQwVghG2vHZt6BE/ydPZNuqIrySQ/6JfliUAMwn5BWs/Ky7ouSwKHCyAVeukusSvg=="],
-
-    "@opencode-ai/pty-linux-x64-gnu": ["@opencode-ai/pty-linux-x64-gnu@0.1.13", "", { "os": "linux", "cpu": "x64" }, "sha512-jliNgsevGuxfIeX7eyzjHhrJkF8uEUPnDLbF2v16uv69FhEHrraf7jyWkxazMP6rNvn2CGtwMMc4BXPS5pzjhg=="],
-
-    "@opencode-ai/pty-linux-x64-musl": ["@opencode-ai/pty-linux-x64-musl@0.1.13", "", { "os": "linux", "cpu": "x64" }, "sha512-rXDpidW66gz2b2M/NbUN8ZKmAxaJcASnuHATeXevlrFdiPUv8uJwvkRd6Pla1fp01Q65MkBmgRa7Q9c+H1PlzA=="],
-
    "@opencode-ai/schema": ["@opencode-ai/schema@workspace:packages/schema"],

    "@opencode-ai/script": ["@opencode-ai/script@workspace:packages/script"],
@@ -5941,6 +5912,10 @@

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

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

    "@ai-sdk/deepseek/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.38", "", { "dependencies": { "@ai-sdk/provider": "3.0.14", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.8" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-/HHGmtKllqjg1OLc023v9w9kK3laW7Z6TzfZukYQWCsGBbzB9p60zTvvpXFVcs44NZBVXL3viOa1HRKUbeee8g=="],
@@ -2,7 +2,7 @@
 exact = true
 # Only install newly resolved package versions published at least 3 days ago.
 minimumReleaseAge = 259200
-minimumReleaseAgeExcludes = ["@ai-sdk/amazon-bedrock", "@ai-sdk/anthropic", "@brendonovich/vite-plugin-opencode", "@opencode-ai/sdk", "@opencode-ai/pty", "@opencode-ai/pty-darwin-arm64", "@opencode-ai/pty-darwin-x64", "@opencode-ai/pty-linux-arm64-gnu", "@opencode-ai/pty-linux-arm64-musl", "@opencode-ai/pty-linux-x64-gnu", "@opencode-ai/pty-linux-x64-musl", "@opentui/core", "@opentui/core-darwin-arm64", "@opentui/core-darwin-x64", "@opentui/core-linux-arm64", "@opentui/core-linux-arm64-musl", "@opentui/core-linux-x64", "@opentui/core-linux-x64-musl", "@opentui/core-win32-arm64", "@opentui/core-win32-x64", "@opentui/keymap", "@opentui/solid", "opentui-spinner", "gitlab-ai-provider", "opencode-gitlab-auth", "@ff-labs/fff-node", "@ff-labs/fff-bun", "@ff-labs/fff-bin-darwin-arm64", "@ff-labs/fff-bin-darwin-x64", "@ff-labs/fff-bin-linux-arm64-gnu", "@ff-labs/fff-bin-linux-arm64-musl", "@ff-labs/fff-bin-linux-x64-gnu", "@ff-labs/fff-bin-linux-x64-musl", "@ff-labs/fff-bin-win32-arm64", "@ff-labs/fff-bin-win32-x64", "@pierre/diffs", "@pierre/theming", "app-builder-lib", "dmg-builder", "electron-builder", "electron-publish", "blume"]
+minimumReleaseAgeExcludes = ["@ai-sdk/amazon-bedrock", "@ai-sdk/anthropic", "@brendonovich/vite-plugin-opencode", "@opencode-ai/sdk", "@opentui/core", "@opentui/core-darwin-arm64", "@opentui/core-darwin-x64", "@opentui/core-linux-arm64", "@opentui/core-linux-arm64-musl", "@opentui/core-linux-x64", "@opentui/core-linux-x64-musl", "@opentui/core-win32-arm64", "@opentui/core-win32-x64", "@opentui/keymap", "@opentui/solid", "opentui-spinner", "gitlab-ai-provider", "opencode-gitlab-auth", "@ff-labs/fff-node", "@ff-labs/fff-bun", "@ff-labs/fff-bin-darwin-arm64", "@ff-labs/fff-bin-darwin-x64", "@ff-labs/fff-bin-linux-arm64-gnu", "@ff-labs/fff-bin-linux-arm64-musl", "@ff-labs/fff-bin-linux-x64-gnu", "@ff-labs/fff-bin-linux-x64-musl", "@ff-labs/fff-bin-win32-arm64", "@ff-labs/fff-bin-win32-x64", "@pierre/diffs", "@pierre/theming", "app-builder-lib", "dmg-builder", "electron-builder", "electron-publish", "blume"]

 [test]
 root = "./do-not-run-tests-from-root"
@@ -1,8 +1,8 @@
 {
  "nodeModules": {
-    "x86_64-linux": "sha256-agxF1Ab8P84tijHvw7jaG+9UZm/dJZhHw8CGNlQQqYc=",
-    "aarch64-linux": "sha256-Kg6CkhoEEsb+ZOGPN/rnNoirLdGSyQl6HU/Fy8t9Mqs=",
-    "aarch64-darwin": "sha256-SnxBpmcfPL2p3OYmSuG5Y/aFW8PeEOLhfQrxAs/53fs=",
-    "x86_64-darwin": "sha256-Dnxl1EcfiE7/rlHmPkko0BKc6hHVk6rQ/2h/PWuLfNo="
+    "x86_64-linux": "sha256-fOM/kGJJ1cipCHQIxioDZEB7NZykpSiqgwm7gIS6THI=",
+    "aarch64-linux": "sha256-XTY2C33HjsBMWO7VIiWc2MynjJrxbTLrOJ+6pM+afI0=",
+    "aarch64-darwin": "sha256-wX6+bC18djtPZ7A9ch+wryM7tDFfrAlT0xx0QTk6EJQ=",
+    "x86_64-darwin": "sha256-dcRRX4bYq5AmG4GcVmYq/M+06dlf4KJHn+clT2JY48g="
  }
 }
@@ -19,14 +19,6 @@ Per-type constructors live on the type, not as top-level re-exports. Use `Messag
 - Use `testEffect(...)` from `test/lib/effect.ts` for tests requiring Effect layers.
 - Keep provider tests fixture-first. Live provider calls must stay behind `RECORD=true` and required API-key checks.

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

 This package is an Effect Schema-first LLM core. The Schema classes in `src/schema/` are the canonical runtime data model. Convenience functions in `src/llm.ts` are thin constructors that return those same Schema class instances; they should improve callsites without creating a second model.
@@ -36,12 +36,7 @@ const resolve = (policy: CachePolicy | undefined): CachePolicyObject => {
 // Protocols whose wire format ignores inline cache markers (OpenAI's implicit
 // prefix caching, Gemini's implicit + out-of-band CachedContent). Skip the
 // whole policy pass for these — emitting hints would be harmless but pointless.
-const RESPECTS_INLINE_HINTS = new Set([
-  "anthropic-messages",
-  "google-vertex-messages",
-  "bedrock-converse",
-  "openrouter",
-])
+const RESPECTS_INLINE_HINTS = new Set(["anthropic-messages", "google-vertex-messages", "bedrock-converse", "openrouter"])

 const makeHint = (ttlSeconds: number | undefined): CacheHint =>
  ttlSeconds !== undefined ? new CacheHint({ type: "ephemeral", ttlSeconds }) : new CacheHint({ type: "ephemeral" })
@@ -1,7 +1,7 @@
 import { Effect, Schema } from "effect"
 import {
  HttpOptions,
-  InvalidRequestError,
+  InvalidRequestReason,
  AIError,
  ModelID,
  ProviderID,
@@ -158,10 +158,9 @@ export function generate(input: ImageRequest | ImageRequestInput) {
    try: () => (input instanceof ImageRequest ? input : request(input)),
    catch: (error) =>
      new AIError({
-        reason: new InvalidRequestError({
-          message: error instanceof Error ? error.message : String(error),
-          cause: error,
-        }),
+        module: "Image",
+        method: "generate",
+        reason: new InvalidRequestReason({ message: error instanceof Error ? error.message : String(error) }),
      }),
  }).pipe(Effect.flatMap((request) => ImageClient.generate(request as unknown as ImageRequestFor<ImageOptions>)))
 }
@@ -3,7 +3,7 @@ import { LLMClient, Service } from "./route/client.js"
 import {
  GenerationOptions,
  HttpOptions,
-  InvalidProviderOutputError,
+  InvalidProviderOutputReason,
  AIError,
  LLMEvent,
  LLMRequest,
@@ -116,7 +116,9 @@ const runGenerateObject = Effect.fn("LLM.generateObject")(function* (
  )
  if (!call || !LLMEvent.is.toolCall(call))
    return yield* new AIError({
-      reason: new InvalidProviderOutputError({
+      module: "LLM",
+      method: "generateObject",
+      reason: new InvalidProviderOutputReason({
        message: `generateObject: model did not call the forced \`${GENERATE_OBJECT_TOOL_NAME}\` tool`,
      }),
    })
@@ -124,9 +126,10 @@ const runGenerateObject = Effect.fn("LLM.generateObject")(function* (
    Effect.mapError(
      (error) =>
        new AIError({
-          reason: new InvalidProviderOutputError({
+          module: "LLM",
+          method: "generateObject",
+          reason: new InvalidProviderOutputReason({
            message: `generateObject: tool input failed schema decode: ${error.message}`,
-            cause: error,
          }),
        }),
    ),
@@ -69,22 +69,14 @@ export interface OptionsInput {
  // SDK Metadata:2649 {user_id?: string | null}
  readonly metadata?: { readonly user_id?: string | null }
  // SDK MessageCreateParamsContainer:2596 ContainerParams|string
-  readonly container?:
-    | string
-    | { readonly id?: string | null; readonly skills?: ReadonlyArray<Record<string, unknown>> | null }
+  readonly container?: string | { readonly id?: string | null; readonly skills?: ReadonlyArray<Record<string, unknown>> | null }
  readonly inference_geo?: string | null
  readonly inferenceGeo?: string | null
  readonly cache_control?: { readonly type: "ephemeral"; readonly ttl?: "5m" | "1h" }
  readonly cacheControl?: { readonly type: "ephemeral"; readonly ttl?: "5m" | "1h" }
  // SDK OutputConfig:2684 {effort, format: JSONOutputFormat}
-  readonly output_config?: {
-    readonly effort?: string | null
-    readonly format?: { readonly type: "json_schema"; readonly schema: Record<string, unknown> } | null
-  }
-  readonly outputConfig?: {
-    readonly effort?: string | null
-    readonly format?: { readonly type: "json_schema"; readonly schema: Record<string, unknown> } | null
-  }
+  readonly output_config?: { readonly effort?: string | null; readonly format?: { readonly type: "json_schema"; readonly schema: Record<string, unknown> } | null }
+  readonly outputConfig?: { readonly effort?: string | null; readonly format?: { readonly type: "json_schema"; readonly schema: Record<string, unknown> } | null }
 }

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

 const AnthropicThinking = Schema.Union([
@@ -406,7 +394,6 @@ const AnthropicEvent = Schema.Struct({
 type AnthropicEvent = Schema.Schema.Type<typeof AnthropicEvent>

 interface ParserState {
-  readonly providerMetadataKey: string
  readonly tools: ToolStream.State<number>
  readonly reasoningSignatures: Readonly<Record<number, string>>
  readonly usage?: Usage
@@ -441,18 +428,18 @@ const cacheControl = (breakpoints: Cache.Breakpoints, cache: CacheHint | undefin
  return Cache.ttlBucket(cache.ttlSeconds) === "1h" ? EPHEMERAL_1H : EPHEMERAL_5M
 }

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

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

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

 const lowerTool = (breakpoints: Cache.Breakpoints, tool: ToolDefinition, inputSchema: JsonSchema): AnthropicTool => ({
@@ -512,21 +499,14 @@ const serverToolResultType = (name: string): AnthropicServerToolResultType | und
  return undefined
 }

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

 const fileIdFromMetadata = (metadata: MediaPart["metadata"]): string | undefined => {
@@ -574,7 +554,9 @@ const documentContextFromMetadata = (metadata: MediaPart["metadata"]): string |
  return undefined
 }

-const citationsFromMetadata = (metadata: MediaPart["metadata"]): AnthropicDocumentBlock["citations"] | undefined => {
+const citationsFromMetadata = (
+  metadata: MediaPart["metadata"],
+): AnthropicDocumentBlock["citations"] | undefined => {
  if (!ProviderShared.isRecord(metadata)) return undefined
  const raw = ProviderShared.isRecord(metadata.anthropic)
    ? (metadata.anthropic.citations ?? metadata.citations)
@@ -724,7 +706,8 @@ const lowerToolResultContent = Effect.fnUntraced(function* (part: ToolResultPart
 })

 const requireThinkingSignature = (request: LLMRequest) => {
-  if (request.model.compatibility?.requireSignature !== undefined) return request.model.compatibility.requireSignature
+  if (request.model.compatibility?.requireSignature !== undefined)
+    return request.model.compatibility.requireSignature
  const provider = request.model.provider.toLowerCase()
  const model = request.model.id.toLowerCase()
  const baseURL = (request.model.route.endpoint.baseURL ?? "").toLowerCase()
@@ -808,7 +791,6 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
  breakpoints: Cache.Breakpoints,
 ) {
  const messages: AnthropicMessage[] = []
-  const providerMetadataKey = request.model.route.providerMetadataKey ?? String(request.model.provider)

  for (const [index, message] of request.messages.entries()) {
    if (message.role === "system") {
@@ -854,8 +836,8 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
        if (part.type === "reasoning") {
          // A signature marks visible thinking; only signature-less parts carrying
          // redactedData round-trip as opaque redacted_thinking blocks.
-          const signature = part.encrypted ?? signatureFromMetadata(part.providerMetadata, providerMetadataKey)
-          const redactedData = redactedDataFromMetadata(part.providerMetadata, providerMetadataKey)
+          const signature = part.encrypted ?? signatureFromMetadata(part.providerMetadata)
+          const redactedData = redactedDataFromMetadata(part.providerMetadata)
          if (signature === undefined && redactedData !== undefined) {
            content.push({ type: "redacted_thinking", data: redactedData })
            continue
@@ -884,7 +866,7 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
          continue
        }
        if (part.type === "tool-result" && part.providerExecuted) {
-          content.push(yield* lowerServerToolResult(part, providerMetadataKey))
+          content.push(yield* lowerServerToolResult(part))
          continue
        }
        return yield* invalid(
@@ -918,24 +900,21 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (

 const resolveOptions = Effect.fn("AnthropicMessages.resolveOptions")(function* (request: LLMRequest) {
  const input = request.providerOptions as Record<string, unknown> | undefined
-  const rawServiceTier =
-    (input as Record<string, unknown> | undefined)?.service_tier ??
-    (input as Record<string, unknown> | undefined)?.serviceTier
+  const rawServiceTier = (input as Record<string, unknown> | undefined)?.service_tier ?? (input as Record<string, unknown> | undefined)?.serviceTier
  const service_tier =
    rawServiceTier === "auto" || rawServiceTier === "standard_only"
      ? (rawServiceTier as "auto" | "standard_only")
      : undefined
  const rawMetadata = (input as Record<string, unknown> | undefined)?.metadata
  const metadata =
-    ProviderShared.isRecord(rawMetadata) && (typeof rawMetadata.user_id === "string" || rawMetadata.user_id === null)
+    ProviderShared.isRecord(rawMetadata) &&
+    (typeof rawMetadata.user_id === "string" || rawMetadata.user_id === null)
      ? { user_id: rawMetadata.user_id as string | null }
      : undefined
  const container =
    typeof (input as Record<string, unknown> | undefined)?.container === "string" ||
    ProviderShared.isRecord((input as Record<string, unknown> | undefined)?.container)
-      ? ((input as Record<string, unknown>).container as
-          | string
-          | { id?: string | null; skills?: ReadonlyArray<Record<string, unknown>> | null })
+      ? ((input as Record<string, unknown>).container as string | { id?: string | null; skills?: ReadonlyArray<Record<string, unknown>> | null })
      : undefined
  const rawInferenceGeo =
    (input as Record<string, unknown> | undefined)?.inference_geo ??
@@ -986,7 +965,8 @@ const resolveThinking = Effect.fn("AnthropicMessages.resolveThinking")(function*
    input.display === "summarized" || input.display === "omitted"
      ? (input.display as "summarized" | "omitted")
      : undefined
-  if (input.type === "adaptive") return { type: "adaptive" as const, ...(display === undefined ? {} : { display }) }
+  if (input.type === "adaptive")
+    return { type: "adaptive" as const, ...(display === undefined ? {} : { display }) }
  if (input.type === "disabled") return { type: "disabled" as const }
  if (input.type !== "enabled") return undefined
  const budget =
@@ -1074,7 +1054,7 @@ const mapFinishReason = (reason: string | null | undefined): FinishReason => {
 // inclusive `inputTokens` the rest of the contract expects. Extended
 // thinking tokens are included in `output_tokens`; newer responses also
 // expose that subset through `output_tokens_details.thinking_tokens`.
-const mapUsage = (usage: AnthropicUsage | undefined, providerMetadataKey: string): Usage | undefined => {
+const mapUsage = (usage: AnthropicUsage | undefined): Usage | undefined => {
  if (!usage) return undefined
  const nonCached = usage.input_tokens ?? undefined
  const cacheRead = usage.cache_read_input_tokens ?? undefined
@@ -1088,7 +1068,7 @@ const mapUsage = (usage: AnthropicUsage | undefined, providerMetadataKey: string
    cacheWriteInputTokens: cacheWrite,
    reasoningTokens: usage.output_tokens_details?.thinking_tokens,
    totalTokens: ProviderShared.totalTokens(inputTokens, usage.output_tokens, undefined),
-    providerMetadata: { [providerMetadataKey]: usage },
+    providerMetadata: { anthropic: usage },
  })
 }

@@ -1097,7 +1077,7 @@ const mapUsage = (usage: AnthropicUsage | undefined, providerMetadataKey: string
 // field prefers `right` when defined, falls back to `left`. `inputTokens` is
 // recomputed from the merged breakdown so the inclusive total stays
 // consistent with `nonCached + cacheRead + cacheWrite`.
-const mergeUsage = (left: Usage | undefined, right: Usage | undefined, providerMetadataKey: string) => {
+const mergeUsage = (left: Usage | undefined, right: Usage | undefined) => {
  if (!left) return right
  if (!right) return left
  const nonCachedInputTokens = right.nonCachedInputTokens ?? left.nonCachedInputTokens
@@ -1115,9 +1095,7 @@ const mergeUsage = (left: Usage | undefined, right: Usage | undefined, providerM
    reasoningTokens,
    totalTokens: ProviderShared.totalTokens(inputTokens, outputTokens, undefined),
    providerMetadata: {
-      [providerMetadataKey]:
-        mergeJsonRecords(left.providerMetadata?.[providerMetadataKey], right.providerMetadata?.[providerMetadataKey]) ??
-        {},
+      anthropic: mergeJsonRecords(left.providerMetadata?.["anthropic"], right.providerMetadata?.["anthropic"]) ?? {},
    },
  })
 }
@@ -1135,7 +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: AnthropicStreamBlock, providerMetadataKey: string): 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
@@ -1149,7 +1127,7 @@ const serverToolResultEvent = (block: AnthropicStreamBlock, providerMetadataKey:
    providerExecuted: true,
    // The complete payload is irreducible provider replay state: subsequent
    // stateless requests must round-trip the typed result block verbatim.
-    providerMetadata: providerMetadata(providerMetadataKey, { blockType: block.type, result: block.content }),
+    providerMetadata: anthropicMetadata({ blockType: block.type, result: block.content }),
  })
 }

@@ -1158,8 +1136,8 @@ type StepResult = readonly [ParserState, ReadonlyArray<LLMEvent>]
 const NO_EVENTS: StepResult["1"] = []

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

 const onContentBlockStart = (
@@ -1211,16 +1189,14 @@ const onContentBlockStart = (
  if (block.type === "thinking" && block.thinking !== undefined) {
    const events: LLMEvent[] = []
    const id = `reasoning-${event.index ?? 0}`
-    const metadata =
-      block.signature === undefined
-        ? undefined
-        : providerMetadata(state.providerMetadataKey, { signature: block.signature })
-    const lifecycle = Lifecycle.reasoningStart(state.lifecycle, events, id, metadata)
+    const providerMetadata =
+      block.signature === undefined ? undefined : anthropicMetadata({ signature: block.signature })
+    const lifecycle = Lifecycle.reasoningStart(state.lifecycle, events, id, providerMetadata)
    return [
      {
        ...state,
        lifecycle: block.thinking
-          ? Lifecycle.reasoningDelta(lifecycle, events, id, block.thinking, metadata)
+          ? Lifecycle.reasoningDelta(lifecycle, events, id, block.thinking, providerMetadata)
          : lifecycle,
        reasoningSignatures:
          event.index === undefined || block.signature === undefined
@@ -1243,14 +1219,14 @@ const onContentBlockStart = (
          state.lifecycle,
          events,
          `reasoning-${event.index ?? 0}`,
-          providerMetadata(state.providerMetadataKey, { redactedData: block.data }),
+          anthropicMetadata({ redactedData: block.data }),
        ),
      },
      events,
    ]
  }

-  const result = serverToolResultEvent(block, state.providerMetadataKey)
+  const result = serverToolResultEvent(block)
  if (!result) return [state, NO_EVENTS]
  const events: LLMEvent[] = []
  return [{ ...state, lifecycle: Lifecycle.stepStart(state.lifecycle, events) }, [...events, result]]
@@ -1330,7 +1306,7 @@ const onContentBlockStop = Effect.fn("AnthropicMessages.onContentBlockStop")(fun
        Lifecycle.textEnd(state.lifecycle, events, `text-${event.index}`),
        events,
        `reasoning-${event.index}`,
-        signature === undefined ? undefined : providerMetadata(state.providerMetadataKey, { signature }),
+        signature === undefined ? undefined : anthropicMetadata({ signature }),
      )
  events.push(...resultEvents)
  const reasoningSignatures = { ...state.reasoningSignatures }
@@ -1342,7 +1318,7 @@ const onMessageDelta = (
  state: ParserState,
  event: AnthropicEvent & { readonly delta?: AnthropicStreamDelta },
 ): StepResult => {
-  const usage = mergeUsage(state.usage, mapUsage(event.usage, state.providerMetadataKey), state.providerMetadataKey)
+  const usage = mergeUsage(state.usage, mapUsage(event.usage))
  return [
    {
      ...state,
@@ -1355,7 +1331,7 @@ const onMessageDelta = (
        providerMetadata:
          event.delta?.stop_sequence === null || event.delta?.stop_sequence === undefined
            ? undefined
-            : providerMetadata(state.providerMetadataKey, { stopSequence: event.delta.stop_sequence }),
+            : anthropicMetadata({ stopSequence: event.delta.stop_sequence }),
      },
    },
    NO_EVENTS,
@@ -1387,15 +1363,14 @@ const providerErrorMessage = (event: AnthropicEvent): string => {
  return message || type || "Anthropic Messages stream error"
 }

-const onError = (event: AnthropicEvent) => {
-  const message = providerErrorMessage(event)
-  const body = ProviderShared.encodeJson(event)
-  return Effect.fail(
+const onError = (event: AnthropicEvent) =>
+  Effect.fail(
    new AIError({
-      reason: classifyProviderFailure({ message, rawBody: body }),
+      module: ADAPTER,
+      method: "stream",
+      reason: classifyProviderFailure({ message: providerErrorMessage(event), code: event.error?.type }),
    }),
  )
-}

 const isKnownStreamBlockType = (type: string) =>
  type === "text" ||
@@ -1443,7 +1418,9 @@ const step = (state: ParserState, event: AnthropicEvent) => {
      if (event.index === undefined)
        return Effect.fail(ProviderShared.eventError(ADAPTER, `Anthropic ${block.type} missing index`))
      if (!block.id)
-        return Effect.fail(ProviderShared.eventError(ADAPTER, `Anthropic tool_use missing id at index ${event.index}`))
+        return Effect.fail(
+          ProviderShared.eventError(ADAPTER, `Anthropic tool_use missing id at index ${event.index}`),
+        )
    }
    return Effect.succeed(onContentBlockStart(state, { ...event, content_block: block }))
  }
@@ -1482,8 +1459,7 @@ export const protocol = Protocol.make({
  },
  stream: {
    event: Protocol.jsonEvent(AnthropicEvent),
-    initial: (request) => ({
-      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
+    initial: () => ({
      tools: ToolStream.empty<number>(),
      reasoningSignatures: {},
      lifecycle: Lifecycle.initial(),
@@ -1497,9 +1473,10 @@ export const route = Route.make({
  provider: "anthropic",
  providerMetadataKey: "anthropic",
  protocol,
-  endpoint: Endpoint.path((input) => (input.request.model.provider === "anthropic" ? `${PATH}?beta=true` : PATH), {
-    baseURL: DEFAULT_BASE_URL,
-  }),
+  endpoint: Endpoint.path(
+    (input) => (input.request.model.provider === "anthropic" ? `${PATH}?beta=true` : PATH),
+    { baseURL: DEFAULT_BASE_URL },
+  ),
  auth: Auth.none,
  framing,
  headers: () => ({ "anthropic-version": "2023-06-01" }),
@@ -258,21 +258,19 @@ const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>) =>
    tool: (name) => ({ tool: { name } }) as const,
  })

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

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

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

 const lowerToolCall = (part: ToolCallPart): BedrockToolUseBlock => ({
@@ -320,7 +318,6 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (
  breakpoints: BedrockCache.Breakpoints,
 ) {
  const messages: BedrockMessage[] = []
-  const providerMetadataKey = request.model.route.providerMetadataKey ?? String(request.model.provider)

  for (const message of request.messages) {
    if (message.role === "system") {
@@ -368,8 +365,8 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (
          continue
        }
        if (part.type === "reasoning") {
-          const signature = reasoningSignature(part, providerMetadataKey)
-          const redactedData = reasoningRedactedData(part, providerMetadataKey)
+          const signature = reasoningSignature(part)
+          const redactedData = reasoningRedactedData(part)
          if (signature === undefined && redactedData !== undefined) {
            content.push({ reasoningContent: { redactedContent: redactedData } })
            continue
@@ -469,7 +466,7 @@ const mapFinishReason = (reason: string): FinishReason => {

 // AWS reports inputTokens separately from cache reads and writes.
 // Bedrock does not break reasoning out of outputTokens for current models.
-const mapUsage = (usage: BedrockUsageSchema | undefined, providerMetadataKey: string): Usage | undefined => {
+const mapUsage = (usage: BedrockUsageSchema | undefined): Usage | undefined => {
  if (!usage) return undefined
  const inputTokens = ProviderShared.sumTokens(
    usage.inputTokens,
@@ -483,12 +480,11 @@ const mapUsage = (usage: BedrockUsageSchema | undefined, providerMetadataKey: st
    cacheReadInputTokens: usage.cacheReadInputTokens,
    cacheWriteInputTokens: usage.cacheWriteInputTokens,
    totalTokens: ProviderShared.totalTokens(inputTokens, usage.outputTokens, usage.totalTokens),
-    providerMetadata: { [providerMetadataKey]: usage },
+    providerMetadata: { bedrock: usage },
  })
 }

 interface ParserState {
-  readonly providerMetadataKey: string
  readonly tools: ToolStream.State<number>
  // Bedrock splits the finish into `messageStop` (carries `stopReason`) and
  // `metadata` (carries usage). Hold the terminal event in state so `onHalt`
@@ -545,14 +541,20 @@ const step = (state: ParserState, event: BedrockEvent) =>
      const reasoning = event.contentBlockDelta.delta.reasoningContent
      const events: LLMEvent[] = []
      const redactedData = reasoning.redactedContent ?? reasoning.data
-      const metadata = reasoning.signature
-        ? providerMetadata(state.providerMetadataKey, { signature: reasoning.signature })
+      const providerMetadata = reasoning.signature
+        ? bedrockMetadata({ signature: reasoning.signature })
        : redactedData !== undefined
-          ? providerMetadata(state.providerMetadataKey, { redactedData })
+          ? bedrockMetadata({ redactedData })
          : undefined
      const lifecycle =
-        reasoning.text !== undefined || metadata !== undefined
-          ? Lifecycle.reasoningDelta(state.lifecycle, events, `reasoning-${index}`, reasoning.text ?? "", metadata)
+        reasoning.text !== undefined || providerMetadata !== undefined
+          ? Lifecycle.reasoningDelta(
+              state.lifecycle,
+              events,
+              `reasoning-${index}`,
+              reasoning.text ?? "",
+              providerMetadata,
+            )
          : state.lifecycle
      return [
        {
@@ -594,7 +596,7 @@ const step = (state: ParserState, event: BedrockEvent) =>
            events,
            `reasoning-${index}`,
            state.reasoningSignatures[index]
-              ? providerMetadata(state.providerMetadataKey, { signature: state.reasoningSignatures[index] })
+              ? bedrockMetadata({ signature: state.reasoningSignatures[index] })
              : undefined,
          )
      events.push(...resultEvents)
@@ -631,7 +633,7 @@ const step = (state: ParserState, event: BedrockEvent) =>
    }

    if (event.metadata) {
-      const usage = mapUsage(event.metadata.usage, state.providerMetadataKey) ?? state.pendingFinish?.usage
+      const usage = mapUsage(event.metadata.usage) ?? state.pendingFinish?.usage
      return [
        {
          ...state,
@@ -645,13 +647,13 @@ const step = (state: ParserState, event: BedrockEvent) =>
    }

    if (event.exception) {
-      const message =
-        event.exception.details.message ?? event.exception.details.originalMessage ?? "Bedrock Converse stream error"
-      const body = ProviderShared.encodeJson(event)
      return yield* new AIError({
+        module: ADAPTER,
+        method: "stream",
        reason: classifyProviderFailure({
-          message,
-          rawBody: body,
+          message:
+            event.exception.details.message ?? event.exception.details.originalMessage ?? "Bedrock Converse stream error",
+          code: event.exception.type,
        }),
      })
    }
@@ -694,8 +696,7 @@ export const protocol = Protocol.make({
  },
  stream: {
    event: BedrockEvent,
-    initial: (request) => ({
-      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
+    initial: () => ({
      tools: ToolStream.empty<number>(),
      pendingFinish: undefined,
      hasToolCalls: false,
@@ -1,7 +1,6 @@
 import { EventStreamCodec } from "@smithy/eventstream-codec"
 import { fromUtf8, toUtf8 } from "@smithy/util-utf8"
-import { Effect, Encoding, Stream } from "effect"
-import { AIError, AIErrorReason } from "../schema/index.js"
+import { Effect, Stream } from "effect"
 import { Framing } from "../route/framing.js"
 import { ProviderShared } from "./shared.js"

