chore(client): regenerate command API

feat(command): identify command sources
chore(client): regenerate path API
2026-07-23 11:16:13 +00:00 · 2026-07-23 12:04:37 +08:00 · 2026-07-23 12:04:35 +08:00 · 2026-07-23 12:04:25 +08:00 · 2026-07-23 12:04:06 +08:00 · 2026-07-23 12:03:46 +08:00
2028 changed files with 75519 additions and 354650 deletions
@@ -0,0 +1,5 @@
+---
+"@opencode-ai/cli": patch
+---
+
+Expose a TUI plugin slot at the top of the session view.
@@ -0,0 +1,7 @@
+---
+"@opencode-ai/client": patch
+"@opencode-ai/plugin": patch
+"@opencode-ai/protocol": patch
+---
+
+Expose transient, read-only session generation through the HTTP API, generated clients, and V2 plugin session context.
@@ -0,0 +1,5 @@
+---
+"@opencode-ai/cli": patch
+---
+
+Expose a TUI plugin slot above the session composer.
@@ -1,2 +1,3 @@
 packages/core/migration/**/snapshot.json linguist-generated
 packages/core/src/database/migration.gen.ts linguist-generated
+packages/core/src/**/*.txt text eol=lf
@@ -90,11 +90,18 @@ jobs:
          opencode-app-id: ${{ vars.OPENCODE_APP_ID }}
          opencode-app-secret: ${{ secrets.OPENCODE_APP_SECRET }}

-      - name: Build
+      - name: Build legacy CLI
+        if: github.ref_name != 'v2'
+        run: ./packages/opencode/script/build.ts ${{ (github.ref_name == 'beta' && '--sourcemaps') || '' }}
+        env:
+          OPENCODE_VERSION: ${{ needs.version.outputs.version }}
+          OPENCODE_RELEASE: ${{ needs.version.outputs.release }}
+          GH_REPO: ${{ needs.version.outputs.repo }}
+          GH_TOKEN: ${{ steps.committer.outputs.token }}
+
+      - name: Build preview CLI
        id: build
-        run: |
-          ./packages/opencode/script/build.ts ${{ (github.ref_name == 'beta' && '--sourcemaps') || '' }}
-          ./packages/cli/script/build.ts ${{ (github.ref_name == 'beta' && '--sourcemaps') || '' }}
+        run: ./packages/cli/script/build.ts ${{ (github.ref_name == 'beta' && '--sourcemaps') || '' }}
        env:
          OPENCODE_VERSION: ${{ needs.version.outputs.version }}
          OPENCODE_RELEASE: ${{ needs.version.outputs.release }}
@@ -102,6 +109,7 @@ jobs:
          GH_TOKEN: ${{ steps.committer.outputs.token }}

      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+        if: github.ref_name != 'v2'
        with:
          name: opencode-cli
          path: |
@@ -109,6 +117,7 @@ jobs:
            packages/opencode/dist/opencode-linux*

      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+        if: github.ref_name != 'v2'
        with:
          name: opencode-cli-windows
          path: packages/opencode/dist/opencode-windows*
@@ -121,6 +130,55 @@ jobs:
    outputs:
      version: ${{ needs.version.outputs.version }}

+  build-node-cli:
+    needs: version
+    if: github.repository == 'anomalyco/opencode'
+    strategy:
+      fail-fast: false
+      matrix:
+        settings:
+          - target: linux-arm64
+            host: blacksmith-4vcpu-ubuntu-2404-arm
+          - target: linux-x64
+            host: blacksmith-4vcpu-ubuntu-2404
+          - target: darwin-arm64
+            host: macos-26
+          - target: windows-arm64
+            host: blacksmith-4vcpu-windows-2025
+          - target: windows-x64
+            host: blacksmith-4vcpu-windows-2025
+    runs-on: ${{ matrix.settings.host }}
+    defaults:
+      run:
+        shell: bash
+    steps:
+      - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0
+
+      - uses: ./.github/actions/setup-bun
+        with:
+          install-flags: --os=* --cpu=*
+
+      - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
+        with:
+          node-version: "26.4.0"
+
+      - name: Build
+        run: bun packages/cli/script/build-node.ts --target=${{ matrix.settings.target }} --skip-install --outdir=dist/node
+        env:
+          OPENCODE_VERSION: ${{ needs.version.outputs.version }}
+          OPENCODE_RELEASE: ${{ needs.version.outputs.release }}
+
+      - name: Verify service lifecycle
+        if: matrix.settings.target != 'windows-arm64'
+        working-directory: packages/cli
+        run: bun run script/service-smoke.ts --node
+
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+        with:
+          name: opencode-node-cli-${{ matrix.settings.target }}
+          path: packages/cli/dist/node/cli-node-*
+          if-no-files-found: error
+
  sign-cli-windows:
    needs:
      - build-cli
@@ -413,6 +471,7 @@ jobs:
    needs:
      - version
      - build-cli
+      - build-node-cli
      - sign-cli-windows
      - build-electron
    if: always() && !failure() && !cancelled()
@@ -441,11 +500,13 @@ jobs:
          registry-url: "https://registry.npmjs.org"

      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
+        if: github.ref_name != 'v2'
        with:
          name: opencode-cli
          path: packages/opencode/dist

      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
+        if: github.ref_name != 'v2'
        with:
          name: opencode-cli-windows
          path: packages/opencode/dist
@@ -461,6 +522,12 @@ jobs:
          name: opencode-preview-cli
          path: packages/cli/dist

+      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
+        with:
+          pattern: opencode-node-cli-*
+          path: packages/cli/dist/node
+          merge-multiple: true
+
      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
        if: needs.version.outputs.release
        with:
@@ -78,11 +78,30 @@ jobs:
          bun run script/build.ts --single --skip-install
          bun run script/service-smoke.ts

+      - name: Setup Node build runtime
+        if: always()
+        uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
+        with:
+          node-version: "26.4.0"
+
+      - name: Verify Node build
+        if: always()
+        timeout-minutes: 15
+        working-directory: packages/cli
+        run: |
+          bun run script/build-node.ts --single --skip-install --outdir=dist/node
+          bun run script/service-smoke.ts --node
+
      - name: Check generated client
        if: runner.os == 'Linux'
        working-directory: packages/client
        run: bun run check:generated

+      - name: Check generated documentation
+        if: runner.os == 'Linux'
+        working-directory: packages/docs
+        run: bun run check:generated
+
  e2e:
    name: e2e (${{ matrix.settings.name }})
    if: github.ref_name != 'v2' && github.head_ref != 'v2'
@@ -11,6 +11,7 @@ node_modules
 playground
 tmp
 dist
+dist-node
 ts-dist
 .turbo
 .typecheck-profiles
@@ -25,6 +26,7 @@ Session.vim
 a.out
 target
 .scripts
+.cache
 .direnv/

 # Local dev files
@@ -1,6 +1,6 @@
 ---
 description: translate English to other languages
-model: opencode/claude-opus-4-8
+model: opencode/gpt-5.6-sol
 ---

 run git diff and translate changed english doc and UI copy files to other international languages. Translate all languages in parallel to save time.
@@ -19,8 +19,6 @@ Valid types are `feat`, `fix`, `docs`, `chore`, `refactor`, and `test`. Scopes a

 Examples: `fix(tui): simplify thinking toggle styling`, `docs: update contributing guide`, `chore(sdk): regenerate types`.

-Never bypass Git hooks. Do not use `--no-verify` or otherwise disable, skip, or circumvent commit or push hooks. If a hook fails, fix the failure or stop and report it to the user.
-
 ## Style Guide

 ### General Principles
@@ -8,6 +8,25 @@ export const zoneID = "430ba34c138cfb5360826c4909f99be8"
 export const awsStage = $app.stage === "production" ? "production" : "dev"
 export const deployAws = $app.stage === awsStage

+if ($app.stage === "production") {
+  new cloudflare.DnsRecord("TrustCenter", {
+    zoneId: zoneID,
+    name: "trust.opencode.ai",
+    type: "CNAME",
+    content: "3a69a5bb27875189.vercel-dns-016.com",
+    proxied: false,
+    ttl: 60,
+  })
+
+  new cloudflare.DnsRecord("TrustCenterVerification", {
+    zoneId: zoneID,
+    name: "opencode.ai",
+    type: "TXT",
+    content: "compai-domain-verification=org_6993a99c6200a2d642bb115d",
+    ttl: 60,
+  })
+}
+
 new cloudflare.RegionalHostname("RegionalHostname", {
  hostname: domain,
  regionKey: "us",
@@ -8,6 +8,8 @@
  makeWrapper,
  writableTmpDirAsHomeHook,
  autoPatchelfHook,
+  copyDesktopItems,
+  makeDesktopItem,
  opencode,
 }:
 let
@@ -27,9 +29,12 @@ stdenv.mkDerivation (finalAttrs: {
    nodejs
    makeWrapper
    writableTmpDirAsHomeHook
-  ] ++ lib.optionals stdenv.hostPlatform.isLinux [
+  ]
+  ++ lib.optionals stdenv.hostPlatform.isLinux [
    autoPatchelfHook
-  ] ++ lib.optionals stdenv.hostPlatform.isDarwin [
+    copyDesktopItems
+  ]
+  ++ lib.optionals stdenv.hostPlatform.isDarwin [
    # Ad-hoc sign the .app: --config.mac.identity=null below skips signing.
    darwin.autoSignDarwinBinariesHook
  ];
@@ -38,20 +43,37 @@ stdenv.mkDerivation (finalAttrs: {
    (lib.getLib stdenv.cc.cc)
  ];

+  desktopItems = lib.optional stdenv.hostPlatform.isLinux (makeDesktopItem {
+    name = "ai.opencode.desktop";
+    desktopName = "OpenCode";
+    exec = "opencode-desktop %U";
+    icon = "ai.opencode.desktop";
+    # Electron 41 derives X11 WM_CLASS from app.name.
+    startupWMClass = "OpenCode";
+    categories = [ "Development" ];
+  });
+
  env = opencode.env // {
    ELECTRON_SKIP_BINARY_DOWNLOAD = "1";
  };

-  # https://github.com/electron/electron/issues/31121
-  # mac builds use a .app bundle which doesnt have this issue
-  postPatch = lib.optionalString stdenv.isLinux ''
-    BASE_PATH=packages/desktop
-    FILES=(src/main/windows.ts)
-    for file in "''${FILES[@]}"; do
-      substituteInPlace $BASE_PATH/$file \
-        --replace-fail "process.resourcesPath" "'$out/opt/opencode-desktop/resources'"
-    done
-  '';
+  postPatch =
+    # NOTE: Relax Bun version check to be a warning instead of an error
+    ''
+      substituteInPlace packages/script/src/index.ts \
+        --replace-fail 'throw new Error(`This script requires bun@''${expectedBunVersionRange}' \
+                       'console.warn(`Warning: This script requires bun@''${expectedBunVersionRange}'
+    ''
+    # https://github.com/electron/electron/issues/31121
+    # mac builds use a .app bundle which doesnt have this issue
+    + lib.optionalString stdenv.isLinux ''
+      BASE_PATH=packages/desktop
+      FILES=(src/main/windows.ts)
+      for file in "''${FILES[@]}"; do
+        substituteInPlace $BASE_PATH/$file \
+          --replace-fail "process.resourcesPath" "'$out/opt/opencode-desktop/resources'"
+      done
+    '';

  preBuild = ''
    cp -r "${electron.dist}" $HOME/.electron-dist
@@ -76,27 +98,38 @@ stdenv.mkDerivation (finalAttrs: {
    runHook postBuild
  '';

-  installPhase =
-    ''
-      runHook preInstall
-    ''
-    + lib.optionalString stdenv.hostPlatform.isDarwin ''
-      mkdir -p $out/Applications
-      mv dist/mac*/*.app $out/Applications
-      makeWrapper "$out/Applications/OpenCode.app/Contents/MacOS/OpenCode" $out/bin/opencode-desktop
-    ''
-    + lib.optionalString stdenv.hostPlatform.isLinux ''
-      mkdir -p $out/opt/opencode-desktop
-      cp -r dist/linux*-unpacked/{resources,LICENSE*} $out/opt/opencode-desktop
-      makeWrapper ${lib.getExe electron} $out/bin/opencode-desktop \
-        --inherit-argv0 \
-        --set ELECTRON_FORCE_IS_PACKAGED 1 \
-        --add-flags $out/opt/opencode-desktop/resources/app.asar \
-        --add-flags "\''${NIXOS_OZONE_WL:+\''${WAYLAND_DISPLAY:+--ozone-platform-hint=auto --enable-features=WaylandWindowDecorations --enable-wayland-ime=true}}"
-    ''
-    + ''
-      runHook postInstall
-    '';
+  installPhase = ''
+    runHook preInstall
+  ''
+  + lib.optionalString stdenv.hostPlatform.isDarwin ''
+    mkdir -p $out/Applications
+    mv dist/mac*/*.app $out/Applications
+    makeWrapper "$out/Applications/OpenCode.app/Contents/MacOS/OpenCode" $out/bin/opencode-desktop
+  ''
+  + lib.optionalString stdenv.hostPlatform.isLinux ''
+    mkdir -p $out/opt/opencode-desktop
+    cp -r dist/linux*-unpacked/{resources,LICENSE*} $out/opt/opencode-desktop
+    install -Dm644 resources/icons/32x32.png \
+      "$out/share/icons/hicolor/32x32/apps/ai.opencode.desktop.png"
+    install -Dm644 resources/icons/64x64.png \
+      "$out/share/icons/hicolor/64x64/apps/ai.opencode.desktop.png"
+    install -Dm644 resources/icons/128x128.png \
+      "$out/share/icons/hicolor/128x128/apps/ai.opencode.desktop.png"
+    install -Dm644 resources/icons/128x128@2x.png \
+      "$out/share/icons/hicolor/256x256/apps/ai.opencode.desktop.png"
+    install -Dm644 resources/icons/icon.png \
+      "$out/share/icons/hicolor/512x512/apps/ai.opencode.desktop.png"
+    install -Dm644 resources/ai.opencode.desktop.metainfo.xml \
+      "$out/share/metainfo/ai.opencode.desktop.metainfo.xml"
+    makeWrapper ${lib.getExe electron} $out/bin/opencode-desktop \
+     --inherit-argv0 \
+     --set ELECTRON_FORCE_IS_PACKAGED 1 \
+     --add-flags $out/opt/opencode-desktop/resources/app.asar \
+     --add-flags "\''${NIXOS_OZONE_WL:+\''${WAYLAND_DISPLAY:+--ozone-platform-hint=auto --enable-features=WaylandWindowDecorations --enable-wayland-ime=true}}"
+  ''
+  + ''
+    runHook postInstall
+  '';

  autoPatchelfIgnoreMissingDeps = [
    "libc.musl-x86_64.so.1"
@@ -1,8 +1,8 @@
 {
  "nodeModules": {
-    "x86_64-linux": "sha256-F1luclnqCPQk9yxfmeSYGaM/nScf28yBu9K3Fv+Xd24=",
-    "aarch64-linux": "sha256-XW0XZnsCRkU3MFJH9TjMRYZHffzVy3cQyiNCkec2gl4=",
-    "aarch64-darwin": "sha256-bf8kvORs3Fs2UYLp3PekF+AJR7NKOcHb+fIQA79RtMk=",
-    "x86_64-darwin": "sha256-sBdQPkzd7JXNW6Lbi9JHiAsfHwdLwTKWY+uPeXAv2Nw="
+    "x86_64-linux": "sha256-qt11SKmOjq0KU542QFbs+u7YyJicn4drCcwCdg325yk=",
+    "aarch64-linux": "sha256-z68doReXTrWS7HeiAjc0btIjAsvzeZZ7hXAlHr0c77Q=",
+    "aarch64-darwin": "sha256-PILYH1Pi8XBvSkuZ+1sNnUTao5kba+m5Z8iJKx6YXPo=",
+    "x86_64-darwin": "sha256-KpcJzP4m0SUavu/WaSffgzOxrHq8ljdy0GOzs9p16lo="
  }
 }
@@ -15,7 +15,7 @@
    "dev:www": "bun run --cwd packages/www dev",
    "dev:storybook": "bun --cwd packages/storybook storybook",
    "lint": "oxlint",
-    "lint:effect-patterns": "ast-grep scan -c script/ast-grep/sgconfig.yml packages/core/src packages/server/src packages/protocol/src packages/cli/src",
+    "lint:effect-patterns": "ast-grep scan -c script/ast-grep/sgconfig.yml packages/util/src packages/core/src packages/server/src packages/protocol/src packages/cli/src",
    "test:lint-rules": "ast-grep test -c script/ast-grep/sgconfig.yml",
    "typecheck": "bun turbo typecheck --concurrency=3",
    "typecheck:profile": "bun script/profile-typecheck.ts",
@@ -37,18 +37,18 @@
      "packages/slack"
    ],
    "catalog": {
-      "@effect/opentelemetry": "4.0.0-beta.83",
-      "@effect/platform-node": "4.0.0-beta.83",
-      "@effect/sql-sqlite-bun": "4.0.0-beta.83",
+      "@effect/opentelemetry": "4.0.0-beta.98",
+      "@effect/platform-node": "4.0.0-beta.98",
+      "@effect/sql-sqlite-bun": "4.0.0-beta.98",
      "@npmcli/arborist": "9.4.0",
      "@types/bun": "1.3.13",
      "@types/cross-spawn": "6.0.6",
      "@octokit/rest": "22.0.0",
      "@hono/standard-validator": "0.2.0",
      "@hono/zod-validator": "0.4.2",
-      "@opentui/core": "0.4.3",
-      "@opentui/keymap": "0.4.3",
-      "@opentui/solid": "0.4.3",
+      "@opentui/core": "0.4.5",
+      "@opentui/keymap": "0.4.5",
+      "@opentui/solid": "0.4.5",
      "@tanstack/solid-virtual": "3.13.32",
      "@shikijs/stream": "4.2.0",
      "ulid": "3.0.1",
@@ -69,12 +69,13 @@
      "dompurify": "3.3.1",
      "drizzle-kit": "1.0.0-rc.2",
      "drizzle-orm": "1.0.0-rc.2",
-      "effect": "4.0.0-beta.83",
+      "effect": "4.0.0-beta.98",
      "ai": "6.0.168",
      "cross-spawn": "7.0.6",
      "hono": "4.10.7",
      "hono-openapi": "1.1.2",
      "fuzzysort": "3.1.0",
+      "get-east-asian-width": "1.6.0",
      "luxon": "3.6.1",
      "marked": "17.0.6",
      "marked-shiki": "1.2.1",
@@ -85,9 +86,11 @@
      "@typescript/native-preview": "7.0.0-dev.20251207.1",
      "zod": "4.1.8",
      "remeda": "2.26.0",
+      "resolve.exports": "2.0.3",
      "sst": "4.13.1",
      "shiki": "4.2.0",
      "solid-list": "0.3.0",
+      "string-width": "7.2.0",
      "tailwindcss": "4.1.11",
      "vite": "7.1.4",
      "@solidjs/meta": "0.29.4",
@@ -163,7 +166,7 @@
    "@ai-sdk/google@3.0.73": "patches/@ai-sdk%2Fgoogle@3.0.73.patch",
    "@pierre/trees@1.0.0-beta.4": "patches/@pierre%2Ftrees@1.0.0-beta.4.patch",
    "@modelcontextprotocol/sdk@1.29.0": "patches/@modelcontextprotocol%2Fsdk@1.29.0.patch",
-    "effect@4.0.0-beta.83": "patches/effect@4.0.0-beta.83.patch",
+    "effect@4.0.0-beta.98": "patches/effect@4.0.0-beta.98.patch",
    "@tanstack/virtual-core@3.17.3": "patches/@tanstack%2Fvirtual-core@3.17.3.patch"
  }
 }
@@ -1,4 +1,4 @@
-# LLM Package Guide
+# AI Package Guide

 ## Effect

@@ -115,10 +115,10 @@ Keep provider facades small and explicit:

 ### Provider Package Entrypoints

-Catalog-selected native providers use package-like export paths from `@opencode-ai/llm`. They are internal entrypoints in one npm package, not separately published provider packages. Every entrypoint implements `ProviderPackage.Definition` and exposes `model(modelID, settings)`, where settings are serializable provider configuration plus common `headers`, `body`, and `limits` overlays.
+Catalog-selected native providers use package-like export paths from `@opencode-ai/ai`. They are internal entrypoints in one npm package, not separately published provider packages. Every entrypoint implements `ProviderPackage.Definition` and exposes `model(modelID, settings)`, where settings are serializable provider configuration plus common `headers`, `body`, and `limits` overlays.

 ```ts
-import { model } from "@opencode-ai/llm/providers/openai/responses"
+import { model } from "@opencode-ai/ai/providers/openai/responses"

 const selected = model("gpt-5", {
  apiKey,
@@ -133,7 +133,7 @@ Do not expose `Route` in provider package settings. Route composition stays an i
 ### Folder layout

 ```
-packages/llm/src/
+packages/ai/src/
  schema/                   canonical Schema model, split by concern
    ids.ts                  branded IDs, literal types, ProviderMetadata
    options.ts              Generation/Provider/Http options, Limits, Model, cache policy
@@ -143,7 +143,7 @@ packages/llm/src/
    index.ts                barrel
  llm.ts                    request constructors and convenience helpers
  route/
-    index.ts                @opencode-ai/llm/route advanced barrel
+    index.ts                @opencode-ai/ai/route advanced barrel
    client.ts               Route.make + LLMClient.prepare/stream/generate
    executor.ts             RequestExecutor service + transport error mapping
    protocol.ts             Protocol type + Protocol.make
@@ -1,7 +1,7 @@
 # AI Library Design

-> Discussion draft. This document describes the intended replacement for the
-> current private `@opencode-ai/llm` API. Names and exact TypeScript signatures
+> Discussion draft. This document describes an intended clean-break redesign of
+> the current private `@opencode-ai/ai` API. Names and exact TypeScript signatures
 > are illustrative until implementation, but the domain boundaries and defaults
 > are deliberate.

