tui: remove the win32 ENABLE_PROCESSED_INPUT ffi shim (#43604)

Signed-off-by: Major Hayden <major@mhtx.net>

.changeset/idempotent-prompt-reuse.md

@@ -0,0 +1,5 @@
+---
+"@opencode-ai/core": patch
+---
+
+Prompt and synthetic inbox ID reuse is now idempotent: reusing an ID within the same Session succeeds and returns the first admission, ignoring the retried payload, metadata, and delivery mode. Previously reuse with a differing payload failed with a conflict. Cross-Session and cross-type reuse still fail, and control items keep their operation-specific conflict behavior.

.changeset/share-session-model-request.md

@@ -0,0 +1,5 @@
+---
+"@opencode-ai/core": patch
+---
+
+Apply shared Session model-request preparation to transient generation.

.changeset/steer-resume-simplified.md

@@ -0,0 +1,5 @@
+---
+"@opencode-ai/core": patch
+---
+
+Simplify interrupt continuation: the steer-scoped resume decision now lives in SessionExecution as a post-cleanup inbox check, and the run coordinator drops its continuation state machine. Wakes arriving during cancellation cleanup now restart a normal full drain, and interrupting an idle session with continue now resumes pending steering input. Recovery-applied moves now end with the same full wake as inbox-admitted moves, retrying any stranded inbox work at the new location. Interrupting with continue now also resumes a next-in-line control item: between-turn manual compaction and moves run under any drain scope, while queued prompts remain parked.

.github/workflows/nix-hashes.yml

@@ -6,7 +6,7 @@ permissions:
 on:
   workflow_dispatch:
   push:
-    branches: [dev, beta]
+    branches: [dev, beta, v2]
     paths:
       - "bun.lock"
       - "package.json"

.opencode/skills/opencode-dev/SKILL.md

@@ -185,6 +185,33 @@ pmap -x <pid> | sort -k3 -nr | head -25
 
 Heap serialization itself can temporarily increase RSS and allocator high-water marks, so record `ps`/`smaps_rollup` both before and after capture. Large anonymous mappings with a comparatively small live heap require native-allocation or allocator investigation; they cannot be explained from JavaScript retainer paths alone.
 
+## CPU profiles
+
+The CLI installs a `SIGPROF` listener on non-Windows processes in `packages/cli/src/cpu-profile.ts`. One signal starts a ten-second CPU profile and stops it automatically; additional signals are ignored while a profile is active. There is no CPU profile CLI flag or environment variable.
+
+1. Get the PID from the health endpoint. For shared-service performance, target the server PID returned here rather than the short wrapper or TUI process:
+
+```bash
+opencode2 api get /api/health
+```
+
+Use `bun dev api get /api/health` instead when targeting the local/dev channel.
+
+2. Start the capture:
+
+```bash
+kill -PROF <server-pid>
+```
+
+3. Wait for `CPU profile written` in the channel's log before opening the file. Profiles are written to the same log directory as `cpu-<pid>-<timestamp>.cpuprofile`; the log's `path=` field is authoritative:
+
+```bash
+grep 'CPU profile' ~/.local/share/opencode/log/opencode.log | tail
+find ~/.local/share/opencode/log -maxdepth 1 -name 'cpu-<server-pid>-*.cpuprofile' -printf '%T@ %s %p\n' | sort -nr | head
+```
+
+Use `opencode-local.log` for a local/dev process. Load the completed `.cpuprofile` in Chrome DevTools or another V8 CPU profile viewer and inspect the hottest functions, call stacks, and self time during the controlled workload.
+
 ## Debugger
 
 - To debug the V2 CLI or TUI with Bun's inspector, launch the CLI entrypoint through Terminal Control with an inspector URL, then attach a debugger to that URL:

AGENTS.md

@@ -176,7 +176,7 @@ const table = sqliteTable("session", {
 
 - Keep durable events minimal: record irreducible new facts and do not repeat state derivable by folding the ordered aggregate history. Enrich projections and read models with previous or derived state when consumers need self-contained views.
 - Keep durable prompt admission separate from model execution. `Session.prompt(...)` publishes `session.inbox.enqueued`, whose projection inserts one durable `session_inbox` row, before scheduling advisory `SessionExecution.wake(sessionID)` unless `resume: false` requests admit-only behavior. Delivery publishes `session.inbox.delivered`; its projection consumes the inbox row and inserts the visible message in the same transaction. `session_inbox` stores only unconsumed work.
-- Reusing a Session ID adopts the existing Session. While a user or synthetic inbox item is pending, reusing its ID reconciles only when Session, type, complete payload, metadata, and delivery match; conflicting reuse fails. Once delivered, retry reconciliation for those message-producing items uses the projected message and does not require retained enqueue history or the original delivery mode. Control items keep their operation-specific conflict behavior.
+- Reusing a Session ID adopts the existing Session. Reusing a user or synthetic inbox item ID is idempotent when Session and type match: the first admission wins and the retried payload, metadata, and delivery mode are ignored, whether the item is still pending or already delivered (reconciled from the projected message without retained enqueue history). Cross-Session or cross-type reuse fails. Control items keep their operation-specific conflict behavior.
 - Keep `SessionExecution` process-global and Session-ID based. Its local implementation owns the process-local Session coordinator and discovers placement through `SessionStore` plus `LocationServiceMap.get(session.location)` only when a drain starts; no layer should take a Session ID. V2 interruption targets the active process-local ownership chain for that Session; interruption of a known but idle or locally unowned Session is a no-op, while the public API rejects an unknown Session.
 - Keep `SessionRunner`, model resolution, tool registry, permissions, and filesystem Location-scoped. Omitted `Location.workspaceID` means implicit-local placement; explicit workspace identity remains reserved for future placement semantics.
 - Preserve one explicit `llm.stream(request)` call per Physical Attempt and reload projected history before durable continuation. A logical Step may use generic pre-output retries, one full-context retry after continuation rejection, incomplete-stream continuation, or one overflow-compaction rebuild. Generic retries retain the logical step number and do not consume another agent-step allowance. Do not delegate orchestration to an in-memory tool loop.