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

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

@@ -6,7 +6,6 @@ import { Endpoint } from "../route/endpoint.js"
 import { Framing } from "../route/framing.js"
 import { Protocol } from "../route/protocol.js"
 import {
-  AIError,
  LLMEvent,
  Usage,
  type FinishReason,
@@ -18,7 +17,6 @@ import {
  type ToolCallPart,
  type ToolDefinition,
 } from "../schema/index.js"
-import { classifyProviderFailure } from "../provider-error.js"
 import { JsonObject, optionalArray, optionalNull, ProviderShared } from "./shared.js"
 import { GeminiToolSchema } from "./utils/gemini-tool-schema.js"
 import { Lifecycle } from "./utils/lifecycle.js"
@@ -223,7 +221,6 @@ const GeminiPromptFeedback = Schema.StructWithRest(
 type GeminiPromptFeedback = Schema.Schema.Type<typeof GeminiPromptFeedback>

 const GeminiEvent = Schema.Struct({
-  error: Schema.optional(Schema.Unknown),
  candidates: optionalNull(Schema.Array(GeminiCandidate)),
  promptFeedback: optionalNull(GeminiPromptFeedback),
  usageMetadata: optionalNull(GeminiUsage),
@@ -232,7 +229,6 @@ type GeminiEvent = Schema.Schema.Type<typeof GeminiEvent>

 interface ParserState {
  readonly route: string
-  readonly providerMetadataKey: string
  readonly finishReason?: string
  readonly hasToolCalls: boolean
  readonly promptFeedback?: GeminiPromptFeedback
@@ -289,23 +285,22 @@ const lowerUserPart = Effect.fn("Gemini.lowerUserPart")(function* (part: TextPar
  return { inlineData: { mimeType: media.mime, data: media.base64 } }
 })

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

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

-const lowerToolCall = (part: ToolCallPart, omitIds: boolean, metadataKey: string) => ({
+const lowerToolCall = (part: ToolCallPart, omitIds: boolean) => ({
  functionCall: { ...(omitIds ? {} : { id: part.id }), name: part.name, args: part.input },
-  thoughtSignature: thoughtSignature(part.providerMetadata, metadataKey),
+  thoughtSignature: thoughtSignature(part.providerMetadata),
 })

 const lowerMessages = Effect.fn("Gemini.lowerMessages")(function* (request: LLMRequest) {
  const contents: GeminiContent[] = []
-  const metadataKey = request.model.route.providerMetadataKey ?? String(request.model.provider)
  const omitCallIds = omitsFunctionCallIds(request.model.id)
  const legacyToolMedia = routesLegacyToolMedia(request.model.id)
  let pendingMedia: GeminiInlineDataPart[] | undefined
@@ -347,19 +342,15 @@ const lowerMessages = Effect.fn("Gemini.lowerMessages")(function* (request: LLMR
        if (!ProviderShared.supportsContent(part, ["text", "reasoning", "tool-call"]))
          return yield* ProviderShared.unsupportedContent("Gemini", "assistant", ["text", "reasoning", "tool-call"])
        if (part.type === "text") {
-          parts.push({ text: part.text, thoughtSignature: thoughtSignature(part.providerMetadata, metadataKey) })
+          parts.push({ text: part.text, thoughtSignature: thoughtSignature(part.providerMetadata) })
          continue
        }
        if (part.type === "reasoning") {
-          parts.push({
-            text: part.text,
-            thought: true,
-            thoughtSignature: thoughtSignature(part.providerMetadata, metadataKey),
-          })
+          parts.push({ text: part.text, thought: true, thoughtSignature: thoughtSignature(part.providerMetadata) })
          continue
        }
        if (part.type === "tool-call") {
-          const lowered = lowerToolCall(part, omitCallIds, metadataKey)
+          const lowered = lowerToolCall(part, omitCallIds)
          const signature = lowered.thoughtSignature
          parts.push({
            ...lowered,
@@ -507,7 +498,7 @@ const fromRequest = Effect.fn("Gemini.fromRequest")(function* (request: LLMReque
 // `cachedContentTokenCount` subset. `candidatesTokenCount` is *exclusive*
 // of `thoughtsTokenCount` — visible-only, not a total — so we sum the two
 // to produce the inclusive `outputTokens` the rest of the contract expects.
-const mapUsage = (usage: GeminiUsage | undefined, metadataKey: string) => {
+const mapUsage = (usage: GeminiUsage | undefined) => {
  if (!usage) return undefined
  // Explicit provider nulls decode as `null`; normalize to `undefined` so the
  // token arithmetic below treats them like absent counts.
@@ -528,7 +519,7 @@ const mapUsage = (usage: GeminiUsage | undefined, metadataKey: string) => {
    cacheReadInputTokens: cached,
    reasoningTokens: thoughts,
    totalTokens: ProviderShared.totalTokens(promptTokens, outputTokens, usage.totalTokenCount ?? undefined),
-    providerMetadata: providerMetadata(metadataKey, usage),
+    providerMetadata: { google: usage },
  })
 }

@@ -576,15 +567,10 @@ const finish = (state: ParserState): ReadonlyArray<LLMEvent> => {
      lifecycle,
      events,
      "reasoning-0",
-      providerMetadata(state.providerMetadataKey, { thoughtSignature: state.reasoningSignature }),
+      googleMetadata({ thoughtSignature: state.reasoningSignature }),
    )
  if (state.textSignature !== undefined)
-    lifecycle = Lifecycle.textEnd(
-      lifecycle,
-      events,
-      "text-0",
-      providerMetadata(state.providerMetadataKey, { thoughtSignature: state.textSignature }),
-    )
+    lifecycle = Lifecycle.textEnd(lifecycle, events, "text-0", googleMetadata({ thoughtSignature: state.textSignature }))
  Lifecycle.finish(lifecycle, events, {
    reason: {
      normalized:
@@ -593,32 +579,16 @@ const finish = (state: ParserState): ReadonlyArray<LLMEvent> => {
    },
    usage: state.usage,
    providerMetadata:
-      state.promptFeedback === undefined
-        ? undefined
-        : providerMetadata(state.providerMetadataKey, { promptFeedback: state.promptFeedback }),
+      state.promptFeedback === undefined ? undefined : googleMetadata({ promptFeedback: state.promptFeedback }),
  })
  return events
 }

 const step = (state: ParserState, event: GeminiEvent) => {
-  if (ProviderShared.isRecord(event.error) && typeof event.error.message === "string") {
-    const body = ProviderShared.encodeJson(event)
-    return Effect.fail(
-      new AIError({
-        reason: classifyProviderFailure({
-          message: event.error.message,
-          status: typeof event.error.code === "number" ? event.error.code : undefined,
-          rawBody: body,
-        }),
-      }),
-    )
-  }
  const nextState = {
    ...state,
    promptFeedback: event.promptFeedback ?? state.promptFeedback,
-    usage: event.usageMetadata
-      ? (mapUsage(event.usageMetadata, state.providerMetadataKey) ?? state.usage)
-      : state.usage,
+    usage: event.usageMetadata ? (mapUsage(event.usageMetadata) ?? state.usage) : state.usage,
  }
  const candidate = event.candidates?.[0]
  if (!candidate?.content)
@@ -662,7 +632,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
          events,
          "reasoning-0",
          part.text,
-          signature ? providerMetadata(state.providerMetadataKey, { thoughtSignature: signature }) : undefined,
+          signature ? googleMetadata({ thoughtSignature: signature }) : undefined,
        )
        continue
      }
@@ -670,16 +640,14 @@ const step = (state: ParserState, event: GeminiEvent) => {
        lifecycle,
        events,
        "reasoning-0",
-        reasoningSignature
-          ? providerMetadata(state.providerMetadataKey, { thoughtSignature: reasoningSignature })
-          : undefined,
+        reasoningSignature ? googleMetadata({ thoughtSignature: reasoningSignature }) : undefined,
      )
      lifecycle = Lifecycle.textDelta(
        lifecycle,
        events,
        "text-0",
        part.text,
-        textSignature ? providerMetadata(state.providerMetadataKey, { thoughtSignature: textSignature }) : undefined,
+        textSignature ? googleMetadata({ thoughtSignature: textSignature }) : undefined,
      )
      textSignature = undefined
      continue
@@ -699,9 +667,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
        lifecycle,
        events,
        "reasoning-0",
-        reasoningSignature
-          ? providerMetadata(state.providerMetadataKey, { thoughtSignature: reasoningSignature })
-          : undefined,
+        reasoningSignature ? googleMetadata({ thoughtSignature: reasoningSignature }) : undefined,
      )
      lifecycle = Lifecycle.stepStart(lifecycle, events)
      events.push(
@@ -709,9 +675,8 @@ const step = (state: ParserState, event: GeminiEvent) => {
          id,
          name: part.functionCall.name,
          input,
-          providerMetadata: part.thoughtSignature
-            ? providerMetadata(state.providerMetadataKey, { thoughtSignature: part.thoughtSignature })
-            : undefined,
+          providerMetadata:
+            part.thoughtSignature ? googleMetadata({ thoughtSignature: part.thoughtSignature }) : undefined,
        }),
      )
      hasToolCalls = true
@@ -749,7 +714,6 @@ export const protocol = Protocol.make({
    event: Protocol.jsonEvent(GeminiEvent),
    initial: (request) => ({
      route: `${request.model.provider}/${request.model.route.id}`,
-      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
      hasToolCalls: false,
      lifecycle: Lifecycle.initial(),
    }),
@@ -9,7 +9,15 @@ import {
  type ImageRoute,
 } from "../image.js"
 import { Auth, type Definition as AuthDefinition } from "../route/auth.js"
-import { AIError, Usage, mergeHttpOptions, mergeJsonRecords, type HttpOptions } from "../schema/index.js"
+import {
+  InvalidProviderOutputReason,
+  AIError,
+  Usage,
+  mergeHttpOptions,
+  mergeJsonRecords,
+  type HttpOptions,
+  type ProviderMetadata,
+} from "../schema/index.js"
 import { ProviderShared } from "./shared.js"
 import { ImageInputs } from "./utils/image-input.js"

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

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

@@ -1,7 +1,6 @@
 import { Effect, Schema, Stream } from "effect"
 import { Headers } from "effect/unstable/http"
 import { Framing } from "../route/framing.js"
-import type { HttpContext } from "../schema/index.js"
 import {
  HttpTransport,
  WebSocketTransport,
@@ -61,8 +60,8 @@ const driver = (options: Options, body: string): WebSocketChannelDriver => {
    observe: (_create, frame) =>
      Effect.gen(function* () {
        const event = yield* decodeEvent(frame).pipe(
-          Effect.mapError((cause) =>
-            ProviderShared.eventError(options.id, `Invalid ${options.name} WebSocket event`, frame, cause),
+          Effect.mapError(() =>
+            ProviderShared.eventError(options.id, `Invalid ${options.name} WebSocket event`, frame),
          ),
        )
        if (terminal)
@@ -74,13 +73,13 @@ const driver = (options: Options, body: string): WebSocketChannelDriver => {
        if (event.type === "error") {
          terminal = true
          yield* OpenResponses.decodeKnownErrorEvent(event).pipe(
-            Effect.mapError((cause) =>
-              ProviderShared.eventError(options.id, `${options.name} returned a malformed error event`, frame, cause),
+            Effect.mapError(() =>
+              ProviderShared.eventError(options.id, `${options.name} returned a malformed error event`, frame),
            ),
          )
          return {
            type: "provider-failure",
-            error: OpenResponses.providerFailure(event, `${options.name} stream error`, frame),
+            error: OpenResponses.providerFailure(options.id, event, `${options.name} stream error`),
          }
        }
        if (event.type === "response.failed") {
@@ -93,7 +92,7 @@ const driver = (options: Options, body: string): WebSocketChannelDriver => {
            )
          return {
            type: "provider-failure",
-            error: OpenResponses.providerFailure(event, `${options.name} response failed`, frame),
+            error: OpenResponses.providerFailure(options.id, event, `${options.name} response failed`),
          }
        }
        if (event.type === "response.created") {
@@ -176,37 +175,23 @@ export const transport = <Body>(options: Options): Transport<Body, Prepared, str
          channel,
        }
      }),
-    execute: (prepared, request, runtime, executeOptions) =>
-      Effect.gen(function* () {
-        if (!executeOptions?.webSocket || !prepared.channel) return yield* http.execute(prepared.http, request, runtime)
-        let fallbackHttp: HttpContext | undefined
-        const exchange: WebSocketChannelExchange = {
-          id: request.id ?? "request",
-          connect: {
-            url: prepared.channel.url,
-            headers: prepared.channel.headers,
-            rotateAfterMs: prepared.channel.rotateAfterMs,
-          },
-          fallback: () =>
-            Stream.unwrap(
-              http.execute(prepared.http, request, runtime).pipe(
-                Effect.map((execution) => {
-                  fallbackHttp = execution.http
-                  return execution.frames
-                }),
-              ),
-            ),
-          driver: prepared.channel.driver,
-        }
-        const execution = yield* executeOptions.webSocket.execute(exchange)
-        return {
-          frames: execution.frames,
-          complete: execution.complete,
-          get http() {
-            return fallbackHttp ?? execution.http
-          },
-        }
-      }),
+    execute: (prepared, request, runtime, executeOptions) => {
+      if (!executeOptions?.webSocket || !prepared.channel) return http.execute(prepared.http, request, runtime)
+      const exchange: WebSocketChannelExchange = {
+        id: request.id ?? "request",
+        connect: {
+          url: prepared.channel.url,
+          headers: prepared.channel.headers,
+          rotateAfterMs: prepared.channel.rotateAfterMs,
+        },
+        fallback: () =>
+          Stream.unwrap(
+            http.execute(prepared.http, request, runtime).pipe(Effect.map((execution) => execution.frames)),
+          ),
+        driver: prepared.channel.driver,
+      }
+      return executeOptions.webSocket.execute(exchange)
+    },
  }
 }

@@ -1,4 +1,4 @@
-import { AIError, TransportError } from "../schema/index.js"
+import { AIError, TransportReason } from "../schema/index.js"
 import type { ChannelCheckpoint, ChannelObservation, WebSocketChannelDriver } from "../route/transport/index.js"
 import { Effect, Option, Schema } from "effect"
 import * as ProviderShared from "./shared.js"
@@ -99,17 +99,17 @@ const incremental = (
 const code = (event: OpenResponses.Event) => event.code || event.error?.code || event.response?.error?.code || undefined

 const rejected = (
+  input: DriverInput,
  observation: Extract<ChannelObservation, { readonly type: "provider-failure" }>,
  recovery: "retry-full" | "rotate-and-retry-full",
 ): ChannelObservation => ({
  type: "rejected",
  recovery,
  error: new AIError({
-    reason: new TransportError({
+    module: input.id,
+    method: "stream",
+    reason: new TransportReason({
      message: observation.error.message,
-      body: observation.error.reason.body,
-      http: observation.error.reason.http,
-      cause: observation.error.reason.cause,
      transport: "websocket",
      operation: "read",
      phase: "receive",
@@ -137,16 +137,15 @@ export const driver = (input: DriverInput): WebSocketChannelDriver => {
    observe: (create, frame) =>
      Effect.gen(function* () {
        const event = yield* decodeEvent(frame).pipe(
-          Effect.mapError((cause) =>
-            ProviderShared.eventError(input.id, `Invalid ${input.name} WebSocket event`, frame, cause),
-          ),
+          Effect.mapError(() => ProviderShared.eventError(input.id, `Invalid ${input.name} WebSocket event`, frame)),
        )
        const observation = yield* input.base.observe(create, frame)
        if (event.type === "response.output_item.done" && event.item) output.push(event.item)
        if (observation.type === "provider-failure") {
          const rejection = code(event)
-          if (rejection === "previous_response_not_found") return rejected(observation, "retry-full")
-          if (rejection === "websocket_connection_limit_reached") return rejected(observation, "rotate-and-retry-full")
+          if (rejection === "previous_response_not_found") return rejected(input, observation, "retry-full")
+          if (rejection === "websocket_connection_limit_reached")
+            return rejected(input, observation, "rotate-and-retry-full")
        }
        if (observation.type !== "completed") return observation
        const responseID = event.response?.id
@@ -5,7 +5,7 @@ import { Protocol } from "../route/protocol.js"
 import {
  AIError,
  LLMEvent,
-  ProviderInternalError,
+  ProviderInternalReason,
  Usage,
  type FinishReason,
  type JsonSchema,
@@ -196,7 +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
 }
@@ -346,8 +346,7 @@ export const Event = Schema.StructWithRest(
    item_id: Schema.optional(Schema.String),
    output_index: Schema.optional(Schema.Number),
    summary_index: Schema.optional(Schema.Number),
-    // OutputItemAdded/Done permit a null item in the Open Responses OpenAPI schema.
-    item: optionalNull(StreamItem),
+    item: Schema.optional(StreamItem),
    response: Schema.optional(
      Schema.StructWithRest(
        Schema.Struct({
@@ -373,6 +372,9 @@ export const Event = Schema.StructWithRest(
 )
 export type Event = Schema.Schema.Type<typeof Event>

+// Which lowered input item a persisted item id is about to be attached to.
+export type ItemKind = "message" | "reasoning" | "function-call" | "hosted-tool"
+
 export interface Extension {
  readonly id: string
  readonly name: string
@@ -382,6 +384,10 @@ export interface Extension {
    readonly request: LLMRequest
  }) => MediaInput | undefined
  readonly lowerHostedToolItem?: (item: unknown) => ExtendedHostedToolItem | undefined
+  // Optional grammar check applied before a persisted item id is resent as
+  // part of replayed history. Returning false drops the id; every lowered
+  // item treats a dropped id the same as an absent one.
+  readonly acceptsItemID?: (kind: ItemKind, id: string) => boolean
 }

 const BASE: Extension = { id: ADAPTER, name: NAME }
@@ -441,37 +447,53 @@ export const lowerToolChoice = (protocolName: string, toolChoice: NonNullable<LL
    tool: (toolName) => ({ type: "function" as const, name: toolName }),
  })

-// Server-issued item ids need a nonempty prefix and suffix, but the prefix is
-// provider-defined and does not necessarily identify the item's semantic type.
+// Servers validate item ids on replayed history, and a malformed or oversized
+// id can fail an otherwise valid request. Only server-issued tokens are worth
+// resending; anything else is treated as absent so the item is resent without
+// an id (or skipped, for items that cannot be expressed without one).
+const ITEM_ID_PATTERN = /^[A-Za-z0-9_-]{1,64}$/
 const itemID = (providerMetadata: ProviderMetadata | undefined, providerMetadataKey: string) => {
  const metadata = providerMetadata?.[providerMetadataKey]
-  if (!ProviderShared.isRecord(metadata) || typeof metadata.itemId !== "string") return undefined
-  const separator = metadata.itemId.indexOf("_")
-  return separator > 0 && separator < metadata.itemId.length - 1 ? metadata.itemId : undefined
+  return ProviderShared.isRecord(metadata) &&
+    typeof metadata.itemId === "string" &&
+    ITEM_ID_PATTERN.test(metadata.itemId)
+    ? metadata.itemId
+    : undefined
 }

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

-const lowerReasoning = (part: ReasoningPart, providerMetadataKey: string): OpenResponsesReasoningInput | undefined => {
+const lowerReasoning = (
+  part: ReasoningPart,
+  providerMetadataKey: string,
+  extension: Extension,
+): OpenResponsesReasoningInput | undefined => {
  const metadata = part.providerMetadata?.[providerMetadataKey]
-  if (!ProviderShared.isRecord(metadata)) return undefined
  const id = itemID(part.providerMetadata, providerMetadataKey)
+  if (!ProviderShared.isRecord(metadata) || !acceptsItemID(extension, "reasoning", id)) return undefined
  const encryptedContent =
    typeof metadata.reasoningEncryptedContent === "string" || metadata.reasoningEncryptedContent === null
      ? metadata.reasoningEncryptedContent
      : undefined
  return {
    type: "reasoning",
-    ...(id === undefined ? {} : { id }),
+    id,
    summary: part.text.length > 0 ? [{ type: "summary_text", text: part.text }] : [],
    encrypted_content: encryptedContent,
  }
@@ -592,7 +614,8 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
          Array<{ id: string | undefined; phase: MessagePhase | null | undefined; parts: TextPart[] }>
        >((groups, part) => {
          const metadata = part.providerMetadata?.[providerMetadataKey]
-          const id = itemID(part.providerMetadata, providerMetadataKey)
+          const rawID = itemID(part.providerMetadata, providerMetadataKey)
+          const id = acceptsItemID(extension, "message", rawID) ? rawID : undefined
          const phase = ProviderShared.isRecord(metadata) ? messagePhase(metadata.phase) : undefined
          const group = groups.at(-1)
          if (group && group.id === id && group.phase === phase) group.parts.push(part)
@@ -617,23 +640,23 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
        }
        if (part.type === "reasoning") {
          flushText()
-          const reasoning = lowerReasoning(part, providerMetadataKey)
+          const reasoning = lowerReasoning(part, providerMetadataKey, extension)
          if (!reasoning) continue
-          const existing = reasoning.id === undefined ? undefined : reasoningItems[reasoning.id]
+          const existing = reasoningItems[reasoning.id]
          if (existing) {
            existing.summary.push(...reasoning.summary)
            if (typeof reasoning.encrypted_content === "string")
              existing.encrypted_content = reasoning.encrypted_content
            continue
          }
-          if (reasoning.id !== undefined) reasoningItems[reasoning.id] = reasoning
+          reasoningItems[reasoning.id] = reasoning
          input.push(reasoning)
          continue
        }
        if (part.type === "tool-call") {
          flushText()
          if (part.providerExecuted === true) continue
-          input.push(lowerToolCall(part, providerMetadataKey))
+          input.push(lowerToolCall(part, providerMetadataKey, extension))
          continue
        }
        if (part.type === "tool-result" && part.providerExecuted === true) {
@@ -645,7 +668,7 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
              : Schema.is(HostedToolItem)(part.result.value)
                ? part.result.value
                : extension.lowerHostedToolItem?.(part.result.value)
-          if (id !== undefined && hosted?.id === id) {
+          if (acceptsItemID(extension, "hosted-tool", id) && hosted?.id === id) {
            if (!hostedToolItems.has(id)) {
              input.push(hosted)
              hostedToolItems.add(id)
@@ -813,7 +836,7 @@ export const providerMetadata = (state: ParserState, metadata: Record<string, un
 })

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

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

@@ -892,7 +915,7 @@ const reasoningMetadata = (state: ParserState, item: StreamItem & { id: string }
 // best-effort, not guaranteed.
 const onOutputItemAdded = (state: ParserState, event: Event): StepResult => {
  const item = event.item
-  if (item?.type === "message" && item.id !== undefined) {
+  if (item?.type === "message" && item.id) {
    const phase = messagePhase(item.phase)
    return [
      {
@@ -923,7 +946,7 @@ const onOutputItemAdded = (state: ParserState, event: Event): StepResult => {
  }
  if (item?.type !== "function_call" || !item.call_id) return [state, NO_EVENTS]
  const id = item.id ?? item.call_id
-  const metadata = item.id !== undefined ? providerMetadata(state, { itemId: item.id }) : undefined
+  const metadata = item.id ? providerMetadata(state, { itemId: item.id }) : undefined
  const events: LLMEvent[] = []
  const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
  return [
@@ -942,7 +965,7 @@ const onOutputItemAdded = (state: ParserState, event: Event): StepResult => {
 }

 const onReasoningSummaryPartAdded = (state: ParserState, event: Event): StepResult => {
-  if (event.item_id === undefined || event.summary_index === undefined) return [state, NO_EVENTS]
+  if (!event.item_id || event.summary_index === undefined) return [state, NO_EVENTS]
  const item = state.reasoningItems[event.item_id]
  if (!item) return [state, NO_EVENTS]
  if (event.summary_index === 0) return [state, NO_EVENTS]
@@ -989,7 +1012,7 @@ const onReasoningSummaryPartAdded = (state: ParserState, event: Event): StepResu
 }

 const onReasoningSummaryPartDone = (state: ParserState, event: Event): StepResult => {
-  if (event.item_id === undefined || event.summary_index === undefined) return [state, NO_EVENTS]
+  if (!event.item_id || event.summary_index === undefined) return [state, NO_EVENTS]
  const item = state.reasoningItems[event.item_id]
  if (!item) return [state, NO_EVENTS]
  return [
@@ -1014,7 +1037,7 @@ const onFunctionCallArgumentsDelta = Effect.fn("OpenResponses.onFunctionCallArgu
  state: ParserState,
  event: Event,
 ) {
-  if (event.item_id === undefined) return [state, NO_EVENTS] satisfies StepResult
+  if (!event.item_id) return [state, NO_EVENTS] satisfies StepResult
  const tool = state.tools[event.item_id]
  if (!tool) return [state, NO_EVENTS] satisfies StepResult
  const final = event.type === "response.function_call_arguments.done" ? event.arguments : undefined
@@ -1045,7 +1068,7 @@ const onOutputItemDone = Effect.fn("OpenResponses.onOutputItemDone")(function* (
  const item = event.item
  if (!item) return [state, NO_EVENTS] satisfies StepResult

-  if (item.type === "message" && item.id !== undefined) {
+  if (item.type === "message" && item.id) {
    const itemPhase = messagePhase(item.phase)
    const phase = itemPhase === undefined ? state.messagePhases[item.id] : itemPhase
    const events: LLMEvent[] = []
@@ -1076,7 +1099,7 @@ const onOutputItemDone = Effect.fn("OpenResponses.onOutputItemDone")(function* (
      : ToolStream.start(state.tools, id, {
          id: item.call_id,
          name: item.name,
-          providerMetadata: item.id !== undefined ? providerMetadata(state, { itemId: item.id }) : undefined,
+          providerMetadata: item.id ? providerMetadata(state, { itemId: item.id }) : undefined,
        })
    const result =
      item.arguments === undefined
@@ -1137,7 +1160,7 @@ const onResponseFinish = Effect.fn("OpenResponses.onResponseFinish")(function* (
          ([current, events], item) => {
            const id = item.id ?? (item.type === "function_call" ? item.call_id : undefined)
            if (
-              id === undefined ||
+              !id ||
              ((item.type !== "function_call" || !current.tools[id]) &&
                (item.type !== "reasoning" || !current.reasoningItems[id]))
            )
@@ -1188,8 +1211,11 @@ const providerErrorMessage = (event: Event, nested: OpenResponsesErrorPayload |
  return message || code
 }