@@ -1074,7 +1074,6 @@ The redesign intentionally removes or changes these current concepts:

 | Current                                 | Proposed                                                    |
 | --------------------------------------- | ----------------------------------------------------------- |
-| `@opencode-ai/llm`                      | `@opencode-ai/ai`                                           |
 | Mandatory `LLM.request({ model, ... })` | Inline calls or model-free portable requests                |
 | `LLM.generate` means one turn           | `LLM.generate` means complete run                           |
 | `LLMClient.generate/stream`             | `LLM.generateTurn/streamTurn` for one turn                  |
@@ -1,11 +1,11 @@
-# @opencode-ai/llm
+# @opencode-ai/ai

-Schema-first LLM core for opencode. One typed request, response, event, and tool language; provider quirks live in adapters, not in calling code.
+Schema-first AI primitives for opencode. Provider quirks live in adapters, not in calling code.

 ```ts
 import { Effect } from "effect"
-import { LLM, LLMClient } from "@opencode-ai/llm"
-import { OpenAI } from "@opencode-ai/llm/providers"
+import { LLM, LLMClient } from "@opencode-ai/ai"
+import { OpenAI } from "@opencode-ai/ai/providers"

 const model = OpenAI.configure({ apiKey: process.env.OPENAI_API_KEY }).responses("gpt-4o-mini")

@@ -24,6 +24,172 @@ const program = Effect.gen(function* () {

 Run `LLMClient.stream(request)` instead of `generate` when you want incremental `LLMEvent`s. The event stream is provider-neutral — same shape across OpenAI Chat, OpenAI Responses, Anthropic Messages, Gemini, Bedrock Converse, and any OpenAI-compatible deployment.

+## Image generation
+
+Use `Image.generate` with an image model for direct asset generation:
+
+```ts
+import { Image, ImageInput } from "@opencode-ai/ai"
+import { OpenAI } from "@opencode-ai/ai/providers"
+
+const program = Effect.gen(function* () {
+  const response = yield* Image.generate({
+    model: OpenAI.configure({ apiKey: process.env.OPENAI_API_KEY }).image("gpt-image-2"),
+    prompt: "A robot tending a rooftop garden",
+    options: {
+      n: 2,
+      size: "1024x1024",
+      quality: "high", // inferred from the OpenAI image model
+      outputFormat: "webp",
+      future_option: true, // unknown native options pass through unchanged
+    },
+  })
+
+  return response.images // GeneratedImage[] with owned bytes or a provider URL
+})
+```
+
+Pass ordered image inputs to the same method for editing, composition, or image-conditioned generation:
+
+```ts
+const response =
+  yield *
+  Image.generate({
+    model,
+    prompt: "Combine these product photos into one studio scene",
+    images: [
+      ImageInput.bytes(firstBytes, "image/png"),
+      ImageInput.url("https://example.com/second.webp"),
+      ImageInput.file("file_123"),
+    ],
+    options,
+    http,
+  })
+```
+
+`ImageInput.fileUri(uri, mediaType)` represents provider file URIs such as Gemini Files. Raw strings are not
+accepted as image inputs, avoiding ambiguity between base64, URLs, and provider IDs. Empty or omitted `images`
+uses text-to-image generation; a non-empty array selects the provider's edit behavior without enforcing provider
+image-count limits locally. `images` is the only common image-editing field. OpenAI uses multipart for byte/data-URL
+edits and its JSON reference body for URL or file-ID edits. Its provider-specific `options.mask` accepts an
+`ImageInput` for inpainting:
+
+```ts
+yield *
+  Image.generate({
+    model: OpenAI.configure({ apiKey }).image("gpt-image-2"),
+    prompt,
+    images: [ImageInput.bytes(sourceBytes, "image/png")],
+    options: { mask: ImageInput.bytes(maskBytes, "image/png") },
+  })
+```
+
+The OpenAI adapter extracts this helper value into the edit request's native `mask` field rather than passing the
+tagged `ImageInput` object through as an ordinary option. On multipart requests, `http.body` can override option
+fields but not structural `model`, `prompt`, `image[]`, or `mask` fields, and the transport owns the multipart
+`Content-Type` boundary. For JSON requests, `http.body` remains the final raw-native overlay. Gemini does not fetch
+public HTTP URLs, and hosted Z.ai image generation does not accept image inputs. These cases fail with
+`InvalidRequest` before network I/O.
+
+Provider-native image options belong to each request. Raw `http.body` fields have final precedence over them:
+
+```ts
+const model = OpenAI.configure({ apiKey }).image("gpt-image-2")
+
+yield *
+  Image.generate({
+    model,
+    prompt,
+    options: { quality: "medium" },
+    http,
+  })
+```
+
+xAI image models use the same request API with xAI-native controls:
+
+```ts
+yield *
+  Image.generate({
+    model: XAI.configure({ apiKey }).image("any-model-id"),
+    prompt,
+    options: {
+      n: 2,
+      aspectRatio: "16:9",
+      resolution: "1k",
+      responseFormat: "b64_json",
+      future_option: true,
+    },
+    http,
+  })
+```
+
+Google's current Gemini image models use the same direct API:
+
+```ts
+import { Google } from "@opencode-ai/ai/providers"
+
+const googleProgram = Effect.gen(function* () {
+  const response = yield* Image.generate({
+    model: Google.configure({ apiKey }).image("any-model-id"),
+    prompt: "A robot tending a rooftop garden",
+    options: {
+      aspectRatio: "16:9",
+      imageSize: "2K",
+      seed: 42,
+      thinkingLevel: "HIGH",
+      includeThoughts: true,
+      futureOption: true,
+    },
+    http,
+  })
+
+  return response.images
+})
+```
+
+Google image options are request-scoped and inferred from the selected model. Known fields autocomplete while
+future string values and arbitrary native Gemini `generationConfig` fields remain available. Native fields override
+their mapped aliases, and `http.body` is the final deep overlay. The selected model ID is sent to Gemini
+`generateContent` without a local allowlist.
+
+Z.ai image models infer open Z.ai-native options from the selected model:
+
+```ts
+yield *
+  Image.generate({
+    model: ZAI.configure({ apiKey }).image("any-model-id"),
+    prompt,
+    options: {
+      quality: "hd",
+      userID: "user-123",
+      future_option: true,
+    },
+    http,
+  })
+```
+
+Z.ai does not include trustworthy MIME metadata for output URLs, so generated images use
+`application/octet-stream`. Output URLs expire after 30 days; download and persist them promptly if they must
+remain available.
+
+Conversational image generation remains part of the LLM interaction. OpenAI Responses exposes it through its hosted image tool:
+
+```ts
+const program = Effect.gen(function* () {
+  const response = yield* LLM.generate(
+    LLM.request({
+      model: OpenAI.configure({ apiKey }).responses("gpt-5"),
+      prompt: "Design a solarpunk rooftop garden, then show me.",
+      tools: [OpenAI.imageGeneration({ quality: "high" })],
+    }),
+  )
+
+  return response.message
+})
+```
+
+The hosted result is represented as a provider-executed tool call and tool result. Its image is a `file` content item with a data URI, so retaining `response.message` preserves the generated image for continuation.
+
 ## Public API

 - **`LLM.request({...})`** — build a provider-neutral `LLMRequest`. Accepts ergonomic inputs (`system: string`, `prompt: string`) that normalize into the canonical Schema classes.
@@ -32,6 +198,8 @@ Run `LLMClient.stream(request)` instead of `generate` when you want incremental
 - **`Model.make(...)` / `ToolCallPart.make(...)` / `ToolResultPart.make(...)` / `ToolDefinition.make(...)`** — model and tool-related constructors from the canonical schema model.
 - **`LLMClient.prepare(request)`** — compile a request through protocol body construction, validation, and HTTP preparation without sending. Useful for inspection and testing.
 - **`LLMEvent.is.*`** — typed guards (`is.textDelta`, `is.toolCall`, `is.finish`, …) for filtering streams.
+- **`Image.generate({...})`** — generate images through a provider-neutral image request and response model.
+- **`ImageClient`** — Effect service and layer for image execution, parallel to `LLMClient`.

 ## Caching

@@ -95,7 +263,7 @@ Normalized cache usage is read back into `response.usage.cacheReadInputTokens` a
 Provider facades configure endpoint/auth/deployment details first, then expose model selectors that take only a model or deployment id. The selected model carries the executable route value used at runtime.

 ```ts
-import { OpenAI, CloudflareAIGateway } from "@opencode-ai/llm/providers"
+import { OpenAI, CloudflareAIGateway } from "@opencode-ai/ai/providers"

 const openai = OpenAI.configure({ apiKey: process.env.OPENAI_API_KEY }).responses("gpt-4o-mini")
 const gateway = CloudflareAIGateway.configure({
@@ -104,14 +272,14 @@ const gateway = CloudflareAIGateway.configure({
 }).model("workers-ai/@cf/meta/llama-3.1-8b-instruct")
 ```

-Included providers: OpenAI, Anthropic, Google (Gemini), Google Vertex Gemini and Anthropic, Amazon Bedrock, Azure OpenAI, Cloudflare AI Gateway, Cloudflare Workers AI, GitHub Copilot, OpenRouter, xAI, plus generic OpenAI-compatible Chat and Responses entrypoints and an Anthropic Messages-compatible entrypoint.
+Included providers: OpenAI, Anthropic, Google (Gemini), Google Vertex Gemini and Anthropic, Amazon Bedrock, Azure OpenAI, Cloudflare AI Gateway, Cloudflare Workers AI, GitHub Copilot, OpenRouter, xAI, Z.ai, plus generic OpenAI-compatible Chat and Responses entrypoints and an Anthropic Messages-compatible entrypoint.

 ### Package-like entrypoints

-Native catalog integrations load provider behavior through package-like entrypoints. These are export paths from the same `@opencode-ai/llm` npm package, not independently published packages. Each entrypoint exports the same `model(modelID, settings)` contract, and `settings` contains serializable provider configuration plus common `headers`, `body`, and `limits` overlays.
+Native catalog integrations load provider behavior through package-like entrypoints. These are export paths from the same `@opencode-ai/ai` npm package, not independently published packages. Each entrypoint exports the same `model(modelID, settings)` contract, and `settings` contains serializable provider configuration plus common `headers`, `body`, and `limits` overlays.

 ```ts
-import { model } from "@opencode-ai/llm/providers/openai/responses"
+import { model } from "@opencode-ai/ai/providers/openai/responses"

 const selected = model("gpt-5", {
  apiKey: process.env.OPENAI_API_KEY,
@@ -123,27 +291,41 @@ const selected = model("gpt-5", {

 OpenAI Chat and OpenAI Responses are separate semantic entrypoints:

- `@opencode-ai/llm/providers/openai/chat`
- `@opencode-ai/llm/providers/openai/responses`
- `@opencode-ai/llm/providers/openai-compatible/responses`
- `@opencode-ai/llm/providers/anthropic-compatible`
- `@opencode-ai/llm/providers/google-vertex`
- `@opencode-ai/llm/providers/google-vertex/anthropic`
+- `@opencode-ai/ai/providers/openai/chat`
+- `@opencode-ai/ai/providers/openai/responses`
+- `@opencode-ai/ai/providers/openai-compatible/responses`
+- `@opencode-ai/ai/providers/anthropic-compatible`
+- `@opencode-ai/ai/providers/google-vertex/gemini`
+- `@opencode-ai/ai/providers/google-vertex/chat`
+- `@opencode-ai/ai/providers/google-vertex/responses`
+- `@opencode-ai/ai/providers/google-vertex/messages`

 Responses HTTP versus WebSocket is a scoped `transport` setting on the OpenAI Responses entrypoint, not another entrypoint. Azure follows the same Chat/Responses split at `providers/azure/chat` and `providers/azure/responses`. Generic OpenAI-compatible Chat remains at `providers/openai-compatible`; compatible Responses is separate at `providers/openai-compatible/responses`. Generic Anthropic Messages-compatible providers use `providers/anthropic-compatible`, which the named Anthropic provider composes. Google Gemini and Amazon Bedrock expose their single native API through their existing provider paths.

-Vertex Gemini and Vertex Anthropic are separate products with separate entrypoints. Both accept `project`, `location`, and an optional `accessToken`; when no explicit token or auth override is supplied they lazily use Google Application Default Credentials. Vertex Gemini instead selects express mode when `apiKey` or `GOOGLE_VERTEX_API_KEY` is present.
+Vertex Gemini, Vertex Chat, Vertex Responses, and Vertex Messages are separate API entrypoints. All accept `project`, `location`, and an optional `accessToken`; when no explicit token or auth override is supplied they lazily use Google Application Default Credentials. Vertex Gemini instead selects express mode when `apiKey` or `GOOGLE_VERTEX_API_KEY` is present. Vertex Chat targets MaaS models through the OpenAI-compatible Chat Completions endpoint, while Vertex Responses targets Grok models and defaults `store` to `false` as required by Vertex. `providers/google-vertex` remains the default alias for `providers/google-vertex/gemini`.

 Tuned Vertex Gemini deployments use model ids shaped like `endpoints/1234567890` and require OAuth or ADC; Vertex express-mode API keys support publisher models only.

 ```ts
-import { model } from "@opencode-ai/llm/providers/google-vertex"
+import { model } from "@opencode-ai/ai/providers/google-vertex/gemini"

 model("gemini-3.5-flash", { project: "my-project", location: "global" })
 ```

 ```ts
-import { model } from "@opencode-ai/llm/providers/google-vertex/anthropic"
+import { model } from "@opencode-ai/ai/providers/google-vertex/chat"
+
+model("deepseek-ai/deepseek-v3.2-maas", { project: "my-project", location: "global" })
+```
+
+```ts
+import { model } from "@opencode-ai/ai/providers/google-vertex/responses"
+
+model("xai/grok-4.20-reasoning", { project: "my-project", location: "global" })
+```
+
+```ts
+import { model } from "@opencode-ai/ai/providers/google-vertex/messages"

 model("claude-sonnet-4-6", { project: "my-project", location: "global" })
 ```
@@ -168,7 +350,7 @@ Adding a new model or deployment is usually 5-15 lines using `Route.make({ proto

 ## Effect

-This package is built on Effect. Public methods return `Effect` or `Stream`; provide `LLMClient.layer` for runtime dispatch and import the provider/protocol modules for the routes you use. The example at `example/tutorial.ts` is a runnable walkthrough.
+This package is built on Effect. Public methods return `Effect` or `Stream`; provide `LLMClient.layer` for LLM dispatch and `ImageClient.layer` for image dispatch, then import the provider/protocol modules for the routes you use. The example at `example/tutorial.ts` is a runnable walkthrough.

 ## See also

@@ -1,15 +1,15 @@
 # LLM Provider Parity Status

-Last reviewed: 2026-07-15
+Last reviewed: 2026-07-17

-This file tracks the gap between the native `@opencode-ai/llm` package and the AI SDK provider packages that opencode still depends on for many catalog/runtime paths.
+This file tracks the gap between the native `@opencode-ai/ai` package and the AI SDK provider packages that opencode still depends on for many catalog/runtime paths.

 ## Existing Status Sources

-| File                                 | What it tracks                                                                   | Limitation                                              |
-| ------------------------------------ | -------------------------------------------------------------------------------- | ------------------------------------------------------- |
-| `packages/llm/DESIGN.md`             | Future clean-break API proposal, currently named `@opencode-ai/ai` in the draft. | Not a provider parity tracker.                          |
-| `packages/llm/example/call-sites.md` | Route/value/provider-facade migration checklist and call-site sketches.          | Architecture migration only; not AI SDK package parity. |
+| File                                | What it tracks                                                          | Limitation                                              |
+| ----------------------------------- | ----------------------------------------------------------------------- | ------------------------------------------------------- |
+| `packages/ai/DESIGN.md`             | Future clean-break API proposal for `@opencode-ai/ai`.                  | Not a provider parity tracker.                          |
+| `packages/ai/example/call-sites.md` | Route/value/provider-facade migration checklist and call-site sketches. | Architecture migration only; not AI SDK package parity. |

 ## Current Implementation Snapshot

@@ -20,11 +20,13 @@ This file tracks the gap between the native `@opencode-ai/llm` package and the A
 | OpenAI Responses WebSocket         | `src/protocols/openai-responses.ts`, `src/route/transport/websocket.ts`                        | Present as `OpenAI.responsesWebSocket(...)`.                                                                                                                             | Runner/catalog support explicitly must not downgrade WebSocket routes; broader runtime selection is not complete.                                                                                                                                         |
 | OpenAI-compatible Chat             | `src/protocols/openai-compatible-chat.ts`, `src/providers/openai-compatible.ts`                | Usable for generic Chat and several profiles: Baseten, Cerebras, DeepInfra, DeepSeek, Fireworks, Groq, TogetherAI.                                                       | Family quirks are mostly endpoint defaults, not full typed behavior.                                                                                                                                                                                      |
 | OpenAI-compatible Responses        | `src/protocols/openai-compatible-responses.ts`, `src/providers/openai-compatible-responses.ts` | Usable for deployments that implement the OpenAI Responses wire protocol.                                                                                                | No named family profiles or recorded deployment coverage yet.                                                                                                                                                                                             |
-| Anthropic-compatible Messages      | `src/protocols/anthropic-messages.ts`, `src/providers/anthropic-compatible.ts`                 | Usable for deployments that implement the Anthropic Messages wire protocol. Named Anthropic composes this base.                                                          | No named compatible family profiles or recorded deployment coverage yet.                                                                                                                                                                                  |
+| Anthropic-compatible Messages      | `src/protocols/anthropic-messages.ts`, `src/providers/anthropic-compatible.ts`                 | Usable for deployments that implement the Anthropic Messages wire protocol. Named Anthropic composes this base; MiniMax M3 has recorded text and tool-loop coverage.       | No named compatible family profiles yet.                                                                                                                                                                                                                  |
 | Anthropic Messages                 | `src/protocols/anthropic-messages.ts`, `src/providers/anthropic.ts`                            | Usable. Supports tools, thinking, cache control, images, server-hosted tool events, and usage.                                                                           | Provider option surface is small. Beta/header handling, metadata, and newer Messages fields need a typed parity pass.                                                                                                                                     |
 | Gemini Developer API               | `src/protocols/gemini.ts`, `src/providers/google.ts`                                           | Usable for Google API key flow. Supports text, images, tools, thinking signatures, and cache usage.                                                                      | This is not Vertex. Typed provider options are narrow; many Gemini request fields currently require raw `http.body` overlays.                                                                                                                             |
-| Vertex Gemini                      | `src/protocols/google-vertex-gemini.ts`, `src/providers/google-vertex.ts`                      | Usable through API-key express mode, explicit OAuth tokens, or ADC with project/location endpoint derivation, including tuned `endpoints/...` deployments.               | Core runner/catalog mapping and recorded provider coverage are missing.                                                                                                                                                                                   |
-| Vertex Anthropic Messages          | `src/protocols/google-vertex-anthropic.ts`, `src/providers/google-vertex-anthropic.ts`         | Usable through explicit OAuth tokens or ADC, including global, regional, and `eu`/`us` multi-region endpoints.                                                           | Core runner/catalog mapping and recorded provider coverage are missing; Vertex-specific hosted-tool parity needs review.                                                                                                                                  |
+| Vertex Gemini                      | `src/protocols/gemini.ts`, `src/providers/google-vertex.ts`                                    | Usable through API-key express mode, explicit OAuth tokens, or ADC with project/location endpoint derivation, including tuned `endpoints/...` deployments.               | Core runner/catalog mapping and recorded provider coverage are missing.                                                                                                                                                                                   |
+| Vertex Chat                        | `src/protocols/openai-chat.ts`, `src/providers/google-vertex-chat.ts`                          | Usable for MaaS models through OpenAI-compatible Chat Completions with explicit OAuth tokens or ADC and project/location endpoint derivation.                            | Core runner/catalog mapping and recorded provider coverage are missing; MaaS family-specific request parity needs review.                                                                                                                                 |
+| Vertex Responses                   | `src/protocols/openai-responses.ts`, `src/providers/google-vertex-responses.ts`                | Usable for Grok models through OpenAI-compatible Responses with explicit OAuth tokens or ADC, project/location endpoint derivation, and storage disabled by default.     | Core runner/catalog mapping and recorded provider coverage are missing; stateful continuation is not supported by Vertex.                                                                                                                                 |
+| Vertex Messages                    | `src/protocols/anthropic-messages.ts`, `src/providers/google-vertex-messages.ts`               | Usable through explicit OAuth tokens or ADC, including global, regional, and `eu`/`us` multi-region endpoints.                                                           | Core runner/catalog mapping and recorded provider coverage are missing; Vertex-specific hosted-tool parity needs review.                                                                                                                                  |
 | Bedrock Converse                   | `src/protocols/bedrock-converse.ts`, `src/providers/amazon-bedrock.ts`                         | Partial but real. Supports AWS event-stream framing, SigV4 with supplied credentials, bearer auth, tools, reasoning signatures, media, cache points, and recorded tests. | Native facade does not mirror the AI SDK plugin's default AWS credential chain/profile behavior. Runner/catalog mapping is missing. Guardrails, inference profiles, region-specific model ID fixes, and model-specific request fields need a parity pass. |
 | Azure OpenAI                       | `src/providers/azure.ts` using OpenAI Chat/Responses protocols                                 | Partial. Supports resource/base URL setup, API key auth, API version query, Chat, and Responses selectors.                                                               | Core runner does not map `@ai-sdk/azure` to this native facade. AAD/token auth and Azure-specific endpoint variants need review.                                                                                                                          |
 | Cloudflare AI Gateway / Workers AI | `src/providers/cloudflare.ts`                                                                  | Present via OpenAI-compatible Chat routes.                                                                                                                               | Useful but not part of the critical AI SDK replacement set yet. Needs per-product recorded coverage before relying on it broadly.                                                                                                                         |
@@ -54,8 +56,8 @@ Other `aisdk:` packages, including Google Vertex, Azure, and Bedrock, currently
 | `@ai-sdk/google`                  | Gemini Developer API                                           | Partial / usable | Add typed options for safety, response schema/modalities, cached content, grounding/search/code execution, and non-text output modes where supported.                                                               |
 | `@ai-sdk/google-vertex`           | Vertex Gemini namespace/facade                                 | Partial / usable | Add runner/catalog mapping, recorded coverage, and broader provider-option parity.                                                                                                                                  |
 | `@ai-sdk/google-vertex/anthropic` | Anthropic Messages over Vertex namespace/facade                | Partial / usable | Add runner/catalog mapping, recorded coverage, and Vertex-specific hosted-tool parity.                                                                                                                              |
-| `@ai-sdk/google-vertex/maas`      | Vertex MaaS OpenAI-compatible namespace/facade                 | Missing          | Decide native Chat/Responses selection, endpoint derivation, auth, and catalog mapping.                                                                                                                             |
-| `@ai-sdk/google-vertex/xai`       | Vertex xAI OpenAI-compatible namespace/facade                  | Missing          | Decide whether this composes the generic compatible bases or the xAI facade, then add endpoint/auth mapping and tests.                                                                                              |
+| `@ai-sdk/google-vertex/maas`      | Vertex Chat                                                    | Partial / usable | Add runner/catalog mapping, recorded coverage, and MaaS family-specific request parity.                                                                                                                             |
+| `@ai-sdk/google-vertex/xai`       | Vertex Chat / Responses                                        | Partial / usable | Decide Chat/Responses selection for catalog models, add runner mapping and recorded coverage, and review xAI-specific request options.                                                                              |
 | `@ai-sdk/azure`                   | Azure OpenAI Chat/Responses facade                             | Partial          | Map runner/catalog metadata to native Azure, handle resourceName/baseURL/apiVersion variants, add AAD/token auth story, and verify Chat vs Responses deployment selection.                                          |
 | `@ai-sdk/amazon-bedrock`          | Bedrock Converse                                               | Partial          | Add default AWS credential chain/profile support, region/inference-profile model ID handling, provider option parity via `additionalModelRequestFields`, guardrails/performance config, and runner/catalog mapping. |
 | `@ai-sdk/amazon-bedrock/mantle`   | Bedrock Mantle OpenAI-compatible Chat/Responses namespace      | Missing          | Decide native Mantle shape, likely separate from Converse because it uses OpenAI-compatible Chat/Responses semantics over Bedrock. Add package mapping and tests.                                                   |
@@ -65,42 +67,42 @@ Other `aisdk:` packages, including Google Vertex, Azure, and Bedrock, currently
 1. Runner support is narrower than the LLM package. The package has native provider facades for Google, Azure, and Bedrock, but the V2 Session runner only maps OpenAI, Anthropic, and explicit OpenAI-compatible Chat from `aisdk` catalog metadata.
 2. OpenAI-compatible Responses is available as a separate package entrypoint, but the V2 runner still maps `@ai-sdk/openai-compatible` to Chat only. Catalog selection must become API-aware before Responses deployments can use it.
 3. Bedrock native auth is not AI SDK parity. The AI SDK plugin uses the default AWS provider chain, profile, container credentials, and Bedrock bearer token env behavior. Native Bedrock currently expects explicit credentials or bearer auth on the facade.
-4. Vertex Gemini and Vertex Anthropic now have native package entrypoints, but the core runner does not map catalog metadata to them yet and recorded provider coverage is still missing.
+4. Vertex Gemini, Vertex Chat, Vertex Responses, and Vertex Messages now have native package entrypoints, but the core runner does not map catalog metadata to them yet and recorded provider coverage is still missing.
 5. Azure is only a provider facade, not a full runtime replacement. Native Azure exists, but the catalog runner does not select it, and token auth/resource variants need review.
 6. Provider option typing is uneven. OpenAI, Anthropic, Gemini, Bedrock, and OpenRouter each expose a small typed subset plus raw HTTP overlays; this is useful but not equivalent to AI SDK provider option coverage.
 7. Structured output is not provider-native yet. `LLM.generateObject` still uses a synthetic tool strategy, while the future design expects native structured output where reliable and tool fallback where needed.
-8. Package/namespace boundaries for the current native loading set are explicit in docs and exports. Other exported provider facades are not catalog package entrypoints until they implement the contract. Missing native boundaries remain for Vertex MaaS, Vertex xAI, and Bedrock Mantle.
+8. Package/namespace boundaries for the current native loading set are explicit in docs and exports. Other exported provider facades are not catalog package entrypoints until they implement the contract. Vertex xAI still needs catalog API selection; the missing native boundary is Bedrock Mantle.
 9. Recorded coverage is uneven. OpenAI, Anthropic, Gemini, Bedrock Converse, Cloudflare, OpenRouter, and several OpenAI-compatible Chat providers have cassettes. Azure, Vertex, and Mantle need first-class recorded scenarios before switching defaults.

 ## Native Namespace Shape

 These are implementation/API slices, not separate npm packages.