-export const providerFailure = (event: Event, fallback: string, body = ProviderShared.encodeJson(event)) => {
+export const providerFailure = (id: string, event: Event, fallback: string) => {
  const nested = event.error ?? event.response?.error ?? undefined
+  const code = event.code || nested?.code || undefined
+  // Keep the full raw payload on the error even when the message is a summary.
+  const body = JSON.stringify(nested ?? event) ?? ""
  const summary = providerErrorMessage(event, nested)
  const message = summary ?? (body === "{}" ? fallback : body)
  const status =
@@ -1204,19 +1230,25 @@ export const providerFailure = (event: Event, fallback: string, body = ProviderS
    event.response === undefined &&
    summary === undefined &&
    status === undefined
-      ? new ProviderInternalError({ message, body })
-      : classifyProviderFailure({ message, status, rawBody: body })
-  return new AIError({ reason })
+      ? new ProviderInternalReason({ message })
+      : classifyProviderFailure({ message, code, status, rawBody: body })
+  return new AIError({
+    module: id,
+    method: "stream",
+    body,
+    reason,
+  })
 }

+const providerError = (state: ParserState, event: Event, fallback: string) => providerFailure(state.id, event, fallback)
+
 export const step = (state: ParserState, input: Event) => {
-  // The OpenAPI requires string IDs but imposes no minLength; empty is not missing.
  const event =
-    input.item_id !== undefined && outputItemID(state, input) !== input.item_id
+    input.item_id && outputItemID(state, input) !== input.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 === undefined) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
+    if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
    return Effect.succeed(
      event.type === "response.output_text.delta"
        ? onOutputTextDelta(state, event, event.item_id)
@@ -1225,7 +1257,7 @@ export const step = (state: ParserState, input: Event) => {
  }
  if (event.type === "response.refusal.delta" || event.type === "response.refusal.done") {
    const value = event.type === "response.refusal.delta" ? event.delta : event.refusal
-    if (event.item_id === undefined || typeof value !== "string")
+    if (!event.item_id || typeof value !== "string")
      return ProviderShared.eventError(state.id, `${event.type} is malformed`)
    return Effect.succeed(
      event.type === "response.refusal.delta"
@@ -1234,7 +1266,7 @@ export const step = (state: ParserState, input: Event) => {
    )
  }
  if (event.type === "response.reasoning.delta" || event.type === "response.reasoning_summary_text.delta") {
-    if (event.item_id === undefined) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
+    if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
    return Effect.succeed(onReasoningDelta(state, event, event.item_id))
  }
  if (
@@ -1242,24 +1274,24 @@ export const step = (state: ParserState, input: Event) => {
    event.type === "response.reasoning_summary_text.done" ||
    event.type === "response.reasoning_text.done"
  ) {
-    if (event.item_id === undefined) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
+    if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
    return Effect.succeed(onReasoningDone(state, event, event.item_id))
  }
  if (event.type === "response.reasoning_summary_part.added")
-    return event.item_id !== undefined
+    return event.item_id
      ? Effect.succeed(onReasoningSummaryPartAdded(state, event))
      : ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
  if (event.type === "response.reasoning_summary_part.done")
-    return event.item_id !== undefined
+    return event.item_id
      ? Effect.succeed(onReasoningSummaryPartDone(state, event))
      : ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
  if (event.type === "response.output_item.added") {
-    if (event.item?.type === "message" && event.item.id === undefined)
+    if (event.item?.type === "message" && !event.item.id)
      return ProviderShared.eventError(state.id, `${event.type} message is missing id`)
    const id = event.item?.id ?? (event.item?.type === "function_call" ? event.item.call_id : undefined)
    return Effect.succeed(
      onOutputItemAdded(
-        event.output_index !== undefined && id !== undefined
+        event.output_index !== undefined && id
          ? { ...state, outputItems: { ...state.outputItems, [event.output_index]: id } }
          : state,
        event,
@@ -1267,27 +1299,20 @@ export const step = (state: ParserState, input: Event) => {
    )
  }
  if (event.type === "response.function_call_arguments.delta" || event.type === "response.function_call_arguments.done")
-    return event.item_id !== undefined
+    return event.item_id
      ? onFunctionCallArgumentsDelta(state, event)
      : ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
  if (event.type === "response.output_item.done") {
-    if (event.item?.type === "message" && event.item.id === undefined)
+    if (event.item?.type === "message" && !event.item.id)
      return ProviderShared.eventError(state.id, `${event.type} message is missing id`)
    return onOutputItemDone(state, event)
  }
  if (event.type === "response.completed" || event.type === "response.incomplete") return onResponseFinish(state, event)
-  if (event.type === "response.failed") return providerFailure(event, `${state.name} response failed`)
+  if (event.type === "response.failed") return providerError(state, event, `${state.name} response failed`)
  if (event.type === "error")
    return decodeKnownErrorEvent(event).pipe(
-      Effect.mapError((cause) =>
-        ProviderShared.eventError(
-          state.id,
-          `${state.name} returned a malformed error event`,
-          ProviderShared.encodeJson(event),
-          cause,
-        ),
-      ),
-      Effect.flatMap(() => providerFailure(event, `${state.name} stream error`)),
+      Effect.mapError(() => ProviderShared.eventError(state.id, `${state.name} returned a malformed error event`)),
+      Effect.flatMap(() => providerError(state, event, `${state.name} stream error`)),
    )
  return Effect.succeed<StepResult>([state, NO_EVENTS])
 }
@@ -7,11 +7,10 @@ import { HttpTransport } from "../route/transport/index.js"
 import { Protocol } from "../route/protocol.js"
 import {
  AIError,
-  AIErrorReason,
-  InvalidProviderOutputError,
+  InvalidProviderOutputReason,
  LLMEvent,
-  ProviderInternalError,
-  UnknownProviderError,
+  ProviderInternalReason,
+  UnknownProviderReason,
  Usage,
  type FinishReason,
  type FinishReasonDetails,
@@ -254,7 +253,6 @@ interface PendingToolDelta {
 }

 export interface ParserState {
-  readonly providerMetadataKey: string
  readonly tools: ToolStream.State<number>
  readonly pendingTools: Partial<Record<number, PendingToolDelta>>
  readonly toolCallEvents: ReadonlyArray<LLMEvent>
@@ -326,18 +324,17 @@ const lowerMedia = Effect.fn("OpenAIChat.lowerMedia")(function* (part: MediaPart
 const openAICompatibleReasoningContent = (native: unknown) =>
  isRecord(native) && typeof native.reasoning_content === "string" ? native.reasoning_content : undefined

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

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

@@ -369,7 +366,7 @@ const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(func
  message: OpenAIChatRequestMessage,
  configuredField: string | undefined,
  requireReasoning: boolean,
-  options: LoweringOptions & { readonly providerMetadataKey: string },
+  options: LoweringOptions,
 ) {
  const content: TextPart[] = []
  const reasoning: ReasoningPart[] = []
@@ -391,14 +388,10 @@ const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(func
    }
  }
  const text = reasoning.map((part) => part.text).join("")
-  const details = reasoningDetails(reasoning, message.native?.openaiCompatible, options.providerMetadataKey)
-  const observedField = reasoning
-    .map((part) => reasoningField(part, options.providerMetadataKey))
-    .find((value) => value !== undefined)
+  const details = reasoningDetails(reasoning, message.native?.openaiCompatible)
+  const observedField = reasoning.map(reasoningField).find((value) => value !== undefined)
  const nativeReasoning = openAICompatibleReasoningContent(message.native?.openaiCompatible)
-  const fullyStructured = reasoning.every((part) =>
-    Array.isArray(part.providerMetadata?.[options.providerMetadataKey]?.reasoningDetails),
-  )
+  const fullyStructured = reasoning.every((part) => Array.isArray(part.providerMetadata?.openai?.reasoningDetails))
  const field = (() => {
    if (configuredField !== undefined && (requireReasoning || reasoning.length > 0 || nativeReasoning !== undefined))
      return configuredField
@@ -466,7 +459,7 @@ const lowerMessage = Effect.fn("OpenAIChat.lowerMessage")(function* (
  message: OpenAIChatRequestMessage,
  reasoningField: string | undefined,
  requireReasoning: boolean,
-  options: LoweringOptions & { readonly providerMetadataKey: string },
+  options: LoweringOptions,
 ) {
  if (message.role === "user") return [yield* lowerUserMessage(message, options)]
  if (message.role === "assistant")
@@ -502,13 +495,8 @@ const lowerMessages = Effect.fn("OpenAIChat.lowerMessages")(function* (request:
  const mistral = ["mistral", "devstral", "codestral", "pixtral", "mixtral"].some((family) => modelID.includes(family))
  const lowering = {
    ...options,
-    providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
    toolCallID: (id: string) => {
-      if (mistral)
-        return id
-          .replace(/[^a-zA-Z0-9]/g, "")
-          .slice(0, 9)
-          .padEnd(9, "0")
+      if (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)
@@ -517,8 +505,7 @@ const lowerMessages = Effect.fn("OpenAIChat.lowerMessages")(function* (request:
  }
  const requireAssistantAfterTool = request.model.compatibility?.requireAssistantAfterTool ?? mistral
  const bridgeTools = () => {
-    if (requireAssistantAfterTool && messages.at(-1)?.role === "tool")
-      messages.push({ role: "assistant", content: "Done." })
+    if (requireAssistantAfterTool && messages.at(-1)?.role === "tool") messages.push({ role: "assistant", content: "Done." })
  }
  const pendingImages: Array<Schema.Schema.Type<typeof OpenAIChatUserContent>> = []
  const flushImages = () => {
@@ -570,10 +557,7 @@ const lowerMessages = Effect.fn("OpenAIChat.lowerMessages")(function* (request:
        )
      continue
    }
-    if (
-      message.role === "assistant" &&
-      message.content.every((part) => part.type === "text" && part.text.trim() === "")
-    )
+    if (message.role === "assistant" && message.content.every((part) => part.type === "text" && part.text.trim() === ""))
      continue
    if (message.role === "tool") {
      const lowered = yield* lowerToolMessages(message, lowering)
@@ -604,10 +588,7 @@ const hasToolHistory = (messages: ReadonlyArray<LLMRequest["messages"][number]>)
 // models.dev provider naming: DeepSeek, Moonshot AI, Together AI, ZAI
 // (Zhipu + Coding Plan variants), Nvidia, Cerebras, Chutes, etc. still
 // require `max_tokens`.
-const detectMaxTokensField = (
-  provider: string,
-  baseURL: string | undefined,
-): "max_tokens" | "max_completion_tokens" => {
+const detectMaxTokensField = (provider: string, baseURL: string | undefined): "max_tokens" | "max_completion_tokens" => {
  const p = provider.toLowerCase()
  const url = (baseURL ?? "").toLowerCase()
  if (
@@ -657,8 +638,7 @@ const detectSupportsStore = (provider: string, baseURL: string | undefined): boo
  const isChutes = p === "chutes" || url.includes("chutes.ai")
  const isCloudflareWorkersAI = p === "cloudflare-workers-ai" || url.includes("api.cloudflare.com")
  const isCloudflareAiGateway = p === "cloudflare-ai-gateway" || url.includes("gateway.ai.cloudflare.com")
-  const isVercelAiGateway =
-    p === "vercel-ai-gateway" || url.includes("ai-gateway.vercel.sh") || url.includes("vercel.sh")
+  const isVercelAiGateway = p === "vercel-ai-gateway" || url.includes("ai-gateway.vercel.sh") || url.includes("vercel.sh")
  const isAntLing = p === "ant-ling" || url.includes("api.ant-ling.com")
  const isOpencode = p === "opencode" || url.includes("opencode.ai")
  const isNonStandard =
@@ -690,7 +670,11 @@ const detectSupportsStrictMode = (provider: string, baseURL: string | undefined)
  return !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia
 }

-const detectZaiToolStream = (provider: string, baseURL: string | undefined, modelID: string): boolean => {
+const detectZaiToolStream = (
+  provider: string,
+  baseURL: string | undefined,
+  modelID: string,
+): boolean => {
  const p = provider.toLowerCase()
  const url = (baseURL ?? "").toLowerCase()
  const isZai =
@@ -740,10 +724,10 @@ export const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (
  const supportsStore = request.model.compatibility?.supportsStore ?? detectSupportsStore(provider, baseURL)
  const supportsUsageInStreaming =
    request.model.compatibility?.supportsUsageInStreaming ?? detectSupportsUsageInStreaming()
-  const supportsStrictMode =
-    request.model.compatibility?.supportsStrictMode ?? detectSupportsStrictMode(provider, baseURL)
+  const supportsStrictMode = request.model.compatibility?.supportsStrictMode ?? detectSupportsStrictMode(provider, baseURL)
  const zaiToolStream =
-    request.model.compatibility?.zaiToolStream ?? detectZaiToolStream(provider, baseURL, request.model.id)
+    request.model.compatibility?.zaiToolStream ??
+    detectZaiToolStream(provider, baseURL, request.model.id)
  const hasHistory = hasToolHistory(request.messages)
  const hasActiveTools = request.tools.length > 0
  return {
@@ -785,22 +769,26 @@ 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 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* new AIError({
-        reason: new UnknownProviderError({
-          message: "Provider reported an error (finish_reason: error)",
-          body: ProviderShared.encodeJson(event),
-        }),
-      })
+      return yield* finishReasonError(
+        event,
+        new UnknownProviderReason({ message: "Provider reported an error (finish_reason: error)" }),
+      )
    case "network_error":
-      return yield* new AIError({
-        reason: new ProviderInternalError({
-          message: "Provider reported a network error (finish_reason: network_error)",
-          body: ProviderShared.encodeJson(event),
-        }),
-      })
+      return yield* finishReasonError(
+        event,
+        new ProviderInternalReason({ message: "Provider reported a network error (finish_reason: network_error)" }),
+      )
    case "stop":
    case "end":
      return "stop" as const
@@ -824,14 +812,15 @@ const mapFinishReason = Effect.fn("OpenAIChat.mapFinishReason")(function* (event
 // Providers differ on cache-hit location: OpenAI uses
 // `prompt_tokens_details.cached_tokens`, DeepSeek uses
 // `prompt_cache_hit_tokens`, and Zai uses top-level `cached_tokens`.
-const mapUsage = (usage: OpenAIChatEvent["usage"], providerMetadataKey: string): Usage | undefined => {
+const mapUsage = (usage: OpenAIChatEvent["usage"]): Usage | undefined => {
  if (!usage) return undefined
  const input = usage.prompt_tokens ?? undefined
  const output = usage.completion_tokens ?? undefined
-  const cached = (usage.prompt_tokens_details?.cached_tokens ??
-    (usage as { prompt_cache_hit_tokens?: number | null }).prompt_cache_hit_tokens ??
-    (usage as { cached_tokens?: number | null }).cached_tokens ??
-    undefined) as number | undefined
+  const cached =
+    (usage.prompt_tokens_details?.cached_tokens ??
+      (usage as { prompt_cache_hit_tokens?: number | null }).prompt_cache_hit_tokens ??
+      (usage as { cached_tokens?: number | null }).cached_tokens ??
+      undefined) as number | undefined
  const cacheWrite = usage.prompt_tokens_details?.cache_write_tokens ?? undefined
  const reasoning = usage.completion_tokens_details?.reasoning_tokens ?? undefined
  const nonCached = ProviderShared.subtractTokens(input, ProviderShared.sumTokens(cached, cacheWrite))
@@ -843,7 +832,7 @@ const mapUsage = (usage: OpenAIChatEvent["usage"], providerMetadataKey: string):
    cacheWriteInputTokens: cacheWrite,
    reasoningTokens: reasoning,
    totalTokens: ProviderShared.totalTokens(input, output, usage.total_tokens ?? undefined),
-    providerMetadata: { [providerMetadataKey]: usage },
+    providerMetadata: { openai: usage },
  })
 }

@@ -917,12 +906,8 @@ const conflictingReasoningTextDetails = (previous: Record<string, unknown>, curr
 const conflictingDetailValue = (previous: unknown, current: unknown) =>
  previous !== undefined && previous !== null && current !== undefined && current !== null && previous !== current

-const reasoningMetadata = (
-  providerMetadataKey: string,
-  field: ParserState["reasoningField"],
-  details?: ReadonlyArray<unknown>,
-) => ({
-  [providerMetadataKey]: {
+const reasoningMetadata = (field: ParserState["reasoningField"], details?: ReadonlyArray<unknown>) => ({
+  openai: {
    ...(field ? { reasoningField: field } : {}),
    ...(details ? { reasoningDetails: details } : {}),
  },
@@ -933,8 +918,12 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
    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,
        }),
@@ -945,17 +934,15 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
    // Moonshot (and a few other OpenAI-compatible providers) attach usage to
    // `choice.usage` instead of the top-level `usage` field.
    const choiceUsage = (choice as unknown as { usage?: OpenAIChatEvent["usage"] })?.usage
-    const usage =
-      mapUsage(event.usage, state.providerMetadataKey) ??
-      (choiceUsage ? mapUsage(choiceUsage, state.providerMetadataKey) : undefined) ??
-      state.usage
+    const usage = mapUsage(event.usage) ?? (choiceUsage ? mapUsage(choiceUsage) : undefined) ?? state.usage
    const rawFinishReason = choice?.finish_reason
-    const finishReason = rawFinishReason
-      ? {
-          normalized: yield* mapFinishReason(event, rawFinishReason),
-          raw: choice?.native_finish_reason ?? rawFinishReason,
-        }
-      : state.finishReason
+    const finishReason =
+      rawFinishReason
+        ? {
+            normalized: yield* mapFinishReason(event, rawFinishReason),
+            raw: choice?.native_finish_reason ?? rawFinishReason,
+          }
+        : state.finishReason
    const delta = choice?.delta
    const toolDeltas = delta?.tool_calls ?? []
    let tools = state.tools
@@ -986,7 +973,7 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
    const detailDelta = Array.isArray(delta?.reasoning_details) ? delta.reasoning_details : undefined
    if (detailDelta !== undefined) appendReasoningDetails(state.reasoningDetails, detailDelta)
    const reasoningDetailsObserved = state.reasoningDetailsObserved || detailDelta !== undefined
-    const deltaMetadata = reasoningMetadata(state.providerMetadataKey, reasoningField)
+    const deltaMetadata = reasoningMetadata(reasoningField)
    const text = detailDelta?.length ? (detailText(detailDelta) ?? reasoning?.text) : reasoning?.text
    if (text !== undefined) lifecycle = Lifecycle.reasoningDelta(lifecycle, events, "reasoning-0", text, deltaMetadata)
    else if (
@@ -1002,11 +989,7 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
        lifecycle,
        events,
        "reasoning-0",
-        reasoningMetadata(
-          state.providerMetadataKey,
-          reasoningField,
-          reasoningDetailsObserved ? state.reasoningDetails : undefined,
-        ),
+        reasoningMetadata(reasoningField, reasoningDetailsObserved ? state.reasoningDetails : undefined),
      )
      lifecycle = Lifecycle.textDelta(lifecycle, events, "text-0", delta.content)
    }
@@ -1016,11 +999,7 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
        lifecycle,
        events,
        "reasoning-0",
-        reasoningMetadata(
-          state.providerMetadataKey,
-          reasoningField,
-          reasoningDetailsObserved ? state.reasoningDetails : undefined,
-        ),
+        reasoningMetadata(reasoningField, reasoningDetailsObserved ? state.reasoningDetails : undefined),
      )
      lifecycle = Lifecycle.textDelta(lifecycle, events, "text-0", delta.refusal)
    }
@@ -1059,14 +1038,7 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
        "OpenAI Chat tool call delta is missing id or name",
      )
      if (ToolStream.isError(result))
-        return yield* new AIError({
-          reason: AIErrorReason.make({
-            ...result.reason,
-            message: result.message,
-            cause: result.reason.cause,
-            body: ProviderShared.encodeJson(event),
-          }),
-        })
+        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)
@@ -1088,7 +1060,6 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>

    return [
      {
-        providerMetadataKey: state.providerMetadataKey,
        tools: finished?.tools ?? tools,
        pendingTools,
        toolCallEvents: finished?.events ?? state.toolCallEvents,
@@ -1110,9 +1081,11 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
 const finishEvents = Effect.fn("OpenAIChat.finishEvents")(function* (state: ParserState) {
  if (state.finishReason === undefined && state.requireFinishReason)
    return yield* new AIError({
-      reason: new InvalidProviderOutputError({
-        message: "OpenAI Chat stream ended without finish_reason",
+      module: ADAPTER,
+      method: "stream",
+      reason: new InvalidProviderOutputReason({
        classification: "incomplete-stream",
+        message: "OpenAI Chat stream ended without finish_reason",
        route: ADAPTER,
      }),
    })
@@ -1130,18 +1103,12 @@ const finishEvents = Effect.fn("OpenAIChat.finishEvents")(function* (state: Pars
      }
    : { normalized: hasToolCalls ? ("tool-calls" as const) : ("stop" as const) }
  const metadata = reasoningMetadata(
-    state.providerMetadataKey,
    state.reasoningField,
    state.reasoningDetailsObserved ? state.reasoningDetails : undefined,
  )
  const started =
    state.reasoningDetailsObserved && !state.reasoningEmitted
-      ? Lifecycle.reasoningStart(
-          state.lifecycle,
-          events,
-          "reasoning-0",
-          reasoningMetadata(state.providerMetadataKey, state.reasoningField),
-        )
+      ? Lifecycle.reasoningStart(state.lifecycle, events, "reasoning-0", reasoningMetadata(state.reasoningField))
      : state.lifecycle
  const ended = Lifecycle.reasoningEnd(started, events, "reasoning-0", metadata)
  const lifecycle = toolCallEvents.length ? Lifecycle.stepStart(ended, events) : ended
@@ -1168,7 +1135,6 @@ export const protocol = Protocol.make({
  stream: {
    event: Protocol.jsonEvent(OpenAIChatEvent),
    initial: (request) => ({
-      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
      tools: ToolStream.empty<number>(),
      pendingTools: {},
      toolCallEvents: [],
@@ -9,7 +9,14 @@ import {
  type ImageRoute,
 } from "../image.js"
 import { Auth, type Definition as AuthDefinition } from "../route/auth.js"
-import { Usage, mergeHttpOptions, mergeJsonRecords, type HttpOptions } from "../schema/index.js"
+import {
+  InvalidProviderOutputReason,
+  AIError,
+  Usage,
+  mergeHttpOptions,
+  mergeJsonRecords,
+  type HttpOptions,
+} from "../schema/index.js"
 import { ProviderShared } from "./shared.js"
 import { ImageInputs } from "./utils/image-input.js"
 import { OpenAIImage } from "./utils/openai-image.js"
@@ -77,6 +84,13 @@ const nativeOptions = (options: OpenAIImageOptions | undefined) => {
  }
 }

+const invalidOutput = (message: string) =>
+  new AIError({
+    module: ADAPTER,
+    method: "generate",
+    reason: new InvalidProviderOutputReason({ message, route: ADAPTER }),
+  })
+
 const applyQuery = (url: string, query: Record<string, string> | undefined) => {
  if (!query) return url
  const next = new URL(url)
@@ -90,12 +104,12 @@ export const model = (input: ModelInput) => {
    generate: Effect.fn("OpenAIImages.generate")(function* (request: ImageRequestFor<OpenAIImageOptions>, execute) {
      const mask = request.options?.mask
      if (mask !== undefined && (request.images?.length ?? 0) === 0)
-        return yield* ImageInputs.invalid("An OpenAI image mask requires at least one input image")
+        return yield* ImageInputs.invalid(ADAPTER, "An OpenAI image mask requires at least one input image")
      const http = mergeHttpOptions(request.model.http, request.http)
      const sourceImages = request.images ?? []
      const multipartImages = yield* Effect.forEach(sourceImages, (image) => {
        if (image.type === "bytes") return Effect.succeed({ data: image.data, mediaType: image.mediaType })
-        if (image.type === "url") return ImageInputs.decodeDataUrl(image.url)
+        if (image.type === "url") return ImageInputs.decodeDataUrl(image.url, ADAPTER)
        return Effect.undefined
      })
      const multipartMask =
@@ -104,7 +118,7 @@ export const model = (input: ModelInput) => {
          : mask.type === "bytes"
            ? { data: mask.data, mediaType: mask.mediaType }
            : mask.type === "url"
-              ? yield* ImageInputs.decodeDataUrl(mask.url)
+              ? yield* ImageInputs.decodeDataUrl(mask.url, ADAPTER)
              : undefined
      const useMultipart =
        sourceImages.length > 0 &&
@@ -147,7 +161,7 @@ export const model = (input: ModelInput) => {
        return undefined
      })
      if (references.some((image) => image === undefined))
-        return yield* ImageInputs.invalid("OpenAI Images accepts image URLs, data URLs, bytes, and file IDs")
+        return yield* ImageInputs.invalid(ADAPTER, "OpenAI Images accepts image URLs, data URLs, bytes, and file IDs")
      const maskReference =
        mask === undefined
          ? undefined
@@ -159,7 +173,7 @@ export const model = (input: ModelInput) => {
                ? { file_id: mask.id }
                : undefined
      if (mask !== undefined && maskReference === undefined)
-        return yield* ImageInputs.invalid("OpenAI Images accepts masks as URLs, data URLs, bytes, or file IDs")
+        return yield* ImageInputs.invalid(ADAPTER, "OpenAI Images accepts masks as URLs, data URLs, bytes, or file IDs")
      const requestBody = mergeJsonRecords(
        {
          model: request.model.id,
@@ -195,9 +209,11 @@ const parseResponse = Effect.fn("OpenAIImages.parseResponse")(function* (
  options: OpenAIImageOptions | undefined,
  overlay: Record<string, unknown> | undefined,
 ) {
-  const output = yield* ProviderShared.imageResponse(ADAPTER, "OpenAI Images", response)
-  const decoded = yield* Schema.decodeUnknownEffect(Schema.fromJsonString(OpenAIImageResponse))(output.body).pipe(
-    Effect.mapError((cause) => output.invalid("OpenAI Images returned an invalid response", cause)),
+  const payload = yield* response.json.pipe(
+    Effect.mapError(() => invalidOutput("Failed to read the OpenAI Images response")),
+  )
+  const decoded = yield* Schema.decodeUnknownEffect(OpenAIImageResponse)(payload).pipe(
+    Effect.mapError(() => invalidOutput("OpenAI Images returned an invalid response")),
  )
  const requestBody = mergeJsonRecords(nativeOptions(options), overlay)
  const format =
@@ -205,7 +221,7 @@ const parseResponse = Effect.fn("OpenAIImages.parseResponse")(function* (
  const images = yield* Effect.forEach(decoded.data, (item, index) => {
    if (item.b64_json)
      return Effect.fromResult(Encoding.decodeBase64(item.b64_json)).pipe(
-        Effect.mapError((cause) => output.invalid(`OpenAI Images result ${index} contains invalid base64 data`, cause)),
+        Effect.mapError(() => invalidOutput(`OpenAI Images result ${index} contains invalid base64 data`)),
        Effect.map(
          (data) =>
            new GeneratedImage({
@@ -225,9 +241,9 @@ const parseResponse = Effect.fn("OpenAIImages.parseResponse")(function* (
            item.revised_prompt === undefined ? undefined : { openai: { revisedPrompt: item.revised_prompt } },
        }),
      )
-    return Effect.fail(output.invalid(`OpenAI Images result ${index} has neither image data nor a URL`))
+    return Effect.fail(invalidOutput(`OpenAI Images result ${index} has neither image data nor a URL`))
  })
-  if (images.length === 0) return yield* output.invalid("OpenAI Images returned no images")
+  if (images.length === 0) return yield* invalidOutput("OpenAI Images returned no images")
  return new ImageResponse({
    images,
    usage:
@@ -86,10 +86,28 @@ const OpenAIResponsesBody = Schema.Struct({
 })
 export type OpenAIResponsesBody = Schema.Schema.Type<typeof OpenAIResponsesBody>