-| API slice                     | Package-like entrypoint                                  | Purpose                                                                      |
-| ----------------------------- | -------------------------------------------------------- | ---------------------------------------------------------------------------- |
-| OpenAI Chat                   | `@opencode-ai/llm/providers/openai/chat`                 | OpenAI `/chat/completions` semantics.                                        |
-| OpenAI Responses              | `@opencode-ai/llm/providers/openai/responses`            | OpenAI `/responses` semantics with HTTP/WebSocket selected through settings. |
-| OpenAI-compatible Chat        | `@opencode-ai/llm/providers/openai-compatible`           | Generic OpenAI-compatible `/chat/completions`.                               |
-| OpenAI-compatible Responses   | `@opencode-ai/llm/providers/openai-compatible/responses` | Generic OpenAI-compatible `/responses`.                                      |
-| Anthropic-compatible Messages | `@opencode-ai/llm/providers/anthropic-compatible`        | Generic Anthropic-compatible `/messages`.                                    |
-| Anthropic Messages            | `@opencode-ai/llm/providers/anthropic`                   | Anthropic Messages API.                                                      |
-| Gemini Developer API          | `@opencode-ai/llm/providers/google`                      | Google AI Studio Gemini API.                                                 |
-| Vertex Gemini                 | `@opencode-ai/llm/providers/google-vertex`               | Vertex Gemini API.                                                           |
-| Vertex Anthropic Messages     | `@opencode-ai/llm/providers/google-vertex/anthropic`     | Vertex-hosted Anthropic Messages API.                                        |
-| Vertex MaaS                   | Missing                                                  | Vertex OpenAI-compatible MaaS APIs.                                          |
-| Vertex xAI                    | Missing                                                  | Vertex-hosted xAI APIs.                                                      |
-| Bedrock Converse              | `@opencode-ai/llm/providers/amazon-bedrock`              | AWS Bedrock Converse API.                                                    |
-| Bedrock Mantle                | Missing                                                  | AWS Bedrock Mantle OpenAI-compatible APIs.                                   |
-| Azure OpenAI Chat             | `@opencode-ai/llm/providers/azure/chat`                  | Azure specialization of OpenAI Chat.                                         |
-| Azure OpenAI Responses        | `@opencode-ai/llm/providers/azure/responses`             | Azure specialization of OpenAI Responses.                                    |
+| API slice                     | Package-like entrypoint                                 | Purpose                                                                      |
+| ----------------------------- | ------------------------------------------------------- | ---------------------------------------------------------------------------- |
+| OpenAI Chat                   | `@opencode-ai/ai/providers/openai/chat`                 | OpenAI `/chat/completions` semantics.                                        |
+| OpenAI Responses              | `@opencode-ai/ai/providers/openai/responses`            | OpenAI `/responses` semantics with HTTP/WebSocket selected through settings. |
+| OpenAI-compatible Chat        | `@opencode-ai/ai/providers/openai-compatible`           | Generic OpenAI-compatible `/chat/completions`.                               |
+| OpenAI-compatible Responses   | `@opencode-ai/ai/providers/openai-compatible/responses` | Generic OpenAI-compatible `/responses`.                                      |
+| Anthropic-compatible Messages | `@opencode-ai/ai/providers/anthropic-compatible`        | Generic Anthropic-compatible `/messages`.                                    |
+| Anthropic Messages            | `@opencode-ai/ai/providers/anthropic`                   | Anthropic Messages API.                                                      |
+| Gemini Developer API          | `@opencode-ai/ai/providers/google`                      | Google AI Studio Gemini API.                                                 |
+| Vertex Gemini                 | `@opencode-ai/ai/providers/google-vertex/gemini`        | Vertex Gemini API; `providers/google-vertex` is the default alias.           |
+| Vertex Chat                   | `@opencode-ai/ai/providers/google-vertex/chat`          | Vertex OpenAI-compatible Chat Completions for MaaS models.                   |
+| Vertex Responses              | `@opencode-ai/ai/providers/google-vertex/responses`     | Vertex OpenAI-compatible Responses for Grok models.                          |
+| Vertex Messages               | `@opencode-ai/ai/providers/google-vertex/messages`      | Vertex-hosted Anthropic Messages API.                                        |
+| Bedrock Converse              | `@opencode-ai/ai/providers/amazon-bedrock`              | AWS Bedrock Converse API.                                                    |
+| Bedrock Mantle                | Missing                                                 | AWS Bedrock Mantle OpenAI-compatible APIs.                                   |
+| Azure OpenAI Chat             | `@opencode-ai/ai/providers/azure/chat`                  | Azure specialization of OpenAI Chat.                                         |
+| Azure OpenAI Responses        | `@opencode-ai/ai/providers/azure/responses`             | Azure specialization of OpenAI Responses.                                    |

 ## Suggested Next Work Slices

 1. Add native runner/catalog mappings for `@ai-sdk/azure`, `@ai-sdk/google`, and `@ai-sdk/amazon-bedrock` where the existing native facades are already close.
 2. Add API-aware runner/catalog selection between OpenAI-compatible Chat and Responses.
 3. Bring Bedrock native auth/config to AI SDK parity: region, profile, default AWS credential chain, bearer token env, endpoint override, and cross-region inference profile handling.
-4. Add runner/catalog mappings and recorded scenarios for the native Vertex Gemini and Vertex Anthropic entrypoints.
-5. Add native Vertex MaaS and Vertex xAI entrypoints by composing the compatible bases and shared Vertex auth/endpoint setup.
+4. Add runner/catalog mappings and recorded scenarios for the native Vertex Gemini, Chat, Responses, and Messages entrypoints.
+5. Decide Chat/Responses selection for `@ai-sdk/google-vertex/xai` catalog models.
 6. Add Bedrock Mantle as a separate OpenAI-compatible Bedrock namespace after deciding whether it uses Chat, Responses, or both by model.
 7. Expand typed provider options from the existing V1 lowerer knowledge in `packages/core/src/v1/config/provider-options.ts` before adding more raw overlay examples.
-8. Add recorded provider tests for Azure, Vertex Gemini, Vertex Anthropic, Bedrock credential-chain behavior, and Mantle before making native runtime the default for those packages.
+8. Add recorded provider tests for Azure, Vertex Gemini, Vertex Chat, Vertex Responses, Vertex Messages, Bedrock credential-chain behavior, and Mantle before making native runtime the default for those packages.
@@ -7,7 +7,7 @@ values directly.
 ## Conversation Summary

 Kit and Aidan want provider-specific LLM behavior to move out of opencode's AI
-SDK transform path and into `packages/llm` where possible. The goal is not a big
+SDK transform path and into `packages/ai` where possible. The goal is not a big
 generic transform layer; the goal is small composable route definitions backed by
 recorded golden tests.

@@ -355,22 +355,34 @@ The package-like OpenAI Responses entrypoint instead keeps transport scoped to
 Responses settings while preserving the same `model(...)` contract:

 ```ts
-import { model } from "@opencode-ai/llm/providers/openai/responses"
+import { model } from "@opencode-ai/ai/providers/openai/responses"

 model("gpt-4o", { apiKey, transport: "websocket" })
 ```

-Vertex keeps Gemini and Anthropic Messages as separate package-like entrypoints,
+Vertex keeps Gemini, Chat, Responses, and Messages as separate package-like entrypoints,
 while sharing project/location resolution and ADC authentication internally:

 ```ts
-import { model } from "@opencode-ai/llm/providers/google-vertex"
+import { model } from "@opencode-ai/ai/providers/google-vertex/gemini"

 model("gemini-3.5-flash", { project, location: "global" })
 ```

 ```ts
-import { model } from "@opencode-ai/llm/providers/google-vertex/anthropic"
+import { model } from "@opencode-ai/ai/providers/google-vertex/chat"
+
+model("deepseek-ai/deepseek-v3.2-maas", { project, location: "global" })
+```
+
+```ts
+import { model } from "@opencode-ai/ai/providers/google-vertex/responses"
+
+model("xai/grok-4.20-reasoning", { project, location: "global" })
+```
+
+```ts
+import { model } from "@opencode-ai/ai/providers/google-vertex/messages"

 model("claude-sonnet-4-6", { project, location: "global" })
 ```
@@ -1,12 +1,12 @@
 import { Config, Effect, Formatter, Layer, Schema, Stream } from "effect"
-import { LLM, LLMClient, Message, ProviderID, Tool, ToolRuntime } from "@opencode-ai/llm"
-import { Route, Auth, Endpoint, Framing, Protocol, RequestExecutor, WebSocketExecutor } from "@opencode-ai/llm/route"
-import { OpenAI } from "@opencode-ai/llm/providers"
+import { LLM, LLMClient, Message, ProviderID, Tool, ToolRuntime } from "@opencode-ai/ai"
+import { Route, Auth, Endpoint, Framing, Protocol, RequestExecutor, WebSocketExecutor } from "@opencode-ai/ai/route"
+import { OpenAI } from "@opencode-ai/ai/providers"

 /**
 * A runnable walkthrough of the LLM package use-site API.
 *
- * Run from `packages/llm` with an OpenAI key in the environment:
+ * Run from `packages/ai` with an OpenAI key in the environment:
 *
 *   OPENAI_API_KEY=... bun example/tutorial.ts
 *
@@ -0,0 +1,37 @@
+{
+  "$schema": "https://json.schemastore.org/package.json",
+  "version": "1.17.20",
+  "name": "@opencode-ai/ai",
+  "type": "module",
+  "license": "MIT",
+  "scripts": {
+    "setup:recording-env": "bun run script/setup-recording-env.ts",
+    "test": "bun test --timeout 30000 --only-failures",
+    "typecheck": "tsgo --noEmit && tsgo --noEmit -p tsconfig.types.json",
+    "build": "tsc -p tsconfig.build.json"
+  },
+  "files": [
+    "dist"
+  ],
+  "exports": {
+    ".": "./src/index.ts",
+    "./*": "./src/*.ts"
+  },
+  "devDependencies": {
+    "@clack/prompts": "1.0.0-alpha.1",
+    "@effect/platform-node": "catalog:",
+    "@opencode-ai/http-recorder": "workspace:*",
+    "@tsconfig/bun": "catalog:",
+    "@types/bun": "catalog:",
+    "@typescript/native-preview": "catalog:",
+    "typescript": "catalog:"
+  },
+  "dependencies": {
+    "@smithy/eventstream-codec": "4.2.14",
+    "@smithy/util-utf8": "4.2.2",
+    "@opencode-ai/schema": "workspace:*",
+    "aws4fetch": "1.0.20",
+    "effect": "catalog:",
+    "google-auth-library": "10.5.0"
+  }
+}
@@ -0,0 +1,38 @@
+#!/usr/bin/env bun
+import { Script } from "@opencode-ai/script"
+import { $ } from "bun"
+import { fileURLToPath } from "url"
+
+const dir = fileURLToPath(new URL("..", import.meta.url))
+process.chdir(dir)
+
+async function published(name: string, version: string) {
+  return (await $`npm view ${name}@${version} version`.nothrow()).exitCode === 0
+}
+
+await $`bun run build`
+const originalText = await Bun.file("package.json").text()
+const pkg = JSON.parse(originalText) as {
+  name: string
+  version: string
+  exports: Record<string, string>
+}
+if (await published(pkg.name, pkg.version)) {
+  console.log(`already published ${pkg.name}@${pkg.version}`)
+} else {
+  for (const [key, value] of Object.entries(pkg.exports)) {
+    const file = value.replace("./src/", "./dist/").replace(".ts", "")
+    // @ts-ignore
+    pkg.exports[key] = {
+      import: file + ".js",
+      types: file + ".d.ts",
+    }
+  }
+  await Bun.write("package.json", JSON.stringify(pkg, null, 2))
+  try {
+    await $`bun pm pack`
+    await $`npm publish *.tgz --tag ${Script.channel} --access public`
+  } finally {
+    await Bun.write("package.json", originalText)
+  }
+}
@@ -161,6 +161,18 @@ const PROVIDERS: ReadonlyArray<Provider> = [
    vars: [{ name: "TOGETHER_AI_API_KEY" }],
    validate: (env) => validateBearer("https://api.together.xyz/v1/models", Redacted.make(env.TOGETHER_AI_API_KEY)),
  },
+  {
+    id: "minimax",
+    label: "MiniMax",
+    tier: "compatible",
+    note: "Anthropic-compatible Messages text/tool recorded tests",
+    vars: [{ name: "MINIMAX_API_KEY" }],
+    validate: (env) =>
+      HttpClientRequest.get("https://api.minimax.io/anthropic/v1/models").pipe(
+        HttpClientRequest.setHeader("x-api-key", Redacted.value(Redacted.make(env.MINIMAX_API_KEY))),
+        executeRequest,
+      ),
+  },
  {
    id: "mistral",
    label: "Mistral",
@@ -0,0 +1,38 @@
+import { Context, Effect, Layer } from "effect"
+import { RequestExecutor } from "./route/executor"
+import type { ImageOptions, ImageRequest, ImageRequestFor, ImageResponse } from "./image"
+import type { LLMError } from "./schema"
+
+export type Execute = RequestExecutor.Interface["execute"]
+
+export interface Interface {
+  readonly generate: <Options extends ImageOptions>(
+    request: ImageRequestFor<Options>,
+  ) => Effect.Effect<ImageResponse, LLMError>
+}
+
+export class Service extends Context.Service<Service, Interface>()("@opencode/ImageClient") {}
+
+export const generate = <Options extends ImageOptions>(
+  request: ImageRequestFor<Options>,
+): Effect.Effect<ImageResponse, LLMError> =>
+  Effect.gen(function* () {
+    const client = yield* Service
+    return yield* client.generate(request)
+  }) as Effect.Effect<ImageResponse, LLMError>
+
+export const layer: Layer.Layer<Service, never, RequestExecutor.Service> = Layer.effect(
+  Service,
+  Effect.gen(function* () {
+    const executor = yield* RequestExecutor.Service
+    return Service.of({
+      generate: (request) => request.model.route.generate(request, executor.execute),
+    })
+  }),
+)
+
+export const ImageClient = {
+  Service,
+  layer,
+  generate,
+} as const
@@ -0,0 +1,166 @@
+import { Effect, Schema } from "effect"
+import { HttpOptions, InvalidRequestReason, LLMError, ModelID, ProviderID, ProviderMetadata, Usage } from "./schema"
+import { ImageClient, Service, type Execute as ImageExecute } from "./image-client"
+
+export interface ImageRoute<Options extends ImageOptions = ImageOptions> {
+  readonly id: string
+  readonly generate: (
+    request: ImageRequestFor<Options>,
+    execute: ImageExecute,
+  ) => Effect.Effect<ImageResponse, LLMError>
+}
+
+export type ImageOptions = Record<string, unknown>
+
+export class ImageModel<Options extends ImageOptions = ImageOptions> {
+  declare protected readonly _Options: (options: Options) => Options
+  readonly id: ModelID
+  readonly provider: ProviderID
+  readonly route: ImageRoute<Options>
+  readonly http?: HttpOptions
+
+  constructor(input: ImageModel.Input<Options>) {
+    this.id = input.id
+    this.provider = input.provider
+    this.route = input.route
+    this.http = input.http
+  }
+
+  static make<Options extends ImageOptions = ImageOptions>(input: ImageModel.MakeInput<Options>) {
+    return new ImageModel<Options>({
+      id: ModelID.make(input.id),
+      provider: ProviderID.make(input.provider),
+      route: input.route,
+      http: input.http,
+    })
+  }
+}
+
+export namespace ImageModel {
+  export interface Input<Options extends ImageOptions = ImageOptions> {
+    readonly id: ModelID
+    readonly provider: ProviderID
+    readonly route: ImageRoute<Options>
+    readonly http?: HttpOptions
+  }
+
+  export interface MakeInput<Options extends ImageOptions = ImageOptions>
+    extends Omit<Input<Options>, "id" | "provider"> {
+    readonly id: string | ModelID
+    readonly provider: string | ProviderID
+  }
+}
+
+export const ImageModelSchema = Schema.declare((value): value is ImageModel => value instanceof ImageModel, {
+  expected: "Image.Model",
+})
+
+const ImageBytesInput = Schema.Struct({
+  type: Schema.Literal("bytes"),
+  data: Schema.Uint8Array,
+  mediaType: Schema.String,
+})
+const ImageUrlInput = Schema.Struct({
+  type: Schema.Literal("url"),
+  url: Schema.String,
+})
+const ImageFileIDInput = Schema.Struct({
+  type: Schema.Literal("file-id"),
+  id: Schema.String,
+})
+const ImageFileURIInput = Schema.Struct({
+  type: Schema.Literal("file-uri"),
+  uri: Schema.String,
+  mediaType: Schema.String,
+})
+
+export const ImageInputSchema = Schema.Union([
+  ImageBytesInput,
+  ImageUrlInput,
+  ImageFileIDInput,
+  ImageFileURIInput,
+]).pipe(Schema.toTaggedUnion("type"))
+export type ImageInput = Schema.Schema.Type<typeof ImageInputSchema>
+
+export const ImageInput = {
+  bytes: (data: Uint8Array, mediaType: string): ImageInput => ({ type: "bytes", data, mediaType }),
+  url: (url: string): ImageInput => ({ type: "url", url }),
+  file: (id: string): ImageInput => ({ type: "file-id", id }),
+  fileUri: (uri: string, mediaType: string): ImageInput => ({ type: "file-uri", uri, mediaType }),
+} as const
+
+export class ImageRequest extends Schema.Class<ImageRequest>("Image.Request")({
+  model: ImageModelSchema,
+  prompt: Schema.String,
+  images: Schema.optional(Schema.Array(ImageInputSchema)),
+  options: Schema.optional(Schema.Record(Schema.String, Schema.Unknown)),
+  http: Schema.optional(HttpOptions),
+}) {
+  declare protected readonly _ImageRequest: void
+}
+
+export type ImageRequestFor<Options extends ImageOptions = ImageOptions> = Omit<ImageRequest, "model" | "options"> & {
+  readonly model: ImageModel<Options>
+  readonly options?: Options
+}
+
+export type ImageModelOptions<Model> = Model extends ImageModel<infer Options> ? Options : never
+
+export type ImageRequestInput<Model extends object = ImageModel> = Omit<
+  ConstructorParameters<typeof ImageRequest>[0],
+  "model" | "options" | "http"
+> & {
+  readonly model: Model
+  readonly options?: NoInfer<ImageModelOptions<Model>>
+  readonly http?: HttpOptions.Input
+} & (Model extends ImageModel<ImageModelOptions<Model>> ? unknown : never)
+
+export class GeneratedImage extends Schema.Class<GeneratedImage>("Image.Generated")({
+  mediaType: Schema.String,
+  data: Schema.Union([Schema.String, Schema.Uint8Array]),
+  providerMetadata: Schema.optional(ProviderMetadata),
+}) {}
+
+export class ImageResponse extends Schema.Class<ImageResponse>("Image.Response")({
+  images: Schema.Array(GeneratedImage),
+  usage: Schema.optional(Usage),
+  providerMetadata: Schema.optional(ProviderMetadata),
+}) {
+  get image() {
+    return this.images[0]
+  }
+}
+
+export function request<const Model extends object>(
+  input: ImageRequestInput<Model>,
+): ImageRequestFor<ImageModelOptions<Model>>
+export function request(input: ImageRequest): ImageRequest
+export function request(input: ImageRequest | ImageRequestInput) {
+  if (input instanceof ImageRequest) return input
+  return new ImageRequest({
+    ...input,
+    model: input.model as unknown as ImageModel,
+    http: input.http === undefined ? undefined : HttpOptions.make(input.http),
+  })
+}
+
+export function generate<const Model extends object>(
+  input: ImageRequestInput<Model>,
+): Effect.Effect<ImageResponse, LLMError, Service>
+export function generate(input: ImageRequest): Effect.Effect<ImageResponse, LLMError, Service>
+export function generate(input: ImageRequest | ImageRequestInput) {
+  return Effect.try({
+    try: () => (input instanceof ImageRequest ? input : request(input)),
+    catch: (error) =>
+      new LLMError({
+        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>)))
+}
+
+export const Image = {
+  request,
+  generate,
+} as const
@@ -1,4 +1,5 @@
 export { LLMClient } from "./route/client"
+export { ImageClient } from "./image-client"
 export { Auth } from "./route/auth"
 export { Provider } from "./provider"
 export { ProviderPackage } from "./provider-package"
@@ -10,6 +11,9 @@ export type {
  Service as LLMClientService,
 } from "./route/client"
 export * from "./schema"
+export { GeneratedImage, ImageInput, ImageInputSchema, ImageModel, ImageRequest, ImageResponse } from "./image"
+export type { ImageModelOptions, ImageOptions, ImageRequestFor, ImageRequestInput, ImageRoute } from "./image"
+export { Image } from "./image"
 export { Tool, ToolFailure, toDefinitions } from "./tool"
 export { ToolRuntime } from "./tool-runtime"
 export type { DispatchResult as ToolDispatchResult, ToolSettlement } from "./tool-runtime"
@@ -18,7 +22,7 @@ export type {
  AnyTool,
  ExecutableTool,
  ExecutableTools,
-  Tool as ToolShape,
+  Definition as ToolShape,
  ToolExecute,
  ToolExecuteContext,
  ToolModelOutputInput,
@@ -0,0 +1 @@
+export * from "./protocols/index"
@@ -27,6 +27,7 @@ import { ToolSchemaProjection } from "./utils/tool-schema"
 import { ToolStream } from "./utils/tool-stream"

 const ADAPTER = "anthropic-messages"
+const MEDIA_MIMES = new Set<string>([...ProviderShared.IMAGE_MIMES, ...ProviderShared.PDF_MIMES])
 export const DEFAULT_BASE_URL = "https://api.anthropic.com/v1"
 export const PATH = "/messages"

@@ -56,6 +57,17 @@ const AnthropicImageBlock = Schema.Struct({
 })
 type AnthropicImageBlock = Schema.Schema.Type<typeof AnthropicImageBlock>

+const AnthropicDocumentBlock = Schema.Struct({
+  type: Schema.tag("document"),
+  source: Schema.Struct({
+    type: Schema.tag("base64"),
+    media_type: Schema.Literal("application/pdf"),
+    data: Schema.String,
+  }),
+  cache_control: Schema.optional(AnthropicCacheControl),
+})
+type AnthropicDocumentBlock = Schema.Schema.Type<typeof AnthropicDocumentBlock>
+
 const AnthropicThinkingBlock = Schema.Struct({
  type: Schema.tag("thinking"),
  thinking: Schema.String,
@@ -101,13 +113,10 @@ const AnthropicServerToolResultBlock = Schema.Struct({
 })
 type AnthropicServerToolResultBlock = Schema.Schema.Type<typeof AnthropicServerToolResultBlock>

-// Anthropic accepts either a plain string or an ordered array of text/image
-// blocks inside `tool_result.content`. The array form is required when a tool
-// returns image bytes (screenshot, image search, etc.) so they can be passed
-// to the model as proper image inputs instead of being JSON-stringified into
-// the prompt — which silently inflates context by megabytes and can push the
-// conversation over the model's token limit.
-const AnthropicToolResultContent = Schema.Union([AnthropicTextBlock, AnthropicImageBlock])
+// Anthropic accepts either a plain string or an ordered array of text, image, and
+// document blocks inside `tool_result.content`. The array form keeps media as native
+// model input instead of JSON-stringifying base64 into prompt text.
+const AnthropicToolResultContent = Schema.Union([AnthropicTextBlock, AnthropicImageBlock, AnthropicDocumentBlock])

 const AnthropicToolResultBlock = Schema.Struct({
  type: Schema.tag("tool_result"),
@@ -117,7 +126,12 @@ const AnthropicToolResultBlock = Schema.Struct({
  cache_control: Schema.optional(AnthropicCacheControl),
 })

-const AnthropicUserBlock = Schema.Union([AnthropicTextBlock, AnthropicImageBlock, AnthropicToolResultBlock])
+const AnthropicUserBlock = Schema.Union([
+  AnthropicTextBlock,
+  AnthropicImageBlock,
+  AnthropicDocumentBlock,
+  AnthropicToolResultBlock,
+])
 type AnthropicUserBlock = Schema.Schema.Type<typeof AnthropicUserBlock>
 const AnthropicAssistantBlock = Schema.Union([
  AnthropicTextBlock,
@@ -319,12 +333,17 @@ const lowerServerToolResult = Effect.fn("AnthropicMessages.lowerServerToolResult
  return { type: wireType, tool_use_id: part.id, content: part.result.value } satisfies AnthropicServerToolResultBlock
 })

-const lowerImage = Effect.fn("AnthropicMessages.lowerImage")(function* (part: MediaPart) {
-  const media = yield* ProviderShared.validateMedia(
-    "Anthropic Messages",
-    part,
-    new Set<string>(ProviderShared.IMAGE_MIMES),
-  )
+const lowerMedia = Effect.fn("AnthropicMessages.lowerMedia")(function* (part: MediaPart) {
+  const media = yield* ProviderShared.validateMedia("Anthropic Messages", part, MEDIA_MIMES)
+  if (media.mime === "application/pdf")
+    return {
+      type: "document" as const,
+      source: {
+        type: "base64" as const,
+        media_type: "application/pdf" as const,
+        data: media.base64,
+      },
+    } satisfies AnthropicDocumentBlock
  return {
    type: "image" as const,
    source: {
@@ -335,25 +354,13 @@ const lowerImage = Effect.fn("AnthropicMessages.lowerImage")(function* (part: Me
  } satisfies AnthropicImageBlock
 })