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

 const nativeImageToolInput = (tool: ToolDefinition) => {
@@ -149,9 +167,7 @@ const hostedToolResult = Effect.fn("OpenAIResponses.hostedToolResult")(function*
  const isError = item.error !== undefined && item.error !== null
  if (item.type === "image_generation_call" && item.result) {
    yield* Effect.fromResult(Encoding.decodeBase64(item.result)).pipe(
-      Effect.mapError((cause) =>
-        ProviderShared.eventError(ADAPTER, "OpenAI Responses returned invalid image base64", undefined, cause),
-      ),
+      Effect.mapError(() => ProviderShared.eventError(ADAPTER, "OpenAI Responses returned invalid image base64")),
    )
    const format = item.output_format ?? "png"
    return {
@@ -186,7 +202,7 @@ const HOSTED_TOOLS = {

 const step = (state: OpenResponses.ParserState, event: OpenResponses.Event) => {
  if (event.type === "response.reasoning_text.delta")
-    return event.item_id !== undefined
+    return event.item_id
      ? Effect.succeed(
          OpenResponses.onReasoningDelta(state, event, OpenResponses.outputItemID(state, event) ?? event.item_id),
        )
@@ -2,12 +2,11 @@ import { Buffer } from "node:buffer"
 import { Tool } from "@opencode-ai/schema/tool"
 import { Effect, Schema, Stream } from "effect"
 import * as Sse from "effect/unstable/encoding/Sse"
-import { Headers, HttpClientRequest, HttpClientResponse } from "effect/unstable/http"
+import { Headers, HttpClientRequest } from "effect/unstable/http"
 import {
-  InvalidProviderOutputError,
-  InvalidRequestError,
+  InvalidProviderOutputReason,
+  InvalidRequestReason,
  AIError,
-  HttpContext,
  type ContentPart,
  type LLMRequest,
  type MediaPart,
@@ -97,15 +96,17 @@ export const sumTokens = (...values: ReadonlyArray<number | undefined>): number
  return values.reduce((acc: number, value) => acc + (value ?? 0), 0)
 }

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

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

 /**
@@ -232,7 +233,7 @@ export const sseFraming = (
      (state, chunk) =>
        Effect.gen(function* () {
          const error = state.parser.feed(chunk)
-          if (error) return yield* eventError("sse", error.message, chunk, error)
+          if (error) return yield* eventError("sse", error.message)
          return [state, state.output.splice(0)] as const
        }),
    ),
@@ -248,39 +249,13 @@ export const sseFraming = (
 /**
 * Canonical invalid-request constructor shared by protocol lowering.
 */
-export const invalidRequest = (message: string, cause?: unknown) =>
+export const invalidRequest = (message: string) =>
  new AIError({
-    reason: new InvalidRequestError({ message, cause }),
+    module: "ProviderShared",
+    method: "request",
+    reason: new InvalidRequestReason({ message }),
  })

-export const imageResponse = Effect.fn("ProviderShared.imageResponse")(function* (
-  route: string,
-  name: string,
-  response: HttpClientResponse.HttpClientResponse,
-) {
-  const http = new HttpContext({ url: response.request.url, status: response.status, headers: response.headers })
-  const body = yield* response.text.pipe(
-    Effect.mapError(
-      (cause) =>
-        new AIError({
-          reason: new InvalidProviderOutputError({
-            route,
-            message: `Failed to read the ${name} response`,
-            http,
-            cause,
-          }),
-        }),
-    ),
-  )
-  return {
-    body,
-    invalid: (message: string, cause?: unknown) =>
-      new AIError({
-        reason: new InvalidProviderOutputError({ route, message, body, http, cause }),
-      }),
-  }
-})
-
 export const matchToolChoice = <Auto, None, Required, Tool>(
  route: string,
  toolChoice: NonNullable<LLMRequest["toolChoice"]>,
@@ -327,7 +302,7 @@ export const unsupportedContent = (
 export const validateWith =
  <A, I, E extends { readonly message: string }>(decode: (input: I) => Effect.Effect<A, E>) =>
  (payload: I) =>
-    decode(payload).pipe(Effect.mapError((error) => invalidRequest(error.message, error)))
+    decode(payload).pipe(Effect.mapError((error) => invalidRequest(error.message)))

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

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

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

 export const decodeDataUrl = (
  url: string,
+  module: string,
 ): Effect.Effect<{ readonly mediaType: string; readonly data: Uint8Array } | undefined, AIError> => {
  if (!url.startsWith("data:")) return Effect.undefined
  const match = /^data:([^;,]+);base64,(.*)$/s.exec(url)
-  if (!match) return Effect.fail(invalid("Image data URLs must contain a MIME type and base64 data"))
+  if (!match) return Effect.fail(invalid(module, "Image data URLs must contain a MIME type and base64 data"))
  return Effect.fromResult(Encoding.decodeBase64(match[2])).pipe(
-    Effect.mapError((cause) => invalid("Image data URL contains invalid base64 data", cause)),
+    Effect.mapError(() => invalid(module, "Image data URL contains invalid base64 data")),
    Effect.map((data) => ({ mediaType: match[1], data })),
  )
 }
@@ -29,9 +29,10 @@ export type ResponseIncludable = (typeof ResponseIncludables)[number] | (string

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

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

 export * as ResponsesHostedTools from "./responses-hosted-tools.js"
@@ -2,7 +2,14 @@ import { Effect, Encoding, Schema } from "effect"
 import { Headers, HttpClientRequest } from "effect/unstable/http"
 import { GeneratedImage, ImageModel, ImageResponse, type ImageRequestFor, type ImageRoute } from "../image.js"
 import { Auth, type Definition as AuthDefinition } from "../route/auth.js"
-import { Usage, mergeHttpOptions, mergeJsonRecords, type HttpOptions } from "../schema/index.js"
+import {
+  InvalidProviderOutputReason,
+  AIError,
+  Usage,
+  mergeHttpOptions,
+  mergeJsonRecords,
+  type HttpOptions,
+} from "../schema/index.js"
 import { ProviderShared, optionalNull } from "./shared.js"
 import { ImageInputs } from "./utils/image-input.js"

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

+const invalidOutput = (message: string) =>
+  new AIError({
+    module: ADAPTER,
+    method: "generate",
+    reason: new InvalidProviderOutputReason({ message, route: ADAPTER }),
+  })
+
 const applyQuery = (url: string, query: Record<string, string> | undefined) => {
  if (!query) return url
  const next = new URL(url)
@@ -106,7 +120,7 @@ export const model = (input: ModelInput) => {
        return undefined
      })
      if (imageReferences.some((image) => image === undefined))
-        return yield* ImageInputs.invalid("xAI Images accepts image URLs, data URLs, bytes, and file IDs")
+        return yield* ImageInputs.invalid(ADAPTER, "xAI Images accepts image URLs, data URLs, bytes, and file IDs")
      const requestBody = mergeJsonRecords(
        {
          model: request.model.id,
@@ -135,17 +149,17 @@ export const model = (input: ModelInput) => {
          HttpClientRequest.bodyText(text, "application/json"),
        ),
      )
-      const output = yield* ProviderShared.imageResponse(ADAPTER, "xAI Images", response)
-      const decoded = yield* Schema.decodeUnknownEffect(Schema.fromJsonString(XAIImageResponse))(output.body).pipe(
-        Effect.mapError((cause) => output.invalid("xAI Images returned an invalid response", cause)),
+      const payload = yield* response.json.pipe(
+        Effect.mapError(() => invalidOutput("Failed to read the xAI Images response")),
+      )
+      const decoded = yield* Schema.decodeUnknownEffect(XAIImageResponse)(payload).pipe(
+        Effect.mapError(() => invalidOutput("xAI Images returned an invalid response")),
      )
      const images = yield* Effect.forEach(decoded.data, (item, index) => {
        const mediaType = item.mime_type ?? "application/octet-stream"
        if (item.b64_json)
          return Effect.fromResult(Encoding.decodeBase64(item.b64_json)).pipe(
-            Effect.mapError((cause) =>
-              output.invalid(`xAI Images result ${index} contains invalid base64 data`, cause),
-            ),
+            Effect.mapError(() => invalidOutput(`xAI Images result ${index} contains invalid base64 data`)),
            Effect.map(
              (data) =>
                new GeneratedImage({
@@ -169,9 +183,9 @@ export const model = (input: ModelInput) => {
                  : { xai: { revisedPrompt: item.revised_prompt } },
            }),
          )
-        return Effect.fail(output.invalid(`xAI Images result ${index} has neither image data nor a URL`))
+        return Effect.fail(invalidOutput(`xAI Images result ${index} has neither image data nor a URL`))
      })
-      if (images.length === 0) return yield* output.invalid("xAI Images returned no images")
+      if (images.length === 0) return yield* invalidOutput("xAI Images returned no images")
      const usage = ProviderShared.isRecord(decoded.usage) ? decoded.usage : undefined
      return new ImageResponse({
        images,
@@ -2,7 +2,13 @@ import { Effect, Schema } from "effect"
 import { Headers, HttpClientRequest } from "effect/unstable/http"
 import { GeneratedImage, ImageModel, ImageResponse, type ImageRequestFor, type ImageRoute } from "../image.js"
 import { Auth, type Definition as AuthDefinition } from "../route/auth.js"
-import { mergeHttpOptions, mergeJsonRecords, type HttpOptions } from "../schema/index.js"
+import {
+  InvalidProviderOutputReason,
+  AIError,
+  mergeHttpOptions,
+  mergeJsonRecords,
+  type HttpOptions,
+} from "../schema/index.js"
 import { ProviderShared } from "./shared.js"
 import { ImageInputs } from "./utils/image-input.js"

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

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

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

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

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

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

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

@@ -79,96 +79,91 @@ const NETWORK_ERROR_TEXT = /network[-_\s]error/i
 export interface ProviderFailure {
  readonly message: string
  readonly status?: number | undefined
+  readonly code?: string | undefined
  // Raw wire payload, scanned for failure signals (codes, overflow phrases)
  // that the summary message does not carry. Not shown to users.
  readonly rawBody?: string | undefined
-  // Some SDKs supply parsed error data separately from the original response text.
-  readonly data?: unknown
-  readonly http?: HttpContext | undefined
-  readonly cause?: unknown
  readonly retryAfterMs?: number | undefined
  readonly rateLimit?: HttpRateLimitDetails | undefined
+  readonly http?: HttpContext | undefined
+  readonly providerMetadata?: ProviderMetadata | undefined
 }

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

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

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

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

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

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

 export const routes = [responsesRoute, chatRoute]

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

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

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

@@ -68,7 +68,7 @@ const protocol = {
 const route = Route.make({
  id: "google-vertex-gemini",
  provider: id,
-  providerMetadataKey: "vertex",
+  providerMetadataKey: "google",
  protocol,
  endpoint: Endpoint.path(({ request }) => {
    const model = String(request.model.id)
@@ -1,116 +0,0 @@
-import { Effect, Schema } from "effect"
-import type { ProviderPackage } from "../provider-package.js"
-import { OpenAIChat } from "../protocols/openai-chat.js"
-import { ProviderShared } from "../protocols/shared.js"
-import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
-import { Route, type RouteDefaultsInput } from "../route/client.js"
-import { Endpoint } from "../route/endpoint.js"
-import { Framing } from "../route/framing.js"
-import { Protocol } from "../route/protocol.js"
-import { ProviderID, type ModelID, type LLMRequest } from "../schema/index.js"
-import { profiles } from "./openai-compatible-profile.js"
-import type { OpenAIProviderOptionsInput } from "./openai-options.js"
-
-export const id = ProviderID.make("groq")
-
-export type ProviderOptions = Pick<OpenAIProviderOptionsInput, "reasoningEffort"> & {
-  /** Controls visible reasoning on GPT-OSS; other models always use parsed reasoning. */
-  readonly includeReasoning?: boolean
-  readonly parallelToolCalls?: boolean
-  readonly serviceTier?: "on_demand" | "flex" | "auto" | "performance" | (string & {})
-  readonly user?: string
-}
-
-export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
-  ProviderAuthOption<"optional"> & {
-    readonly baseURL?: string
-    readonly providerOptions?: ProviderOptions
-  }
-
-export interface Settings extends ProviderPackage.Settings {
-  readonly apiKey?: string
-  readonly baseURL?: string
-  readonly providerOptions?: ProviderOptions
-}
-
-const Options = Schema.Struct({
-  includeReasoning: Schema.optional(Schema.Boolean),
-  parallelToolCalls: Schema.optional(Schema.Boolean),
-  serviceTier: Schema.optional(Schema.String),
-  user: Schema.optional(Schema.String),
-})
-
-export const protocol = Protocol.make({
-  id: "groq-chat",
-  body: {
-    schema: Schema.Struct({
-      ...OpenAIChat.bodyFields,
-      reasoning_format: Schema.optional(Schema.Literal("parsed")),
-      include_reasoning: Schema.optional(Schema.Boolean),
-      parallel_tool_calls: Schema.optional(Schema.Boolean),
-      service_tier: Schema.optional(Schema.String),
-      user: Schema.optional(Schema.String),
-    }),
-    from: Effect.fn("Groq.fromRequest")(function* (request: LLMRequest) {
-      const options = yield* ProviderShared.validateWith(Schema.decodeUnknownEffect(Options))(
-        request.providerOptions ?? {},
-      )
-      const gptOSS = request.model.id.startsWith("openai/gpt-oss-")
-      return {
-        ...(yield* OpenAIChat.fromRequest(request)),
-        reasoning_format: gptOSS ? undefined : ("parsed" as const),
-        include_reasoning: gptOSS ? options.includeReasoning : undefined,
-        parallel_tool_calls: options.parallelToolCalls,
-        service_tier: options.serviceTier,
-        user: options.user,
-      }
-    }),
-  },
-  stream: OpenAIChat.protocol.stream,
-})
-
-export const route = Route.make({
-  id: "groq-chat",
-  provider: id,
-  providerMetadataKey: "openai",
-  protocol,
-  endpoint: Endpoint.path("/chat/completions", { baseURL: profiles.groq.baseURL }),
-  framing: Framing.sse,
-})
-
-export const configure = (input: LanguageModelOptions = {}) => {
-  const { apiKey: _apiKey, auth: _auth, baseURL, ...defaults } = input
-  const configured = route.with({
-    ...defaults,
-    endpoint: { baseURL: baseURL ?? profiles.groq.baseURL },
-    auth: AuthOptions.bearer(input, "GROQ_API_KEY"),
-  })
-  return {
-    id,
-    model: (modelID: string | ModelID) =>
-      configured.model<ProviderOptions>({
-        id: modelID,
-        compatibility: {
-          maxTokensField: "max_completion_tokens",
-          reasoningField: "reasoning",
-          requireReasoning: false,
-          supportsStore: false,
-          supportsStrictMode: false,
-        },
-      }),
-    configure,
-  }
-}
-
-export const provider = configure()
-
-export const model: ProviderPackage.Definition<Settings, ProviderOptions>["model"] = (modelID, settings) =>
-  configure({
-    apiKey: settings.apiKey,
-    baseURL: settings.baseURL,
-    headers: settings.headers === undefined ? undefined : { ...settings.headers },
-    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    providerOptions: settings.providerOptions,
-  }).model(modelID)
-
-export * as Groq from "./groq.js"
@@ -6,13 +6,11 @@ export * as Azure from "./azure.js"
 export * as Cerebras from "./cerebras.js"
 export * as Cloudflare from "./cloudflare.js"
 export { CloudflareAIGateway, CloudflareWorkersAI } from "./cloudflare.js"
-export * as DeepInfra from "./deepinfra.js"
 export * as Google from "./google.js"
 export * as GoogleVertex from "./google-vertex.js"
 export * as GoogleVertexChat from "./google-vertex-chat.js"
 export * as GoogleVertexMessages from "./google-vertex-messages.js"
 export * as GoogleVertexResponses from "./google-vertex-responses.js"
-export * as Groq from "./groq.js"
 export * as OpenAI from "./openai.js"
 export * as OpenAICompatible from "./openai-compatible.js"
 export * as OpenAICompatibleResponses from "./openai-compatible-responses.js"
@@ -164,7 +164,6 @@ const bodyOptions = (input: unknown) => {
 export const route = Route.make({
  id: ADAPTER,
  provider: profile.provider,
-  providerMetadataKey: "openrouter",
  protocol,
  endpoint: Endpoint.path("/chat/completions", { baseURL: profile.baseURL }),
  framing: Framing.sse,
@@ -1,6 +1,6 @@
 import { Config, Effect, Redacted } from "effect"
 import { Headers } from "effect/unstable/http"
-import { AuthenticationError, InvalidRequestError, AIError, type HttpOptions } from "../schema/index.js"
+import { AuthenticationReason, InvalidRequestReason, AIError, type HttpOptions } from "../schema/index.js"

 export class MissingCredentialError extends Error {
  readonly _tag = "MissingCredentialError"
@@ -137,10 +137,12 @@ export function bearerHeader(name: string, source?: Secret | Credential) {
 const toAIError = (error: AuthError): AIError => {
  if (error instanceof MissingCredentialError || error instanceof Config.ConfigError) {
    return new AIError({
+      module: "Auth",
+      method: "apply",
      reason:
        error instanceof MissingCredentialError
-          ? new AuthenticationError({ message: error.message, cause: error, kind: "missing" })
-          : new InvalidRequestError({ message: `Failed to resolve auth config: ${error.message}`, cause: error }),
+          ? new AuthenticationReason({ message: error.message, kind: "missing" })
+          : new InvalidRequestReason({ message: `Failed to resolve auth config: ${error.message}` }),
    })
  }
  return error
@@ -12,14 +12,13 @@ import * as ProviderShared from "../protocols/shared.js"
 import type { ProtocolID, ProviderOptions } from "../schema/index.js"
 import {
  AIError,
-  AIErrorReason,
  GenerationOptions,
  HttpOptions,
  LLMRequest,
  LLMResponse,
  LanguageModel,
  LLMEvent,
-  InvalidProviderOutputError,
+  InvalidProviderOutputReason,
  ProviderID,
  mergeGenerationOptions,
  mergeHttpOptions,
@@ -90,7 +89,6 @@ export interface RouteDefaultsInput {
 export interface RoutePatch<Body, Prepared> extends RouteDefaultsInput {
  readonly id?: string
  readonly provider?: string | ProviderID
-  readonly providerMetadataKey?: string
  readonly auth?: Auth.Definition
  readonly transport?: Transport<Body, Prepared, unknown>
  readonly endpoint?: EndpointPatch<Body>
@@ -228,14 +226,16 @@ export interface MakeTransportInput<Body, Prepared, Frame, Event, State> {
 const streamError = (route: string, message: string, cause: Cause.Cause<unknown>) => {
  const failed = cause.reasons.find(Cause.isFailReason)?.error
  if (failed instanceof AIError) return failed
-  return ProviderShared.eventError(route, message, undefined, cause)
+  return ProviderShared.eventError(route, message, Cause.pretty(cause))
 }

 const incompleteStreamError = (route: string) =>
  new AIError({
-    reason: new InvalidProviderOutputError({
-      message: "The provider response ended unexpectedly.",
+    module: "LLMClient",
+    method: "stream",
+    reason: new InvalidProviderOutputReason({
      classification: "incomplete-stream",
+      message: "The provider response ended unexpectedly.",
      route,
    }),
  })
@@ -264,12 +264,11 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
  const decodeEventEffect = Schema.decodeUnknownEffect(protocol.stream.event)
  const decodeEvent = (route: string) => (frame: Frame) =>
    decodeEventEffect(frame).pipe(
-      Effect.mapError((cause) =>
+      Effect.mapError(() =>
        ProviderShared.eventError(
          input.id,
          `Invalid ${route} stream event`,
          typeof frame === "string" ? frame : ProviderShared.encodeJson(frame),
-          cause,
        ),
      ),
    )
@@ -290,16 +289,11 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
      defaults: routeInput.defaults ?? {},
      body: protocol.body,
      with: (patch: RoutePatch<Body, Prepared>) => {
-        const { id, provider, providerMetadataKey, auth, transport, endpoint, ...defaults } = patch
+        const { id, provider, auth, transport, endpoint, ...defaults } = patch
        return build({
          ...routeInput,
          id: id ?? routeInput.id,
          provider: provider ?? routeInput.provider,
-          providerMetadataKey:
-            providerMetadataKey ??
-            (provider !== undefined && String(provider) !== String(routeInput.provider)
-              ? String(provider)
-              : routeInput.providerMetadataKey),
          auth: auth ?? routeInput.auth,
          endpoint: endpoint ? Endpoint.merge(routeInput.endpoint, endpoint) : routeInput.endpoint,
          transport: (transport as Transport<Body, Prepared, Frame> | undefined) ?? routeInput.transport,
@@ -324,48 +318,19 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
        return Stream.unwrap(
          routeInput.transport.execute(prepared, request, runtime, options).pipe(
            Effect.map((execution) => {
-              const terminal = protocol.stream.terminal
-              // Preserve assembled inputs; replace only serialized event fallbacks with their original wire data.
-              const frameError =
-                (frame: Frame, event: Frame | Event = frame) =>
-                (error: AIError) =>
-                  new AIError({
-                    reason: AIErrorReason.make({
-                      ...error.reason,
-                      message: error.reason.message,
-                      cause: error.reason.cause,
-                      body:
-                        error.reason.body !== undefined && error.reason.body !== ProviderShared.encodeJson(event)
-                          ? error.reason.body
-                          : (execution.body?.(frame) ??
-                            (typeof frame === "string" ? frame : ProviderShared.encodeJson(frame))),
-                    }),
-                  })
              const events = execution.frames.pipe(
-                Stream.mapEffect((frame) =>
-                  decodeEvent(route)(frame).pipe(
-                    Effect.catchCause((cause) =>
-                      Effect.fail(streamError(route, `Failed to decode ${route} event`, cause)),
-                    ),
-                    Effect.map((event) => ({ event, frame })),
-                    Effect.mapError(frameError(frame)),
-                  ),
-                ),
-                terminal ? Stream.takeUntil(({ event }) => terminal(event)) : (stream) => stream,
+                Stream.mapEffect(decodeEvent(route)),
+                protocol.stream.terminal ? Stream.takeUntil(protocol.stream.terminal) : (stream) => stream,
              )
              const stream = Stream.suspend(() => {
                let state = protocol.stream.initial(request)
                const parsed = events.pipe(
-                  Stream.mapEffect(({ event, frame }) =>
+                  Stream.mapEffect((event) =>
                    protocol.stream.step(state, event).pipe(
-                      Effect.catchCause((cause) =>
-                        Effect.fail(streamError(route, `Failed to parse ${route} event`, cause)),
-                      ),
                      Effect.map(([next, output]) => {
                        state = next
                        return output
                      }),
-                      Effect.mapError(frameError(frame, event)),
                    ),
                  ),
                  Stream.flatMap(Stream.fromIterable),
@@ -374,24 +339,15 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
                return onHalt
                  ? parsed.pipe(
                      Stream.concat(
-                        Stream.suspend(() => Stream.unwrap(onHalt(state).pipe(Effect.map(Stream.fromIterable)))),
+                        Stream.suspend(() =>
+                          Stream.unwrap(onHalt(state).pipe(Effect.map(Stream.fromIterable))),
+                        ),
                      ),
                    )
                  : parsed
              }).pipe(
                Stream.catchCause((cause) => Stream.fail(streamError(route, `Failed to read ${route} stream`, cause))),
                requireTerminalEvent(route),
-                Stream.mapError(
-                  (error) =>
-                    new AIError({
-                      reason: AIErrorReason.make({
-                        ...error.reason,
-                        message: error.reason.message,
-                        cause: error.reason.cause,
-                        http: error.reason.http ?? execution.http,
-                      }),
-                    }),
-                ),
              )
              return execution.complete ? stream.pipe(Stream.onEnd(execution.complete)) : stream
            }),
@@ -7,7 +7,14 @@ import {
  HttpClientRequest,
  HttpClientResponse,
 } from "effect/unstable/http"
-import { HttpContext, HttpRateLimitDetails, AIError, TransportError } from "../schema/index.js"
+import {
+  HttpContext,
+  HttpRateLimitDetails,
+  HttpRequestDetails,
+  HttpResponseDetails,
+  AIError,
+  TransportReason,
+} from "../schema/index.js"
 import { classifyProviderFailure } from "../provider-error.js"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

-export class TransportError extends Schema.TaggedError<TransportError>("AI.Error.Transport")("Transport", {
-  ...ReasonFields,
+export class TransportReason extends Schema.Class<TransportReason>("AI.Error.Transport")({
+  _tag: Schema.tag("Transport"),
+  message: Schema.String,
  transport: TransportType,
  operation: TransportOperation,
  code: Schema.optional(Schema.String),
  url: Schema.optional(Schema.String),
+  http: Schema.optional(HttpContext),
  phase: Schema.optional(
    Schema.Literals(["prepare", "queue", "connect", "send", "receive", "decode", "complete", "fallback", "close"]),
  ),
@@ -95,40 +116,51 @@ export class TransportError extends Schema.TaggedError<TransportError>("AI.Error
  ),
 }) {}

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

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

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

 export class AIError extends Schema.TaggedError<AIError>()("AI.Error", {
+  module: Schema.String,
+  method: Schema.String,
  reason: AIErrorReason,
+  // Raw provider payload as a string, so classified failures never lose the
+  // original error detail even when the pretty message is a summary.
+  body: Schema.optional(Schema.String),
 }) {
  override readonly cause = this.reason

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

@@ -1,7 +1,7 @@
 import { describe, expect } from "bun:test"
 import { Deferred, Effect, Fiber, Layer, Ref, Stream } from "effect"
 import { Headers, HttpClient, HttpClientError, HttpClientRequest, HttpClientResponse } from "effect/unstable/http"
-import { LLM, AIError, HttpContext, InvalidProviderOutputError, TransportError } from "../src/index.js"
+import { LLM, AIError } from "../src/index.js"
 import { LLMClient, RequestExecutor, WebSocketTransport, type WebSocketChannelExecutor } from "../src/route.js"
 import * as OpenAIChat from "../src/protocols/openai-chat.js"
 import * as OpenAI from "../src/providers/openai.js"
@@ -62,85 +62,22 @@ const countedResponsesLayer = (attempts: Ref.Ref<number>, responses: ReadonlyArr
 const expectAIError = (error: unknown) => {
  expect(error).toBeInstanceOf(AIError)
  if (!(error instanceof AIError)) throw new Error("expected AIError")
-  expect(error.reason).toBeInstanceOf(Error)
-  expect(error.cause).toBe(error.reason)
  return error
 }