-// Tool results may carry structured text/images. Keep media as provider-native
+// Tool results may carry structured text, images, and documents. Keep media as provider-native
 // content instead of JSON-stringifying base64 into a prompt string.
 const lowerToolResultContentItem = Effect.fn("AnthropicMessages.lowerToolResultContentItem")(function* (
  item: ToolContent,
 ) {
  if (item.type === "text") return { type: "text" as const, text: item.text } satisfies AnthropicTextBlock
-  const media = yield* ProviderShared.validateToolFile(
-    "Anthropic Messages",
-    item,
-    new Set<string>(ProviderShared.IMAGE_MIMES),
-  )
-  return {
-    type: "image" as const,
-    source: {
-      type: "base64" as const,
-      media_type: media.mime,
-      data: media.base64,
-    },
-  } satisfies AnthropicImageBlock
+  return yield* lowerMedia({ type: "media", mediaType: item.mime, data: item.uri, filename: item.name })
 })

 const lowerToolResultContent = Effect.fn("AnthropicMessages.lowerToolResultContent")(function* (part: ToolResultPart) {
@@ -445,7 +452,7 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
          continue
        }
        if (part.type === "media") {
-          content.push(yield* lowerImage(part))
+          content.push(yield* lowerMedia(part))
          continue
        }
        return yield* ProviderShared.unsupportedContent("Anthropic Messages", "user", ["text", "media"])
@@ -703,7 +710,14 @@ const onContentBlockStart = (state: ParserState, event: AnthropicEvent): StepRes
          providerExecuted: block.type === "server_tool_use",
        }),
      },
-      [...events, LLMEvent.toolInputStart({ id: block.id ?? String(event.index), name: block.name ?? "" })],
+      [
+        ...events,
+        LLMEvent.toolInputStart({
+          id: block.id ?? String(event.index),
+          name: block.name ?? "",
+          providerExecuted: block.type === "server_tool_use" ? true : undefined,
+        }),
+      ],
    ]
  }

@@ -52,6 +52,7 @@ const BedrockToolResultContentItem = Schema.Union([
  Schema.Struct({ text: Schema.String }),
  Schema.Struct({ json: Schema.Unknown }),
  BedrockMedia.ImageBlock,
+  BedrockMedia.DocumentBlock,
 ])

 const BedrockToolResultBlock = Schema.Struct({
@@ -283,8 +284,6 @@ const lowerToolResultContent = Effect.fn("BedrockConverse.lowerToolResultContent
      data: item.uri,
      filename: item.name,
    })
-    if (!("image" in media))
-      return yield* ProviderShared.invalidRequest("Bedrock Converse only supports image media in tool results")
    content.push(media)
  }
  return content
@@ -437,21 +436,22 @@ const mapFinishReason = (reason: string): FinishReason => {
  return "unknown"
 }

-// AWS Bedrock Converse reports `inputTokens` (inclusive total) with
-// `cacheReadInputTokens` and `cacheWriteInputTokens` as subsets. Pass
-// the total through and derive the non-cached breakdown. Bedrock does
-// not break reasoning out of `outputTokens` for any current model.
+// 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): Usage | undefined => {
  if (!usage) return undefined
-  const cacheTotal = (usage.cacheReadInputTokens ?? 0) + (usage.cacheWriteInputTokens ?? 0)
-  const nonCached = ProviderShared.subtractTokens(usage.inputTokens, cacheTotal)
+  const inputTokens = ProviderShared.sumTokens(
+    usage.inputTokens,
+    usage.cacheReadInputTokens,
+    usage.cacheWriteInputTokens,
+  )
  return new Usage({
-    inputTokens: usage.inputTokens,
+    inputTokens,
    outputTokens: usage.outputTokens,
-    nonCachedInputTokens: nonCached,
+    nonCachedInputTokens: usage.inputTokens,
    cacheReadInputTokens: usage.cacheReadInputTokens,
    cacheWriteInputTokens: usage.cacheWriteInputTokens,
-    totalTokens: ProviderShared.totalTokens(usage.inputTokens, usage.outputTokens, usage.totalTokens),
+    totalTokens: ProviderShared.totalTokens(inputTokens, usage.outputTokens, usage.totalTokens),
    providerMetadata: { bedrock: usage },
  })
 }
@@ -561,7 +561,9 @@ const step = (state: ParserState, event: BedrockEvent) =>
      return [
        {
          ...state,
-          hasToolCalls: resultEvents.some(LLMEvent.is.toolCall) ? true : state.hasToolCalls,
+          hasToolCalls:
+            resultEvents.some((event) => LLMEvent.is.toolCall(event) || LLMEvent.is.toolInputError(event)) ||
+            state.hasToolCalls,
          lifecycle,
          tools: result.tools,
          reasoningSignatures: Object.fromEntries(
@@ -1,7 +1,7 @@
 import { EventStreamCodec } from "@smithy/eventstream-codec"
 import { fromUtf8, toUtf8 } from "@smithy/util-utf8"
 import { Effect, Stream } from "effect"
-import type { Framing } from "../route/framing"
+import { Framing } from "../route/framing"
 import { ProviderShared } from "./shared"

 // Bedrock streams responses using the AWS event stream binary protocol — each
@@ -79,7 +79,7 @@ const consumeFrames = (route: string) => (state: FrameBufferState, chunk: Uint8A
 * under its `:event-type` header so the chunk schema can match the JSON
 * payload directly.
 */
-export const framing = (route: string): Framing<object> => ({
+export const framing = (route: string): Framing.Definition<object> => ({
  id: "aws-event-stream",
  frame: (bytes) => bytes.pipe(Stream.mapAccumEffect(() => initialFrameBuffer, consumeFrames(route))),
 })
@@ -41,9 +41,11 @@ const GeminiInlineDataPart = Schema.Struct({
    data: Schema.String,
  }),
 })
+type GeminiInlineDataPart = Schema.Schema.Type<typeof GeminiInlineDataPart>

 const GeminiFunctionCallPart = Schema.Struct({
  functionCall: Schema.Struct({
+    id: Schema.optional(Schema.String),
    name: Schema.String,
    args: Schema.Unknown,
  }),
@@ -52,8 +54,10 @@ const GeminiFunctionCallPart = Schema.Struct({

 const GeminiFunctionResponsePart = Schema.Struct({
  functionResponse: Schema.Struct({
+    id: Schema.optional(Schema.String),
    name: Schema.String,
    response: Schema.Unknown,
+    parts: Schema.optional(Schema.Array(GeminiInlineDataPart)),
  }),
 })

@@ -197,8 +201,13 @@ const thoughtSignature = (providerMetadata: ProviderMetadata | undefined) => {
    : undefined
 }

+const functionCallId = (providerMetadata: ProviderMetadata | undefined) => {
+  const google = providerMetadata?.google
+  return ProviderShared.isRecord(google) && typeof google.functionCallId === "string" ? google.functionCallId : undefined
+}
+
 const lowerToolCall = (part: ToolCallPart) => ({
-  functionCall: { name: part.name, args: part.input },
+  functionCall: { id: functionCallId(part.providerMetadata), name: part.name, args: part.input },
  thoughtSignature: thoughtSignature(part.providerMetadata),
 })

@@ -255,6 +264,7 @@ const lowerMessages = Effect.fn("Gemini.lowerMessages")(function* (request: LLMR
      if (part.result.type !== "content") {
        parts.push({
          functionResponse: {
+            id: functionCallId(part.providerMetadata),
            name: part.name,
            response: {
              name: part.name,
@@ -266,20 +276,23 @@ const lowerMessages = Effect.fn("Gemini.lowerMessages")(function* (request: LLMR
      }
      const content: ReadonlyArray<ToolContent> = part.result.value
      const text = content.filter((item) => item.type === "text").map((item) => item.text)
+      const media: GeminiInlineDataPart[] = []
+      for (const item of content) {
+        if (item.type === "text") continue
+        const value = yield* ProviderShared.validateToolFile("Gemini", item, MEDIA_MIMES)
+        media.push({ inlineData: { mimeType: value.mime, data: value.base64 } })
+      }
      parts.push({
        functionResponse: {
+          id: functionCallId(part.providerMetadata),
          name: part.name,
          response: {
            name: part.name,
            content: text.join("\n"),
          },
+          parts: media.length > 0 ? media : undefined,
        },
      })
-      for (const item of content) {
-        if (item.type === "text") continue
-        const media = yield* ProviderShared.validateToolFile("Gemini", item, MEDIA_MIMES)
-        parts.push({ inlineData: { mimeType: media.mime, data: media.base64 } })
-      }
    }
    contents.push({ role: "user", parts })
  }
@@ -441,6 +454,10 @@ const step = (state: ParserState, event: GeminiEvent) => {
    if ("functionCall" in part) {
      const input = part.functionCall.args
      const id = `tool_${nextToolCallId++}`
+      const metadata = {
+        ...(part.functionCall.id === undefined ? {} : { functionCallId: part.functionCall.id }),
+        ...(part.thoughtSignature === undefined ? {} : { thoughtSignature: part.thoughtSignature }),
+      }
      lifecycle = Lifecycle.reasoningEnd(
        lifecycle,
        events,
@@ -453,9 +470,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
          id,
          name: part.functionCall.name,
          input,
-          providerMetadata: part.thoughtSignature
-            ? googleMetadata({ thoughtSignature: part.thoughtSignature })
-            : undefined,
+          providerMetadata: Object.keys(metadata).length > 0 ? googleMetadata(metadata) : undefined,
        }),
      )
      hasToolCalls = true
@@ -0,0 +1,314 @@
+import { Effect, Encoding, Schema } from "effect"
+import { Headers, HttpClientRequest } from "effect/unstable/http"
+import {
+  GeneratedImage,
+  ImageModel,
+  ImageResponse,
+  type ImageInput,
+  type ImageRequestFor,
+  type ImageRoute,
+} from "../image"
+import { Auth, type Definition as AuthDefinition } from "../route/auth"
+import {
+  InvalidProviderOutputReason,
+  LLMError,
+  Usage,
+  mergeHttpOptions,
+  mergeJsonRecords,
+  type HttpOptions,
+  type ProviderMetadata,
+} from "../schema"
+import { ProviderShared } from "./shared"
+import { ImageInputs } from "./utils/image-input"
+
+const ADAPTER = "google-images"
+export const DEFAULT_BASE_URL = "https://generativelanguage.googleapis.com/v1beta"
+
+export type GoogleImageString<Known extends string> = Known | (string & {})
+
+export type GoogleImageOptions = {
+  readonly aspectRatio?: GoogleImageString<
+    "1:1" | "2:3" | "3:2" | "3:4" | "4:3" | "4:5" | "5:4" | "9:16" | "16:9" | "21:9"
+  >
+  readonly imageSize?: GoogleImageString<"1K" | "2K" | "4K">
+  readonly seed?: number
+  readonly thinkingLevel?: GoogleImageString<"MINIMAL" | "LOW" | "MEDIUM" | "HIGH">
+  readonly includeThoughts?: boolean
+} & Record<string, unknown>
+
+export type GoogleImageBody = Record<string, unknown> & {
+  readonly contents: ReadonlyArray<{
+    readonly role: "user"
+    readonly parts: ReadonlyArray<Record<string, unknown>>
+  }>
+  readonly generationConfig: Record<string, unknown>
+}
+
+const GoogleUsage = Schema.StructWithRest(
+  Schema.Struct({
+    cachedContentTokenCount: Schema.optional(Schema.Number),
+    thoughtsTokenCount: Schema.optional(Schema.Number),
+    promptTokenCount: Schema.optional(Schema.Number),
+    candidatesTokenCount: Schema.optional(Schema.Number),
+    totalTokenCount: Schema.optional(Schema.Number),
+    promptTokensDetails: Schema.optional(Schema.Unknown),
+    candidatesTokensDetails: Schema.optional(Schema.Unknown),
+  }),
+  [Schema.Record(Schema.String, Schema.Unknown)],
+)
+
+const GoogleImageResponse = Schema.Struct({
+  candidates: Schema.optional(
+    Schema.Array(
+      Schema.Struct({
+        index: Schema.optional(Schema.Number),
+        content: Schema.optional(
+          Schema.Struct({
+            parts: Schema.Array(
+              Schema.Struct({
+                text: Schema.optional(Schema.String),
+                thought: Schema.optional(Schema.Boolean),
+                thoughtSignature: Schema.optional(Schema.String),
+                inlineData: Schema.optional(
+                  Schema.Struct({
+                    mimeType: Schema.String,
+                    data: Schema.String,
+                  }),
+                ),
+              }),
+            ),
+          }),
+        ),
+        finishReason: Schema.optional(Schema.String),
+        finishMessage: Schema.optional(Schema.String),
+        safetyRatings: Schema.optional(Schema.Unknown),
+        citationMetadata: Schema.optional(Schema.Unknown),
+        groundingMetadata: Schema.optional(Schema.Unknown),
+      }),
+    ),
+  ),
+  usageMetadata: Schema.optional(GoogleUsage),
+  modelVersion: Schema.optional(Schema.String),
+  responseId: Schema.optional(Schema.String),
+  promptFeedback: Schema.optional(Schema.Unknown),
+})
+
+export interface ModelInput {
+  readonly id: string
+  readonly auth: AuthDefinition
+  readonly baseURL?: string
+  readonly headers?: Record<string, string>
+  readonly http?: HttpOptions
+}
+
+const nativeOptions = (options: GoogleImageOptions | undefined) => {
+  const { aspectRatio, imageSize, seed, thinkingLevel, includeThoughts, ...native } = options ?? {}
+  const image = {
+    aspectRatio,
+    imageSize,
+  }
+  const thinkingConfig = {
+    thinkingLevel,
+    includeThoughts,
+  }
+  return (
+    mergeJsonRecords(
+      {
+        responseModalities: ["IMAGE"],
+        imageConfig: Object.values(image).some((value) => value !== undefined) ? image : undefined,
+        seed,
+        thinkingConfig: Object.values(thinkingConfig).some((value) => value !== undefined) ? thinkingConfig : undefined,
+      },
+      native,
+    ) ?? { responseModalities: ["IMAGE"] }
+  )
+}
+
+const invalidOutput = (message: string, providerMetadata?: ProviderMetadata) =>
+  new LLMError({
+    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)
+  Object.entries(query).forEach(([key, value]) => next.searchParams.set(key, value))
+  return next.toString()
+}
+
+export const model = (input: ModelInput) => {
+  const route: ImageRoute<GoogleImageOptions> = {
+    id: ADAPTER,
+    generate: Effect.fn("GoogleImages.generate")(function* (request: ImageRequestFor<GoogleImageOptions>, execute) {
+      const imageParts = yield* Effect.forEach(request.images ?? [], googleImagePart)
+      const http = mergeHttpOptions(request.model.http, request.http)
+      const requestBody = mergeJsonRecords(
+        {
+          contents: [{ role: "user", parts: [{ text: request.prompt }, ...imageParts] }],
+          generationConfig: nativeOptions(request.options),
+        },
+        http?.body,
+      ) as GoogleImageBody
+      const text = ProviderShared.encodeJson(requestBody)
+      const url = applyQuery(
+        `${(input.baseURL ?? DEFAULT_BASE_URL).replace(/\/$/, "")}/models/${request.model.id}:generateContent`,
+        http?.query,
+      )
+      const headers = yield* Auth.toEffect(input.auth)({
+        request,
+        method: "POST",
+        url,
+        body: text,
+        headers: Headers.fromInput({ ...input.headers, ...http?.headers }),
+      })
+      const response = yield* execute(
+        HttpClientRequest.post(url).pipe(
+          HttpClientRequest.setHeaders(headers),
+          HttpClientRequest.bodyText(text, "application/json"),
+        ),
+      )
+      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) => ({
+        index: candidate.index ?? candidateIndex,
+        finishReason: candidate.finishReason,
+        finishMessage: candidate.finishMessage,
+        safetyRatings: candidate.safetyRatings,
+        citationMetadata: candidate.citationMetadata,
+        groundingMetadata: candidate.groundingMetadata,
+        parts: (candidate.content?.parts ?? []).map((part) =>
+          part.inlineData === undefined
+            ? {
+                type: "text",
+                text: part.text,
+                thought: part.thought,
+                thoughtSignature: part.thoughtSignature,
+              }
+            : {
+                type: "inlineData",
+                mediaType: part.inlineData.mimeType,
+                thought: part.thought,
+                thoughtSignature: part.thoughtSignature,
+              },
+        ),
+      }))
+      const encoded = candidates.flatMap((candidate, candidateIndex) =>
+        (candidate.content?.parts ?? []).flatMap((part, partIndex) =>
+          part.inlineData === undefined || part.thought === true
+            ? []
+            : [{ candidate, candidateIndex, partIndex, inlineData: part.inlineData }],
+        ),
+      )
+      const images = yield* Effect.forEach(encoded, (item) =>
+        Effect.fromResult(Encoding.decodeBase64(item.inlineData.data)).pipe(
+          Effect.mapError(() =>
+            invalidOutput(
+              `Google Images candidate ${item.candidateIndex} part ${item.partIndex} contains invalid base64 data`,
+            ),
+          ),
+          Effect.map(
+            (data) =>
+              new GeneratedImage({
+                mediaType: item.inlineData.mimeType,
+                data,
+                providerMetadata: {
+                  google: {
+                    candidateIndex: item.candidate.index ?? item.candidateIndex,
+                    partIndex: item.partIndex,
+                    finishReason: item.candidate.finishReason,
+                    safetyRatings: item.candidate.safetyRatings,
+                    citationMetadata: item.candidate.citationMetadata,
+                    groundingMetadata: item.candidate.groundingMetadata,
+                    thoughtSignature: item.candidate.content?.parts[item.partIndex]?.thoughtSignature,
+                  },
+                },
+              }),
+          ),
+        ),
+      )
+      if (images.length === 0) {
+        const finishReasons = candidates.flatMap((candidate) =>
+          candidate.finishReason === undefined ? [] : [candidate.finishReason],
+        )
+        return yield* invalidOutput(
+          `Google Images returned no final images${
+            finishReasons.length === 0 ? "" : ` (finish reasons: ${finishReasons.join(", ")})`
+          }; inspect reason.providerMetadata.google for prompt feedback and candidate details`,
+          {
+            google: {
+              promptFeedback: decoded.promptFeedback,
+              candidates: candidateMetadata,
+            },
+          },
+        )
+      }
+      const usage = decoded.usageMetadata
+      const outputTokens =
+        usage?.candidatesTokenCount === undefined
+          ? undefined
+          : usage.candidatesTokenCount + (usage.thoughtsTokenCount ?? 0)
+      return new ImageResponse({
+        images,
+        usage:
+          usage === undefined
+            ? undefined
+            : new Usage({
+                inputTokens: usage.promptTokenCount,
+                outputTokens,
+                nonCachedInputTokens: ProviderShared.subtractTokens(
+                  usage.promptTokenCount,
+                  usage.cachedContentTokenCount,
+                ),
+                cacheReadInputTokens: usage.cachedContentTokenCount,
+                reasoningTokens: usage.thoughtsTokenCount,
+                totalTokens: ProviderShared.totalTokens(usage.promptTokenCount, outputTokens, usage.totalTokenCount),
+                providerMetadata: { google: usage },
+              }),
+        providerMetadata: {
+          google: {
+            modelVersion: decoded.modelVersion,
+            responseId: decoded.responseId,
+            promptFeedback: decoded.promptFeedback,
+            candidates: candidateMetadata,
+          },
+        },
+      })
+    }),
+  }
+  return ImageModel.make<GoogleImageOptions>({ id: input.id, provider: "google", route, http: input.http })
+}
+
+const googleImagePart = (image: ImageInput): Effect.Effect<Record<string, unknown>, LLMError> => {
+  if (image.type === "bytes")
+    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, 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",
+            ),
+          )
+        return Effect.succeed({
+          inlineData: { mimeType: decoded.mediaType, data: Encoding.encodeBase64(decoded.data) },
+        })
+      }),
+    )
+  return Effect.fail(
+    ImageInputs.invalid(ADAPTER, "Google generateContent requires Gemini file URIs rather than provider file IDs"),
+  )
+}
+
+export const GoogleImages = {
+  model,
+} as const
@@ -1,9 +1,8 @@
 export * as AnthropicMessages from "./anthropic-messages"
-export * as GoogleVertexAnthropic from "./google-vertex-anthropic"
-export * as GoogleVertexGemini from "./google-vertex-gemini"
 export * as BedrockConverse from "./bedrock-converse"
 export * as Gemini from "./gemini"
 export * as OpenAIChat from "./openai-chat"
+export * as OpenAIImages from "./openai-images"
 export * as OpenAICompatibleChat from "./openai-compatible-chat"
 export * as OpenAICompatibleResponses from "./openai-compatible-responses"
 export * as OpenAIResponses from "./openai-responses"
@@ -25,6 +25,7 @@ import { ToolStream } from "./utils/tool-stream"

 const ADAPTER = "openai-chat"
 const IMAGE_MIMES = new Set<string>(ProviderShared.IMAGE_MIMES)
+const RESERVED_REASONING_FIELDS = new Set(["role", "content", "tool_calls"])
 export const DEFAULT_BASE_URL = "https://api.openai.com/v1"
 export const PATH = "/chat/completions"

@@ -70,12 +71,18 @@ const OpenAIChatMessage = Schema.Union([
    role: Schema.Literal("user"),
    content: Schema.Union([Schema.String, Schema.Array(OpenAIChatUserContent)]),
  }),
-  Schema.Struct({
-    role: Schema.Literal("assistant"),
-    content: Schema.NullOr(Schema.String),
-    tool_calls: optionalArray(OpenAIChatAssistantToolCall),
-    reasoning_content: Schema.optional(Schema.String),
-  }),
+  Schema.StructWithRest(
+    Schema.Struct({
+      role: Schema.Literal("assistant"),
+      content: Schema.NullOr(Schema.String),
+      tool_calls: optionalArray(OpenAIChatAssistantToolCall),
+      reasoning_content: Schema.optional(Schema.String),
+      reasoning: Schema.optional(Schema.String),
+      reasoning_text: Schema.optional(Schema.String),
+      reasoning_details: Schema.optional(Schema.Unknown),
+    }),
+    [Schema.Record(Schema.String, Schema.Unknown)],
+  ),
  Schema.Struct({ role: Schema.Literal("tool"), tool_call_id: Schema.String, content: Schema.String }),
 ]).pipe(Schema.toTaggedUnion("role"))
 type OpenAIChatMessage = Schema.Schema.Type<typeof OpenAIChatMessage>
@@ -142,11 +149,17 @@ const OpenAIChatToolCallDelta = Schema.Struct({
 })
 type OpenAIChatToolCallDelta = Schema.Schema.Type<typeof OpenAIChatToolCallDelta>

-const OpenAIChatDelta = Schema.Struct({
-  content: optionalNull(Schema.String),
-  reasoning_content: optionalNull(Schema.String),
-  tool_calls: optionalNull(Schema.Array(OpenAIChatToolCallDelta)),
-})
+const OpenAIChatDelta = Schema.StructWithRest(
+  Schema.Struct({
+    content: optionalNull(Schema.String),
+    reasoning_content: optionalNull(Schema.String),
+    reasoning: optionalNull(Schema.String),
+    reasoning_text: optionalNull(Schema.String),
+    reasoning_details: optionalNull(Schema.Unknown),
+    tool_calls: optionalNull(Schema.Array(OpenAIChatToolCallDelta)),
+  }),
+  [Schema.Record(Schema.String, Schema.Unknown)],
+)

 const OpenAIChatChoice = Schema.Struct({
  delta: optionalNull(OpenAIChatDelta),
@@ -160,12 +173,23 @@ export const OpenAIChatEvent = Schema.Struct({
 export type OpenAIChatEvent = Schema.Schema.Type<typeof OpenAIChatEvent>
 type OpenAIChatRequestMessage = LLMRequest["messages"][number]

-interface ParserState {
+interface PendingToolDelta {
+  readonly id?: string
+  readonly name?: string
+  readonly input: string
+}
+
+export interface ParserState {
  readonly tools: ToolStream.State<number>
+  readonly pendingTools: Partial<Record<number, PendingToolDelta>>
  readonly toolCallEvents: ReadonlyArray<LLMEvent>
  readonly usage?: Usage
  readonly finishReason?: FinishReason
  readonly lifecycle: Lifecycle.State
+  readonly reasoningField?: string
+  readonly reasoningDetails: Array<unknown>
+  readonly reasoningDetailsObserved: boolean
+  readonly reasoningEmitted: boolean
 }