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

-  it.effect("preserves close reasons and native event causes without fabricated HTTP metadata", () =>
-    Effect.gen(function* () {
-      class TestSocket extends EventTarget {
-        readyState = globalThis.WebSocket.OPEN
-        send() {}
-        close() {}
-      }
-      const socket = new TestSocket()
-      const connection = yield* WebSocketTransport.fromWebSocket(
-        // oxlint-disable-next-line typescript-eslint/no-unsafe-type-assertion
-        socket as unknown as globalThis.WebSocket,
-        { url: "wss://provider.test/responses", headers: Headers.empty },
-      )
-      const event = new CloseEvent("close", { code: 1011, reason: "upstream trace: req_close" })
-      socket.dispatchEvent(event)
-      const error = yield* connection.messages.pipe(Stream.runDrain, Effect.flip)
-
-      expect(error.reason).toMatchObject({ _tag: "Transport", code: "1011", phase: "close" })
-      expect(error.message).toBe("WebSocket closed with code 1011")
-      expect(error.reason.body).toBe(event.reason)
-      expect(error.reason.cause).toBe(event)
-      expect(error.reason.http).toBeUndefined()
-      yield* connection.close
-    }),
-  )
-
-  it.effect("preserves opening event errors and native send exceptions", () =>
-    Effect.gen(function* () {
-      const cause = new Error("native send failed")
-      class TestSocket extends EventTarget {
-        readyState = globalThis.WebSocket.CONNECTING
-        send() {
-          throw cause
-        }
-        close() {}
-      }
-      const socket = new TestSocket()
-      const open = WebSocketTransport.fromWebSocket(
-        // oxlint-disable-next-line typescript-eslint/no-unsafe-type-assertion
-        socket as unknown as globalThis.WebSocket,
-        { url: "wss://provider.test/responses", headers: Headers.empty },
-      )
-      const fiber = yield* open.pipe(Effect.flip, Effect.forkChild({ startImmediately: true }))
-      const event = new ErrorEvent("error", { message: "handshake rejected", error: cause })
-      socket.dispatchEvent(event)
-      const error = yield* Fiber.join(fiber)
-      expect(error.reason.cause).toBe(cause)
-      expect(error.message).toContain("handshake rejected")
-      expect(error.reason.http).toBeUndefined()
-
-      socket.readyState = globalThis.WebSocket.OPEN
-      const connection = yield* open
-      const sent = yield* connection.sendText("create").pipe(Effect.flip)
-      expect(sent.reason.cause).toBe(cause)
-      expect(sent.message).toBe(cause.message)
-      yield* connection.close
-    }),
-  )
-
-  it.effect("preserves raw driver failures and known upgrade metadata", () =>
-    Effect.gen(function* () {
-      const cause = new Error("driver validation failed")
-      const frame = '{ "error": "failed", "trace": "original" }'
-      const http = new HttpContext({
-        url: "https://provider.test/responses",
-        status: 101,
-        headers: { upgrade: "websocket" },
-      })
-      const executor = WebSocketTransport.makeDirect({
-        open: () =>
-          Effect.succeed({
-            http,
-            sendText: () => Effect.void,
-            messages: Stream.make(frame),
-            close: Effect.void,
-          }),
-      })
-      const execution = yield* executor.execute({
-        id: "exchange_error",
-        connect: { url: "wss://provider.test/responses", headers: Headers.empty },
-        fallback: () => Stream.empty,
-        driver: {
-          create: () => Effect.succeed({ message: "create", mode: "full" }),
-          observe: () =>
-            Effect.succeed({
-              type: "provider-failure",
-              error: new AIError({
-                reason: new InvalidProviderOutputError({
-                  message: "Driver failed",
-                  cause,
-                  body: "narrowed",
-                }),
-              }),
-            }),
-        },
-      })
-      const error = yield* execution.frames.pipe(Stream.runDrain, Effect.flip)
-
-      expect(error.message).toBe("Driver failed")
-      expect(error.reason.body).toBe(frame)
-      expect(error.reason.cause).toBe(cause)
-      expect(error.reason.http).toBe(http)
-      expect(execution.http).toBe(http)
-    }),
-  )
-
-  it.effect("retains diagnostic fields when annotating transport delivery", () =>
-    Effect.gen(function* () {
-      const cause = new Error("connection closed")
-      const executor = WebSocketTransport.makeDirect({
-        open: () =>
-          Effect.succeed({
-            sendText: () => Effect.void,
-            messages: Stream.fail(
-              new AIError({
-                reason: new TransportError({
-                  message: "Socket closed",
-                  transport: "websocket",
-                  operation: "read",
-                  phase: "close",
-                  recovery: "retry-full",
-                  body: "server close detail",
-                  cause,
-                }),
-              }),
-            ),
-            close: Effect.void,
-          }),
-      })
-      const execution = yield* executor.execute({
-        id: "exchange_closed",
-        connect: { url: "wss://provider.test/responses", headers: Headers.empty },
-        fallback: () => Stream.empty,
-        driver: {
-          create: () => Effect.succeed({ message: "create", mode: "full" }),
-          observe: (_create, frame) => Effect.succeed({ type: "frame", frame }),
-        },
-      })
-      const error = yield* execution.frames.pipe(Stream.runDrain, Effect.flip)
-
-      expect(error.message).toBe("Socket closed")
-      expect(error.reason.body).toBe("server close detail")
-      expect(error.reason.cause).toBe(cause)
-      expect(error.reason).toMatchObject({ phase: "close", delivery: "ambiguous", recovery: "retry-full" })
-      expect(error.reason.http).toBeUndefined()
-    }),
-  )
-
  it.effect("runs a channel driver through the direct executor", () =>
    Effect.gen(function* () {
      const sent = yield* Ref.make("")
@@ -1,7 +1,10 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:azure", "provider:azure"],
+    "tags": [
+      "prefix:azure",
+      "provider:azure"
+    ],
    "name": "azure/chat-streams-text",
    "recordedAt": "2026-08-23T17:21:53.198Z"
  },
@@ -1,7 +1,10 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:azure", "provider:azure"],
+    "tags": [
+      "prefix:azure",
+      "provider:azure"
+    ],
    "name": "azure/responses-calls-a-tool",
    "recordedAt": "2026-08-23T17:21:55.170Z"
  },
@@ -1,7 +1,10 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:azure", "provider:azure"],
+    "tags": [
+      "prefix:azure",
+      "provider:azure"
+    ],
    "name": "azure/responses-continues-after-a-tool-result",
    "recordedAt": "2026-08-23T17:21:56.397Z"
  },
@@ -1,7 +1,10 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:azure", "provider:azure"],
+    "tags": [
+      "prefix:azure",
+      "provider:azure"
+    ],
    "name": "azure/responses-streams-text",
    "recordedAt": "2026-08-23T17:21:54.158Z"
  },
@@ -2,7 +2,11 @@
  "version": 1,
  "metadata": {
    "model": "openai.gpt-oss-120b",
-    "tags": ["prefix:bedrock-mantle", "provider:amazon-bedrock", "protocol:openai-responses"],
+    "tags": [
+      "prefix:bedrock-mantle",
+      "provider:amazon-bedrock",
+      "protocol:openai-responses"
+    ],
    "name": "bedrock-mantle/streams-text",
    "recordedAt": "2026-08-25T03:29:02.968Z"
  },
@@ -7,7 +7,13 @@
    "route": "cloudflare-workers-ai",
    "transport": "http",
    "model": "@cf/openai/gpt-oss-20b",
-    "tags": ["prefix:cloudflare-workers-ai", "provider:cloudflare-workers-ai", "tool", "tool-call", "golden"]
+    "tags": [
+      "prefix:cloudflare-workers-ai",
+      "provider:cloudflare-workers-ai",
+      "tool",
+      "tool-call",
+      "golden"
+    ]
  },
  "interactions": [
    {
@@ -29,4 +35,4 @@
      }
    }
  ]
-}
+}
@@ -1,32 +0,0 @@
-{
-  "version": 1,
-  "metadata": {
-    "provider": "deepinfra",
-    "route": "deepinfra-chat",
-    "transport": "http",
-    "model": "meta-llama/Llama-3.3-70B-Instruct-Turbo",
-    "tags": ["prefix:deepinfra-chat", "provider:deepinfra", "text", "golden"],
-    "name": "deepinfra-chat/deepinfra-llama-3-3-70b-text",
-    "recordedAt": "2026-08-26T00:34:03.019Z"
-  },
-  "interactions": [
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://api.deepinfra.com/v1/openai/chat/completions",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"messages\":[{\"role\":\"system\",\"content\":\"You are concise.\"},{\"role\":\"user\",\"content\":\"Reply exactly with: Hello!\"}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":40,\"temperature\":0}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream; charset=utf-8"
-        },
-        "body": "data: {\"service_tier\":null,\"id\":\"chatcmpl-RbZ8MyY5pos2MShihSmoXoRe\",\"object\":\"chat.completion.chunk\",\"created\":1787704442,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RbZ8MyY5pos2MShihSmoXoRe\",\"object\":\"chat.completion.chunk\",\"created\":1787704442,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"Hello\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RbZ8MyY5pos2MShihSmoXoRe\",\"object\":\"chat.completion.chunk\",\"created\":1787704442,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"!\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RbZ8MyY5pos2MShihSmoXoRe\",\"object\":\"chat.completion.chunk\",\"created\":1787704442,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":25,\"total_tokens\":28,\"completion_tokens\":3,\"estimated_cost\":null,\"prompt_tokens_details\":null}}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RbZ8MyY5pos2MShihSmoXoRe\",\"object\":\"chat.completion.chunk\",\"created\":1787704442,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[],\"usage\":{\"prompt_tokens\":25,\"total_tokens\":28,\"completion_tokens\":3,\"estimated_cost\":3.46e-6,\"prompt_tokens_details\":null}}\n\ndata: [DONE]\n\n"
-      }
-    }
-  ]
-}
@@ -1,32 +0,0 @@
-{
-  "version": 1,
-  "metadata": {
-    "provider": "deepinfra",
-    "route": "deepinfra-chat",
-    "transport": "http",
-    "model": "meta-llama/Llama-3.3-70B-Instruct-Turbo",
-    "tags": ["prefix:deepinfra-chat", "provider:deepinfra", "tool", "tool-call", "golden"],
-    "name": "deepinfra-chat/deepinfra-llama-3-3-70b-tool-call",
-    "recordedAt": "2026-08-26T00:34:04.173Z"
-  },
-  "interactions": [
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://api.deepinfra.com/v1/openai/chat/completions",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"messages\":[{\"role\":\"system\",\"content\":\"Call tools exactly as requested.\"},{\"role\":\"user\",\"content\":\"Call get_weather with city exactly Paris.\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}}],\"tool_choice\":{\"type\":\"function\",\"function\":{\"name\":\"get_weather\"}},\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":80,\"temperature\":0}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream; charset=utf-8"
-        },
-        "body": "data: {\"service_tier\":null,\"id\":\"chatcmpl-RFOwhlB2PhZrgMLviGrx5BQf\",\"object\":\"chat.completion.chunk\",\"created\":1787704443,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RFOwhlB2PhZrgMLviGrx5BQf\",\"object\":\"chat.completion.chunk\",\"created\":1787704443,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":[{\"index\":0,\"id\":\"call_SMfBjXa8eCmHLyjfeyARxe3a\",\"function\":{\"arguments\":\"{\\\"city\\\": \\\"Paris\\\"}\",\"name\":\"get_weather\"},\"type\":\"function\"}]},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RFOwhlB2PhZrgMLviGrx5BQf\",\"object\":\"chat.completion.chunk\",\"created\":1787704443,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":230,\"total_tokens\":244,\"completion_tokens\":14,\"estimated_cost\":null,\"prompt_tokens_details\":null}}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RFOwhlB2PhZrgMLviGrx5BQf\",\"object\":\"chat.completion.chunk\",\"created\":1787704443,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[],\"usage\":{\"prompt_tokens\":230,\"total_tokens\":244,\"completion_tokens\":14,\"estimated_cost\":0.000027480000000000005,\"prompt_tokens_details\":null}}\n\ndata: [DONE]\n\n"
-      }
-    }
-  ]
-}
@@ -1,50 +0,0 @@
-{
-  "version": 1,
-  "metadata": {
-    "provider": "deepinfra",
-    "route": "deepinfra-chat",
-    "transport": "http",
-    "model": "meta-llama/Llama-3.3-70B-Instruct-Turbo",
-    "tags": ["prefix:deepinfra-chat", "provider:deepinfra", "tool", "tool-loop", "golden"],
-    "name": "deepinfra-chat/deepinfra-llama-3-3-70b-tool-loop",
-    "recordedAt": "2026-08-26T00:34:05.656Z"
-  },
-  "interactions": [
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://api.deepinfra.com/v1/openai/chat/completions",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"messages\":[{\"role\":\"system\",\"content\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"},{\"role\":\"user\",\"content\":\"What is the weather in Paris?\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":80,\"temperature\":0}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream; charset=utf-8"
-        },
-        "body": "data: {\"service_tier\":null,\"id\":\"chatcmpl-RxlHFSnlLbUUz6XSxqBQj7TC\",\"object\":\"chat.completion.chunk\",\"created\":1787704444,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RxlHFSnlLbUUz6XSxqBQj7TC\",\"object\":\"chat.completion.chunk\",\"created\":1787704444,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":[{\"index\":0,\"id\":\"call_W3stxe7FNHozlB4tDxlTVZou\",\"function\":{\"arguments\":\"{\\\"city\\\": \\\"Paris\\\"}\",\"name\":\"get_weather\"},\"type\":\"function\"}]},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RxlHFSnlLbUUz6XSxqBQj7TC\",\"object\":\"chat.completion.chunk\",\"created\":1787704444,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":243,\"total_tokens\":257,\"completion_tokens\":14,\"estimated_cost\":null,\"prompt_tokens_details\":null}}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RxlHFSnlLbUUz6XSxqBQj7TC\",\"object\":\"chat.completion.chunk\",\"created\":1787704444,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[],\"usage\":{\"prompt_tokens\":243,\"total_tokens\":257,\"completion_tokens\":14,\"estimated_cost\":0.000028780000000000002,\"prompt_tokens_details\":null}}\n\ndata: [DONE]\n\n"
-      }
-    },
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://api.deepinfra.com/v1/openai/chat/completions",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"messages\":[{\"role\":\"system\",\"content\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"},{\"role\":\"user\",\"content\":\"What is the weather in Paris?\"},{\"role\":\"assistant\",\"content\":null,\"tool_calls\":[{\"id\":\"call_W3stxe7FNHozlB4tDxlTVZou\",\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"arguments\":\"{\\\"city\\\":\\\"Paris\\\"}\"}}],\"reasoning_content\":\"\"},{\"role\":\"tool\",\"tool_call_id\":\"call_W3stxe7FNHozlB4tDxlTVZou\",\"content\":\"{\\\"temperature\\\":22,\\\"condition\\\":\\\"sunny\\\"}\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":80,\"temperature\":0}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream; charset=utf-8"
-        },
-        "body": "data: {\"service_tier\":null,\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"Paris\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\" is\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\" sunny\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\".\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":276,\"total_tokens\":281,\"completion_tokens\":5,\"estimated_cost\":null,\"prompt_tokens_details\":null}}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[],\"usage\":{\"prompt_tokens\":276,\"total_tokens\":281,\"completion_tokens\":5,\"estimated_cost\":0.0000292,\"prompt_tokens_details\":null}}\n\ndata: [DONE]\n\n"
-      }
-    }
-  ]
-}
@@ -1,7 +1,11 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:google-vertex", "provider:google-vertex", "protocol:gemini"],
+    "tags": [
+      "prefix:google-vertex",
+      "provider:google-vertex",
+      "protocol:gemini"
+    ],
    "name": "google-vertex/calls-a-tool",
    "recordedAt": "2026-08-23T17:21:51.036Z"
  },
@@ -1,7 +1,11 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:google-vertex", "provider:google-vertex", "protocol:gemini"],
+    "tags": [
+      "prefix:google-vertex",
+      "provider:google-vertex",
+      "protocol:gemini"
+    ],
    "name": "google-vertex/continues-after-a-tool-result",
    "recordedAt": "2026-08-23T17:21:51.853Z"
  },
@@ -1,7 +1,11 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:google-vertex", "provider:google-vertex", "protocol:gemini"],
+    "tags": [
+      "prefix:google-vertex",
+      "provider:google-vertex",
+      "protocol:gemini"
+    ],
    "name": "google-vertex/streams-text",
    "recordedAt": "2026-08-23T17:21:50.112Z"
  },
@@ -1,29 +0,0 @@
-{
-  "version": 1,
-  "metadata": {
-    "model": "openai/gpt-oss-20b",
-    "tags": ["prefix:groq-chat", "provider:groq", "protocol:groq-chat", "text", "usage"],
-    "name": "groq-chat/streams-text-with-usage",
-    "recordedAt": "2026-08-26T14:40:09.833Z"
-  },
-  "interactions": [
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://api.groq.com/openai/v1/chat/completions",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"model\":\"openai/gpt-oss-20b\",\"messages\":[{\"role\":\"user\",\"content\":\"Reply with exactly one word: hello\"}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"reasoning_effort\":\"low\",\"max_completion_tokens\":512,\"include_reasoning\":false,\"service_tier\":\"on_demand\",\"user\":\"recorded-test\"}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream"
-        },
-        "body": "data: {\"id\":\"chatcmpl-660048b0-3c99-4215-a582-e116b77eb881\",\"object\":\"chat.completion.chunk\",\"created\":1787755209,\"model\":\"openai/gpt-oss-20b\",\"system_fingerprint\":\"fp_66891002f6\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\"},\"logprobs\":null,\"finish_reason\":null}],\"x_groq\":{\"id\":\"req_01m0z87915eep9bpf10gg7331e\",\"seed\":94036161}}\n\ndata: {\"id\":\"chatcmpl-660048b0-3c99-4215-a582-e116b77eb881\",\"object\":\"chat.completion.chunk\",\"created\":1787755209,\"model\":\"openai/gpt-oss-20b\",\"system_fingerprint\":\"fp_66891002f6\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\"hello\"},\"logprobs\":null,\"finish_reason\":null}]}\n\ndata: {\"id\":\"chatcmpl-660048b0-3c99-4215-a582-e116b77eb881\",\"object\":\"chat.completion.chunk\",\"created\":1787755209,\"model\":\"openai/gpt-oss-20b\",\"system_fingerprint\":\"fp_66891002f6\",\"choices\":[{\"index\":0,\"delta\":{},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"x_groq\":{\"id\":\"req_01m0z87915eep9bpf10gg7331e\",\"usage\":{\"queue_time\":0.10886435,\"prompt_tokens\":78,\"prompt_time\":0.003693734,\"completion_tokens\":20,\"completion_time\":0.020459983,\"total_tokens\":98,\"total_time\":0.024153717,\"completion_tokens_details\":{\"reasoning_tokens\":10}}},\"usage\":{\"queue_time\":0.10886435,\"prompt_tokens\":78,\"prompt_time\":0.003693734,\"completion_tokens\":20,\"completion_time\":0.020459983,\"total_tokens\":98,\"total_time\":0.024153717,\"completion_tokens_details\":{\"reasoning_tokens\":10}}}\n\ndata: {\"id\":\"chatcmpl-660048b0-3c99-4215-a582-e116b77eb881\",\"object\":\"chat.completion.chunk\",\"created\":1787755209,\"model\":\"openai/gpt-oss-20b\",\"system_fingerprint\":\"fp_66891002f6\",\"choices\":[],\"usage\":{\"queue_time\":0.10886435,\"prompt_tokens\":78,\"prompt_time\":0.003693734,\"completion_tokens\":20,\"completion_time\":0.020459983,\"total_tokens\":98,\"total_time\":0.024153717,\"completion_tokens_details\":{\"reasoning_tokens\":10}},\"service_tier\":\"on_demand\"}\n\ndata: [DONE]\n\n"
-      }
-    }
-  ]
-}
@@ -1,7 +1,12 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:openai-responses-cache", "provider:openai", "protocol:openai-responses", "cache"],
+    "tags": [
+      "prefix:openai-responses-cache",
+      "provider:openai",
+      "protocol:openai-responses",
+      "cache"
+    ],
    "name": "openai-responses-cache/reports-cached-tokens-on-identical-second-call",
    "recordedAt": "2026-08-25T03:29:25.124Z"
  },
@@ -5,7 +5,14 @@
    "route": "openai-responses",
    "transport": "http",
    "model": "gpt-5.5",
-    "tags": ["prefix:openai-responses", "provider:openai", "flagship", "tool", "tool-loop", "golden"],
+    "tags": [
+      "prefix:openai-responses",
+      "provider:openai",
+      "flagship",
+      "tool",
+      "tool-loop",
+      "golden"
+    ],
    "name": "openai-responses/openai-responses-gpt-5-5-tool-loop",
    "recordedAt": "2026-08-20T06:30:22.262Z"
  },
@@ -1,7 +1,14 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:pdf", "pdf", "provider:openai", "protocol:openai-responses", "tool", "tool-result"],
+    "tags": [
+      "prefix:pdf",
+      "pdf",
+      "provider:openai",
+      "protocol:openai-responses",
+      "tool",
+      "tool-result"
+    ],
    "name": "pdf/openai-tool-result",
    "recordedAt": "2026-08-25T03:29:08.297Z"
  },
@@ -1,7 +1,13 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:pdf", "pdf", "provider:openai", "protocol:openai-responses", "user-input"],
+    "tags": [
+      "prefix:pdf",
+      "pdf",
+      "provider:openai",
+      "protocol:openai-responses",
+      "user-input"
+    ],
    "name": "pdf/openai-user-input",
    "recordedAt": "2026-08-25T03:29:05.645Z"
  },
@@ -1,7 +1,14 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:pdf", "pdf", "provider:xai", "protocol:xai-responses", "tool", "tool-result"],
+    "tags": [
+      "prefix:pdf",
+      "pdf",
+      "provider:xai",
+      "protocol:xai-responses",
+      "tool",
+      "tool-result"
+    ],
    "name": "pdf/xai-tool-result",
    "recordedAt": "2026-08-25T03:29:11.774Z"
  },
@@ -1,7 +1,13 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:pdf", "pdf", "provider:xai", "protocol:xai-responses", "user-input"],
+    "tags": [
+      "prefix:pdf",
+      "pdf",
+      "provider:xai",
+      "protocol:xai-responses",
+      "user-input"
+    ],
    "name": "pdf/xai-user-input",
    "recordedAt": "2026-08-25T03:29:10.612Z"
  },
@@ -52,4 +52,4 @@
      }
    }
  ]
-}
+}
@@ -2,7 +2,12 @@
  "version": 1,
  "metadata": {
    "model": "anthropic/claude-sonnet-4.6",
-    "tags": ["prefix:openai-compatible-chat", "provider:vercel-ai-gateway", "protocol:openai-chat", "reasoning"],
+    "tags": [
+      "prefix:openai-compatible-chat",
+      "provider:vercel-ai-gateway",
+      "protocol:openai-chat",
+      "reasoning"
+    ],
    "name": "vercel-ai-gateway-reasoning",
    "recordedAt": "2026-07-18T11:28:42.077Z"
  },
@@ -26,4 +31,4 @@
      }
    }
  ]
-}
+}
@@ -7,37 +7,6 @@ import { it } from "./lib/effect.js"
 import { dynamicResponse } from "./lib/http.js"

 describe("Image", () => {
-  for (const provider of [OpenAI, Google, XAI, ZAI]) {
-    const model = provider.configure({ apiKey: "test", baseURL: "https://image.test" }).image("image-model")
-    for (const body of ['{"data":42,"candidates":42,"opaque":{"nested":[1,2]},"trace":"outer"}', '{"invalid":']) {
-      it.effect(`retains ${model.provider} image response body and decode cause: ${body}`, () =>
-        Effect.gen(function* () {
-          const error = yield* Image.generate({ model, prompt: "hello" }).pipe(Effect.flip)
-          expect(error.reason._tag).toBe("InvalidProviderOutput")
-          expect(error.message).toContain("invalid response")
-          expect(error.reason.body).toBe(body)
-          expect(error.reason.http).toMatchObject({ status: 200, headers: { "x-image-trace": "trace-1" } })
-          expect(error.reason.http?.url).toStartWith("https://image.test/")
-          expect(error.reason.cause).toBeInstanceOf(Error)
-        }).pipe(
-          Effect.provide(
-            ImageClient.layer.pipe(
-              Layer.provide(
-                dynamicResponse((input) =>
-                  Effect.succeed(
-                    input.respond(body, {
-                      headers: { "content-type": "application/json", "x-image-trace": "trace-1" },
-                    }),
-                  ),
-                ),
-              ),
-            ),
-          ),
-        ),
-      )
-    }
-  }
-
  it.effect("generates images through the OpenAI Images API", () =>
    Effect.gen(function* () {
      const response = yield* Image.generate({
@@ -561,17 +530,22 @@ describe("Image", () => {
        Effect.sync(() => {
          expect(error.reason._tag).toBe("InvalidProviderOutput")
          if (error.reason._tag !== "InvalidProviderOutput") return
-          expect(error.message).toContain("finish reasons: IMAGE_SAFETY")
-          expect(JSON.parse(error.reason.body ?? "")).toEqual({
-            promptFeedback: { blockReason: "SAFETY" },
-            candidates: [
-              {
-                finishReason: "IMAGE_SAFETY",
-                finishMessage: "The generated image was blocked by safety filters.",
-                safetyRatings: [{ category: "HARM_CATEGORY_DANGEROUS_CONTENT", blocked: true }],
-                content: { parts: [{ text: "blocked", thought: false }] },
-              },
-            ],
+          expect(error.reason.message).toContain("finish reasons: IMAGE_SAFETY")
+          expect(error.reason.providerMetadata).toEqual({
+            google: {
+              promptFeedback: { blockReason: "SAFETY" },
+              candidates: [
+                {
+                  index: 0,
+                  finishReason: "IMAGE_SAFETY",
+                  finishMessage: "The generated image was blocked by safety filters.",
+                  safetyRatings: [{ category: "HARM_CATEGORY_DANGEROUS_CONTENT", blocked: true }],
+                  citationMetadata: undefined,
+                  groundingMetadata: undefined,
+                  parts: [{ type: "text", text: "blocked", thought: false, thoughtSignature: undefined }],
+                },
+              ],
+            },
          })
        }),
      ),
@@ -18,19 +18,13 @@ describe("provider error classification", () => {
    expect(messages.every(isContextOverflow)).toBe(true)
  })

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

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

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

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

@@ -111,39 +108,6 @@ describe("provider error classification", () => {
 })

 describe("provider error rawBody classification", () => {
-  test("classifies provider envelopes without separate code inputs", () => {
-    const cases = [
-      ['{"type":"error","error":{"type":"overloaded_error","message":"Try again"}}', "ProviderInternal"],
-      ['{"error":{"code":"insufficient_quota","message":"Request failed"}}', "QuotaExceeded"],
-      [
-        '{"type":"response.failed","response":{"error":{"code":"authentication_error","message":"Denied"}}}',
-        "Authentication",
-      ],
-      ['{"error":{"code":429,"status":"RESOURCE_EXHAUSTED","message":"Try again"}}', "ProviderInternal"],
-      ['{"exception":{"type":"throttlingException","details":{"message":"Try again"}}}', "RateLimit"],
-    ] as const
-    for (const [rawBody, expected] of cases) {
-      const reason = classifyProviderFailure({ message: "Request failed", rawBody })
-      expect(reason._tag).toBe(expected)
-      expect(reason.body).toBe(rawBody)
-      expect(reason).not.toHaveProperty("code")
-    }
-  })
-
-  test("classifies separately supplied SDK data without replacing the response body", () => {
-    const data = { error: { code: "authentication_error" } }
-    for (const value of [data, JSON.stringify(data)]) {
-      const reason = classifyProviderFailure({
-        message: "Request failed",
-        status: 400,
-        rawBody: '{"message":"Request failed"}',
-        data: value,
-      })
-      expect(reason._tag).toBe("Authentication")
-      expect(reason.body).toBe('{"message":"Request failed"}')
-    }
-  })
-
  test("classifies overflow signals buried in the raw payload when the summary is vague", () => {
    const reason = classifyProviderFailure({
      message: "Request failed",
@@ -28,8 +28,6 @@ describe("provider package entrypoints", () => {
      import("@opencode-ai/ai/providers/amazon-bedrock/mantle/responses"),
      import("@opencode-ai/ai/providers/togetherai"),
      import("@opencode-ai/ai/providers/cerebras"),
-      import("@opencode-ai/ai/providers/deepinfra"),
-      import("@opencode-ai/ai/providers/groq"),
    ])

    for (const module of modules) expect(module.model).toBeFunction()
@@ -39,24 +37,6 @@ describe("provider package entrypoints", () => {
    expect(modules[19].model).toBe(modules[20].model)
  })

-  test("maps DeepInfra package settings onto its native executable model", async () => {
-    const DeepInfra = await import("@opencode-ai/ai/providers/deepinfra")
-    const settings = {
-      apiKey: "fixture",
-      baseURL: "https://provider.example.test/v1/",
-      headers: { "x-application": "opencode" },
-      body: { service_tier: "priority" },
-      providerOptions: { reasoningEffort: "high" as const },
-    }
-    const deepinfra = DeepInfra.model("google/gemma-3-27b-it", settings)
-
-    expect(deepinfra.route.id).toBe("deepinfra-chat")
-    expect(deepinfra.route.endpoint.baseURL).toBe("https://provider.example.test/v1/openai")
-    expect(deepinfra.route.defaults.providerOptions).toEqual(settings.providerOptions)
-    expect(deepinfra.route.defaults.headers).toEqual(settings.headers)
-    expect(deepinfra.route.defaults.http?.body).toEqual(settings.body)
-  })
-
  test("maps OpenRouter and xAI package settings onto executable models", async () => {
    const OpenRouter = await import("@opencode-ai/ai/providers/openrouter")
    const XAI = await import("@opencode-ai/ai/providers/xai")
@@ -39,7 +39,7 @@ describe("Anthropic Messages sad-path recorded", () => {

      expect(error).toBeInstanceOf(AIError)
      expect(error.reason).toMatchObject({ _tag: "InvalidRequest" })
-      expect(error.message).toContain("`tool_use` ids were found without `tool_result` blocks")
+      expect(error.reason.message).toContain("`tool_use` ids were found without `tool_result` blocks")
    }),
  )
 })
@@ -2,7 +2,7 @@ import { describe, expect } from "bun:test"
 import { Effect } from "effect"
 import { HttpClientRequest } from "effect/unstable/http"
 import { CacheHint, LLM, AIError, LLMRequest, Message, ToolCallPart, ToolDefinition, Usage } from "../../src/index.js"
-import { Auth, Endpoint, LLMClient, Route } from "../../src/route.js"
+import { Auth, LLMClient } from "../../src/route.js"
 import { compileRequest } from "../../src/route/client.js"
 import * as AnthropicMessages from "../../src/protocols/anthropic-messages.js"
 import { GoogleVertexMessages } from "../../src/providers.js"
@@ -810,99 +810,6 @@ describe("Anthropic Messages route", () => {
    }),
  )