 // =============================================================================
@@ -208,6 +232,20 @@ 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) => {
+  const field = part.providerMetadata?.openai?.reasoningField
+  return typeof field === "string" ? field : undefined
+}
+
+const reasoningDetails = (parts: ReadonlyArray<ReasoningPart>, native: unknown) => {
+  const observed = parts.flatMap((part) => {
+    const details = part.providerMetadata?.openai?.reasoningDetails
+    return Array.isArray(details) ? details : []
+  })
+  if (parts.some((part) => Array.isArray(part.providerMetadata?.openai?.reasoningDetails))) return observed
+  if (isRecord(native) && Array.isArray(native.reasoning_details)) return native.reasoning_details
+}
+
 const lowerUserMessage = Effect.fn("OpenAIChat.lowerUserMessage")(function* (message: OpenAIChatRequestMessage) {
  const content: Array<Schema.Schema.Type<typeof OpenAIChatUserContent>> = []
  for (const part of message.content) {
@@ -228,6 +266,7 @@ const lowerUserMessage = Effect.fn("OpenAIChat.lowerUserMessage")(function* (mes

 const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(function* (
  message: OpenAIChatRequestMessage,
+  configuredField?: string,
 ) {
  const content: TextPart[] = []
  const reasoning: ReasoningPart[] = []
@@ -248,15 +287,31 @@ const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(func
      continue
    }
  }
-  return {
+  const text = reasoning.map((part) => part.text).join("")
+  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?.openai?.reasoningDetails))
+  const field = (() => {
+    if (configuredField !== undefined) return configuredField
+    if (reasoning.length === 0) return undefined
+    if (observedField !== undefined) return observedField
+    if (nativeReasoning !== undefined) return "reasoning_content"
+    if (!fullyStructured) return "reasoning_content"
+  })()
+  const reasoningText = (() => {
+    if (configuredField !== undefined) return reasoning.length === 0 ? (nativeReasoning ?? "") : text
+    if (reasoning.length === 0) return nativeReasoning
+    return text
+  })()
+  const result = {
    role: "assistant" as const,
    content: content.length === 0 ? null : ProviderShared.joinText(content),
    tool_calls: toolCalls.length === 0 ? undefined : toolCalls,
-    reasoning_content:
-      reasoning.length > 0
-        ? reasoning.map((part) => part.text).join("")
-        : openAICompatibleReasoningContent(message.native?.openaiCompatible),
+    reasoning_details: details,
  }
+  if (field === undefined || reasoningText === undefined) return result
+  return { ...result, [field]: reasoningText }
 })

 const lowerToolMessages = Effect.fn("OpenAIChat.lowerToolMessages")(function* (message: OpenAIChatRequestMessage) {
@@ -282,9 +337,12 @@ const lowerToolMessages = Effect.fn("OpenAIChat.lowerToolMessages")(function* (m
  return { messages, images }
 })

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

@@ -322,7 +380,7 @@ const lowerMessages = Effect.fn("OpenAIChat.lowerMessages")(function* (request:
      continue
    }
    flushImages()
-    messages.push(...(yield* lowerMessage(message)))
+    messages.push(...(yield* lowerMessage(message, request.model.compatibility?.reasoningField)))
  }
  flushImages()
  return messages
@@ -340,6 +398,11 @@ const lowerOptions = Effect.fn("OpenAIChat.lowerOptions")(function* (request: LL
 const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (request: LLMRequest) {
  // `fromRequest` returns the provider body only. Endpoint, auth, framing,
  // validation, and HTTP execution are composed by `Route.make`.
+  const reasoningField = request.model.compatibility?.reasoningField
+  if (reasoningField && RESERVED_REASONING_FIELDS.has(reasoningField))
+    return yield* ProviderShared.invalidRequest(
+      `OpenAI Chat reasoning field conflicts with reserved field ${reasoningField}`,
+    )
  const generation = request.generation
  const toolSchemaCompatibility = request.model.compatibility?.toolSchema
  return {
@@ -400,6 +463,73 @@ const mapUsage = (usage: OpenAIChatEvent["usage"]): Usage | undefined => {
  })
 }

+const reasoningDelta = (
+  delta: Schema.Schema.Type<typeof OpenAIChatDelta> | null | undefined,
+  configuredField?: string,
+) => {
+  if (!delta) return undefined
+  const fields = new Set([configuredField, "reasoning_content", "reasoning", "reasoning_text"])
+  for (const field of fields) {
+    if (field === undefined) continue
+    const text = delta[field]
+    if (typeof text === "string" && text.length > 0) return { field, text }
+  }
+  return undefined
+}
+
+const detailText = (details: ReadonlyArray<unknown>) => {
+  const text = details.flatMap((detail) => {
+    if (!isRecord(detail)) return []
+    if (detail.type === "reasoning.text" && typeof detail.text === "string" && detail.text) return [detail.text]
+    if (detail.type === "reasoning.summary" && typeof detail.summary === "string" && detail.summary)
+      return [detail.summary]
+    return []
+  })
+  if (text.length > 0) return text.join("")
+}
+
+const appendReasoningDetails = (result: Array<unknown>, details: ReadonlyArray<unknown>) => {
+  for (const detail of details) {
+    const previous = result.at(-1)
+    if (
+      !isRecord(previous) ||
+      previous.type !== "reasoning.text" ||
+      !isRecord(detail) ||
+      detail.type !== "reasoning.text" ||
+      conflictingReasoningTextDetails(previous, detail)
+    ) {
+      result.push(detail)
+      continue
+    }
+    result[result.length - 1] = {
+      ...previous,
+      ...Object.fromEntries(Object.entries(detail).filter((entry) => entry[1] !== undefined)),
+      text: `${typeof previous.text === "string" ? previous.text : ""}${typeof detail.text === "string" ? detail.text : ""}`,
+      signature: mergeDetailValue(previous.signature, detail.signature),
+      format: mergeDetailValue(previous.format, detail.format),
+    }
+  }
+}
+
+const mergeDetailValue = (previous: unknown, current: unknown) =>
+  previous || current || (previous !== undefined ? previous : current)
+
+const conflictingReasoningTextDetails = (previous: Record<string, unknown>, current: Record<string, unknown>) =>
+  conflictingDetailValue(previous.id, current.id) ||
+  conflictingDetailValue(previous.index, current.index) ||
+  conflictingDetailValue(previous.format, current.format) ||
+  (Boolean(previous.signature) && Boolean(current.signature) && previous.signature !== current.signature)
+
+const conflictingDetailValue = (previous: unknown, current: unknown) =>
+  previous !== undefined && previous !== null && current !== undefined && current !== null && previous !== current
+
+const reasoningMetadata = (field: ParserState["reasoningField"], details?: ReadonlyArray<unknown>) => ({
+  openai: {
+    ...(field ? { reasoningField: field } : {}),
+    ...(details ? { reasoningDetails: details } : {}),
+  },
+})
+
 const step = (state: ParserState, event: OpenAIChatEvent) =>
  Effect.gen(function* () {
    const events: LLMEvent[] = []
@@ -409,25 +539,56 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
    const delta = choice?.delta
    const toolDeltas = delta?.tool_calls ?? []
    let tools = state.tools
+    let pendingTools = state.pendingTools

    let lifecycle = state.lifecycle

-    if (delta?.reasoning_content)
-      lifecycle = Lifecycle.reasoningDelta(lifecycle, events, "reasoning-0", delta.reasoning_content)
+    const reasoning = reasoningDelta(delta, state.reasoningField)
+    const reasoningField = state.reasoningField ?? (!state.lifecycle.text.has("text-0") ? reasoning?.field : undefined)
+    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(reasoningField)
+    const text = detailDelta?.length ? (detailText(detailDelta) ?? reasoning?.text) : reasoning?.text
+    if (!state.lifecycle.text.has("text-0") && text !== undefined)
+      lifecycle = Lifecycle.reasoningDelta(lifecycle, events, "reasoning-0", text, deltaMetadata)
+    else if (
+      reasoningDetailsObserved &&
+      !lifecycle.reasoning.has("reasoning-0") &&
+      (Boolean(delta?.content) || toolDeltas.length > 0)
+    )
+      lifecycle = Lifecycle.reasoningStart(lifecycle, events, "reasoning-0", deltaMetadata)
+    const reasoningEmitted = state.reasoningEmitted || lifecycle.reasoning.has("reasoning-0")

    if (delta?.content) {
-      lifecycle = Lifecycle.reasoningEnd(lifecycle, events, "reasoning-0")
+      lifecycle = Lifecycle.reasoningEnd(
+        lifecycle,
+        events,
+        "reasoning-0",
+        reasoningMetadata(reasoningField, reasoningDetailsObserved ? state.reasoningDetails : undefined),
+      )
      lifecycle = Lifecycle.textDelta(lifecycle, events, "text-0", delta.content)
    }

-    if (toolDeltas.length) lifecycle = Lifecycle.reasoningEnd(lifecycle, events, "reasoning-0")
-
    for (const tool of toolDeltas) {
+      const current = tools[tool.index]
+      const pending = pendingTools[tool.index]
+      const id = current?.id ?? pending?.id ?? (tool.id || undefined)
+      const name = current?.name ?? pending?.name ?? (tool.function?.name || undefined)
+      const text = `${pending?.input ?? ""}${tool.function?.arguments ?? ""}`
+      if (!current && (!id || !name)) {
+        pendingTools = { ...pendingTools, [tool.index]: { id: id || undefined, name: name || undefined, input: text } }
+        continue
+      }
+      if (pending) {
+        pendingTools = { ...pendingTools }
+        delete pendingTools[tool.index]
+      }
      const result = ToolStream.appendOrStart(
        ADAPTER,
        tools,
        tool.index,
-        { id: tool.id ?? undefined, name: tool.function?.name ?? undefined, text: tool.function?.arguments ?? "" },
+        { id: id || undefined, name: name || undefined, text },
        "OpenAI Chat tool call delta is missing id or name",
      )
      if (ToolStream.isError(result)) return yield* result
@@ -436,8 +597,11 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
      events.push(...result.events)
    }

+    if (finishReason !== undefined && state.finishReason === undefined && Object.keys(pendingTools).length > 0)
+      return yield* ProviderShared.eventError(ADAPTER, "OpenAI Chat tool call delta is missing id or name")
+
    // Finalize accumulated tool inputs eagerly when finish_reason arrives so
-    // JSON parse failures fail the stream at the boundary rather than at halt.
+    // valid calls and malformed local calls settle independently.
    const finished =
      finishReason !== undefined && state.finishReason === undefined && Object.keys(tools).length > 0
        ? yield* ToolStream.finishAll(ADAPTER, tools)
@@ -446,10 +610,15 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
    return [
      {
        tools: finished?.tools ?? tools,
+        pendingTools,
        toolCallEvents: finished?.events ?? state.toolCallEvents,
        usage,
        finishReason,
        lifecycle,
+        reasoningField,
+        reasoningDetails: state.reasoningDetails,
+        reasoningDetailsObserved,
+        reasoningEmitted,
      },
      events,
    ] as const
@@ -459,7 +628,16 @@ const finishEvents = (state: ParserState): ReadonlyArray<LLMEvent> => {
  const events: LLMEvent[] = []
  const hasToolCalls = state.toolCallEvents.length > 0
  const reason = state.finishReason === "stop" && hasToolCalls ? "tool-calls" : state.finishReason
-  const lifecycle = state.toolCallEvents.length ? Lifecycle.stepStart(state.lifecycle, events) : state.lifecycle
+  const metadata = reasoningMetadata(
+    state.reasoningField,
+    state.reasoningDetailsObserved ? state.reasoningDetails : undefined,
+  )
+  const started =
+    state.reasoningDetailsObserved && !state.reasoningEmitted
+      ? Lifecycle.reasoningStart(state.lifecycle, events, "reasoning-0", reasoningMetadata(state.reasoningField))
+      : state.lifecycle
+  const ended = Lifecycle.reasoningEnd(started, events, "reasoning-0", metadata)
+  const lifecycle = state.toolCallEvents.length ? Lifecycle.stepStart(ended, events) : ended
  events.push(...state.toolCallEvents)
  if (reason) Lifecycle.finish(lifecycle, events, { reason, usage: state.usage })
  return events
@@ -482,7 +660,16 @@ export const protocol = Protocol.make({
  },
  stream: {
    event: Protocol.jsonEvent(OpenAIChatEvent),
-    initial: () => ({ tools: ToolStream.empty<number>(), toolCallEvents: [], lifecycle: Lifecycle.initial() }),
+    initial: (request) => ({
+      tools: ToolStream.empty<number>(),
+      pendingTools: {},
+      toolCallEvents: [],
+      lifecycle: Lifecycle.initial(),
+      reasoningField: request.model.compatibility?.reasoningField,
+      reasoningDetails: [],
+      reasoningDetailsObserved: false,
+      reasoningEmitted: false,
+    }),
    step,
    onHalt: finishEvents,
  },
@@ -0,0 +1,270 @@
+import { Effect, Encoding, Schema } from "effect"
+import { Headers, HttpClientRequest, HttpClientResponse } from "effect/unstable/http"
+import {
+  ImageModel,
+  GeneratedImage,
+  ImageResponse,
+  type ImageInput,
+  type ImageRequestFor,
+  type ImageRoute,
+} from "../image"
+import { Auth, type Definition as AuthDefinition } from "../route/auth"
+import {
+  InvalidProviderOutputReason,
+  LLMError,
+  Usage,
+  mergeHttpOptions,
+  mergeJsonRecords,
+  type HttpOptions,
+} from "../schema"
+import { ProviderShared } from "./shared"
+import { ImageInputs } from "./utils/image-input"
+import { OpenAIImage } from "./utils/openai-image"
+
+const ADAPTER = "openai-images"
+export const DEFAULT_BASE_URL = "https://api.openai.com/v1"
+export const PATH = "/images/generations"
+export const EDIT_PATH = "/images/edits"
+
+export type OpenAIImageString<Known extends string> = Known | (string & {})
+
+export type OpenAIImageOptions = {
+  readonly mask?: ImageInput
+  readonly n?: number
+  readonly size?: OpenAIImageString<
+    "auto" | "256x256" | "512x512" | "1024x1024" | "1536x1024" | "1024x1536" | "1792x1024" | "1024x1792"
+  >
+  readonly quality?: OpenAIImageString<"auto" | "low" | "medium" | "high" | "standard" | "hd">
+  readonly background?: OpenAIImageString<"auto" | "opaque" | "transparent">
+  readonly moderation?: OpenAIImageString<"auto" | "low">
+  readonly outputFormat?: OpenAIImageString<"png" | "jpeg" | "webp">
+  readonly outputCompression?: number
+} & Record<string, unknown>
+
+export type OpenAIImageBody = Record<string, unknown> & {
+  readonly model: string
+  readonly prompt: string
+}
+
+const OpenAIImageResponse = Schema.Struct({
+  data: Schema.Array(
+    Schema.Struct({
+      b64_json: Schema.optional(Schema.String),
+      url: Schema.optional(Schema.String),
+      revised_prompt: Schema.optional(Schema.String),
+    }),
+  ),
+  output_format: Schema.optional(Schema.String),
+  usage: Schema.optional(
+    Schema.Struct({
+      input_tokens: Schema.optional(Schema.Number),
+      output_tokens: Schema.optional(Schema.Number),
+      total_tokens: Schema.optional(Schema.Number),
+      input_tokens_details: Schema.optional(Schema.Record(Schema.String, Schema.Unknown)),
+      output_tokens_details: Schema.optional(Schema.Record(Schema.String, Schema.Unknown)),
+    }),
+  ),
+})
+
+export interface ModelInput {
+  readonly id: string
+  readonly auth: AuthDefinition
+  readonly baseURL?: string
+  readonly headers?: Record<string, string>
+  readonly http?: HttpOptions
+}
+
+const nativeOptions = (options: OpenAIImageOptions | undefined) => {
+  if (!options) return undefined
+  const { mask: _, outputFormat, outputCompression, ...native } = options
+  return {
+    output_format: outputFormat,
+    output_compression: outputCompression,
+    ...native,
+  }
+}
+
+const invalidOutput = (message: string) =>
+  new LLMError({
+    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)
+  Object.entries(query).forEach(([key, value]) => next.searchParams.set(key, value))
+  return next.toString()
+}
+
+export const model = (input: ModelInput) => {
+  const route: ImageRoute<OpenAIImageOptions> = {
+    id: ADAPTER,
+    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(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, ADAPTER)
+        return Effect.succeed(undefined)
+      })
+      const multipartMask =
+        mask === undefined
+          ? undefined
+          : mask.type === "bytes"
+            ? { data: mask.data, mediaType: mask.mediaType }
+            : mask.type === "url"
+              ? yield* ImageInputs.decodeDataUrl(mask.url, ADAPTER)
+              : undefined
+      const useMultipart =
+        sourceImages.length > 0 &&
+        multipartImages.every((image) => image !== undefined) &&
+        (mask === undefined || multipartMask !== undefined)
+      const path = sourceImages.length === 0 ? PATH : EDIT_PATH
+      const url = applyQuery(`${(input.baseURL ?? DEFAULT_BASE_URL).replace(/\/$/, "")}${path}`, http?.query)
+
+      if (useMultipart) {
+        const form = new FormData()
+        form.append("model", request.model.id)
+        form.append("prompt", request.prompt)
+        Object.entries(mergeJsonRecords(nativeOptions(request.options), http?.body) ?? {}).forEach(([key, value]) => {
+          if (["model", "prompt", "image", "image[]", "images", "mask"].includes(key)) return
+          form.append(key, typeof value === "string" ? value : ProviderShared.encodeJson(value))
+        })
+        multipartImages.forEach((image, index) => {
+          if (image === undefined) return
+          form.append("image[]", imageBlob(image.data, image.mediaType), `image-${index}`)
+        })
+        if (multipartMask !== undefined)
+          form.append("mask", imageBlob(multipartMask.data, multipartMask.mediaType), "mask")
+        const headers = yield* Auth.toEffect(input.auth)({
+          request,
+          method: "POST",
+          url,
+          body: "[multipart/form-data]",
+          headers: Headers.remove(Headers.fromInput({ ...input.headers, ...http?.headers }), "content-type"),
+        })
+        const response = yield* execute(
+          HttpClientRequest.post(url).pipe(HttpClientRequest.setHeaders(headers), HttpClientRequest.bodyFormData(form)),
+        )
+        return yield* parseResponse(response, request.options, http?.body)
+      }
+
+      const references = sourceImages.map((image) => {
+        if (image.type === "bytes") return { image_url: ImageInputs.dataUrl(image) }
+        if (image.type === "url") return { image_url: image.url }
+        if (image.type === "file-id") return { file_id: image.id }
+        return undefined
+      })
+      if (references.some((image) => image === undefined))
+        return yield* ImageInputs.invalid(ADAPTER, "OpenAI Images accepts image URLs, data URLs, bytes, and file IDs")
+      const maskReference =
+        mask === undefined
+          ? undefined
+          : mask.type === "bytes"
+            ? { image_url: ImageInputs.dataUrl(mask) }
+            : mask.type === "url"
+              ? { image_url: mask.url }
+              : mask.type === "file-id"
+                ? { file_id: mask.id }
+                : undefined
+      if (mask !== undefined && maskReference === undefined)
+        return yield* ImageInputs.invalid(ADAPTER, "OpenAI Images accepts masks as URLs, data URLs, bytes, or file IDs")
+      const requestBody = mergeJsonRecords(
+        {
+          model: request.model.id,
+          prompt: request.prompt,
+          images: references.length === 0 ? undefined : references,
+          mask: maskReference,
+        },
+        nativeOptions(request.options),
+        http?.body,
+      ) as OpenAIImageBody
+      const text = ProviderShared.encodeJson(requestBody)
+      const headers = yield* Auth.toEffect(input.auth)({
+        request,
+        method: "POST",
+        url,
+        body: text,
+        headers: Headers.fromInput({ ...input.headers, ...http?.headers }),
+      })
+      const response = yield* execute(
+        HttpClientRequest.post(url).pipe(
+          HttpClientRequest.setHeaders(headers),
+          HttpClientRequest.bodyText(text, "application/json"),
+        ),
+      )
+      return yield* parseResponse(response, request.options, http?.body)
+    }),
+  }
+  return ImageModel.make<OpenAIImageOptions>({ id: input.id, provider: "openai", route, http: input.http })
+}
+
+const parseResponse = Effect.fn("OpenAIImages.parseResponse")(function* (
+  response: HttpClientResponse.HttpClientResponse,
+  options: OpenAIImageOptions | undefined,
+  overlay: Record<string, unknown> | undefined,
+) {
+  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 =
+    decoded.output_format ?? (typeof requestBody?.output_format === "string" ? requestBody.output_format : "png")
+  const images = yield* Effect.forEach(decoded.data, (item, index) => {
+    if (item.b64_json)
+      return Effect.fromResult(Encoding.decodeBase64(item.b64_json)).pipe(
+        Effect.mapError(() => invalidOutput(`OpenAI Images result ${index} contains invalid base64 data`)),
+        Effect.map(
+          (data) =>
+            new GeneratedImage({
+              mediaType: `image/${format}`,
+              data,
+              providerMetadata:
+                item.revised_prompt === undefined ? undefined : { openai: { revisedPrompt: item.revised_prompt } },
+            }),
+        ),
+      )
+    if (item.url)
+      return Effect.succeed(
+        new GeneratedImage({
+          mediaType: `image/${format}`,
+          data: item.url,
+          providerMetadata:
+            item.revised_prompt === undefined ? undefined : { openai: { revisedPrompt: item.revised_prompt } },
+        }),
+      )
+    return Effect.fail(invalidOutput(`OpenAI Images result ${index} has neither image data nor a URL`))
+  })
+  if (images.length === 0) return yield* invalidOutput("OpenAI Images returned no images")
+  return new ImageResponse({
+    images,
+    usage:
+      decoded.usage === undefined
+        ? undefined
+        : new Usage({
+            inputTokens: decoded.usage.input_tokens,
+            outputTokens: decoded.usage.output_tokens,
+            totalTokens: decoded.usage.total_tokens,
+            providerMetadata: { openai: decoded.usage },
+          }),
+    providerMetadata: { openai: { outputFormat: format } },
+  })
+})
+
+const imageBlob = (data: Uint8Array, mediaType: string) => {
+  const buffer = new ArrayBuffer(data.byteLength)
+  new Uint8Array(buffer).set(data)
+  return new Blob([buffer], { type: mediaType })
+}
+
+export const OpenAIImages = {
+  model,
+} as const
@@ -1,4 +1,4 @@
-import { Effect, Schema } from "effect"
+import { Effect, Encoding, Schema } from "effect"
 import { Route } from "../route/client"
 import { Auth } from "../route/auth"
 import { Endpoint } from "../route/endpoint"
@@ -11,6 +11,7 @@ import {
  type FinishReason,
  type JsonSchema,
  type LLMRequest,
+  type MediaPart,
  type ProviderMetadata,
  type ReasoningPart,
  type TextPart,
@@ -25,8 +26,10 @@ import { OpenAIOptions } from "./utils/openai-options"
 import { Lifecycle } from "./utils/lifecycle"
 import { ToolSchemaProjection } from "./utils/tool-schema"
 import { ToolStream } from "./utils/tool-stream"
+import { OpenAIImage } from "./utils/openai-image"

 const ADAPTER = "openai-responses"
+const MEDIA_MIMES = new Set<string>([...ProviderShared.IMAGE_MIMES, ...ProviderShared.PDF_MIMES])
 export const DEFAULT_BASE_URL = "https://api.openai.com/v1"
 export const PATH = "/responses"

@@ -41,7 +44,17 @@ const OpenAIResponsesInputImage = Schema.Struct({
  type: Schema.tag("input_image"),
  image_url: Schema.String,
 })
-const OpenAIResponsesInputContent = Schema.Union([OpenAIResponsesInputText, OpenAIResponsesInputImage])
+const OpenAIResponsesInputFile = Schema.Struct({
+  type: Schema.tag("input_file"),
+  filename: Schema.String,
+  file_data: Schema.String,
+  mime_type: Schema.optional(Schema.String),
+})
+const OpenAIResponsesInputContent = Schema.Union([
+  OpenAIResponsesInputText,
+  OpenAIResponsesInputImage,
+  OpenAIResponsesInputFile,
+])
 type OpenAIResponsesInputContent = Schema.Schema.Type<typeof OpenAIResponsesInputContent>

 const OpenAIResponsesOutputText = Schema.Struct({
@@ -67,9 +80,13 @@ const OpenAIResponsesItemReference = Schema.Struct({
 })

 // `function_call_output.output` accepts either a plain string or an ordered
-// array of content items so tools can return images in addition to text.
+// array of content items so tools can return images and files in addition to text.
 // https://platform.openai.com/docs/api-reference/responses/object
-const OpenAIResponsesFunctionCallOutputContent = Schema.Union([OpenAIResponsesInputText, OpenAIResponsesInputImage])
+const OpenAIResponsesFunctionCallOutputContent = Schema.Union([
+  OpenAIResponsesInputText,
+  OpenAIResponsesInputImage,
+  OpenAIResponsesInputFile,
+])

 const OpenAIResponsesFunctionCallOutput = Schema.Union([
  Schema.String,
@@ -113,11 +130,24 @@ const OpenAIResponsesTool = Schema.Struct({
  parameters: JsonObject,
  strict: Schema.optional(Schema.Boolean),
 })
-type OpenAIResponsesTool = Schema.Schema.Type<typeof OpenAIResponsesTool>
+const OpenAIResponsesImageGenerationTool = Schema.Struct({
+  type: Schema.tag("image_generation"),
+  action: Schema.optional(Schema.Literals(["auto", "generate", "edit"])),
+  background: Schema.optional(Schema.Literals(["auto", "opaque", "transparent"])),
+  input_fidelity: Schema.optional(Schema.Literals(["low", "high"])),
+  output_compression: Schema.optional(Schema.Int.check(Schema.isBetween({ minimum: 0, maximum: 100 }))),
+  output_format: Schema.optional(Schema.Literals(["png", "jpeg", "webp"])),
+  partial_images: Schema.optional(Schema.Int.check(Schema.isGreaterThanOrEqualTo(0))),
+  quality: Schema.optional(Schema.Literals(["auto", "low", "medium", "high"])),
+  size: Schema.optional(OpenAIImage.Size),
+})
+const OpenAIResponsesTools = Schema.Union([OpenAIResponsesTool, OpenAIResponsesImageGenerationTool])
+type OpenAIResponsesTool = Schema.Schema.Type<typeof OpenAIResponsesTools>

 const OpenAIResponsesToolChoice = Schema.Union([
  Schema.Literals(["auto", "none", "required"]),
  Schema.Struct({ type: Schema.tag("function"), name: Schema.String }),
+  Schema.Struct({ type: Schema.tag("image_generation") }),
 ])

 // Fields shared between the HTTP body and the WebSocket `response.create`
@@ -128,7 +158,7 @@ const OpenAIResponsesCoreFields = {
  model: Schema.String,
  input: Schema.Array(OpenAIResponsesInputItem),
  instructions: Schema.optional(Schema.String),
-  tools: optionalArray(OpenAIResponsesTool),
+  tools: optionalArray(OpenAIResponsesTools),
  tool_choice: Schema.optional(OpenAIResponsesToolChoice),
  store: Schema.optional(Schema.Boolean),
  service_tier: Schema.optional(OpenAIOptions.OpenAIServiceTier),
@@ -194,6 +224,8 @@ const OpenAIResponsesStreamItem = Schema.Struct({
  outputs: Schema.optional(Schema.Unknown),
  server_label: Schema.optional(Schema.String),
  output: Schema.optional(Schema.Unknown),
+  result: Schema.optional(Schema.String),
+  output_format: Schema.optional(Schema.Literals(["png", "jpeg", "webp"])),
  error: Schema.optional(Schema.Unknown),
  encrypted_content: optionalNull(Schema.String),
 })
@@ -221,7 +253,7 @@ const OpenAIResponsesEvent = Schema.Struct({
      Schema.Struct({
        id: Schema.optional(Schema.String),
        service_tier: optionalNull(Schema.String),
-        incomplete_details: optionalNull(Schema.Struct({ reason: Schema.String })),
+        incomplete_details: optionalNull(Schema.Struct({ reason: Schema.optional(Schema.String) })),
        usage: optionalNull(OpenAIResponsesUsage),
        error: optionalNull(OpenAIResponsesErrorPayload),
      }),
@@ -258,21 +290,41 @@ const invalid = ProviderShared.invalidRequest
 // =============================================================================
 // Request Lowering
 // =============================================================================
-const lowerTool = (tool: ToolDefinition, inputSchema: JsonSchema): OpenAIResponsesTool => ({
-  type: "function",
-  name: tool.name,
-  description: tool.description,
-  parameters: ToolSchemaProjection.openAI(inputSchema),
-  // TODO: Read this from OpenAI-specific tool options so direct LLM callers can opt into strict schemas.
-  strict: false,
+const nativeImageToolInput = (tool: ToolDefinition) => {
+  const native = tool.native?.openai
+  return ProviderShared.isRecord(native) && native.type === "image_generation" ? native : undefined
+}
+
+const nativeImageTool = (tool: ToolDefinition) => {
+  const native = nativeImageToolInput(tool)
+  return Schema.is(OpenAIResponsesImageGenerationTool)(native) ? native : undefined
+}
+
+const lowerTool = Effect.fn("OpenAIResponses.lowerTool")(function* (tool: ToolDefinition, inputSchema: JsonSchema) {
+  const native = nativeImageToolInput(tool)
+  if (native !== undefined) {
+    if (Schema.is(OpenAIResponsesImageGenerationTool)(native)) return native
+    return yield* invalid("OpenAI Responses image generation tool options are invalid")
+  }
+  return {
+    type: "function" as const,
+    name: tool.name,
+    description: tool.description,
+    parameters: ToolSchemaProjection.openAI(inputSchema),
+    // TODO: Read this from OpenAI-specific tool options so direct LLM callers can opt into strict schemas.
+    strict: false,
+  }
 })

-const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>) =>
+const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>, tools: ReadonlyArray<ToolDefinition>) =>
  ProviderShared.matchToolChoice("OpenAI Responses", toolChoice, {
    auto: () => "auto" as const,
    none: () => "none" as const,
    required: () => "required" as const,
-    tool: (name) => ({ type: "function" as const, name }),
+    tool: (name) =>
+      tools.some((tool) => tool.name === name && nativeImageTool(tool) !== undefined)
+        ? ({ type: "image_generation" } as const)
+        : { type: "function" as const, name },
  })

 const lowerToolCall = (part: ToolCallPart): OpenAIResponsesInputItem => ({
@@ -307,42 +359,58 @@ const hostedToolItemID = (part: ToolResultPart) => {
    : undefined
 }

-const lowerUserContent = Effect.fn("OpenAIResponses.lowerUserContent")(function* (
-  part: LLMRequest["messages"][number]["content"][number],
-) {
-  if (part.type === "text") return { type: "input_text" as const, text: part.text }
-  if (part.type === "media") {
-    const media = yield* ProviderShared.validateMedia(
-      "OpenAI Responses",
-      part,
-      new Set<string>(ProviderShared.IMAGE_MIMES),
-    )
-    return { type: "input_image" as const, image_url: media.dataUrl }
+const lowerMedia = Effect.fn("OpenAIResponses.lowerMedia")(function* (part: MediaPart, provider: string) {
+  const media = yield* ProviderShared.validateMedia("OpenAI Responses", part, MEDIA_MIMES)
+  if (media.mime === "application/pdf") {
+    // xAI models inline bytes and MIME separately; OpenAI uses a data URL in file_data.
+    if (provider === "xai")
+      return {
+        type: "input_file" as const,
+        filename: part.filename ?? "document.pdf",
+        file_data: media.base64,
+        mime_type: media.mime,
+      }
+    return {
+      type: "input_file" as const,
+      filename: part.filename ?? "document.pdf",
+      file_data: media.dataUrl,
+    }
  }
-  return yield* ProviderShared.unsupportedContent("OpenAI Responses", "user", ["text", "media"])
-})
-
-// Tool results may carry structured text/images. Keep media as provider-native
-// content instead of JSON-stringifying base64 into a prompt string.
-const lowerToolResultContentItem = Effect.fn("OpenAIResponses.lowerToolResultContentItem")(function* (
-  item: ToolContent,
-) {
-  if (item.type === "text") return { type: "input_text" as const, text: item.text }
-  const media = yield* ProviderShared.validateToolFile(
-    "OpenAI Responses",
-    item,
-    new Set<string>(ProviderShared.IMAGE_MIMES),
-  )
  return { type: "input_image" as const, image_url: media.dataUrl }
 })

-const lowerToolResultOutput = Effect.fn("OpenAIResponses.lowerToolResultOutput")(function* (part: ToolResultPart) {
+const lowerUserContent = Effect.fn("OpenAIResponses.lowerUserContent")(function* (
+  part: LLMRequest["messages"][number]["content"][number],
+  provider: string,
+) {
+  if (part.type === "text") return { type: "input_text" as const, text: part.text }
+  if (part.type === "media") return yield* lowerMedia(part, provider)
+  return yield* ProviderShared.unsupportedContent("OpenAI Responses", "user", ["text", "media"])
+})
+
+// Tool results may carry structured text, images, and files. Keep media as provider-native
+// content instead of JSON-stringifying base64 into a prompt string.
+const lowerToolResultContentItem = Effect.fn("OpenAIResponses.lowerToolResultContentItem")(function* (
+  item: ToolContent,
+  provider: string,
+) {
+  if (item.type === "text") return { type: "input_text" as const, text: item.text }
+  return yield* lowerMedia(
+    { type: "media", mediaType: item.mime, data: item.uri, filename: item.name },
+    provider,
+  )
+})
+
+const lowerToolResultOutput = Effect.fn("OpenAIResponses.lowerToolResultOutput")(function* (
+  part: ToolResultPart,
+  provider: string,
+) {
  // Text/json/error results are encoded as a plain string for backward
  // compatibility with existing cassettes and provider expectations.
  if (part.result.type !== "content") return ProviderShared.toolResultText(part)
  // Preserve the narrowed array element type when compiled through a consumer package.
  const content: ReadonlyArray<ToolContent> = part.result.value
-  return yield* Effect.forEach(content, lowerToolResultContentItem)
+  return yield* Effect.forEach(content, (item) => lowerToolResultContentItem(item, provider))
 })

 const lowerMessages = Effect.fn("OpenAIResponses.lowerMessages")(function* (request: LLMRequest) {
@@ -365,7 +433,10 @@ const lowerMessages = Effect.fn("OpenAIResponses.lowerMessages")(function* (requ
    }

    if (message.role === "user") {
-      input.push({ role: "user", content: yield* Effect.forEach(message.content, lowerUserContent) })
+      input.push({
+        role: "user",
+        content: yield* Effect.forEach(message.content, (part) => lowerUserContent(part, request.model.provider)),
+      })
      continue
    }

@@ -420,6 +491,15 @@ const lowerMessages = Effect.fn("OpenAIResponses.lowerMessages")(function* (requ
          const itemID = hostedToolItemID(part)
          if (store !== false && itemID && !hostedToolReferences.has(itemID))
            input.push({ type: "item_reference", id: itemID })
+          if (store === false && part.name === "image_generation" && part.result.type === "content") {
+            const content: ReadonlyArray<ToolContent> = part.result.value
+            input.push({
+              role: "user",
+              content: yield* Effect.forEach(content, (item) =>
+                lowerToolResultContentItem(item, request.model.provider),
+              ),
+            })
+          }
          if (itemID) hostedToolReferences.add(itemID)
          continue
        }
@@ -440,7 +520,7 @@ const lowerMessages = Effect.fn("OpenAIResponses.lowerMessages")(function* (requ
      input.push({
        type: "function_call_output",
        call_id: part.id,
-        output: yield* lowerToolResultOutput(part),
+        output: yield* lowerToolResultOutput(part, request.model.provider),
      })
    }
  }
@@ -485,10 +565,10 @@ const fromRequest = Effect.fn("OpenAIResponses.fromRequest")(function* (request:
    tools:
      request.tools.length === 0
        ? undefined
-        : request.tools.map((tool) =>
+        : yield* Effect.forEach(request.tools, (tool) =>
            lowerTool(tool, ToolSchemaProjection.modelCompatibility(tool.inputSchema, toolSchemaCompatibility)),
          ),
-    tool_choice: request.toolChoice ? yield* lowerToolChoice(request.toolChoice) : undefined,
+    tool_choice: request.toolChoice ? yield* lowerToolChoice(request.toolChoice, request.tools) : undefined,
    stream: true as const,
    max_output_tokens: generation?.maxTokens,
    temperature: generation?.temperature,
@@ -522,7 +602,8 @@ const mapUsage = (usage: OpenAIResponsesUsage | null | undefined) => {

 const mapFinishReason = (event: OpenAIResponsesEvent, hasFunctionCall: boolean): FinishReason => {
  const reason = event.response?.incomplete_details?.reason
-  if (reason === undefined || reason === null) return hasFunctionCall ? "tool-calls" : "stop"
+  if (reason === undefined || reason === null)
+    return hasFunctionCall ? "tool-calls" : event.type === "response.incomplete" ? "unknown" : "stop"
  if (reason === "max_output_tokens") return "length"
  if (reason === "content_filter") return "content-filter"
  return hasFunctionCall ? "tool-calls" : "unknown"
@@ -574,14 +655,29 @@ const isReasoningItem = (

 // Round-trip the full item as the structured result so consumers can extract
 // outputs / sources / status without re-decoding.
-const hostedToolResult = (item: OpenAIResponsesStreamItem) => {
+const hostedToolResult = Effect.fn("OpenAIResponses.hostedToolResult")(function* (item: OpenAIResponsesStreamItem) {
  const isError = typeof item.error !== "undefined" && item.error !== null
+  if (item.type === "image_generation_call" && item.result) {
+    yield* Effect.fromResult(Encoding.decodeBase64(item.result)).pipe(
+      Effect.mapError(() => ProviderShared.eventError(ADAPTER, "OpenAI Responses returned invalid image base64")),
+    )
+    return {
+      type: "content" as const,
+      value: [
+        {
+          type: "file" as const,
+          uri: `data:image/${item.output_format ?? "png"};base64,${item.result}`,
+          mime: `image/${item.output_format ?? "png"}`,
+        },
+      ],
+    }
+  }
  return isError ? { type: "error" as const, value: item.error } : { type: "json" as const, value: item }
-}
+})

-const hostedToolEvents = (
+const hostedToolEvents = Effect.fn("OpenAIResponses.hostedToolEvents")(function* (
  item: OpenAIResponsesStreamItem & { type: HostedToolType; id: string },
-): ReadonlyArray<LLMEvent> => {
+) {
  const tool = HOSTED_TOOLS[item.type]
  const providerMetadata = openaiMetadata({ itemId: item.id })
  return [
@@ -595,12 +691,12 @@ const hostedToolEvents = (
    LLMEvent.toolResult({
      id: item.id,
      name: tool.name,
-      result: hostedToolResult(item),
+      result: yield* hostedToolResult(item),
      providerExecuted: true,
      providerMetadata,
    }),
  ]
-}
+})

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

@@ -835,7 +931,9 @@ const onOutputItemDone = Effect.fn("OpenAIResponses.onOutputItemDone")(function*
      {
        ...state,
        lifecycle,
-        hasFunctionCall: resultEvents.some(LLMEvent.is.toolCall) ? true : state.hasFunctionCall,
+        hasFunctionCall:
+          resultEvents.some((event) => LLMEvent.is.toolCall(event) || LLMEvent.is.toolInputError(event)) ||
+          state.hasFunctionCall,
        tools: result.tools,
      },
      events,
@@ -845,7 +943,7 @@ const onOutputItemDone = Effect.fn("OpenAIResponses.onOutputItemDone")(function*
  if (isHostedToolItem(item)) {
    const events: LLMEvent[] = []
    const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
-    events.push(...hostedToolEvents(item))
+    events.push(...(yield* hostedToolEvents(item)))
    return [{ ...state, lifecycle }, events] satisfies StepResult
  }

@@ -158,7 +158,8 @@ export const parseToolInput = (route: string, name: string, raw: string) =>
 export const IMAGE_MIMES = ["image/png", "image/jpeg", "image/gif", "image/webp"] as const
 export const VIDEO_MIMES = ["video/mp4", "video/webm", "video/quicktime"] as const
 export const AUDIO_MIMES = ["audio/wav", "audio/mp3", "audio/aiff", "audio/aac", "audio/ogg", "audio/flac"] as const
-export const MEDIA_MIMES = [...IMAGE_MIMES, ...VIDEO_MIMES, ...AUDIO_MIMES] as const
+export const PDF_MIMES = ["application/pdf"] as const
+export const MEDIA_MIMES = [...IMAGE_MIMES, ...VIDEO_MIMES, ...AUDIO_MIMES, ...PDF_MIMES] as const
 export const MAX_MEDIA_ENCODED_BYTES = 28 * 1024 * 1024
 export const MAX_MEDIA_DECODED_BYTES = 20 * 1024 * 1024

@@ -49,10 +49,10 @@ const DOCUMENT_FORMATS = {
  "text/markdown": "md",
 } as const satisfies Record<string, DocumentFormat>

-const documentBlock = (part: MediaPart, format: DocumentFormat, bytes: string): DocumentBlock => ({
+const documentBlock = (name: string, format: DocumentFormat, bytes: string): DocumentBlock => ({
  document: {
    format,
-    name: part.filename ?? `document.${format}`,
+    name,
    source: { bytes },
  },
 })
@@ -77,12 +77,14 @@ export const lower = Effect.fn("BedrockMedia.lower")(function* (part: MediaPart)
    return yield* ProviderShared.invalidRequest(`Bedrock Converse does not support image media type ${part.mediaType}`)
  const documentFormat = DOCUMENT_FORMATS[mime as keyof typeof DOCUMENT_FORMATS]
  if (documentFormat) {
+    if (!part.filename)
+      return yield* ProviderShared.invalidRequest("Bedrock Converse document media requires a filename")
    const media = yield* ProviderShared.validateMedia(
      "Bedrock Converse",
      part,
      new Set<string>(Object.keys(DOCUMENT_FORMATS)),
    )
-    return documentBlock(part, documentFormat, media.base64)
+    return documentBlock(part.filename, documentFormat, media.base64)
  }
  return yield* ProviderShared.invalidRequest(`Bedrock Converse does not support media type ${part.mediaType}`)
 })
@@ -0,0 +1,34 @@
+import { Effect, Encoding } from "effect"
+import type { ImageInput } from "../../image"
+import { InvalidRequestReason, LLMError } from "../../schema"
+
+const invalid = (module: string, message: string) =>
+  new LLMError({
+    module,
+    method: "generate",
+    reason: new InvalidRequestReason({ message }),
+  })
+
+export const dataUrl = (input: Extract<ImageInput, { readonly type: "bytes" }>) =>
+  `data:${input.mediaType};base64,${Encoding.encodeBase64(input.data)}`
+
+export const decodeDataUrl = (
+  url: string,
+  module: string,
+): Effect.Effect<{ readonly mediaType: string; readonly data: Uint8Array } | undefined, LLMError> => {
+  if (!url.startsWith("data:")) return Effect.succeed(undefined)
+  const match = /^data:([^;,]+);base64,(.*)$/s.exec(url)
+  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(() => invalid(module, "Image data URL contains invalid base64 data")),
+    Effect.map((data) => ({ mediaType: match[1], data })),
+  )
+}
+
+export const invalidImageInput = invalid
+
+export const ImageInputs = {
+  dataUrl,
+  decodeDataUrl,
+  invalid: invalidImageInput,
+} as const
@@ -44,7 +44,7 @@ export const reasoningDelta = (
  providerMetadata?: ProviderMetadata,
 ): State => {
  const started = reasoningStart(state, events, id, providerMetadata)
-  events.push(LLMEvent.reasoningDelta({ id, text }))
+  events.push(LLMEvent.reasoningDelta({ id, text, providerMetadata }))
  return started
 }

@@ -0,0 +1,20 @@
+import { Schema } from "effect"
+
+const dimensions = (value: string) => {
+  const match = /^(\d+)x(\d+)$/.exec(value)
+  if (!match) return undefined
+  return { width: Number(match[1]), height: Number(match[2]) }
+}
+
+export const Size = Schema.String.check(
+  Schema.makeFilter((value) => {
+    if (value === "auto") return undefined
+    const parsed = dimensions(value)
+    if (!parsed) return "image size must be `auto` or `{width}x{height}`"
+    return parsed.width > 0 && parsed.height > 0 ? undefined : "image dimensions must be positive integers"
+  }),
+)
+
+export const OpenAIImage = {
+  Size,
+} as const
@@ -1,5 +1,5 @@
 import { Effect } from "effect"
-import { LLMError, LLMEvent, type ProviderMetadata, type ToolCall } from "../../schema"
+import { LLMError, LLMEvent, type ProviderMetadata, type ToolCall, type ToolInputError } from "../../schema"
 import { eventError, parseToolInput, type ToolAccumulator } from "../shared"

 type StreamKey = string | number
@@ -53,6 +53,7 @@ const inputStart = (tool: PendingTool) =>
  LLMEvent.toolInputStart({
    id: tool.id,
    name: tool.name,
+    providerExecuted: tool.providerExecuted ? true : undefined,
    providerMetadata: tool.providerMetadata,
  })

@@ -63,19 +64,36 @@ const inputDelta = (tool: PendingTool, text: string) =>
    text,
  })

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

 /** Store the updated tool and produce the optional public delta event. */
 const appendTool = <K extends StreamKey>(
@@ -122,8 +140,8 @@ export const appendOrStart = <K extends StreamKey>(
  missingToolMessage: string,
 ): AppendOutcome<K> | LLMError => {
  const current = tools[key]
-  const id = delta.id ?? current?.id
-  const name = delta.name ?? current?.name
+  const id = current?.id ?? delta.id
+  const name = current?.name ?? delta.name
  if (!id || !name) return eventError(route, missingToolMessage)

  const tool = {
@@ -158,8 +176,9 @@ export const appendExisting = <K extends StreamKey>(

 /**
 * Finalize one pending tool call: parse the accumulated raw JSON, remove it
- * from state, and return the optional public `tool-call` event. Missing keys are
- * a no-op because some providers emit stop events for non-tool content blocks.
+ * from state, and return either a call or a non-executable local input error.
+ * Missing keys are a no-op because some providers emit stop events for
+ * non-tool content blocks.
 */
 export const finish = <K extends StreamKey>(route: string, tools: State<K>, key: K) =>
  Effect.gen(function* () {
@@ -167,10 +186,7 @@ export const finish = <K extends StreamKey>(route: string, tools: State<K>, key:
    if (!tool) return { tools }
    return {
      tools: withoutTool(tools, key),
-      events: [
-        LLMEvent.toolInputEnd({ id: tool.id, name: tool.name, providerMetadata: tool.providerMetadata }),
-        yield* toolCall(route, tool),
-      ],
+      events: finishEvents(tool, yield* toolCall(route, tool)),
    }
  })

@@ -185,17 +201,14 @@ export const finishWithInput = <K extends StreamKey>(route: string, tools: State
    if (!tool) return { tools }
    return {
      tools: withoutTool(tools, key),
-      events: [
-        LLMEvent.toolInputEnd({ id: tool.id, name: tool.name, providerMetadata: tool.providerMetadata }),
-        yield* toolCall(route, tool, input),
-      ],
+      events: finishEvents(tool, yield* toolCall(route, tool, input)),
    }
  })