-  it.effect("round-trips compatible provider metadata in its own namespace", () =>
-    Effect.gen(function* () {
-      const compatible = Route.make({
-        id: "custom-anthropic-messages",
-        provider: "custom-anthropic",
-        protocol: AnthropicMessages.protocol,
-        endpoint: Endpoint.path("/messages", { baseURL: "https://compatible.test/v1" }),
-        auth: Auth.header("x-api-key", "test"),
-        framing: AnthropicMessages.framing,
-      }).model({ id: "custom-model" })
-      const result = [
-        {
-          type: "web_search_result",
-          url: "https://example.com",
-          citations: [{ type: "web_search_result_location", cited_text: "Example" }],
-        },
-      ]
-      const response = yield* LLMClient.generate(LLM.request({ model: compatible, prompt: "Search." })).pipe(
-        Effect.provide(
-          fixedResponse(
-            sseEvents(
-              { type: "message_start", message: { usage: { input_tokens: 5, custom_start: true } } },
-              { type: "content_block_start", index: 0, content_block: { type: "thinking", thinking: "Thinking." } },
-              { type: "content_block_delta", index: 0, delta: { type: "signature_delta", signature: "custom_sig" } },
-              { type: "content_block_stop", index: 0 },
-              {
-                type: "content_block_start",
-                index: 1,
-                content_block: { type: "redacted_thinking", data: "custom_redacted" },
-              },
-              { type: "content_block_stop", index: 1 },
-              {
-                type: "content_block_start",
-                index: 2,
-                content_block: {
-                  type: "server_tool_use",
-                  id: "custom_tool",
-                  name: "web_search",
-                  input: { query: "example" },
-                },
-              },
-              { type: "content_block_stop", index: 2 },
-              {
-                type: "content_block_start",
-                index: 3,
-                content_block: { type: "web_search_tool_result", tool_use_id: "custom_tool", content: result },
-              },
-              { type: "content_block_stop", index: 3 },
-              {
-                type: "message_delta",
-                delta: { stop_reason: "end_turn", stop_sequence: "custom_stop" },
-                usage: { output_tokens: 2, custom_terminal: true },
-              },
-              { type: "message_stop" },
-            ),
-          ),
-        ),
-      )
-
-      expect(response.message.content).toMatchObject([
-        { type: "reasoning", text: "Thinking.", providerMetadata: { "custom-anthropic": { signature: "custom_sig" } } },
-        { type: "reasoning", text: "", providerMetadata: { "custom-anthropic": { redactedData: "custom_redacted" } } },
-        { type: "tool-call", id: "custom_tool", providerExecuted: true },
-        {
-          type: "tool-result",
-          providerExecuted: true,
-          providerMetadata: { "custom-anthropic": { blockType: "web_search_tool_result", result } },
-        },
-      ])
-      expect(response.usage?.providerMetadata).toEqual({
-        "custom-anthropic": { input_tokens: 5, custom_start: true, output_tokens: 2, custom_terminal: true },
-      })
-      expect(response.events.at(-1)).toMatchObject({
-        providerMetadata: { "custom-anthropic": { stopSequence: "custom_stop" } },
-      })
-
-      const prepared = yield* compileRequest(
-        LLM.request({ model: compatible, messages: [response.message], cache: "none" }),
-      )
-      expect(prepared.body.messages).toEqual([
-        {
-          role: "assistant",
-          content: [
-            { type: "thinking", thinking: "Thinking.", signature: "custom_sig" },
-            { type: "redacted_thinking", data: "custom_redacted" },
-            { type: "server_tool_use", id: "custom_tool", name: "web_search", input: { query: "example" } },
-            { type: "web_search_tool_result", tool_use_id: "custom_tool", content: result },
-          ],
-        },
-      ])
-    }),
-  )
-
  it.effect("parses text, reasoning, and usage stream fixtures", () =>
    Effect.gen(function* () {
      const body = sseEvents(
@@ -966,8 +873,9 @@ describe("Anthropic Messages route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({
-        reason: { _tag: "InvalidProviderOutput", classification: "incomplete-stream" },
+      expect(error.reason).toMatchObject({
+        _tag: "InvalidProviderOutput",
+        classification: "incomplete-stream",
        message: "The provider response ended unexpectedly.",
      })
    }),
@@ -1056,8 +964,8 @@ describe("Anthropic Messages route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({
-        reason: { _tag: "InvalidProviderOutput" },
+      expect(error.reason).toMatchObject({
+        _tag: "InvalidProviderOutput",
        message: "Invalid anthropic/anthropic-messages stream event",
      })
    }),
@@ -1078,8 +986,8 @@ describe("Anthropic Messages route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({
-        reason: { _tag: "InvalidProviderOutput" },
+      expect(error.reason).toMatchObject({
+        _tag: "InvalidProviderOutput",
        message: "Invalid anthropic/anthropic-messages stream event",
      })
    }),
@@ -1105,8 +1013,8 @@ describe("Anthropic Messages route", () => {
            Effect.flip,
          )

-          expect(error).toMatchObject({
-            reason: { _tag: "InvalidProviderOutput" },
+          expect(error.reason).toMatchObject({
+            _tag: "InvalidProviderOutput",
            message: "Invalid anthropic/anthropic-messages stream event",
          })
        }),
@@ -1121,8 +1029,8 @@ describe("Anthropic Messages route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({
-        reason: { _tag: "InvalidProviderOutput" },
+      expect(error.reason).toMatchObject({
+        _tag: "InvalidProviderOutput",
        message: "Invalid anthropic/anthropic-messages stream event",
      })
    }),
@@ -1644,7 +1552,7 @@ describe("Anthropic Messages route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({ reason: { _tag: "ProviderInternal" }, message: "overloaded_error: Overloaded" })
+      expect(error.reason).toMatchObject({ _tag: "ProviderInternal", message: "overloaded_error: Overloaded" })
    }),
  )

@@ -1662,9 +1570,10 @@ describe("Anthropic Messages route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({
-        reason: { _tag: "InvalidRequest", classification: "context-overflow" },
+      expect(error.reason).toMatchObject({
+        _tag: "InvalidRequest",
        message: "invalid_request_error: prompt is too long: 210000 tokens",
+        classification: "context-overflow",
      })
    }),
  )
@@ -1676,7 +1585,7 @@ describe("Anthropic Messages route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({ reason: { _tag: "ProviderInternal" }, message: "overloaded_error" })
+      expect(error.reason).toMatchObject({ _tag: "ProviderInternal", message: "overloaded_error" })
    }),
  )

@@ -1687,7 +1596,7 @@ describe("Anthropic Messages route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({ reason: { _tag: "UnknownProvider" }, message: "Anthropic Messages stream error" })
+      expect(error.reason).toMatchObject({ _tag: "UnknownProvider", message: "Anthropic Messages stream error" })
    }),
  )

@@ -1704,7 +1613,7 @@ describe("Anthropic Messages route", () => {
      )

      expect(error).toBeInstanceOf(AIError)
-      expect(error).toMatchObject({ reason: { _tag: "InvalidRequest" }, message: "Bad request" })
+      expect(error.reason).toMatchObject({ _tag: "InvalidRequest", message: "Bad request" })
    }),
  )

@@ -569,57 +569,6 @@ describe("Bedrock Converse route", () => {
    }),
  )

-  it.effect("round-trips reassigned provider reasoning and usage metadata in its own namespace", () =>
-    Effect.gen(function* () {
-      const compatible = model.route.with({ provider: "custom-bedrock" }).model({ id: model.id })
-      const redactedData = "cmVkYWN0ZWQtdGhpbmtpbmc="
-      const response = yield* LLMClient.generate(LLMRequest.update(baseRequest, { model: compatible })).pipe(
-        Effect.provide(
-          fixedBytes(
-            eventStreamBody(
-              ["messageStart", { role: "assistant" }],
-              ["contentBlockDelta", { contentBlockIndex: 0, delta: { reasoningContent: { text: "Let me think." } } }],
-              ["contentBlockDelta", { contentBlockIndex: 0, delta: { reasoningContent: { signature: "custom_sig" } } }],
-              ["contentBlockStop", { contentBlockIndex: 0 }],
-              [
-                "contentBlockDelta",
-                { contentBlockIndex: 1, delta: { reasoningContent: { redactedContent: redactedData } } },
-              ],
-              ["contentBlockStop", { contentBlockIndex: 1 }],
-              ["messageStop", { stopReason: "end_turn" }],
-              ["metadata", { usage: { inputTokens: 5, outputTokens: 2, totalTokens: 7 } }],
-            ),
-          ),
-        ),
-      )
-
-      expect(response.message.content).toEqual([
-        {
-          type: "reasoning",
-          text: "Let me think.",
-          providerMetadata: { "custom-bedrock": { signature: "custom_sig" } },
-        },
-        { type: "reasoning", text: "", providerMetadata: { "custom-bedrock": { redactedData } } },
-      ])
-      expect(response.usage?.providerMetadata).toEqual({
-        "custom-bedrock": { inputTokens: 5, outputTokens: 2, totalTokens: 7 },
-      })
-
-      const prepared = yield* compileRequest(
-        LLM.request({ model: compatible, messages: [response.message], cache: "none" }),
-      )
-      expect(prepared.body.messages).toEqual([
-        {
-          role: "assistant",
-          content: [
-            { reasoningContent: { reasoningText: { text: "Let me think.", signature: "custom_sig" } } },
-            { reasoningContent: { redactedContent: redactedData } },
-          ],
-        },
-      ])
-    }),
-  )
-
  it.effect("preserves reasoning signatures when contentBlockStop is missing", () =>
    Effect.gen(function* () {
      const response = yield* LLMClient.generate(baseRequest).pipe(
@@ -789,32 +738,19 @@ describe("Bedrock Converse route", () => {
      ])
      const error = yield* LLMClient.generate(baseRequest).pipe(Effect.provide(fixedBytes(body)), Effect.flip)

-      expect(error).toMatchObject({ reason: { _tag: "UnknownProvider" }, message: "A future provider failure" })
+      expect(error.reason).toMatchObject({ _tag: "UnknownProvider", message: "A future provider failure" })
    }),
  )

  it.effect("classifies throttlingException as a rate limit", () =>
    Effect.gen(function* () {
-      const payload = { message: "Slow down", details: { opaque: [1, 2] }, trace: "outer", p: "padding" }
      const body = concat([
        eventFrame("messageStart", { role: "assistant" }),
-        exceptionFrame("throttlingException", payload),
+        exceptionFrame("throttlingException", { message: "Slow down" }),
      ])
      const error = yield* LLMClient.generate(baseRequest).pipe(Effect.provide(fixedBytes(body)), Effect.flip)

-      expect(error).toMatchObject({ reason: { _tag: "RateLimit" }, message: "Slow down" })
-      expect(JSON.parse(error.reason.body ?? "")).toEqual({
-        headers: {
-          ":message-type": { type: "string", value: "exception" },
-          ":exception-type": { type: "string", value: "throttlingException" },
-          ":content-type": { type: "string", value: "application/json" },
-        },
-        body: JSON.stringify(payload),
-      })
-      expect(error.reason.http).toMatchObject({
-        status: 200,
-        headers: { "content-type": "application/vnd.amazon.eventstream" },
-      })
+      expect(error.reason).toMatchObject({ _tag: "RateLimit", message: "Slow down" })
    }),
  )

@@ -827,9 +763,10 @@ describe("Bedrock Converse route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({
-        reason: { _tag: "InvalidRequest", classification: "context-overflow" },
+      expect(error.reason).toMatchObject({
+        _tag: "InvalidRequest",
        message: "Input is too long for requested model",
+        classification: "context-overflow",
      })
    }),
  )
@@ -848,7 +785,7 @@ describe("Bedrock Converse route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({ reason: { _tag: "ProviderInternal" }, message: "Upstream model failed" })
+      expect(error.reason).toMatchObject({ _tag: "ProviderInternal", message: "Upstream model failed" })
    }),
  )

@@ -859,32 +796,10 @@ describe("Bedrock Converse route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({
-        reason: { _tag: "InvalidProviderOutput" },
+      expect(error.reason).toMatchObject({
+        _tag: "InvalidProviderOutput",
        message: "BadStream: Stream failed",
      })
-      expect(JSON.parse(error.reason.body ?? "")).toMatchObject({
-        headers: { ":error-code": { value: "BadStream" } },
-        body: "",
-      })
-    }),
-  )
-
-  it.effect("retains malformed AWS payloads with headers and decode cause", () =>
-    Effect.gen(function* () {
-      const headers = {
-        ":message-type": { type: "string" as const, value: "event" },
-        ":event-type": { type: "string" as const, value: "messageStart" },
-      }
-      const body = '{"malformed":'
-      const error = yield* LLMClient.generate(baseRequest).pipe(
-        Effect.provide(fixedBytes(codec.encode({ headers, body: utf8Encoder.encode(body) }))),
-        Effect.flip,
-      )
-      expect(error.reason._tag).toBe("InvalidProviderOutput")
-      expect(JSON.parse(error.reason.body ?? "")).toEqual({ headers, body })
-      expect(error.reason.cause).toBeInstanceOf(Error)
-      expect(error.reason.http?.status).toBe(200)
    }),
  )

@@ -1,13 +1,11 @@
 import { describe, expect } from "bun:test"
 import { Effect } from "effect"
 import { HttpClientRequest } from "effect/unstable/http"
-import { LLM, Message } from "../../src/index.js"
+import { LLM } from "../../src/index.js"
 import { AmazonBedrockMantle } from "../../src/providers.js"
-import { model } from "../../src/providers/amazon-bedrock/mantle.js"
-import { OpenAIResponses } from "../../src/protocols/openai-responses.js"
 import { compileRequest, LLMClient } from "../../src/route/client.js"
 import { it } from "../lib/effect.js"
-import { dynamicResponse, fixedResponse } from "../lib/http.js"
+import { dynamicResponse } from "../lib/http.js"
 import { sseEvents } from "../lib/sse.js"
 import { recordedTests } from "../recorded-test.js"

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

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

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

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

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

      const replay = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
@@ -1,102 +0,0 @@
-import { describe, expect } from "bun:test"
-import { Effect } from "effect"
-import { LLM } from "../../src/index.js"
-import { Anthropic, Google, OpenAI } from "../../src/providers.js"
-import { LLMClient } from "../../src/route.js"
-import { it } from "../lib/effect.js"
-import { fixedResponse } from "../lib/http.js"
-import { sseEvents } from "../lib/sse.js"
-
-describe("provider error retention", () => {
-  const options = { apiKey: "test", baseURL: "https://provider.test" }
-  const cases = [
-    {
-      name: "Anthropic",
-      model: Anthropic.configure(options).model("claude"),
-      event: { type: "error", error: { type: "rate_limit_error", message: "Slow down", details: { opaque: [1, 2] } } },
-    },
-    {
-      name: "OpenAI Chat",
-      model: OpenAI.configure(options).chat("gpt"),
-      event: { error: { code: "rate_limit_exceeded", message: "Slow down", details: { opaque: [1, 2] } } },
-    },
-    {
-      name: "OpenAI Responses",
-      model: OpenAI.configure(options).responses("gpt"),
-      event: {
-        type: "response.failed",
-        response: {
-          id: "resp_error",
-          error: { code: "rate_limit_exceeded", message: "Slow down", details: { opaque: [1, 2] } },
-          opaque: { upstream: true },
-        },
-      },
-    },
-    {
-      name: "Gemini",
-      model: Google.configure(options).model("gemini"),
-      event: { error: { code: 429, status: "RESOURCE_EXHAUSTED", message: "Slow down", details: { opaque: [1, 2] } } },
-    },
-  ]
-
-  for (const entry of cases) {
-    it.effect(`retains the complete ${entry.name} event and HTTP context`, () =>
-      Effect.gen(function* () {
-        const body = JSON.stringify({ ...entry.event, trace: { opaque: "outer" } })
-        const error = yield* LLMClient.generate(LLM.request({ model: entry.model, prompt: "hello" })).pipe(
-          Effect.provide(
-            fixedResponse(sseEvents(body), {
-              headers: { "content-type": "text/event-stream", "x-provider-trace": "trace-1" },
-            }),
-          ),
-          Effect.flip,
-        )
-        expect(error.message).toContain("Slow down")
-        expect(error.reason._tag).toBe(entry.name === "Gemini" ? "ProviderInternal" : "RateLimit")
-        expect(error.reason.body).toBe(body)
-        expect(error.reason.http).toMatchObject({ status: 200, headers: { "x-provider-trace": "trace-1" } })
-        expect(error.reason.http?.url).toStartWith("https://provider.test/")
-        expect(error.reason.cause).toBeUndefined()
-        expect(error.cause).toBe(error.reason)
-      }),
-    )
-  }
-
-  it.effect("retains malformed provider frames and the original decode cause", () =>
-    Effect.gen(function* () {
-      const body = '{"type":"error","error":{"message":42,"opaque":{"nested":true}},"trace":"outer"}'
-      const error = yield* LLMClient.generate(
-        LLM.request({ model: Anthropic.configure(options).model("claude"), prompt: "hello" }),
-      ).pipe(Effect.provide(fixedResponse(sseEvents(body))), Effect.flip)
-      expect(error.reason._tag).toBe("InvalidProviderOutput")
-      expect(error.reason.body).toBe(body)
-      expect(error.reason.cause).toBeInstanceOf(Error)
-      expect(error.reason.http?.status).toBe(200)
-    }),
-  )
-
-  it.effect("retains the HTTP response context when a channel falls back", () =>
-    Effect.gen(function* () {
-      const body = '{"type":"error","error":{"code":"rate_limit_exceeded","message":"Slow down","extra":42}}'
-      const error = yield* LLMClient.generate(
-        LLM.request({ model: OpenAI.configure(options).responses("gpt"), prompt: "hello" }),
-        {
-          webSocket: {
-            execute: (exchange) => Effect.succeed({ frames: exchange.fallback(), complete: Effect.void }),
-          },
-        },
-      ).pipe(
-        Effect.provide(fixedResponse(sseEvents(body), { headers: { "x-provider-trace": "fallback-1" } })),
-        Effect.flip,
-      )
-
-      expect(error.reason._tag).toBe("RateLimit")
-      expect(error.reason.body).toBe(body)
-      expect(error.reason.http).toMatchObject({
-        url: "https://provider.test/responses",
-        status: 200,
-        headers: { "x-provider-trace": "fallback-1" },
-      })
-    }),
-  )
-})
@@ -515,10 +515,7 @@ describe("Gemini route", () => {
        {
          role: "model",
          parts: [
-            {
-              functionCall: { id: "call_image", name: "read", args: { path: "pixel.png" } },
-              thoughtSignature: "sig_1",
-            },
+            { functionCall: { id: "call_image", name: "read", args: { path: "pixel.png" } }, thoughtSignature: "sig_1" },
          ],
        },
        {
@@ -609,7 +606,10 @@ describe("Gemini route", () => {
      expect(prepared.body.contents).toEqual([
        {
          role: "model",
-          parts: [{ functionCall: { name: "shot", args: {} } }, { functionCall: { name: "shot", args: {} } }],
+          parts: [
+            { functionCall: { name: "shot", args: {} } },
+            { functionCall: { name: "shot", args: {} } },
+          ],
        },
        {
          role: "user",
@@ -1071,9 +1071,7 @@ describe("Gemini route", () => {
      const prepared = yield* compileRequest(
        LLM.request({
          model,
-          messages: [
-            Message.assistant([{ type: "text", text: "All done.", providerMetadata: delta?.providerMetadata }]),
-          ],
+          messages: [Message.assistant([{ type: "text", text: "All done.", providerMetadata: delta?.providerMetadata }])],
        }),
      )
      expect(prepared.body.contents).toEqual([
@@ -1574,7 +1572,9 @@ describe("Gemini route", () => {
              { candidates: [{ content: { role: "model", parts: null } }] },
              { candidates: [{ content: null, finishReason: null }] },
              {
-                candidates: [{ content: { role: "model", parts: [{ text: "Hello" }] }, finishReason: "STOP" as const }],
+                candidates: [
+                  { content: { role: "model", parts: [{ text: "Hello" }] }, finishReason: "STOP" as const },
+                ],
              },
            ),
          ),
@@ -1,6 +1,6 @@
 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 { Cerebras, 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"
@@ -55,9 +55,6 @@ const cerebras = Cerebras.configure({ apiKey: process.env.CEREBRAS_API_KEY ?? "f
 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")
@@ -225,13 +222,6 @@ describeRecordedGoldenScenarios([
    requires: ["GROQ_API_KEY"],
    scenarios: ["text", "tool-call", { id: "tool-loop", timeout: 30_000 }],
  },
-  {
-    name: "DeepInfra Llama 3.3 70B",
-    prefix: "deepinfra-chat",
-    model: deepInfra,
-    requires: ["DEEPINFRA_API_KEY"],
-    scenarios: ["text", "tool-call", { id: "tool-loop", timeout: 30_000 }],
-  },
  {
    name: "OpenRouter gpt-4o-mini",
    prefix: "openai-compatible-chat",
@@ -26,7 +26,9 @@ const recorded = recordedTests({
 describe("Google Vertex Gemini recorded", () => {
  recorded.effect("streams text", () =>
    Effect.gen(function* () {
-      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Reply with exactly one word: hello" }))
+      const response = yield* LLMClient.generate(
+        LLM.request({ model, prompt: "Reply with exactly one word: hello" }),
+      )