 /**
 * Finalize every pending tool call at once. OpenAI Chat has this shape: it does
- * not emit per-tool stop events, so all accumulated calls finish when the choice
- * receives a terminal `finish_reason`.
+ * not emit per-tool stop events, so all accumulated calls finish independently
+ * when the choice receives a terminal `finish_reason`.
 */
 export const finishAll = <K extends StreamKey>(route: string, tools: State<K>) =>
  Effect.gen(function* () {
@@ -205,12 +218,7 @@ export const finishAll = <K extends StreamKey>(route: string, tools: State<K>) =
    return {
      tools: empty<K>(),
      events: yield* Effect.forEach(pending, (tool) =>
-        toolCall(route, tool).pipe(
-          Effect.map((call) => [
-            LLMEvent.toolInputEnd({ id: tool.id, name: tool.name, providerMetadata: tool.providerMetadata }),
-            call,
-          ]),
-        ),
+        toolCall(route, tool).pipe(Effect.map((event) => finishEvents(tool, event))),
      ).pipe(Effect.map((events) => events.flat())),
    }
  })
@@ -0,0 +1,202 @@
+import { Effect, Encoding, Schema } from "effect"
+import { Headers, HttpClientRequest } from "effect/unstable/http"
+import { GeneratedImage, ImageModel, ImageResponse, type ImageRequestFor, type ImageRoute } from "../image"
+import { Auth, type Definition as AuthDefinition } from "../route/auth"
+import {
+  InvalidProviderOutputReason,
+  LLMError,
+  Usage,
+  mergeHttpOptions,
+  mergeJsonRecords,
+  type HttpOptions,
+} from "../schema"
+import { ProviderShared, optionalNull } from "./shared"
+import { ImageInputs } from "./utils/image-input"
+
+const ADAPTER = "xai-images"
+export const DEFAULT_BASE_URL = "https://api.x.ai/v1"
+export const PATH = "/images/generations"
+export const EDIT_PATH = "/images/edits"
+
+export type XAIImageString<Known extends string> = Known | (string & {})
+
+export type XAIImageOptions = {
+  readonly n?: number
+  readonly aspectRatio?: XAIImageString<
+    | "1:1"
+    | "3:4"
+    | "4:3"
+    | "9:16"
+    | "16:9"
+    | "2:3"
+    | "3:2"
+    | "9:19.5"
+    | "19.5:9"
+    | "9:20"
+    | "20:9"
+    | "1:2"
+    | "2:1"
+    | "auto"
+  >
+  readonly aspect_ratio?: XAIImageString<
+    | "1:1"
+    | "3:4"
+    | "4:3"
+    | "9:16"
+    | "16:9"
+    | "2:3"
+    | "3:2"
+    | "9:19.5"
+    | "19.5:9"
+    | "9:20"
+    | "20:9"
+    | "1:2"
+    | "2:1"
+    | "auto"
+  >
+  readonly resolution?: XAIImageString<"1k" | "2k">
+  readonly responseFormat?: XAIImageString<"url" | "b64_json">
+  readonly response_format?: XAIImageString<"url" | "b64_json">
+} & Record<string, unknown>
+
+type XAIImageBody = Record<string, unknown> & {
+  readonly model: string
+  readonly prompt: string
+}
+
+const XAIImageResponse = Schema.Struct({
+  data: Schema.Array(
+    Schema.Struct({
+      b64_json: optionalNull(Schema.String),
+      url: optionalNull(Schema.String),
+      revised_prompt: optionalNull(Schema.String),
+      mime_type: optionalNull(Schema.String),
+    }),
+  ),
+  usage: Schema.optional(Schema.Unknown),
+})
+
+export interface ModelInput {
+  readonly id: string
+  readonly auth: AuthDefinition
+  readonly baseURL?: string
+  readonly headers?: Record<string, string>
+  readonly http?: HttpOptions
+}
+
+const nativeOptions = (options: XAIImageOptions | undefined) => {
+  if (!options) return undefined
+  const { aspectRatio, responseFormat, ...native } = options
+  return {
+    aspect_ratio: aspectRatio,
+    response_format: responseFormat,
+    ...native,
+  }
+}
+
+const invalidOutput = (message: string) =>
+  new LLMError({
+    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)
+  Object.entries(query).forEach(([key, value]) => next.searchParams.set(key, value))
+  return next.toString()
+}
+
+export const model = (input: ModelInput) => {
+  const route: ImageRoute<XAIImageOptions> = {
+    id: ADAPTER,
+    generate: Effect.fn("XAIImages.generate")(function* (request: ImageRequestFor<XAIImageOptions>, execute) {
+      const http = mergeHttpOptions(request.model.http, request.http)
+      const imageReferences = (request.images ?? []).map((image) => {
+        if (image.type === "bytes") return { url: ImageInputs.dataUrl(image), type: "image_url" as const }
+        if (image.type === "url") return { url: image.url, type: "image_url" as const }
+        if (image.type === "file-id") return { file_id: image.id }
+        return undefined
+      })
+      if (imageReferences.some((image) => image === undefined))
+        return yield* ImageInputs.invalid(ADAPTER, "xAI Images accepts image URLs, data URLs, bytes, and file IDs")
+      const requestBody = mergeJsonRecords(
+        {
+          model: request.model.id,
+          prompt: request.prompt,
+          image: imageReferences.length === 1 ? imageReferences[0] : undefined,
+          images: imageReferences.length > 1 ? imageReferences : undefined,
+        },
+        nativeOptions(request.options),
+        http?.body,
+      ) as XAIImageBody
+      const text = ProviderShared.encodeJson(requestBody)
+      const url = applyQuery(
+        `${(input.baseURL ?? DEFAULT_BASE_URL).replace(/\/$/, "")}${imageReferences.length === 0 ? PATH : EDIT_PATH}`,
+        http?.query,
+      )
+      const headers = yield* Auth.toEffect(input.auth)({
+        request,
+        method: "POST",
+        url,
+        body: text,
+        headers: Headers.fromInput({ ...input.headers, ...http?.headers }),
+      })
+      const response = yield* execute(
+        HttpClientRequest.post(url).pipe(
+          HttpClientRequest.setHeaders(headers),
+          HttpClientRequest.bodyText(text, "application/json"),
+        ),
+      )
+      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(() => invalidOutput(`xAI Images result ${index} contains invalid base64 data`)),
+            Effect.map(
+              (data) =>
+                new GeneratedImage({
+                  mediaType,
+                  data,
+                  providerMetadata:
+                    item.revised_prompt === undefined || item.revised_prompt === null
+                      ? undefined
+                      : { xai: { revisedPrompt: item.revised_prompt } },
+                }),
+            ),
+          )
+        if (item.url)
+          return Effect.succeed(
+            new GeneratedImage({
+              mediaType,
+              data: item.url,
+              providerMetadata:
+                item.revised_prompt === undefined || item.revised_prompt === null
+                  ? undefined
+                  : { xai: { revisedPrompt: item.revised_prompt } },
+            }),
+          )
+        return Effect.fail(invalidOutput(`xAI Images result ${index} has neither image data nor a URL`))
+      })
+      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,
+        usage: usage === undefined ? undefined : new Usage({ providerMetadata: { xai: usage } }),
+        providerMetadata: usage === undefined ? undefined : { xai: { usage } },
+      })
+    }),
+  }
+  return ImageModel.make<XAIImageOptions>({ id: input.id, provider: "xai", route, http: input.http })
+}
+
+export const XAIImages = {
+  model,
+} as const
@@ -0,0 +1,132 @@
+import { Effect, Schema } from "effect"
+import { Headers, HttpClientRequest } from "effect/unstable/http"
+import { GeneratedImage, ImageModel, ImageResponse, type ImageRequestFor, type ImageRoute } from "../image"
+import { Auth, type Definition as AuthDefinition } from "../route/auth"
+import { InvalidProviderOutputReason, LLMError, mergeHttpOptions, mergeJsonRecords, type HttpOptions } from "../schema"
+import { ProviderShared } from "./shared"
+import { ImageInputs } from "./utils/image-input"
+
+const ADAPTER = "zai-images"
+export const DEFAULT_BASE_URL = "https://api.z.ai/api/paas/v4"
+export const PATH = "/images/generations"
+
+export type ZAIImageString<Known extends string> = Known | (string & {})
+
+export type ZAIImageOptions = {
+  readonly size?: ZAIImageString<
+    "1024x1024" | "768x1344" | "864x1152" | "1344x768" | "1152x864" | "1440x720" | "720x1440"
+  >
+  readonly quality?: ZAIImageString<"hd" | "standard">
+  readonly userID?: string
+} & Record<string, unknown>
+
+type ZAIImageBody = Record<string, unknown> & {
+  readonly model: string
+  readonly prompt: string
+}
+
+const ZAIImageResponse = Schema.Struct({
+  created: Schema.optional(Schema.Int),
+  id: Schema.optional(Schema.String),
+  request_id: Schema.optional(Schema.String),
+  data: Schema.Array(Schema.Struct({ url: Schema.String })),
+  content_filter: Schema.optional(
+    Schema.Array(
+      Schema.Struct({
+        role: Schema.optional(Schema.String),
+        level: Schema.optional(Schema.Number),
+      }),
+    ),
+  ),
+})
+
+export interface ModelInput {
+  readonly id: string
+  readonly auth: AuthDefinition
+  readonly baseURL?: string
+  readonly headers?: Record<string, string>
+  readonly http?: HttpOptions
+}
+
+const nativeOptions = (options: ZAIImageOptions | undefined) => {
+  if (!options) return undefined
+  const { userID, ...native } = options
+  return {
+    user_id: userID,
+    ...native,
+  }
+}
+
+const invalidOutput = (message: string) =>
+  new LLMError({
+    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)
+  Object.entries(query).forEach(([key, value]) => next.searchParams.set(key, value))
+  return next.toString()
+}
+
+export const model = (input: ModelInput) => {
+  const route: ImageRoute<ZAIImageOptions> = {
+    id: ADAPTER,
+    generate: Effect.fn("ZAIImages.generate")(function* (request: ImageRequestFor<ZAIImageOptions>, execute) {
+      if ((request.images?.length ?? 0) > 0)
+        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 },
+        nativeOptions(request.options),
+        http?.body,
+      ) as ZAIImageBody
+      const text = ProviderShared.encodeJson(requestBody)
+      const url = applyQuery(`${(input.baseURL ?? DEFAULT_BASE_URL).replace(/\/$/, "")}${PATH}`, http?.query)
+      const headers = yield* Auth.toEffect(input.auth)({
+        request,
+        method: "POST",
+        url,
+        body: text,
+        headers: Headers.fromInput({ ...input.headers, ...http?.headers }),
+      })
+      const response = yield* execute(
+        HttpClientRequest.post(url).pipe(
+          HttpClientRequest.setHeaders(headers),
+          HttpClientRequest.bodyText(text, "application/json"),
+        ),
+      )
+      const payload = yield* response.json.pipe(
+        Effect.mapError(() => invalidOutput("Failed to read the Z.ai Images response")),
+      )
+      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) =>
+            new GeneratedImage({
+              mediaType: "application/octet-stream",
+              data: item.url,
+            }),
+        ),
+        providerMetadata: {
+          zai: {
+            created: decoded.created,
+            id: decoded.id,
+            requestID: decoded.request_id,
+            contentFilter: decoded.content_filter,
+          },
+        },
+      })
+    }),
+  }
+  return ImageModel.make<ZAIImageOptions>({ id: input.id, provider: "zai", route, http: input.http })
+}
+
+export const ZAIImages = {
+  model,
+} as const
@@ -16,13 +16,17 @@ import {

 const patterns = [
  /prompt is too long/i,
+  /request_too_large/i,
  /input is too long for requested model/i,
  /exceeds the context window/i,
+  /exceeds (?:the )?(?:model'?s )?maximum context length(?: of [\d,]+ tokens?|\s*\([\d,]+\))/i,
  /input token count.*exceeds the maximum/i,
  /tokens in request more than max tokens allowed/i,
  /maximum prompt length is \d+/i,
  /reduce the length of the messages/i,
  /maximum context length is \d+ tokens/i,
+  /exceeds (?:the )?maximum allowed input length of [\d,]+ tokens?/i,
+  /input \(\d+ tokens\) is longer than the model'?s context length \(\d+ tokens\)/i,
  /exceeds the limit of \d+/i,
  /exceeds the available context size/i,
  /greater than the context length/i,
@@ -34,11 +38,17 @@ const patterns = [
  /input length.*exceeds.*context length/i,
  /prompt too long; exceeded (?:max )?context length/i,
  /too large for model with \d+ maximum context length/i,
+  /prompt has [\d,]+ tokens?, but the configured context size is [\d,]+ tokens?/i,
  /model_context_window_exceeded/i,
+  /too many tokens/i,
+  /token limit exceeded/i,
 ]

+const exclusions = [/^(throttling error|service unavailable):/i, /rate limit/i, /too many requests/i]
+
 export const isContextOverflow = (message: string) =>
-  patterns.some((pattern) => pattern.test(message)) || /^4(00|13)\s*(status code)?\s*\(no body\)/i.test(message)
+  !exclusions.some((pattern) => pattern.test(message)) &&
+  (patterns.some((pattern) => pattern.test(message)) || /^4(00|13)\s*(status code)?\s*\(no body\)/i.test(message))

 export const isContextOverflowFailure = (failure: unknown) =>
  failure instanceof LLMError
@@ -0,0 +1 @@
+export * from "./providers/index"
@@ -0,0 +1,81 @@
+import type { ProviderPackage } from "../provider-package"
+import { OpenAICompatibleChat } from "../protocols/openai-compatible-chat"
+import type { RouteDefaultsInput } from "../route/client"
+import { ProviderID, type ModelID, type ProviderOptions } from "../schema"
+import { GoogleVertexShared } from "./google-vertex-shared"
+
+export const id = ProviderID.make("google-vertex")
+
+export type Config = RouteDefaultsInput &
+  GoogleVertexShared.OAuthOptions & {
+    readonly baseURL?: string
+    readonly location?: string
+    readonly project?: string
+  }
+
+export interface Settings extends ProviderPackage.Settings {
+  readonly accessToken?: string
+  readonly apiKey?: never
+  readonly baseURL?: string
+  readonly location?: string
+  readonly project?: string
+  readonly providerOptions?: ProviderOptions
+}
+
+const route = OpenAICompatibleChat.route.with({
+  id: "google-vertex-chat",
+  provider: id,
+})
+
+export const routes = [route]
+
+const configuredRoute = (input: Config) => {
+  if ("apiKey" in input && input.apiKey !== undefined) throw new Error("Google Vertex Chat does not support API keys")
+  const {
+    accessToken: _accessToken,
+    auth: _auth,
+    baseURL,
+    location: inputLocation,
+    project: inputProject,
+    ...rest
+  } = input
+  const location = GoogleVertexShared.location(inputLocation, "global")
+  const project = GoogleVertexShared.project(inputProject)
+  return route.with({
+    ...rest,
+    endpoint: {
+      baseURL:
+        baseURL ??
+        `https://aiplatform.googleapis.com/v1/projects/${GoogleVertexShared.requireProject(project)}/locations/${location}/endpoints/openapi`,
+    },
+    auth: GoogleVertexShared.oauth(input, project),
+  })
+}
+
+export const configure = (input: Config = {}) => {
+  const route = configuredRoute(input)
+  return {
+    id,
+    model: (modelID: string | ModelID) => route.model({ id: modelID }),
+    configure,
+  }
+}
+
+export const provider = {
+  id,
+  configure,
+}
+
+export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) => {
+  if (settings.apiKey !== undefined) throw new Error("Google Vertex Chat does not support API keys")
+  return configure({
+    accessToken: settings.accessToken,
+    baseURL: settings.baseURL,
+    headers: settings.headers === undefined ? undefined : { ...settings.headers },
+    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
+    limits: settings.limits,
+    location: settings.location,
+    project: settings.project,
+    providerOptions: settings.providerOptions,
+  }).model(modelID)
+}
@@ -0,0 +1,111 @@
+import { Effect, Schema, Struct } from "effect"
+import type { ProviderPackage } from "../provider-package"
+import { AnthropicMessages } from "../protocols/anthropic-messages"
+import { Auth } from "../route/auth"
+import { Route, type RouteDefaultsInput } from "../route/client"
+import { Endpoint } from "../route/endpoint"
+import { Framing } from "../route/framing"
+import { Protocol } from "../route/protocol"
+import { ProviderID, type ModelID, type ProviderOptions } from "../schema"
+import { GoogleVertexShared } from "./google-vertex-shared"
+
+const VERSION = "vertex-2023-10-16" as const
+
+// models.dev uses this provider id even though the API contract is Anthropic Messages.
+export const id = ProviderID.make("google-vertex-anthropic")
+
+export type Config = RouteDefaultsInput &
+  GoogleVertexShared.OAuthOptions & {
+    readonly baseURL?: string
+    readonly location?: string
+    readonly project?: string
+  }
+
+export interface Settings extends ProviderPackage.Settings {
+  readonly accessToken?: string
+  readonly apiKey?: never
+  readonly baseURL?: string
+  readonly location?: string
+  readonly project?: string
+  readonly providerOptions?: ProviderOptions
+}
+
+const route = Route.make({
+  id: "google-vertex-messages",
+  provider: id,
+  providerMetadataKey: "anthropic",
+  protocol: Protocol.make({
+    id: AnthropicMessages.protocol.id,
+    body: {
+      schema: Schema.Struct({
+        ...Struct.omit(AnthropicMessages.AnthropicMessagesBody.fields, ["model"]),
+        anthropic_version: Schema.Literal(VERSION),
+      }),
+      from: (request) =>
+        AnthropicMessages.protocol.body.from(request).pipe(
+          Effect.map((body) => ({
+            ...Struct.omit(body, ["model"]),
+            anthropic_version: VERSION,
+          })),
+        ),
+    },
+    stream: AnthropicMessages.protocol.stream,
+  }),
+  endpoint: Endpoint.path(({ request }) => `/${request.model.id}:streamRawPredict`),
+  auth: Auth.none,
+  framing: Framing.sse,
+})
+
+export const routes = [route]
+
+const configuredRoute = (input: Config) => {
+  if ("apiKey" in input && input.apiKey !== undefined)
+    throw new Error("Google Vertex Messages does not support API keys")
+  const {
+    accessToken: _accessToken,
+    auth: _auth,
+    baseURL,
+    location: inputLocation,
+    project: inputProject,
+    ...rest
+  } = input
+  const location = GoogleVertexShared.location(inputLocation, "global")
+  const project = GoogleVertexShared.project(inputProject)
+  return route.with({
+    ...rest,
+    endpoint: {
+      baseURL:
+        baseURL ??
+        `https://${GoogleVertexShared.host(location)}/v1/projects/${GoogleVertexShared.requireProject(project)}/locations/${location}/publishers/anthropic/models`,
+    },
+    auth: GoogleVertexShared.oauth(input, project),
+  })
+}
+
+export const configure = (input: Config = {}) => {
+  const route = configuredRoute(input)
+  return {
+    id,
+    model: (modelID: string | ModelID) => route.model({ id: modelID }),
+    configure,
+  }
+}
+
+export const provider = {
+  id,
+  configure,
+}
+
+export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) => {
+  if (settings.apiKey !== undefined) throw new Error("Google Vertex Messages does not support API keys")
+  return configure({
+    accessToken: settings.accessToken,
+    baseURL: settings.baseURL,
+    headers: settings.headers === undefined ? undefined : { ...settings.headers },
+    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
+    limits: settings.limits,
+    location: settings.location,
+    project: settings.project,
+    providerOptions: settings.providerOptions,
+  }).model(modelID)
+}
@@ -1,10 +1,10 @@
 import type { ProviderPackage } from "../provider-package"
-import { GoogleVertexAnthropic } from "../protocols/google-vertex-anthropic"
+import { OpenAICompatibleResponses } from "../protocols/openai-compatible-responses"
 import type { RouteDefaultsInput } from "../route/client"
 import { ProviderID, type ModelID, type ProviderOptions } from "../schema"
 import { GoogleVertexShared } from "./google-vertex-shared"

-export const id = ProviderID.make("google-vertex-anthropic")
+export const id = ProviderID.make("google-vertex")

 export type Config = RouteDefaultsInput &
  GoogleVertexShared.OAuthOptions & {
@@ -22,11 +22,16 @@ export interface Settings extends ProviderPackage.Settings {
  readonly providerOptions?: ProviderOptions
 }

-export const routes = [GoogleVertexAnthropic.route]
+const route = OpenAICompatibleResponses.route.with({
+  id: "google-vertex-responses",
+  provider: id,
+})
+
+export const routes = [route]

 const configuredRoute = (input: Config) => {
  if ("apiKey" in input && input.apiKey !== undefined)
-    throw new Error("Google Vertex Anthropic does not support API keys")
+    throw new Error("Google Vertex Responses does not support API keys")
  const {
    accessToken: _accessToken,
    auth: _auth,
@@ -37,12 +42,12 @@ const configuredRoute = (input: Config) => {
  } = input
  const location = GoogleVertexShared.location(inputLocation, "global")
  const project = GoogleVertexShared.project(inputProject)
-  return GoogleVertexAnthropic.route.with({
+  return route.with({
    ...rest,
    endpoint: {
      baseURL:
        baseURL ??
-        `https://${GoogleVertexShared.host(location)}/v1/projects/${GoogleVertexShared.requireProject(project)}/locations/${location}/publishers/anthropic/models`,
+        `https://aiplatform.googleapis.com/v1/projects/${GoogleVertexShared.requireProject(project)}/locations/${location}/endpoints/openapi`,
    },
    auth: GoogleVertexShared.oauth(input, project),
  })
@@ -63,7 +68,7 @@ export const provider = {
 }

 export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) => {
-  if (settings.apiKey !== undefined) throw new Error("Google Vertex Anthropic does not support API keys")
+  if (settings.apiKey !== undefined) throw new Error("Google Vertex Responses does not support API keys")
  return configure({
    accessToken: settings.accessToken,
    baseURL: settings.baseURL,
@@ -6,7 +6,7 @@ const SCOPE = "https://www.googleapis.com/auth/cloud-platform"

 export type OAuthOptions =
  | { readonly accessToken?: string; readonly auth?: never }
-  | { readonly accessToken?: never; readonly auth?: Auth }
+  | { readonly accessToken?: never; readonly auth?: Auth.Definition }

 export type ApiKeyOptions =
  | (OAuthOptions & { readonly apiKey?: never })
@@ -1,7 +1,9 @@
 import type { ProviderPackage } from "../provider-package"
-import { GoogleVertexGemini } from "../protocols/google-vertex-gemini"
+import { Gemini } from "../protocols/gemini"
 import { Auth } from "../route/auth"
-import type { RouteDefaultsInput } from "../route/client"
+import { Route, type RouteDefaultsInput } from "../route/client"
+import { Endpoint } from "../route/endpoint"
+import { Framing } from "../route/framing"
 import { ProviderID, type ModelID, type ProviderOptions } from "../schema"
 import { GoogleVertexShared } from "./google-vertex-shared"

@@ -25,7 +27,20 @@ export type Settings = ProviderPackage.Settings &
    readonly providerOptions?: ProviderOptions
  }

-export const routes = [GoogleVertexGemini.route]
+const route = Route.make({
+  id: "google-vertex-gemini",
+  provider: id,
+  providerMetadataKey: "google",
+  protocol: Gemini.protocol,
+  endpoint: Endpoint.path(({ request }) => {
+    const model = String(request.model.id)
+    return `/${model.startsWith("endpoints/") ? model : `models/${model}`}:streamGenerateContent?alt=sse`
+  }),
+  auth: Auth.none,
+  framing: Framing.sse,
+})
+
+export const routes = [route]

 const configuredRoute = (input: Config, modelID: string | ModelID) => {
  const {
@@ -48,7 +63,7 @@ const configuredRoute = (input: Config, modelID: string | ModelID) => {
    (apiKey
      ? "https://aiplatform.googleapis.com/v1/publishers/google"
      : `https://${GoogleVertexShared.host(location)}/v1beta1/projects/${GoogleVertexShared.requireProject(project)}/locations/${location}${endpointModel ? "" : "/publishers/google"}`)
-  return GoogleVertexGemini.route.with({
+  return route.with({
    ...rest,
    endpoint: { baseURL: endpoint },
    auth: apiKey === undefined ? GoogleVertexShared.oauth(input, project) : Auth.header("x-goog-api-key", apiKey),
@@ -0,0 +1,2 @@
+export { model } from "../google-vertex-chat"
+export type { Settings } from "../google-vertex-chat"
@@ -0,0 +1,2 @@
+export { model } from "../google-vertex"
+export type { Settings } from "../google-vertex"
@@ -0,0 +1,2 @@
+export { model } from "../google-vertex-messages"
+export type { Settings } from "../google-vertex-messages"
@@ -0,0 +1,2 @@
+export { model } from "../google-vertex-responses"
+export type { Settings } from "../google-vertex-responses"
@@ -2,14 +2,20 @@ import type { RouteDefaultsInput } from "../route/client"
 import { Auth } from "../route/auth"
 import type { ProviderAuthOption } from "../route/auth-options"
 import type { ProviderPackage } from "../provider-package"
-import { ProviderID, type ModelID, type ProviderOptions } from "../schema"
-import * as Gemini from "../protocols/gemini"
+import { HttpOptions, ProviderID, mergeHttpOptions, type ModelID, type ProviderOptions } from "../schema"
+import { Gemini } from "../protocols/gemini"
+import { GoogleImages } from "../protocols/google-images"
+
+export type { GoogleImageOptions } from "../protocols/google-images"

 export const id = ProviderID.make("google")

 export const routes = [Gemini.route]

-export type Config = RouteDefaultsInput & ProviderAuthOption<"optional"> & { readonly baseURL?: string }
+export type Config = RouteDefaultsInput &
+  ProviderAuthOption<"optional"> & {
+    readonly baseURL?: string
+  }

 export interface Settings extends ProviderPackage.Settings {
  readonly apiKey?: string
@@ -31,9 +37,18 @@ const configuredRoute = (input: Config) => {

 export const configure = (input: Config = {}) => {
  const route = configuredRoute(input)
+  const image = (modelID: string | ModelID) =>
+    GoogleImages.model({
+      id: modelID,
+      auth: auth(input),
+      baseURL: input.baseURL,
+      headers: input.headers,
+      http: mergeHttpOptions(input.http === undefined ? undefined : HttpOptions.make(input.http)),
+    })
  return {
    id,
    model: (modelID: string | ModelID) => route.model({ id: modelID }),
+    image,
    configure,
  }
 }
@@ -48,3 +63,5 @@ export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, se
    limits: settings.limits,
    providerOptions: settings.providerOptions,
  }).model(modelID)
+
+export const image = provider.image
@@ -7,9 +7,12 @@ export { CloudflareAIGateway, CloudflareWorkersAI } from "./cloudflare"
 export * as GitHubCopilot from "./github-copilot"
 export * as Google from "./google"
 export * as GoogleVertex from "./google-vertex"
-export * as GoogleVertexAnthropic from "./google-vertex-anthropic"
+export * as GoogleVertexChat from "./google-vertex-chat"
+export * as GoogleVertexMessages from "./google-vertex-messages"
+export * as GoogleVertexResponses from "./google-vertex-responses"
 export * as OpenAI from "./openai"
 export * as OpenAICompatible from "./openai-compatible"
 export * as OpenAICompatibleResponses from "./openai-compatible-responses"
 export * as OpenRouter from "./openrouter"
 export * as XAI from "./xai"
+export * as ZAI from "./zai"
@@ -0,0 +1 @@
+export * from "../openai-compatible-responses"
@@ -1,12 +1,14 @@
 import { AuthOptions, type ProviderAuthOption } from "../route/auth-options"
 import type { Route, RouteDefaultsInput } from "../route/client"
 import type { ProviderPackage } from "../provider-package"
-import { ProviderID, type ModelID } from "../schema"
+import { HttpOptions, ProviderID, ToolDefinition, mergeHttpOptions, type ModelID } from "../schema"
 import * as OpenAIChat from "../protocols/openai-chat"
 import * as OpenAIResponses from "../protocols/openai-responses"
 import { withOpenAIOptions, type OpenAIProviderOptionsInput } from "./openai-options"
+import { OpenAIImages, type OpenAIImageString } from "../protocols/openai-images"

 export type { OpenAIOptionsInput, OpenAIResponseIncludable } from "./openai-options"
+export type { OpenAIImageOptions } from "../protocols/openai-images"

 export const id = ProviderID.make("openai")

@@ -22,6 +24,39 @@ export type Config = RouteDefaultsInput &
    readonly providerOptions?: OpenAIProviderOptionsInput
  }

+export interface ImageGenerationOptions {
+  readonly action?: OpenAIImageString<"auto" | "generate" | "edit">
+  readonly background?: OpenAIImageString<"auto" | "opaque" | "transparent">
+  readonly inputFidelity?: OpenAIImageString<"low" | "high">
+  readonly outputCompression?: number
+  readonly outputFormat?: OpenAIImageString<"png" | "jpeg" | "webp">
+  readonly partialImages?: number
+  readonly quality?: OpenAIImageString<"auto" | "low" | "medium" | "high" | "standard" | "hd">
+  readonly size?: OpenAIImageString<
+    "auto" | "256x256" | "512x512" | "1024x1024" | "1536x1024" | "1024x1536" | "1792x1024" | "1024x1792"
+  >
+}
+
+export const imageGeneration = (options: ImageGenerationOptions = {}) =>
+  ToolDefinition.make({
+    name: "image_generation",
+    description: "Generate or edit an image using OpenAI's hosted image generation tool.",
+    inputSchema: { type: "object", properties: {}, additionalProperties: false },
+    native: {
+      openai: {
+        type: "image_generation",
+        action: options.action,
+        background: options.background,
+        input_fidelity: options.inputFidelity,
+        output_compression: options.outputCompression,
+        output_format: options.outputFormat,
+        partial_images: options.partialImages,
+        quality: options.quality,
+        size: options.size,
+      },
+    },
+  })
+
 export interface Settings extends ProviderPackage.Settings {
  readonly apiKey?: string
  readonly baseURL?: string
@@ -55,6 +90,17 @@ export const configure = (input: Config = {}) => {
  const responsesWebSocket = (id: string | ModelID) =>
    responsesWebSocketRoute.with(withOpenAIOptions(id, modelDefaults, { textVerbosity: true })).model({ id })
  const chat = (id: string | ModelID) => chatRoute.with(withOpenAIOptions(id, modelDefaults)).model({ id })
+  const image = (modelID: string | ModelID) =>
+    OpenAIImages.model({
+      id: modelID,
+      auth: auth(input),
+      baseURL: input.baseURL,
+      headers: input.headers,
+      http: mergeHttpOptions(
+        input.http === undefined ? undefined : HttpOptions.make(input.http),
+        input.queryParams === undefined ? undefined : new HttpOptions({ query: input.queryParams }),
+      ),
+    })

  return {
    id,
@@ -62,6 +108,7 @@ export const configure = (input: Config = {}) => {
    responses,
    responsesWebSocket,
    chat,
+    image,
    configure,
  }
 }
@@ -97,3 +144,4 @@ export const chatModel: ProviderPackage.Definition<Settings>["model"] = (modelID
 export const responses = provider.responses
 export const responsesWebSocket = provider.responsesWebSocket
 export const chat = provider.chat
+export const image = provider.image
@@ -41,13 +41,31 @@ export const protocol = Protocol.make({
    schema: OpenRouterBody,
    from: (request) =>
      OpenAIChat.protocol.body.from(request).pipe(
-        Effect.map(
-          (body) =>
-            ({
-              ...body,
-              ...bodyOptions(request.providerOptions?.openrouter),
-            }) as OpenRouterBody,
-        ),
+        Effect.map((body) => {
+          const sourceAssistants = request.messages.filter((message) => message.role === "assistant")
+          let assistantIndex = 0
+          const messages = body.messages.map((message) => {
+            if (message.role !== "assistant") return message
+            const source = sourceAssistants[assistantIndex++]
+            const reasoning = source?.content
+              .filter((part) => part.type === "reasoning")
+              .map((part) => part.text)
+              .join("")
+            const reasoningDetails = Array.isArray(message.reasoning_details) ? message.reasoning_details : undefined
+            return {
+              ...message,
+              reasoning_content: undefined,
+              reasoning_text: undefined,
+              reasoning: reasoning && reasoningDetails && reasoningDetails.length > 0 ? reasoning : undefined,
+              reasoning_details: reasoningDetails,
+            }
+          })
+          return {
+            ...body,
+            messages,
+            ...bodyOptions(request.providerOptions?.openrouter),
+          } as OpenRouterBody
+        }),
      ),
  },
  stream: OpenAIChat.protocol.stream,
@@ -1,9 +1,10 @@
 import { AuthOptions, type ProviderAuthOption } from "../route/auth-options"
 import type { RouteDefaultsInput } from "../route/client"
-import { ProviderID, type ModelID } from "../schema"
+import { HttpOptions, ProviderID, type ModelID } from "../schema"
 import * as OpenAICompatibleProfiles from "./openai-compatible-profile"
 import * as OpenAICompatibleChat from "../protocols/openai-compatible-chat"
 import * as OpenAIResponses from "../protocols/openai-responses"
+import { XAIImages } from "../protocols/xai-images"

 export const id = ProviderID.make("xai")

@@ -12,6 +13,8 @@ export type ModelOptions = RouteDefaultsInput &
    readonly baseURL?: string
  }

+export type { XAIImageOptions } from "../protocols/xai-images"
+
 export const routes = [OpenAIResponses.route, OpenAICompatibleChat.route]

 const auth = (options: ProviderAuthOption<"optional">) => AuthOptions.bearer(options, "XAI_API_KEY")
@@ -41,11 +44,20 @@ export const configure = (input: ModelOptions = {}) => {
  const chatRoute = configuredChatRoute(input)
  const responses = (modelID: string | ModelID) => responsesRoute.model({ id: modelID })
  const chat = (modelID: string | ModelID) => chatRoute.model({ id: modelID })
+  const image = (modelID: string | ModelID) =>
+    XAIImages.model({
+      id: modelID,
+      auth: auth(input),
+      baseURL: input.baseURL ?? OpenAICompatibleProfiles.profiles.xai.baseURL,
+      headers: input.headers,
+      http: input.http === undefined ? undefined : HttpOptions.make(input.http),
+    })
  return {
    id,
    model: responses,
    responses,
    chat,
+    image,
    configure,
  }
 }
@@ -54,3 +66,4 @@ export const provider = configure()
 export const model = provider.model
 export const responses = provider.responses
 export const chat = provider.chat
+export const image = provider.image
@@ -0,0 +1,35 @@
+import { ZAIImages } from "../protocols/zai-images"
+import { AuthOptions, type ProviderAuthOption } from "../route/auth-options"
+import { HttpOptions, ProviderID, type ModelID } from "../schema"
+
+export const id = ProviderID.make("zai")
+
+export type Config = ProviderAuthOption<"optional"> & {
+  readonly baseURL?: string
+  readonly headers?: Record<string, string>
+  readonly http?: HttpOptions.Input
+}
+
+export type { ZAIImageOptions } from "../protocols/zai-images"
+
+const auth = (options: ProviderAuthOption<"optional">) => AuthOptions.bearer(options, "ZAI_API_KEY")
+
+export const configure = (input: Config = {}) => {
+  const image = (modelID: string | ModelID) =>
+    ZAIImages.model({
+      id: modelID,
+      auth: auth(input),
+      baseURL: input.baseURL,
+      headers: input.headers,
+      http: input.http === undefined ? undefined : HttpOptions.make(input.http),
+    })
+
+  return {
+    id,
+    image,
+    configure,
+  }
+}
+
+export const provider = configure()
+export const image = provider.image
@@ -0,0 +1 @@
+export * from "./route/index"
@@ -4,7 +4,7 @@ import { Auth } from "./auth"
 export type ApiKeyMode = "optional" | "required"

 export type AuthOverride = {
-  readonly auth: Auth
+  readonly auth: Auth.Definition
  readonly apiKey?: never
 }

@@ -44,7 +44,10 @@ export type AtLeastOne<T> = {
 * override, otherwise resolve `apiKey` (option > config var) and apply it as
 * a bearer token.
 */
-export const bearer = (options: ProviderAuthOption<"optional">, envVar: string | ReadonlyArray<string>): Auth => {
+export const bearer = (
+  options: ProviderAuthOption<"optional">,
+  envVar: string | ReadonlyArray<string>,
+): Auth.Definition => {
  if ("auth" in options && options.auth) return options.auth
  return (Array.isArray(envVar) ? envVar : [envVar])
    .reduce(
@@ -1,6 +1,6 @@
 import { Config, Effect, Redacted } from "effect"
 import { Headers } from "effect/unstable/http"
-import { AuthenticationReason, InvalidRequestReason, LLMError, type LLMRequest } from "../schema"
+import { AuthenticationReason, InvalidRequestReason, LLMError, type HttpOptions } from "../schema"

 export class MissingCredentialError extends Error {
  readonly _tag = "MissingCredentialError"
@@ -15,7 +15,7 @@ export type AuthError = CredentialError | LLMError
 type Secret = string | Redacted.Redacted | Config.Config<string | Redacted.Redacted>

 export interface AuthInput {
-  readonly request: LLMRequest
+  readonly request: { readonly http?: HttpOptions }
  readonly method: "POST" | "GET"
  readonly url: string
  readonly body: string
@@ -25,19 +25,19 @@ export interface AuthInput {
 export interface Credential {
  readonly load: Effect.Effect<Redacted.Redacted, CredentialError>
  readonly orElse: (that: Credential) => Credential
-  readonly bearer: () => Auth
-  readonly header: (name: string) => Auth
+  readonly bearer: () => Definition
+  readonly header: (name: string) => Definition
  readonly pipe: <A>(f: (self: Credential) => A) => A
 }

-export interface Auth {
+export interface Definition {
  readonly apply: (input: AuthInput) => Effect.Effect<Headers.Headers, AuthError>
-  readonly andThen: (that: Auth) => Auth
-  readonly orElse: (that: Auth) => Auth
-  readonly pipe: <A>(f: (self: Auth) => A) => A
+  readonly andThen: (that: Definition) => Definition
+  readonly orElse: (that: Definition) => Definition
+  readonly pipe: <A>(f: (self: Definition) => A) => A
 }

-export const isAuth = (input: unknown): input is Auth =>
+export const isAuth = (input: unknown): input is Definition =>
  typeof input === "object" && input !== null && "apply" in input && typeof input.apply === "function"

 const credential = (load: Effect.Effect<Redacted.Redacted, CredentialError>): Credential => {
@@ -51,8 +51,8 @@ const credential = (load: Effect.Effect<Redacted.Redacted, CredentialError>): Cr
  return self
 }

-const auth = (apply: Auth["apply"]): Auth => {
-  const self: Auth = {
+const auth = (apply: Definition["apply"]): Definition => {
+  const self: Definition = {
    apply,
    andThen: (that) =>
      auth((input) => apply(input).pipe(Effect.flatMap((headers) => that.apply({ ...input, headers })))),
@@ -109,15 +109,15 @@ const credentialInput = (source: Secret | Credential) =>
    ? credentialFromSecret(source, "value")
    : source

-export function bearer(source: Secret | Credential): Auth
+export function bearer(source: Secret | Credential): Definition
 export function bearer(source: Secret | Credential) {
  return credentialInput(source).bearer()
 }

 export const apiKey = bearer

-export function header(name: string): (source: Secret | Credential) => Auth
-export function header(name: string, source: Secret | Credential): Auth
+export function header(name: string): (source: Secret | Credential) => Definition
+export function header(name: string, source: Secret | Credential): Definition
 export function header(name: string, source?: Secret | Credential) {
  if (source === undefined) {
    return (next: Secret | Credential) => credentialInput(next).header(name)
@@ -125,8 +125,8 @@ export function header(name: string, source?: Secret | Credential) {
  return credentialInput(source).header(name)
 }

-export function bearerHeader(name: string): (source: Secret | Credential) => Auth
-export function bearerHeader(name: string, source: Secret | Credential): Auth
+export function bearerHeader(name: string): (source: Secret | Credential) => Definition
+export function bearerHeader(name: string, source: Secret | Credential): Definition
 export function bearerHeader(name: string, source?: Secret | Credential) {
  const render = (input: Secret | Credential) =>
    fromCredential(credentialInput(input), (secret) => ({ [name]: `Bearer ${secret}` }))
@@ -149,7 +149,7 @@ const toLLMError = (error: AuthError): LLMError => {
 }

 export const toEffect =
-  (input: Auth) =>
+  (input: Definition) =>
  (authInput: AuthInput): Effect.Effect<Headers.Headers, LLMError> =>
    input.apply(authInput).pipe(Effect.mapError(toLLMError))

@@ -1,9 +1,9 @@
 import { Cause, Context, Effect, Layer, Schema, Stream } from "effect"
 import * as Option from "effect/Option"
-import { Auth, type Auth as AuthDef } from "./auth"
+import { Auth } from "./auth"
 import { Endpoint, type EndpointPatch } from "./endpoint"
 import { RequestExecutor } from "./executor"
-import type { Framing } from "./framing"
+import { Framing } from "./framing"
 import { HttpTransport } from "./transport"
 import type { Transport, TransportRuntime } from "./transport"
 import { WebSocketExecutor } from "./transport"
@@ -40,8 +40,8 @@ export interface Route<Body, Prepared = unknown> {
  /** ProviderMetadata namespace emitted and consumed by this route. */
  readonly providerMetadataKey?: string
  readonly protocol: ProtocolID
-  readonly endpoint: Endpoint<Body>
-  readonly auth: AuthDef
+  readonly endpoint: Endpoint.Definition<Body>
+  readonly auth: Auth.Definition
  readonly transport: Transport<Body, Prepared, unknown>
  readonly defaults: RouteDefaults
  readonly body: RouteBody<Body>
@@ -86,7 +86,7 @@ export interface RouteDefaultsInput {
 export interface RoutePatch<Body, Prepared> extends RouteDefaultsInput {
  readonly id?: string
  readonly provider?: string | ProviderID
-  readonly auth?: AuthDef
+  readonly auth?: Auth.Definition
  readonly transport?: Transport<Body, Prepared, unknown>
  readonly endpoint?: EndpointPatch<Body>
 }
@@ -122,7 +122,7 @@ const mergeRouteDefaults = (base: RouteDefaults | undefined, patch: RouteDefault
  }
 }

-const endpointBaseURL = <Body>(endpoint: Endpoint<Body>) =>
+const endpointBaseURL = <Body>(endpoint: Endpoint.Definition<Body>) =>
  typeof endpoint.baseURL === "string" ? endpoint.baseURL : undefined

 const mergeHeaders = (...items: ReadonlyArray<Record<string, string> | undefined>) => {
@@ -192,11 +192,11 @@ export interface MakeInput<Body, Frame, Event, State> {
  /** Semantic API contract — owns body construction, body schema, and parsing. */
  readonly protocol: Protocol<Body, Frame, Event, State>
  /** Where the request is sent. */
-  readonly endpoint: Endpoint<Body>
+  readonly endpoint: Endpoint.Definition<Body>
  /** Per-request transport auth. Provider facades override this via `route.with(...)`. */
-  readonly auth?: AuthDef
+  readonly auth?: Auth.Definition
  /** Stream framing — bytes -> frames before `protocol.stream.event` decoding. */
-  readonly framing: Framing<Frame>
+  readonly framing: Framing.Definition<Frame>
  /** Static / per-request headers added before `auth` runs. */
  readonly headers?: (input: { readonly request: LLMRequest }) => Record<string, string>
  /** Route/request defaults used when compiling requests for this route. */
@@ -213,9 +213,9 @@ export interface MakeTransportInput<Body, Prepared, Frame, Event, State> {
  /** Semantic API contract — owns body construction, body schema, and parsing. */
  readonly protocol: Protocol<Body, Frame, Event, State>
  /** Where the request is sent. */
-  readonly endpoint: Endpoint<Body>
+  readonly endpoint: Endpoint.Definition<Body>
  /** Per-request transport auth. Provider facades override this via `route.with(...)`. */
-  readonly auth?: AuthDef
+  readonly auth?: Auth.Definition
  /** Static / per-request headers added before `auth` runs. */
  readonly headers?: (input: { readonly request: LLMRequest }) => Record<string, string>
  /** Runnable transport route. */
@@ -19,21 +19,24 @@ export type EndpointPart<Body> = string | ((input: EndpointInput<Body>) => strin
 * URL embeds the model id, region, or another body field (e.g. Bedrock,
 * Gemini).
 */
-export interface Endpoint<Body> {
+export interface Definition<Body> {
  readonly baseURL?: string
  readonly path: EndpointPart<Body>
  readonly query?: Record<string, string>
 }

-export type EndpointPatch<Body> = Partial<Endpoint<Body>>
+export type EndpointPatch<Body> = Partial<Definition<Body>>

 /** Construct an `Endpoint` from a path string or path function. */
-export const path = <Body>(value: EndpointPart<Body>, options: Omit<Endpoint<Body>, "path"> = {}): Endpoint<Body> => ({
+export const path = <Body>(
+  value: EndpointPart<Body>,
+  options: Omit<Definition<Body>, "path"> = {},
+): Definition<Body> => ({
  ...options,
  path: value,
 })

-export const merge = <Body>(base: Endpoint<Body>, patch: EndpointPatch<Body>): Endpoint<Body> => ({
+export const merge = <Body>(base: Definition<Body>, patch: EndpointPatch<Body>): Definition<Body> => ({
  ...base,
  ...patch,
  baseURL: patch.baseURL ?? base.baseURL,
@@ -44,7 +47,7 @@ export const merge = <Body>(base: Endpoint<Body>, patch: EndpointPatch<Body>): E
 const renderPart = <Body>(part: EndpointPart<Body>, input: EndpointInput<Body>) =>
  typeof part === "function" ? part(input) : part

-export const render = <Body>(endpoint: Endpoint<Body>, input: EndpointInput<Body>) => {
+export const render = <Body>(endpoint: Definition<Body>, input: EndpointInput<Body>) => {
  const url = new URL(`${ProviderShared.trimBaseUrl(endpoint.baseURL ?? "")}${renderPart(endpoint.path, input)}`)
  for (const [key, value] of Object.entries(endpoint.query ?? {})) url.searchParams.set(key, value)
  return url
@@ -16,12 +16,12 @@ import type { LLMError } from "../schema"
 * The frame type is opaque to this layer; the protocol's `decode` step turns
 * a frame into a typed chunk.
 */
-export interface Framing<Frame> {
+export interface Definition<Frame> {
  readonly id: string
  readonly frame: (bytes: Stream.Stream<Uint8Array, LLMError>) => Stream.Stream<Frame, LLMError>
 }

 /** Server-Sent Events framing. Used by every JSON-streaming HTTP provider. */
-export const sse: Framing<string> = { id: "sse", frame: ProviderShared.sseFraming }
+export const sse: Definition<string> = { id: "sse", frame: ProviderShared.sseFraming }

 export * as Framing from "./framing"
@@ -17,9 +17,9 @@ export { Framing } from "./framing"
 export { Protocol } from "./protocol"
 export { HttpTransport, WebSocketExecutor, WebSocketTransport } from "./transport"
 export * as Transport from "./transport"
-export type { Auth as AuthShape, AuthInput, Credential, CredentialError } from "./auth"
+export type { Definition as AuthShape, AuthInput, Credential, CredentialError } from "./auth"
 export type { ApiKeyMode, AuthOverride, ProviderAuthOption } from "./auth-options"
-export type { Endpoint as EndpointFn, EndpointInput } from "./endpoint"
-export type { Framing as FramingDef } from "./framing"
+export type { Definition as EndpointFn, EndpointInput } from "./endpoint"
+export type { Definition as FramingDef } from "./framing"
 export type { Protocol as ProtocolDef } from "./protocol"
 export type { Transport as TransportDef, TransportRuntime } from "./transport"
@@ -2,7 +2,7 @@ import { Effect, Stream } from "effect"
 import { Headers, HttpClientRequest } from "effect/unstable/http"
 import { Auth } from "../auth"
 import { render as renderEndpoint } from "../endpoint"
-import { Framing, type Framing as FramingDef } from "../framing"
+import { Framing } from "../framing"
 import type { Transport, TransportPrepareInput } from "./index"
 import * as ProviderShared from "../../protocols/shared"
 import { mergeJsonRecords, type LLMRequest } from "../../schema"
@@ -18,7 +18,7 @@ export interface JsonRequestParts<Body = unknown> {

 export interface HttpPrepared<Frame> {
  readonly request: HttpClientRequest.HttpClientRequest
-  readonly framing: FramingDef<Frame>
+  readonly framing: Framing.Definition<Frame>
 }

 const applyQuery = (url: string, query: Record<string, string> | undefined) => {
@@ -107,7 +107,7 @@ export const jsonRequestParts = <Body>(input: JsonRequestInput<Body>) =>
  })

 export interface HttpJsonInput<_Body, Frame> {
-  readonly framing: FramingDef<Frame>
+  readonly framing: Framing.Definition<Frame>
 }

 export type HttpJsonPatch<Body, Frame> = Partial<HttpJsonInput<Body, Frame>>
@@ -1,6 +1,6 @@
 import type { Effect, Stream } from "effect"
-import type { Endpoint } from "../endpoint"
-import type { Auth } from "../auth"
+import { Endpoint } from "../endpoint"
+import { Auth } from "../auth"
 import type { Interface as RequestExecutorInterface } from "../executor"
 import type { Interface as WebSocketExecutorInterface } from "./websocket"
 import type { LLMError, LLMRequest } from "../../schema"
@@ -23,8 +23,8 @@ export interface Transport<Body, Prepared, Frame> {
 export interface TransportPrepareInput<Body> {
  readonly body: Body
  readonly request: LLMRequest
-  readonly endpoint: Endpoint<Body>
-  readonly auth: Auth
+  readonly endpoint: Endpoint.Definition<Body>
+  readonly auth: Auth.Definition
  readonly encodeBody: (body: Body) => string
  readonly headers?: (input: { readonly request: LLMRequest }) => Record<string, string>
 }
--- a/Show More
+++ b/Show More
				`@@ -0,0 +1 @@`
				`export * from "../openai-compatible-responses"`