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

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

-  it.effect("round-trips Vertex Gemini metadata through signed content, tool calls, and usage", () =>
-    Effect.gen(function* () {
-      const model = GoogleVertex.configure({
-        accessToken: "vertex-token",
-        project: "vertex-project",
-      }).model("gemini-3.5-flash")
-      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Check the weather." })).pipe(
-        Effect.provide(
-          fixedResponse(
-            sseEvents({
-              candidates: [
-                {
-                  content: {
-                    role: "model",
-                    parts: [
-                      { text: "Thinking.", thought: true, thoughtSignature: "reasoning_sig" },
-                      { text: "Checking.", thoughtSignature: "text_sig" },
-                      {
-                        functionCall: { id: "provider_call_1", name: "lookup", args: { query: "weather" } },
-                        thoughtSignature: "tool_sig",
-                      },
-                    ],
-                  },
-                  finishReason: "STOP",
-                },
-              ],
-              promptFeedback: { blockReasonMessage: "Reviewed" },
-              usageMetadata: { promptTokenCount: 5, candidatesTokenCount: 2, thoughtsTokenCount: 1 },
-            }),
-          ),
-        ),
-      )
-      const reasoning = response.events.find((event) => event.type === "reasoning-end")
-      const text = response.events.find((event) => event.type === "text-delta")
-      const toolCall = response.toolCalls[0]
-
-      expect(reasoning?.providerMetadata).toEqual({ vertex: { thoughtSignature: "reasoning_sig" } })
-      expect(text?.providerMetadata).toEqual({ vertex: { thoughtSignature: "text_sig" } })
-      expect(toolCall).toMatchObject({
-        id: "provider_call_1",
-        providerMetadata: { vertex: { thoughtSignature: "tool_sig" } },
-      })
-      expect(response.usage?.providerMetadata).toEqual({
-        vertex: { promptTokenCount: 5, candidatesTokenCount: 2, thoughtsTokenCount: 1 },
-      })
-      expect(response.events.at(-1)?.providerMetadata).toEqual({
-        vertex: { promptFeedback: { blockReasonMessage: "Reviewed" } },
-      })
-
-      const prepared = yield* compileRequest(
-        LLM.request({
-          model,
-          messages: [
-            Message.assistant([
-              { type: "reasoning", text: "Thinking.", providerMetadata: reasoning?.providerMetadata },
-              { type: "text", text: "Checking.", providerMetadata: text?.providerMetadata },
-              ToolCallPart.make({
-                id: toolCall.id,
-                name: toolCall.name,
-                input: toolCall.input,
-                providerMetadata: toolCall.providerMetadata,
-              }),
-            ]),
-            Message.tool({ id: toolCall.id, name: toolCall.name, result: "sunny", resultType: "text" }),
-          ],
-        }),
-      )
-
-      expect(prepared.body.contents).toEqual([
-        {
-          role: "model",
-          parts: [
-            { text: "Thinking.", thought: true, thoughtSignature: "reasoning_sig" },
-            { text: "Checking.", thoughtSignature: "text_sig" },
-            { functionCall: { name: "lookup", args: { query: "weather" } }, thoughtSignature: "tool_sig" },
-          ],
-        },
-        {
-          role: "user",
-          parts: [{ functionResponse: { name: "lookup", response: { name: "lookup", content: "sunny" } } }],
-        },
-      ])
-    }),
-  )
-
  it.effect("projects Anthropic Messages onto the Vertex raw-predict API", () =>
    Effect.gen(function* () {
      const model = GoogleVertexMessages.configure({
@@ -1,185 +0,0 @@
-import { configure } from "@opencode-ai/ai/providers/groq"
-import { describe, expect } from "bun:test"
-import { Effect } from "effect"
-import { LLM, LLMEvent, LLMRequest, LLMResponse, Message, ToolChoice, ToolDefinition } from "../../src/index.js"
-import { LLMClient } from "../../src/route.js"
-import { compileRequest } from "../../src/route/client.js"
-import { recordedTests } from "../recorded-test.js"
-
-const apiKey = process.env.GROQ_API_KEY ?? "fixture"
-const recorded = recordedTests({
-  prefix: "groq-chat",
-  provider: "groq",
-  protocol: "groq-chat",
-  requires: ["GROQ_API_KEY"],
-})
-
-const weather = ToolDefinition.make({
-  name: "lookup_weather",
-  description: "Look up the current weather for a city",
-  inputSchema: {
-    type: "object",
-    properties: { city: { type: "string", enum: ["Paris", "London"] } },
-    required: ["city"],
-    additionalProperties: false,
-  },
-})
-
-describe("Groq recorded", () => {
-  recorded.effect.with(
-    "streams text with usage",
-    { tags: ["text", "usage"], metadata: { model: "openai/gpt-oss-20b" } },
-    () =>
-      Effect.gen(function* () {
-        const request = LLM.request({
-          model: configure({
-            apiKey,
-            providerOptions: {
-              includeReasoning: false,
-              reasoningEffort: "low",
-              serviceTier: "on_demand",
-              user: "recorded-test",
-            },
-          }).model("openai/gpt-oss-20b"),
-          prompt: "Reply with exactly one word: hello",
-          generation: { maxTokens: 512 },
-        })
-        const compiled = yield* compileRequest(request)
-        expect(compiled.body).toMatchObject({
-          max_completion_tokens: 512,
-          stream_options: { include_usage: true },
-          include_reasoning: false,
-          service_tier: "on_demand",
-          user: "recorded-test",
-        })
-        expect(compiled.body.max_tokens).toBeUndefined()
-        expect(compiled.body.store).toBeUndefined()
-        expect(compiled.body.reasoning_format).toBeUndefined()
-
-        const response = yield* LLMClient.generate(request)
-        expect(response.text.toLowerCase().trim()).toBe("hello")
-        expect(response.reasoning).toBe("")
-        expect(response.events.some(LLMEvent.is.textDelta)).toBe(true)
-        expectUsage(response)
-      }),
-    60_000,
-  )
-
-  for (const item of [
-    {
-      name: "continues Qwen parallel tool calls",
-      model: configure({ apiKey, providerOptions: { parallelToolCalls: true, reasoningEffort: "none" } }).model(
-        "qwen/qwen3.6-27b",
-      ),
-      cities: ["Paris", "London"],
-      reasoning: false,
-    },
-    {
-      name: "replays GPT OSS reasoning through a tool loop",
-      model: configure({ apiKey, providerOptions: { includeReasoning: true, reasoningEffort: "low" } }).model(
-        "openai/gpt-oss-20b",
-      ),
-      cities: ["Paris"],
-      reasoning: true,
-    },
-  ]) {
-    recorded.effect.with(
-      item.name,
-      {
-        tags: ["tool", "tool-loop", "usage", item.reasoning ? "reasoning" : "parallel"],
-        metadata: { model: item.model.id },
-      },
-      () =>
-        Effect.gen(function* () {
-          const request = LLM.request({
-            model: item.model,
-            prompt: `Look up the current weather in ${item.cities.join(" and ")}. Call lookup_weather once for each city in the same response before answering. After receiving all results, report each city's weather in one short sentence.`,
-            tools: [weather],
-            toolChoice: "required",
-            generation: { maxTokens: 1536 },
-          })
-          const compiled = yield* compileRequest(request)
-          expect(compiled.body.stream_options).toEqual({ include_usage: true })
-          expect(compiled.body.store).toBeUndefined()
-          expect(compiled.body.reasoning_format).toBe(item.reasoning ? undefined : "parsed")
-          expect(compiled.body.tools[0].function.strict).toBeUndefined()
-          if (!item.reasoning) expect(compiled.body.parallel_tool_calls).toBe(true)
-
-          const first = yield* LLMClient.generate(request)
-          expect(first.finishReason.normalized).toBe("tool-calls")
-          expect(first.toolCalls).toHaveLength(item.cities.length)
-          expect(new Set(first.toolCalls.map((call) => call.id)).size).toBe(item.cities.length)
-          expect(first.toolCalls.map((call) => call.input)).toEqual(
-            expect.arrayContaining(item.cities.map((city) => ({ city }))),
-          )
-          expect(first.toolCalls.every((call) => call.name === "lookup_weather")).toBe(true)
-          expectUsage(first)
-          if (item.reasoning) {
-            expect(first.reasoning.length).toBeGreaterThan(0)
-            expect(first.events.some(LLMEvent.is.reasoningDelta)).toBe(true)
-          }
-
-          const followUp = LLMRequest.update(request, {
-            toolChoice: ToolChoice.make("none"),
-            messages: [
-              ...request.messages,
-              first.message,
-              ...first.toolCalls.map((call) =>
-                Message.tool({ id: call.id, name: call.name, result: { condition: "sunny", temperature: "18C" } }),
-              ),
-            ],
-          })
-          const replay = yield* compileRequest(followUp)
-          if (item.reasoning) {
-            expect(replay.body.messages).toEqual(
-              expect.arrayContaining([expect.objectContaining({ role: "assistant", reasoning: first.reasoning })]),
-            )
-          }
-          expect(replay.body.reasoning_format).toBe(item.reasoning ? undefined : "parsed")
-
-          const second = yield* LLMClient.generate(followUp)
-          expect(second.finishReason.normalized).toBe("stop")
-          expect(second.toolCalls).toHaveLength(0)
-          expect(second.text.toLowerCase()).toContain("sunny")
-          item.cities.forEach((city) => expect(second.text).toContain(city))
-          expectUsage(second)
-        }),
-      60_000,
-    )
-  }
-
-  recorded.effect.with(
-    "streams Qwen parsed reasoning",
-    { tags: ["reasoning", "usage"], metadata: { model: "qwen/qwen3.6-27b" } },
-    () =>
-      Effect.gen(function* () {
-        const request = LLM.request({
-          model: configure({
-            apiKey,
-            providerOptions: { reasoningEffort: "default" },
-          }).model("qwen/qwen3.6-27b"),
-          prompt:
-            "What is 173 multiplied by 219? Think through the arithmetic, then reply with only the final integer.",
-          generation: { maxTokens: 2048 },
-        })
-        const compiled = yield* compileRequest(request)
-        expect(compiled.body).toMatchObject({ reasoning_format: "parsed", reasoning_effort: "default" })
-        expect(compiled.body.include_reasoning).toBeUndefined()
-
-        const response = yield* LLMClient.generate(request)
-        expect(response.text.replaceAll(",", "").trim()).toBe("37887")
-        expect(response.text).not.toContain("<think>")
-        expect(response.reasoning.length).toBeGreaterThan(0)
-        expect(response.events.some(LLMEvent.is.reasoningDelta)).toBe(true)
-        expectUsage(response)
-      }),
-    60_000,
-  )
-})
-
-function expectUsage(response: LLMResponse) {
-  expect(response.usage).toBeDefined()
-  expect(response.usage?.inputTokens).toBeGreaterThan(0)
-  expect(response.usage?.outputTokens).toBeGreaterThan(0)
-  expect(response.events.filter(LLMEvent.is.finish)).toHaveLength(1)
-}
@@ -1,112 +0,0 @@
-import { expect } from "bun:test"
-import { Effect } from "effect"
-import { LanguageModel, LLM, Message } from "../../src/index.js"
-import { OpenAIChat } from "../../src/protocols/openai-chat.js"
-import { Groq } from "../../src/providers/groq.js"
-import { compileRequest } from "../../src/route/client.js"
-import { it } from "../lib/effect.js"
-import { weatherTool } from "../recorded-scenarios.js"
-
-it.effect("Groq reuses Chat streaming and defaults to parsed reasoning", () =>
-  Effect.gen(function* () {
-    expect(Groq.protocol.stream).toBe(OpenAIChat.protocol.stream)
-    const model = Groq.configure({ apiKey: "fixture" }).model("llama-3.3-70b-versatile")
-    expect(model.route.endpoint.baseURL).toBe("https://api.groq.com/openai/v1")
-    const compiled = yield* compileRequest(
-      LLM.request({ model, prompt: "Hello", tools: [weatherTool], generation: { maxTokens: 64 } }),
-    )
-    expect(compiled.body).toMatchObject({
-      max_completion_tokens: 64,
-      stream_options: { include_usage: true },
-      reasoning_format: "parsed",
-    })
-    for (const key of ["store", "max_tokens", "include_reasoning", "parallel_tool_calls", "service_tier", "user"])
-      expect(compiled.body[key]).toBeUndefined()
-    expect(compiled.body.tools?.[0]?.function).not.toHaveProperty("strict")
-  }),
-)
-
-it.effect("Groq lowers its own options for custom catalog identities and endpoints", () =>
-  Effect.gen(function* () {
-    const model = LanguageModel.update(
-      Groq.model("qwen/qwen3.6-27b", {
-        apiKey: "fixture",
-        baseURL: "https://gateway.example/v1",
-        headers: { "x-client": "test" },
-        body: { custom: "value" },
-        providerOptions: {
-          reasoningEffort: "default",
-          parallelToolCalls: true,
-          serviceTier: "flex",
-          user: "test-user",
-        },
-      }),
-      { provider: "custom-groq" },
-    )
-    const compiled = yield* compileRequest(
-      LLM.request({ model, prompt: "Hello", providerOptions: { parallelToolCalls: false, includeReasoning: false } }),
-    )
-    expect(model.route.endpoint.baseURL).toBe("https://gateway.example/v1")
-    expect(model.route.defaults.headers).toEqual({ "x-client": "test" })
-    expect(model.route.defaults.http?.body).toEqual({ custom: "value" })
-    expect(compiled.body).toMatchObject({
-      reasoning_effort: "default",
-      reasoning_format: "parsed",
-      parallel_tool_calls: false,
-      service_tier: "flex",
-      user: "test-user",
-    })
-    expect(compiled.body.include_reasoning).toBeUndefined()
-    for (const key of ["reasoningFormat", "reasoningEffort", "parallelToolCalls", "serviceTier"])
-      expect(compiled.body).not.toHaveProperty(key)
-  }),
-)
-
-it.effect("Groq replays reasoning only when present and preserves explicit reasoning exclusion", () =>
-  Effect.gen(function* () {
-    const compiled = yield* compileRequest(
-      LLM.request({
-        model: Groq.configure({ apiKey: "fixture" }).model("openai/gpt-oss-20b"),
-        messages: [
-          Message.user("Think"),
-          Message.assistant([
-            { type: "reasoning", text: "Thinking" },
-            { type: "text", text: "Answer" },
-          ]),
-          Message.user("Again"),
-          Message.assistant("Answer only"),
-          Message.user("Continue"),
-        ],
-        providerOptions: { reasoningEffort: "low", includeReasoning: false },
-      }),
-    )
-    expect(compiled.body).toMatchObject({ reasoning_effort: "low", include_reasoning: false })
-    expect(compiled.body.reasoning_format).toBeUndefined()
-    expect(compiled.body.messages[1]).toMatchObject({ reasoning: "Thinking", content: "Answer" })
-    expect(compiled.body.messages[1]).not.toHaveProperty("reasoning_content")
-    expect(compiled.body.messages[3]).not.toHaveProperty("reasoning")
-  }),
-)
-
-it.effect("Groq omits reasoning_format for the GPT-OSS family by default", () =>
-  Effect.gen(function* () {
-    for (const id of ["openai/gpt-oss-20b", "openai/gpt-oss-120b", "openai/gpt-oss-safeguard-20b"]) {
-      const compiled = yield* compileRequest(
-        LLM.request({ model: Groq.configure({ apiKey: "fixture" }).model(id), prompt: "Hello" }),
-      )
-      expect(compiled.body.reasoning_format).toBeUndefined()
-      expect(compiled.body.include_reasoning).toBeUndefined()
-    }
-  }),
-)
-
-it.effect("Groq validates option types", () =>
-  Effect.gen(function* () {
-    for (const providerOptions of [{ includeReasoning: "false" }, { parallelToolCalls: "false" }]) {
-      const error = yield* compileRequest(
-        LLM.request({ model: Groq.configure({ apiKey: "fixture" }).model("qwen"), prompt: "Hello", providerOptions }),
-      ).pipe(Effect.flip)
-      expect(error.reason._tag).toBe("InvalidRequest")
-    }
-  }),
-)
@@ -2,80 +2,13 @@ import { describe, expect } from "bun:test"
 import { ConfigProvider, Effect } from "effect"
 import { HttpClientRequest } from "effect/unstable/http"
 import { LLM, Message, ToolDefinition } from "../../src/index.js"
-import {
-  AmazonBedrock,
-  AmazonBedrockMantle,
-  Anthropic,
-  AnthropicCompatible,
-  Azure,
-  Cerebras,
-  CloudflareAIGateway,
-  CloudflareWorkersAI,
-  DeepInfra,
-  Google,
-  GoogleVertex,
-  GoogleVertexChat,
-  GoogleVertexMessages,
-  GoogleVertexResponses,
-  Groq,
-  OpenAI,
-  OpenAICompatible,
-  OpenAICompatibleResponses,
-  OpenRouter,
-  TogetherAI,
-  XAI,
-} from "../../src/providers/index.js"
+import { Cerebras, 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("assigns provider-owned metadata namespaces across native routes", () =>
-    Effect.gen(function* () {
-      const vertex = { project: "project", accessToken: "token" }
-      const providers = [
-        [OpenAI.configure({ apiKey: "test" }).chat("model"), "openai"],
-        [OpenAI.configure({ apiKey: "test" }).responses("model"), "openai"],
-        [Azure.configure({ resourceName: "resource", apiKey: "test" }).chat("model"), "azure"],
-        [Azure.configure({ resourceName: "resource", apiKey: "test" }).responses("model"), "azure"],
-        [AmazonBedrock.configure({ apiKey: "test" }).model("model"), "bedrock"],
-        [AmazonBedrockMantle.configure({ apiKey: "test" }).chat("model"), "mantle"],
-        [AmazonBedrockMantle.configure({ apiKey: "test" }).responses("model"), "mantle"],
-        [Google.configure({ apiKey: "test" }).model("model"), "google"],
-        [GoogleVertex.configure(vertex).model("model"), "vertex"],
-        [GoogleVertexChat.configure(vertex).model("model"), "vertex"],
-        [GoogleVertexResponses.configure(vertex).model("model"), "vertex"],
-        [GoogleVertexMessages.configure(vertex).model("model"), "anthropic"],
-        [Anthropic.configure({ apiKey: "test" }).model("model"), "anthropic"],
-        [
-          AnthropicCompatible.configure({ baseURL: "https://example.test/v1", provider: "minimax" }).model("model"),
-          "minimax",
-        ],
-        [
-          OpenAICompatible.configure({ baseURL: "https://example.test/v1", provider: "custom" }).model("model"),
-          "custom",
-        ],
-        [
-          OpenAICompatibleResponses.configure({ baseURL: "https://example.test/v1", provider: "custom" }).model(
-            "model",
-          ),
-          "custom",
-        ],
-        [Cerebras.configure({ apiKey: "test" }).model("model"), "cerebras"],
-        [DeepInfra.configure({ apiKey: "test" }).model("model"), "deepinfra"],
-        [TogetherAI.configure({ apiKey: "test" }).model("model"), "togetherai"],
-        [CloudflareAIGateway.configure({ accountId: "account" }).model("model"), "cloudflare-ai-gateway"],
-        [CloudflareWorkersAI.configure({ accountId: "account" }).model("model"), "cloudflare-workers-ai"],
-        [OpenRouter.configure({ apiKey: "test" }).model("model"), "openrouter"],
-        [XAI.configure({ apiKey: "test" }).chat("model"), "xai"],
-        [XAI.configure({ apiKey: "test" }).responses("model"), "xai"],
-      ] as const
-
-      for (const [model, key] of providers) expect(model.route.providerMetadataKey).toBe(key)
-    }),
-  )
-
  it.effect("preserves native Together AI and Cerebras provider and route identities", () =>
    Effect.gen(function* () {
      const together = TogetherAI.configure({ apiKey: "fixture" }).model("meta-llama/Llama-3.3-70B")
@@ -96,18 +29,6 @@ describe("native OpenAI-compatible providers", () => {
    }),
  )

-  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(
@@ -161,21 +82,6 @@ describe("native OpenAI-compatible providers", () => {
    }),
  )

-  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]) {
@@ -216,18 +122,6 @@ describe("native OpenAI-compatible providers", () => {
          token: "cerebras-secret",
          url: "https://api.cerebras.ai/v1/chat/completions",
        },
-        {
-          model: DeepInfra.configure().model("gemma"),
-          env: { DEEPINFRA_API_KEY: "deepinfra-secret" },
-          token: "deepinfra-secret",
-          url: "https://api.deepinfra.com/v1/openai/chat/completions",
-        },
-        {
-          model: Groq.configure().model("llama"),
-          env: { GROQ_API_KEY: "groq-secret" },
-          token: "groq-secret",
-          url: "https://api.groq.com/openai/v1/chat/completions",
-        },
      ]

      yield* Effect.forEach(scenarios, (scenario) =>
@@ -68,13 +68,11 @@ for (const item of cases) {
          expect(response.text.replaceAll(",", "").trim()).toBe("37887")
          expect(response.reasoning.length).toBeGreaterThan(0)
          expect(response.events.some(LLMEvent.is.reasoningDelta)).toBe(true)
-          const metadata = response.message.content.find((part) => part.type === "reasoning")?.providerMetadata?.[
-            item.model.route.providerMetadataKey ?? String(item.model.provider)
-          ]
-          expect(metadata?.reasoningField).toBe(item.structured ? "reasoning" : "reasoning_content")
-          expect(Array.isArray(metadata?.reasoningDetails)).toBe(item.structured)
+          const metadata = response.message.content.find((part) => part.type === "reasoning")?.providerMetadata
+          expect(metadata?.openai?.reasoningField).toBe(item.structured ? "reasoning" : "reasoning_content")
+          expect(Array.isArray(metadata?.openai?.reasoningDetails)).toBe(item.structured)
          if (!item.structured) return
-          const details = metadata?.reasoningDetails
+          const details = metadata?.openai?.reasoningDetails
          if (!Array.isArray(details)) return
          expect(
            details.some(
@@ -128,11 +126,7 @@ for (const item of cases) {
          ).toMatch(/^Paris is sunny\.?$/)
          const details = events
            .filter(LLMEvent.is.reasoningEnd)
-            .map(
-              (event) =>
-                event.providerMetadata?.[item.model.route.providerMetadataKey ?? String(item.model.provider)]
-                  ?.reasoningDetails,
-            )
+            .map((event) => event.providerMetadata?.openai?.reasoningDetails)
            .find(Array.isArray)
          expect(Array.isArray(details)).toBe(item.structured)
          if (!item.structured || !Array.isArray(details)) return
@@ -903,70 +903,6 @@ describe("OpenAI Chat route", () => {
    }),
  )

-  it.effect("uses the configured provider metadata namespace for reasoning and usage", () =>
-    Effect.gen(function* () {
-      const selected = LanguageModel.update(model, {
-        route: { ...model.route, providerMetadataKey: "vendor" },
-      })
-      const details = [{ type: "reasoning.text", text: "thinking", signature: "signed" }]
-      const response = yield* LLMClient.generate(LLMRequest.update(request, { model: selected })).pipe(
-        Effect.provide(
-          fixedResponse(
-            sseEvents(
-              { choices: [{ delta: { reasoning: "thinking", reasoning_details: details } }] },
-              deltaChunk({ content: "Hello" }),
-              deltaChunk({}, "stop"),
-              usageChunk({ prompt_tokens: 5, completion_tokens: 2, total_tokens: 7 }),
-            ),
-          ),
-        ),
-      )
-
-      expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
-        vendor: { reasoningField: "reasoning", reasoningDetails: details },
-      })
-      expect(response.usage?.providerMetadata).toEqual({
-        vendor: { prompt_tokens: 5, completion_tokens: 2, total_tokens: 7 },
-      })
-
-      const replay = yield* compileRequest(LLM.request({ model: selected, messages: [response.message] }))
-      expect(replay.body.messages).toEqual([
-        { role: "assistant", content: "Hello", reasoning: "thinking", reasoning_details: details },
-      ])
-    }),
-  )
-
-  it.effect("falls back to the selected provider for the metadata namespace", () =>
-    Effect.gen(function* () {
-      const compatible = model.route.with({ provider: "deepseek" }).model({ id: "deepseek-chat" })
-      const selected = LanguageModel.update(compatible, {
-        route: { ...compatible.route, providerMetadataKey: undefined },
-      })
-      const response = yield* LLMClient.generate(LLMRequest.update(request, { model: selected })).pipe(
-        Effect.provide(
-          fixedResponse(
-            sseEvents(
-              deltaChunk({ reasoning_content: "thinking" }),
-              deltaChunk({ content: "Hello" }),
-              deltaChunk({}, "stop"),
-              usageChunk({ prompt_tokens: 5, completion_tokens: 2, total_tokens: 7 }),
-            ),
-          ),
-        ),
-      )
-
-      expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
-        deepseek: { reasoningField: "reasoning_content" },
-      })
-      expect(response.usage?.providerMetadata).toEqual({
-        deepseek: { prompt_tokens: 5, completion_tokens: 2, total_tokens: 7 },
-      })
-
-      const replay = yield* compileRequest(LLM.request({ model: selected, messages: [response.message] }))
-      expect(replay.body.messages).toEqual([{ role: "assistant", content: "Hello", reasoning_content: "thinking" }])
-    }),
-  )
-
  it.effect("parses and replays a configured custom reasoning field", () =>
    Effect.gen(function* () {
      const custom = LanguageModel.update(model, { compatibility: { reasoningField: "vendor_reasoning" } })
@@ -1457,7 +1393,7 @@ describe("OpenAI Chat route", () => {
      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.body ?? "")).toMatchObject({
+      expect(decodeJson(error.reason.raw ?? "")).toMatchObject({
        choices: [{ finish_reason: "tool_calls" }],
      })
    }),
@@ -1534,9 +1470,9 @@ describe("OpenAI Chat route", () => {
      )

      expect((yield* Ref.get(events)).some((event) => event.type === "text-delta")).toBeTrue()
-      expect(error.message).toBe("ECONNRESET: socket closed unexpectedly")
      expect(error.reason).toMatchObject({
        _tag: "Transport",
+        message: "ECONNRESET: socket closed unexpectedly",
        transport: "http",
        operation: "read",
        code: "ECONNRESET",
@@ -1552,9 +1488,9 @@ describe("OpenAI Chat route", () => {
        Effect.flip,
      )

-      expect(error.message).toBe("ECONNRESET: socket closed before output")
      expect(error.reason).toMatchObject({
        _tag: "Transport",
+        message: "ECONNRESET: socket closed before output",
        transport: "http",
        operation: "read",
        code: "ECONNRESET",
@@ -1575,7 +1511,7 @@ describe("OpenAI Chat route", () => {
      )

      expect(error).toBeInstanceOf(AIError)
-      expect(error).toMatchObject({ reason: { _tag: "InvalidRequest" }, message: "Bad request" })
+      expect(error.reason).toMatchObject({ _tag: "InvalidRequest", message: "Bad request" })
    }),
  )

@@ -437,7 +437,7 @@ describe("OpenAI-compatible Chat route", () => {
        outputTokens: undefined,
        totalTokens: undefined,
        providerMetadata: {
-          deepseek: {
+          openai: {
            prompt_tokens: null,
            completion_tokens: null,
            total_tokens: null,
@@ -489,8 +489,9 @@ describe("OpenAI-compatible Chat route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({
-        reason: { _tag: "InvalidProviderOutput", classification: "incomplete-stream" },
+      expect(error.reason).toMatchObject({
+        _tag: "InvalidProviderOutput",
+        classification: "incomplete-stream",
        message: "OpenAI Chat stream ended without finish_reason",
      })
    }),
@@ -526,11 +527,11 @@ describe("OpenAI-compatible Chat route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({
-        reason: { _tag: "ProviderInternal" },
+      expect(error.reason).toMatchObject({
+        _tag: "ProviderInternal",
        message: "Provider reported a network error (finish_reason: network_error)",
      })
-      expect(decodeJson(error.reason.body ?? "")).toMatchObject({
+      expect(decodeJson(error.body ?? "")).toMatchObject({
        id: "chatcmpl_fixture",
        choices: [{ finish_reason: "network_error" }],
      })
@@ -539,8 +540,8 @@ describe("OpenAI-compatible Chat route", () => {
        Effect.provide(fixedResponse(sseEvents(deltaChunk({}, "error")))),
        Effect.flip,
      )
-      expect(generic).toMatchObject({
-        reason: { _tag: "UnknownProvider" },
+      expect(generic.reason).toMatchObject({
+        _tag: "UnknownProvider",
        message: "Provider reported an error (finish_reason: error)",
      })
    }),
@@ -561,8 +562,8 @@ describe("OpenAI-compatible Chat route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({ reason: { _tag: "ProviderInternal" }, message: "Provider disconnected" })
-      expect(decodeJson(error.reason.body ?? "")).toMatchObject({
+      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",
@@ -602,7 +603,7 @@ 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.body ?? "")).toMatchObject({
+      expect(decodeJson(error.reason.raw ?? "")).toMatchObject({
        choices: [{ delta: { tool_calls: [{ id: "call_1" }] } }],
      })
    }),
@@ -195,19 +195,15 @@ describe("Open Responses-compatible route", () => {
          model,
          messages: [
            Message.assistant([
-              { type: "text", text: "Kept.", providerMetadata: { "openai-compatible": { itemId: "history_1" } } },
+              // The baseline does not enforce a provider id grammar, so a
+              // non-OpenAI but well-formed token is resent as-is.
+              { type: "text", text: "Kept.", providerMetadata: { openresponses: { itemId: "history_1" } } },
+              // Shape violations are dropped even without a grammar policy.
              {
                type: "text",
-                text: "Long.",
-                providerMetadata: { "openai-compatible": { itemId: `history_${"a".repeat(64)}` } },
+                text: "Dropped.",
+                providerMetadata: { openresponses: { itemId: `m${"a".repeat(64)}` } },
              },
-              {
-                type: "text",
-                text: "Opaque.",
-                providerMetadata: { "openai-compatible": { itemId: "provider_value/with+symbols" } },
-              },
-              { type: "text", text: "No suffix.", providerMetadata: { "openai-compatible": { itemId: "msg_" } } },
-              { type: "text", text: "No prefix.", providerMetadata: { "openai-compatible": { itemId: "_item" } } },
            ]),
          ],
        }),
@@ -222,23 +218,8 @@ describe("Open Responses-compatible route", () => {
        },
        {
          type: "message",
-          id: `history_${"a".repeat(64)}`,
          role: "assistant",
-          content: [{ type: "output_text", text: "Long." }],
-        },
-        {
-          type: "message",
-          id: "provider_value/with+symbols",
-          role: "assistant",
-          content: [{ type: "output_text", text: "Opaque." }],
-        },
-        {
-          type: "message",
-          role: "assistant",
-          content: [
-            { type: "output_text", text: "No suffix." },
-            { type: "output_text", text: "No prefix." },
-          ],
+          content: [{ type: "output_text", text: "Dropped." }],
        },
      ])
    }),
@@ -267,7 +248,7 @@ describe("Open Responses-compatible route", () => {
              name: item.type,
              result: { type: "json", value: item },
              providerExecuted: true,
-              providerMetadata: { example: { itemId: item.id } },
+              providerMetadata: { openresponses: { itemId: item.id } },
            }),
          ),
        }),
@@ -302,296 +283,11 @@ describe("Open Responses-compatible route", () => {
      )

      expect(response.message.content).toEqual([
-        { type: "text", text: "Indexed", providerMetadata: { "openai-compatible": { itemId: "msg_1" } } },
+        { type: "text", text: "Indexed", providerMetadata: { openresponses: { itemId: "msg_1" } } },
      ])
    }),
  )

-  describe("stream validation", () => {
-    const request = LLM.request({
-      model: configure({ apiKey: "test-key", baseURL: "https://responses.example.test/v1" }).model("example-model"),
-      prompt: "Respond.",
-    })
-
-    const fixtures = [
-      {
-        item: { type: "message" },
-        events: [
-          { type: "response.output_text.delta", delta: "Preserved" },
-          { type: "response.output_text.done", text: "Preserved" },
-          { type: "response.refusal.delta", delta: "Preserved" },
-          { type: "response.refusal.done", refusal: "Preserved" },
-        ],
-      },
-      {
-        item: { type: "reasoning", encrypted_content: "encrypted-state" },
-        events: [
-          { type: "response.reasoning.delta", delta: "Preserved" },
-          { type: "response.reasoning.done", text: "Preserved" },
-          { type: "response.reasoning_summary_text.delta", delta: "Preserved" },
-          { type: "response.reasoning_summary_text.done", text: "Preserved" },
-          { type: "response.reasoning_text.done", text: "Preserved" },
-        ],
-      },
-      {
-        item: { type: "function_call", call_id: "call_1", name: "lookup" },
-        events: [
-          { type: "response.function_call_arguments.delta", delta: '{"query":"Preserved"}' },
-          { type: "response.function_call_arguments.done", arguments: '{"query":"Preserved"}' },
-        ],
-      },
-    ]
-
-    const routings = [
-      { name: "empty item and event IDs", id: "", item_id: "" },
-      { name: "empty event ID with registered index", id: "item_1", item_id: "", output_index: 2 },
-      { name: "empty stored ID with registered index", id: "", item_id: "wrong_item", output_index: 2 },
-      { name: "empty item and event IDs with registered index", id: "", item_id: "", output_index: 2 },
-    ]
-
-    fixtures.forEach((fixture) => {
-      fixture.events.forEach((event) => {
-        routings.forEach((routing) => {
-          it.effect(`${event.type} preserves content with ${routing.name}`, () =>
-            Effect.gen(function* () {
-              const item = { ...fixture.item, id: routing.id }
-              const response = yield* LLMClient.generate(request).pipe(
-                Effect.provide(
-                  fixedResponse(
-                    sseEvents(
-                      { type: "response.output_item.added", output_index: routing.output_index, item },
-                      { ...event, item_id: routing.item_id, output_index: routing.output_index },
-                      { type: "response.output_item.done", output_index: routing.output_index, item },
-                      { type: "response.completed", response: { id: "resp_1" } },
-                    ),
-                  ),
-                ),
-              )
-
-              const metadata = { "openai-compatible": { itemId: routing.id } }
-              if (fixture.item.type === "function_call") {
-                expect(response.toolCalls).toEqual([
-                  expect.objectContaining({
-                    id: "call_1",
-                    name: "lookup",
-                    input: { query: "Preserved" },
-                    providerMetadata: metadata,
-                  }),
-                ])
-                return
-              }
-              if (fixture.item.type === "reasoning") {
-                expect(response.message.content).toEqual([
-                  {
-                    type: "reasoning",
-                    text: "Preserved",
-                    providerMetadata: {
-                      "openai-compatible": { itemId: routing.id, reasoningEncryptedContent: "encrypted-state" },
-                    },
-                  },
-                ])
-                expect(response.events.filter(LLMEvent.is.reasoningEnd)).toHaveLength(1)
-                return
-              }
-              expect(response.message.content).toEqual([
-                { type: "text", text: "Preserved", providerMetadata: metadata },
-              ])
-              expect(response.events.filter(LLMEvent.is.textEnd)).toEqual([
-                expect.objectContaining({ id: routing.id, providerMetadata: metadata }),
-              ])
-            }),
-          )
-        })
-      })
-    })
-
-    routings.forEach((routing) => {
-      it.effect(`preserves reasoning summary boundaries and terminal metadata with ${routing.name}`, () =>
-        Effect.gen(function* () {
-          const address = { item_id: routing.item_id, output_index: routing.output_index }
-          const response = yield* LLMClient.generate(request).pipe(
-            Effect.provide(
-              fixedResponse(
-                sseEvents(
-                  {
-                    type: "response.output_item.added",
-                    output_index: routing.output_index,
-                    item: { type: "reasoning", id: routing.id },
-                  },
-                  { type: "response.reasoning_summary_part.added", ...address, summary_index: 0 },
-                  { type: "response.reasoning_summary_text.delta", ...address, summary_index: 0, delta: "First." },
-                  { type: "response.reasoning_summary_text.done", ...address, summary_index: 0, text: "First." },
-                  { type: "response.reasoning_summary_part.done", ...address, summary_index: 0 },
-                  { type: "response.reasoning_summary_part.added", ...address, summary_index: 1 },
-                  { type: "response.reasoning_summary_text.done", ...address, summary_index: 1, text: "Second." },
-                  { type: "response.reasoning_summary_part.done", ...address, summary_index: 1 },
-                  {
-                    type: "response.completed",
-                    response: { output: [{ type: "reasoning", id: routing.id, encrypted_content: "final-state" }] },
-                  },
-                ),
-              ),
-            ),
-          )
-
-          expect(response.message.content).toEqual([
-            {
-              type: "reasoning",
-              text: "First.",
-              providerMetadata: { "openai-compatible": { itemId: routing.id } },
-            },
-            {
-              type: "reasoning",
-              text: "Second.",
-              providerMetadata: {
-                "openai-compatible": { itemId: routing.id, reasoningEncryptedContent: "final-state" },
-              },
-            },
-          ])
-          expect(response.events.filter(LLMEvent.is.reasoningEnd)).toEqual([
-            expect.objectContaining({
-              id: `${routing.id}:0`,
-              providerMetadata: { "openai-compatible": { itemId: routing.id } },
-            }),
-            expect.objectContaining({
-              id: `${routing.id}:1`,
-              providerMetadata: {
-                "openai-compatible": { itemId: routing.id, reasoningEncryptedContent: "final-state" },
-              },
-            }),
-          ])
-        }),
-      )
-    })
-
-    it.effect("reconciles pending empty-ID function arguments from completed output", () =>
-      Effect.gen(function* () {
-        const item = { type: "function_call", id: "", call_id: "call_1", name: "lookup" }
-        const response = yield* LLMClient.generate(request).pipe(
-          Effect.provide(
-            fixedResponse(
-              sseEvents(
-                { type: "response.output_item.added", item },
-                { type: "response.function_call_arguments.delta", item_id: "", delta: '{"query":"partial' },
-                {
-                  type: "response.completed",
-                  response: { output: [{ ...item, arguments: '{"query":"complete"}' }] },
-                },
-              ),
-            ),
-          ),
-        )
-
-        expect(response.toolCalls).toEqual([
-          expect.objectContaining({
-            id: "call_1",
-            name: "lookup",
-            input: { query: "complete" },
-            providerMetadata: { "openai-compatible": { itemId: "" } },
-          }),
-        ])
-      }),
-    )
-
-    it.effect("treats null output items as no-ops without disturbing registered items", () =>
-      Effect.gen(function* () {
-        const response = yield* LLMClient.generate(request).pipe(
-          Effect.provide(
-            fixedResponse(
-              sseEvents(
-                { type: "response.output_item.added", output_index: 0, item: null },
-                { type: "response.output_item.done", output_index: 0, item: null },
-                { type: "response.output_item.added", output_index: 0, item: { type: "message", id: "msg_1" } },
-                { type: "response.output_text.delta", output_index: 0, item_id: "wrong_item", delta: "Before " },
-                { type: "response.output_item.added", output_index: 0, item: null },
-                { type: "response.output_item.done", output_index: 0, item: null },
-                { type: "response.output_text.delta", output_index: 0, item_id: "wrong_item", delta: "after" },
-                { type: "response.output_item.done", output_index: 0, item: { type: "message", id: "msg_1" } },
-                { type: "response.completed", response: { id: "resp_1" } },
-              ),
-            ),
-          ),
-        )
-
-        expect(response.message.content).toEqual([
-          { type: "text", text: "Before after", providerMetadata: { "openai-compatible": { itemId: "msg_1" } } },
-        ])
-        expect(response.events.map((event) => event.type)).toEqual([
-          "step-start",
-          "text-start",
-          "text-delta",
-          "text-delta",
-          "text-end",
-          "step-finish",
-          "finish",
-        ])
-      }),
-    )
-
-    it.effect("rejects missing, null, and non-string event IDs even with a registered output index", () =>
-      Effect.gen(function* () {
-        yield* Effect.forEach(
-          [
-            ...fixtures.flatMap((fixture) => fixture.events.map((event) => ({ item: fixture.item, event }))),
-            ...["response.reasoning_summary_part.added", "response.reasoning_summary_part.done"].map((type) => ({
-              item: { type: "reasoning" },
-              event: { type, summary_index: 0 },
-            })),
-          ],
-          (fixture) =>
-            Effect.forEach([undefined, null, 0, false, {}, []], (item_id) =>
-              Effect.gen(function* () {
-                const error = yield* LLMClient.generate(request).pipe(
-                  Effect.provide(
-                    fixedResponse(
-                      sseEvents(
-                        {
-                          type: "response.output_item.added",
-                          output_index: 0,
-                          item: { ...fixture.item, id: "item_1" },
-                        },
-                        { ...fixture.event, output_index: 0, item_id },
-                        { type: "response.completed", response: { id: "resp_1" } },
-                      ),
-                    ),
-                  ),
-                  Effect.flip,
-                )
-                expect(error.reason._tag).toBe("InvalidProviderOutput")
-              }),
-            ),
-        )
-      }),
-    )
-
-    it.effect("keeps malformed output item IDs invalid", () =>
-      Effect.gen(function* () {
-        yield* Effect.forEach(["response.output_item.added", "response.output_item.done"], (type) =>
-          Effect.forEach(fixtures, (fixture) =>
-            Effect.forEach(
-              fixture.item.type === "message" ? [undefined, null, 0, false, {}, []] : [null, 0, false, {}, []],
-              (id) =>
-                Effect.gen(function* () {
-                  const error = yield* LLMClient.generate(request).pipe(
-                    Effect.provide(
-                      fixedResponse(
-                        sseEvents(
-                          { type, item: { ...fixture.item, id } },
-                          { type: "response.completed", response: { id: "resp_1" } },
-                        ),
-                      ),
-                    ),
-                    Effect.flip,
-                  )
-                  expect(error.reason._tag).toBe("InvalidProviderOutput")
-                }),
-            ),
-          ),
-        )
-      }),
-    )
-  })
-
  it.effect("streams function calls without optional item ids through the shared baseline", () =>
    Effect.gen(function* () {
      const model = configure({
@@ -666,7 +362,7 @@ describe("Open Responses-compatible route", () => {

      expect(response.events.find(LLMEvent.is.toolCall)).toMatchObject({
        input: { query: "complete" },
-        providerMetadata: { example: { itemId: "item_1" } },
+        providerMetadata: { openresponses: { itemId: "item_1" } },
      })
    }),
  )
@@ -698,7 +394,7 @@ describe("Open Responses-compatible route", () => {
      )

      expect(response.events.find((event) => event.type === "reasoning-end")).toMatchObject({
-        providerMetadata: { "openai-compatible": { itemId: "rs_raw", reasoningEncryptedContent: "raw-state" } },
+        providerMetadata: { openresponses: { itemId: "rs_raw", reasoningEncryptedContent: "raw-state" } },
      })
    }),
  )
@@ -747,7 +443,7 @@ describe("Open Responses-compatible route", () => {
            Message.assistant({
              type: "text",
              text: "Unclassified.",
-              providerMetadata: { "openai-compatible": { phase: null } },
+              providerMetadata: { openresponses: { phase: null } },
            }),
          ],
        }),
@@ -806,7 +502,7 @@ describe("Open Responses-compatible route", () => {
        {
          type: "text",
          text: "I can't help with that.",
-          providerMetadata: { example: { itemId: "msg_refusal" } },
+          providerMetadata: { openresponses: { itemId: "msg_refusal" } },
        },
      ])

@@ -895,7 +591,7 @@ describe("Open Responses-compatible route", () => {

      expect(response.toolCalls).toEqual([])
      expect(response.events.find(LLMEvent.is.finish)).toMatchObject({
-        providerMetadata: { example: { responseId: "resp_1" } },
+        providerMetadata: { openresponses: { responseId: "resp_1" } },
      })
    }),
  )
@@ -4,7 +4,6 @@ import { Headers, HttpClientRequest } from "effect/unstable/http"
 import {
  LLM,
  AIError,
-  HttpContext,
  HttpOptions,
  LLMEvent,
  LLMRequest,
@@ -13,7 +12,7 @@ import {
  ToolCallPart,
  ToolDefinition,
  ToolResultPart,
-  TransportError,
+  TransportReason,
  Usage,
 } from "../../src/index.js"
 import {
@@ -53,11 +52,10 @@ const baseChannelDriver = (message: string): WebSocketChannelDriver => ({
      return Effect.succeed({
        type: "provider-failure",
        error: new AIError({
-          reason: new TransportError({
+          module: "test",
+          method: "stream",
+          reason: new TransportReason({
            message: "provider rejected request",
-            body: frame,
-            cause: new Error("provider cause"),
-            http: new HttpContext({ url: "https://provider.test", status: 200, headers: { "x-trace": "trace-1" } }),
            transport: "websocket",
            operation: "read",
            phase: "receive",
@@ -746,18 +744,6 @@ describe("OpenAI Responses route", () => {
          reason: { _tag: "Transport", delivery: "rejected", recovery: "rotate-and-retry-full" },
        },
      })
-      for (const observation of [missing, limit]) {
-        expect(observation.type).toBe("rejected")
-        if (observation.type !== "rejected") continue
-        expect(observation.error.message).toBe("provider rejected request")
-        expect(observation.error.reason.cause).toBeInstanceOf(Error)
-        expect(observation.error.reason.cause).toMatchObject({ message: "provider cause" })
-        expect(observation.error.reason.http).toMatchObject({ status: 200, headers: { "x-trace": "trace-1" } })
-        expect(ProviderShared.decodeJson(observation.error.reason.body ?? "")).toMatchObject({
-          type: "error",
-          error: { code: expect.any(String) },
-        })
-      }
    }),
  )

@@ -1124,7 +1110,9 @@ describe("OpenAI Responses route", () => {
  it.effect("marks post-send WebSocket failures with delivery state", () =>
    Effect.gen(function* () {
      const failure = new AIError({
-        reason: new TransportError({
+        module: "test",
+        method: "receive",
+        reason: new TransportReason({
          message: "socket closed",
          transport: "websocket",
          operation: "read",
@@ -2241,35 +2229,6 @@ describe("OpenAI Responses route", () => {
    }),
  )

-  it.effect("accepts empty IDs for native reasoning text deltas", () =>
-    Effect.gen(function* () {
-      const response = yield* LLMClient.generate(request).pipe(
-        Effect.provide(
-          fixedResponse(
-            sseEvents(
-              { type: "response.output_item.added", output_index: 1, item: { type: "reasoning", id: "" } },
-              { type: "response.reasoning_text.delta", output_index: 1, item_id: "", delta: "Raw" },
-              {
-                type: "response.output_item.done",
-                output_index: 1,
-                item: { type: "reasoning", id: "", encrypted_content: "state" },
-              },
-              { type: "response.completed", response: { id: "resp_1" } },
-            ),
-          ),
-        ),
-      )
-
-      expect(response.message.content).toEqual([
-        {
-          type: "reasoning",
-          text: "Raw",
-          providerMetadata: { openai: { itemId: "", reasoningEncryptedContent: "state" } },
-        },
-      ])
-    }),
-  )
-
  it.effect("falls back to item ids when an output index was not registered", () =>
    Effect.gen(function* () {
      const response = yield* LLMClient.generate(request).pipe(
@@ -3191,50 +3150,38 @@ describe("OpenAI Responses route", () => {
    }),
  )

-  it.effect("preserves provider-issued item ids and removes malformed ids without dropping items", () =>
+  it.effect("drops replayed item ids outside the server's grammar", () =>
    Effect.gen(function* () {
      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
            Message.assistant([
+              // Fails the message id prefix.
              {
                type: "text",
                text: "Hello",
                providerMetadata: { openai: { itemId: "history_1" } },
              },
+              // Oversized for the Responses item id limit.
              {
                type: "text",
                text: "World",
-                providerMetadata: { openai: { itemId: `message_${"a".repeat(64)}` } },
+                providerMetadata: { openai: { itemId: `m${"a".repeat(64)}` } },
              },
+              // Fails the reasoning id prefix, so the whole item is unreplayable
+              // statelessly and is skipped rather than sent malformed.
              {
                type: "reasoning",
                text: "Checked the diff.",
                providerMetadata: { openai: { itemId: "thinking_1", reasoningEncryptedContent: "encrypted-state" } },
              },
-              {
-                type: "reasoning",
-                text: "Missing suffix.",
-                providerMetadata: { openai: { itemId: "rs_", reasoningEncryptedContent: "another-state" } },
-              },
-              {
-                type: "reasoning",
-                text: "No prefix separator.",
-                providerMetadata: { openai: { itemId: "550e8400-e29b-41d4-a716-446655440000" } },
-              },
              ToolCallPart.make({
                id: "call_1",
                name: "lookup",
                input: { query: "weather" },
                providerMetadata: { openai: { itemId: "toolu_01A" } },
              }),
-              ToolCallPart.make({
-                id: "call_2",
-                name: "lookup",
-                input: { query: "news" },
-                providerMetadata: { openai: { itemId: "fc_" } },
-              }),
            ]),
          ],
        }),
@@ -3243,44 +3190,18 @@ describe("OpenAI Responses route", () => {
      expect(prepared.body.input).toEqual([
        {
          type: "message",
-          id: "history_1",
          role: "assistant",
-          content: [{ type: "output_text", text: "Hello" }],
-        },
-        {
-          type: "message",
-          id: `message_${"a".repeat(64)}`,
-          role: "assistant",
-          content: [{ type: "output_text", text: "World" }],
-        },
-        {
-          type: "reasoning",
-          id: "thinking_1",
-          summary: [{ type: "summary_text", text: "Checked the diff." }],
-          encrypted_content: "encrypted-state",
-        },
-        {
-          type: "reasoning",
-          summary: [{ type: "summary_text", text: "Missing suffix." }],
-          encrypted_content: "another-state",
-        },
-        {
-          type: "reasoning",
-          summary: [{ type: "summary_text", text: "No prefix separator." }],
+          content: [
+            { type: "output_text", text: "Hello" },
+            { type: "output_text", text: "World" },
+          ],
        },
        {
          type: "function_call",
-          id: "toolu_01A",
          call_id: "call_1",
          name: "lookup",
          arguments: '{"query":"weather"}',
        },
-        {
-          type: "function_call",
-          call_id: "call_2",
-          name: "lookup",
-          arguments: '{"query":"news"}',
-        },
      ])
    }),
  )
@@ -4447,7 +4368,7 @@ describe("OpenAI Responses route", () => {
      )

      expect(error).toBeInstanceOf(AIError)
-      expect(error).toMatchObject({ reason: { _tag: "RateLimit" }, message: "rate_limit_exceeded: Slow down" })
+      expect(error.reason).toMatchObject({ _tag: "RateLimit", message: "rate_limit_exceeded: Slow down" })
    }),
  )

@@ -4458,7 +4379,7 @@ describe("OpenAI Responses route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({ reason: { _tag: "ProviderInternal" }, message: "internal_error" })
+      expect(error.reason).toMatchObject({ _tag: "ProviderInternal", message: "internal_error" })
    }),
  )

@@ -4469,7 +4390,7 @@ describe("OpenAI Responses route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({ reason: { _tag: "ProviderInternal" }, message: "internal_error" })
+      expect(error.reason).toMatchObject({ _tag: "ProviderInternal", message: "internal_error" })
    }),
  )

@@ -4494,8 +4415,8 @@ describe("OpenAI Responses route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({
-        reason: { _tag: "ProviderInternal" },
+      expect(error.reason).toMatchObject({
+        _tag: "ProviderInternal",
        message: "server_error: Upstream model unavailable",
      })
    }),
@@ -4515,7 +4436,7 @@ describe("OpenAI Responses route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({ reason: { _tag: "InvalidRequest" }, message: "invalid_prompt" })
+      expect(error.reason).toMatchObject({ _tag: "InvalidRequest", message: "invalid_prompt" })
    }),
  )

@@ -4538,9 +4459,10 @@ describe("OpenAI Responses route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({
-        reason: { _tag: "InvalidRequest", classification: "context-overflow" },
+      expect(error.reason).toMatchObject({
+        _tag: "InvalidRequest",
        message: "context_length_exceeded: prompt too long",
+        classification: "context-overflow",
      })
    }),
  )
@@ -4565,9 +4487,10 @@ describe("OpenAI Responses route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({
-        reason: { _tag: "InvalidRequest", classification: "context-overflow" },
+      expect(error.reason).toMatchObject({
+        _tag: "InvalidRequest",
        message: "context_length_exceeded: prompt too long",
+        classification: "context-overflow",
      })
    }),
  )
@@ -4589,7 +4512,7 @@ describe("OpenAI Responses route", () => {
        Effect.flip,
      )

-      expect(error).toMatchObject({ reason: { _tag: "UnknownProvider" }, message: "Something went wrong" })
+      expect(error.reason).toMatchObject({ _tag: "UnknownProvider", message: "Something went wrong" })
    }),
  )

@@ -4601,8 +4524,8 @@ describe("OpenAI Responses route", () => {
      )

      expect(error.reason).toMatchObject({ _tag: "UnknownProvider" })
-      expect(error.message).toContain('"error":null')
-      expect(error.reason.body).toBe(error.message)
+      expect(error.reason.message).toContain('"error":null')
+      expect(error.body).toBe(error.reason.message)
    }),
  )

@@ -4614,8 +4537,8 @@ describe("OpenAI Responses route", () => {
      )

      expect(error.reason).toMatchObject({ _tag: "ProviderInternal" })
-      expect(error.message).toContain('"type":"error"')
-      expect(error.reason.body).toBe(error.message)
+      expect(error.reason.message).toContain('"type":"error"')
+      expect(error.body).toBe(error.reason.message)
    }),
  )

@@ -4627,8 +4550,8 @@ describe("OpenAI Responses route", () => {
      )

      expect(error.reason).toMatchObject({ _tag: "UnknownProvider" })
-      expect(error.message).toContain('"resp_failed_3"')
-      expect(error.reason.body).toBe(error.message)
+      expect(error.reason.message).toContain('"resp_failed_3"')
+      expect(error.body).toBe(error.reason.message)
    }),
  )

@@ -4645,7 +4568,7 @@ describe("OpenAI Responses route", () => {
      )

      expect(error).toBeInstanceOf(AIError)
-      expect(error).toMatchObject({ reason: { _tag: "InvalidRequest" }, message: "Bad request" })
+      expect(error.reason).toMatchObject({ _tag: "InvalidRequest", message: "Bad request" })
    }),
  )
 })
@@ -295,7 +295,7 @@ describe("OpenRouter", () => {
              {
                type: "reasoning",
                text: "Thinking",
-                providerMetadata: { openrouter: { reasoningField: "reasoning", reasoningDetails: details } },
+                providerMetadata: { openai: { reasoningField: "reasoning", reasoningDetails: details } },
              },
            ]),
          ],
@@ -328,7 +328,7 @@ describe("OpenRouter", () => {
            Message.assistant({
              type: "reasoning",
              text: "Thinking",
-              providerMetadata: { openrouter: { reasoningField: "reasoning", reasoningDetails: details } },
+              providerMetadata: { openai: { reasoningField: "reasoning", reasoningDetails: details } },
            }),
          ],
        }),
@@ -354,7 +354,7 @@ describe("OpenRouter", () => {
            Message.assistant({
              type: "reasoning",
              text: "AB",
-              providerMetadata: { openrouter: { reasoningField: "reasoning", reasoningDetails: details } },
+              providerMetadata: { openai: { reasoningField: "reasoning", reasoningDetails: details } },
            }),
          ],
        }),
@@ -80,9 +80,7 @@ describe("Z.ai Images", () => {

  it.effect("sanitizes unpaired surrogates in outbound image requests", () =>
    Image.generate({
-      model: ZAI.configure({ apiKey: "test", http: { body: { metadata: { source: "default\uDC00" } } } }).image(
-        "model",
-      ),
+      model: ZAI.configure({ apiKey: "test", http: { body: { metadata: { source: "default\uDC00" } } } }).image("model"),
      prompt: "A red circle \uD800 on a white background \u{1F600}",
    }).pipe(
      Effect.provide(
@@ -1,12 +1,6 @@
 import { describe, expect, test } from "bun:test"
-import { Effect, Schema, Stream } from "effect"
 import * as OpenAIChat from "../src/protocols/openai-chat.js"
-import { AIError, HttpContext, InvalidProviderOutputError, LLM } from "../src/index.js"
-import { Anthropic } from "../src/providers.js"
-import { Auth, Framing, HttpTransport, LLMClient, Route } from "../src/route.js"
-import { fixedResponse, truncatedStream } from "./lib/http.js"
-import { sseEvents } from "./lib/sse.js"
-import { it } from "./lib/effect.js"
+import { Auth } from "../src/route.js"

 describe("Route.with", () => {
  test("merges endpoint query and header defaults while replacing auth and id", () => {
@@ -46,185 +40,4 @@ describe("Route.with", () => {
      "x-patch": "patch",
    })
  })
-
-  test("assigns metadata ownership to a replacement provider and preserves explicit overrides", () => {
-    const route = OpenAIChat.route.with({ provider: "azure" })
-    const overridden = route.with({ providerMetadataKey: "custom-azure" }).with({ headers: { "x-test": "value" } })
-
-    expect(route.providerMetadataKey).toBe("azure")
-    expect(overridden.providerMetadataKey).toBe("custom-azure")
-    expect(overridden.defaults).not.toHaveProperty("providerMetadataKey")
-  })
-})
-
-describe("Route diagnostics", () => {
-  const route = OpenAIChat.route.with({ endpoint: { baseURL: "https://provider.test/v1" } })
-  const request = LLM.request({ model: route.model({ id: "test" }), prompt: "Hello" })
-  const headers = { "content-type": "text/event-stream", "x-request-id": "req_stream" }
-
-  it.effect("retains an entire invalid event and its validation cause", () =>
-    Effect.gen(function* () {
-      const frame = '{ "choices": "invalid", "diagnostic": { "detail": "original" } }'
-      const error = yield* LLMClient.generate(request).pipe(
-        Effect.provide(fixedResponse(sseEvents(frame), { headers })),
-        Effect.flip,
-      )
-
-      expect(error.reason._tag).toBe("InvalidProviderOutput")
-      expect(error.reason.body).toBe(frame)
-      expect(error.reason.cause).toMatchObject({ _tag: "SchemaError" })
-      expect(error.reason.http).toEqual(
-        new HttpContext({ url: "https://provider.test/v1/chat/completions", status: 200, headers }),
-      )
-    }),
-  )
-
-  it.effect("retains original provider error fields discarded by the event schema", () =>
-    Effect.gen(function* () {
-      const frame =
-        '{ "error": { "message": "Rate limit exceeded", "code": "rate_limit_exceeded", "debug": { "trace": "original" } }, "request_id": "req_original" }'
-      const error = yield* LLMClient.generate(request).pipe(
-        Effect.provide(fixedResponse(sseEvents(frame), { headers })),
-        Effect.flip,
-      )
-
-      expect(error.reason._tag).toBe("RateLimit")
-      expect(error.message).toBe("Rate limit exceeded")
-      expect(error.reason.body).toBe(frame)
-      expect(error.reason.http?.headers["x-request-id"]).toBe("req_stream")
-    }),
-  )
-
-  it.effect("preserves semantic parser bodies while replacing serialized event fallbacks", () =>
-    Effect.forEach([undefined, '{"query": BROKEN}', ""], (body) =>
-      Effect.gen(function* () {
-        const cause = new Error("parser failure")
-        const http = new HttpContext({
-          url: "https://upstream.test/v1",
-          status: 202,
-          headers: { "x-trace": "precise" },
-        })
-        const frame = '{ "type": "failure", "debug": "not in schema" }'
-        const failing = Route.make({
-          id: "diagnostics",
-          provider: "test",
-          endpoint: route.endpoint,
-          framing: Framing.sse,
-          protocol: {
-            ...OpenAIChat.protocol,
-            stream: {
-              event: Schema.fromJsonString(Schema.Struct({ type: Schema.String })),
-              initial: () => undefined,
-              step: (_state, event) =>
-                Effect.fail(
-                  new AIError({
-                    reason: new InvalidProviderOutputError({
-                      message: "Parser failed",
-                      body: body ?? JSON.stringify(event),
-                      http,
-                      cause,
-                    }),
-                  }),
-                ),
-            },
-          },
-        })
-        const error = yield* LLMClient.generate(
-          LLM.request({ model: failing.model({ id: "test" }), prompt: "Hello" }),
-        ).pipe(Effect.provide(fixedResponse(sseEvents(frame), { headers })), Effect.flip)
-
-        expect(error.message).toBe("Parser failed")
-        expect(error.reason.body).toBe(body ?? frame)
-        expect(error.reason.cause).toBe(cause)
-        expect(error.reason.http).toBe(http)
-      }),
-    ),
-  )
-
-  it.effect("retains malformed assembled Anthropic hosted-tool arguments", () =>
-    Effect.gen(function* () {
-      const body = '{"query": BROKEN}'
-      const error = yield* LLMClient.generate(
-        LLM.request({
-          model: Anthropic.configure({ apiKey: "test", baseURL: "https://provider.test" }).model("claude"),
-          prompt: "Hello",
-        }),
-      ).pipe(
-        Effect.provide(
-          fixedResponse(
-            sseEvents(
-              { type: "message_start", message: { usage: { input_tokens: 5 } } },
-              {
-                type: "content_block_start",
-                index: 0,
-                content_block: { type: "server_tool_use", id: "srv1", name: "web_search" },
-              },
-              {
-                type: "content_block_delta",
-                index: 0,
-                delta: { type: "input_json_delta", partial_json: '{"query": ' },
-              },
-              { type: "content_block_delta", index: 0, delta: { type: "input_json_delta", partial_json: "BROKEN}" } },
-              { type: "content_block_stop", index: 0 },
-            ),
-            { headers },
-          ),
-        ),
-        Effect.flip,
-      )
-
-      expect(error.reason._tag).toBe("InvalidProviderOutput")
-      expect(error.message).toContain("Invalid JSON input for anthropic-messages tool call web_search")
-      expect(error.reason.body).toBe(body)
-      expect(error.reason.cause).toBeInstanceOf(Error)
-      expect(error.reason.http).toMatchObject({ status: 200, headers })
-    }),
-  )
-
-  it.effect("adds successful HTTP metadata without replacing framing diagnostics", () =>
-    Effect.gen(function* () {
-      const cause = new Error("frame checksum mismatch")
-      const failure = new AIError({
-        reason: new InvalidProviderOutputError({
-          message: "Invalid frame",
-          body: "original frame representation",
-          cause,
-        }),
-      })
-      const failing = route.with({
-        transport: HttpTransport.httpJson({ framing: { id: "failure", frame: () => Stream.fail(failure) } }),
-      })
-      const error = yield* LLMClient.generate(
-        LLM.request({ model: failing.model({ id: "test" }), prompt: "Hello" }),
-      ).pipe(Effect.provide(fixedResponse("wire bytes", { headers })), Effect.flip)
-
-      expect(error.reason.body).toBe(failure.reason.body)
-      expect(error.reason.cause).toBe(cause)
-      expect(error.message).toBe(failure.message)
-      expect(error.reason.http?.status).toBe(200)
-      expect(error.reason.http?.headers).toEqual(headers)
-    }),
-  )
-
-  it.effect("retains the original read error after successful response headers", () =>
-    Effect.gen(function* () {
-      const cause = new Error("socket disconnected")
-      const error = yield* LLMClient.generate(request).pipe(Effect.provide(truncatedStream([], cause)), Effect.flip)
-
-      expect(error.reason).toMatchObject({ _tag: "Transport", operation: "read" })
-      expect(error.reason.cause).toBe(cause)
-      expect(error.reason.http?.status).toBe(200)
-      expect(error.reason.body).toBeUndefined()
-    }),
-  )
-
-  it.effect("retains successful response headers on an incomplete stream", () =>
-    Effect.gen(function* () {
-      const error = yield* LLMClient.generate(request).pipe(Effect.provide(fixedResponse("", { headers })), Effect.flip)
-
-      expect(error.reason).toMatchObject({ _tag: "InvalidProviderOutput", classification: "incomplete-stream" })
-      expect(error.reason.http?.headers).toEqual(headers)
-      expect(error.reason.body).toBeUndefined()
-    }),
-  )
 })
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Kit Langton	4f6ea3e8ef	refactor(tui): simplify plugin update controls	2026-08-25 22:02:54 -04:00
Kit Langton	f74d7972eb	fix(tui): refine plugin update states	2026-08-25 22:02:54 -04:00
Kit Langton	0395985b58	docs(cli): document plugin update controls	2026-08-25 22:02:54 -04:00
Kit Langton	7a0944cdd6	refactor(tui): use generated plugin update client	2026-08-25 22:02:54 -04:00
Kit Langton	fe2583cd26	fix(tui): harden plugin update controls	2026-08-25 22:02:54 -04:00
Kit Langton	b63167558c	feat(tui): add plugin update controls	2026-08-25 22:02:54 -04:00
Kit Langton	88ad378d83	refactor(core): simplify plugin updates	2026-08-25 22:02:53 -04:00
Kit Langton	f19ad6fa31	fix(core): harden explicit plugin updates	2026-08-25 22:02:53 -04:00
Kit Langton	1b7332467e	chore(sdk): regenerate types	2026-08-25 22:02:53 -04:00
Kit Langton	e6b6f30914	feat(core): support explicit plugin updates	2026-08-25 22:02:53 -04:00