nix/hashes.json

@@ -1,8 +1,8 @@
 {
   "nodeModules": {
-    "x86_64-linux": "sha256-uduwrM143NDSc+tXsi4lVVfoMll2a3BDHRUjuO7GB68=",
+    "x86_64-linux": "sha256-JEqi00PCle+o5OfBlJJaZtXd+4sYB3o+rvYiESlN4dY=",
-    "aarch64-linux": "sha256-6DUda78XdXY6DP86lIUkweSjys3iG4Y4mo1PiaNuXbg=",
+    "aarch64-linux": "sha256-zk3Uk1SQyeRrQ7BuFwlOnQAptUHIkq+oPdfd+sTEq5U=",
-    "aarch64-darwin": "sha256-AkJwfLULLZVwwz+XU1QcFUZoIS7oVPCn+n/MXEaxrqE=",
+    "aarch64-darwin": "sha256-3BOd3EcqimoG3rTI6lTHe91YVlCoEi8/68eT1lbOi0c=",
-    "x86_64-darwin": "sha256-hAxKGdiITTxQ2uujQt6prNjo3NxGAMMeo+9HlMWK6GU="
+    "x86_64-darwin": "sha256-X7wGmjiMloF5Zhuc20kAxLC+tl613YNXRgA+dQjP2WM="
   }
 }

nix/node_modules.nix

@@ -10,7 +10,7 @@
     ]).nodeModules.${stdenvNoCC.hostPlatform.system},
 }:
 let
-  packageJson = lib.pipe ../packages/opencode/package.json [
+  packageJson = lib.pipe ../packages/cli/package.json [
     builtins.readFile
     builtins.fromJSON
   ];
@@ -52,7 +52,7 @@ stdenvNoCC.mkDerivation {
       --cpu="${bunCpu}" \
       --os="${bunOs}" \
       --filter '!./' \
-      --filter './packages/opencode' \
+      --filter './packages/cli' \
       --filter './packages/desktop' \
       --filter './packages/app' \
       --frozen-lockfile \

nix/opencode.nix

@@ -48,13 +48,13 @@ stdenvNoCC.mkDerivation (finalAttrs: {
   env.OPENCODE_DISABLE_MODELS_FETCH = true;
   env.OPENCODE_VERSION = finalAttrs.version;
   env.OPENCODE_CHANNEL = "prod";
+  env.NODE_OPTIONS = "--max-old-space-size=4096";
 
   buildPhase = ''
     runHook preBuild
 
-    cd ./packages/opencode
+    cd ./packages/cli
     bun --bun ./script/build.ts --single --skip-install
-    bun --bun ./script/schema.ts schema.json
 
     runHook postBuild
   '';
@@ -62,10 +62,9 @@ stdenvNoCC.mkDerivation (finalAttrs: {
   installPhase = ''
     runHook preInstall
 
-    install -Dm755 dist/opencode-*/bin/opencode $out/bin/opencode
+    install -Dm755 dist/cli-*/bin/opencode2 $out/bin/opencode2
-    install -Dm644 schema.json $out/share/opencode/schema.json
 
-    wrapProgram $out/bin/opencode \
+    wrapProgram $out/bin/opencode2 \
       --prefix PATH : ${
         lib.makeBinPath (
           [
@@ -81,9 +80,9 @@ stdenvNoCC.mkDerivation (finalAttrs: {
 
   postInstall = lib.optionalString (stdenvNoCC.buildPlatform.canExecute stdenvNoCC.hostPlatform) ''
     # trick yargs into also generating zsh completions
-    installShellCompletion --cmd opencode \
+    installShellCompletion --cmd opencode2 \
-      --bash <($out/bin/opencode completion) \
+      --bash <($out/bin/opencode2 completion) \
-      --zsh <(SHELL=/bin/zsh $out/bin/opencode completion)
+      --zsh <(SHELL=/bin/zsh $out/bin/opencode2 completion)
   '';
 
   nativeInstallCheckInputs = [
@@ -95,7 +94,6 @@ stdenvNoCC.mkDerivation (finalAttrs: {
   versionCheckProgramArg = "--version";
 
   passthru = {
-    jsonschema = "${placeholder "out"}/share/opencode/schema.json";
     env = finalAttrs.env;
   };
 
@@ -103,7 +101,7 @@ stdenvNoCC.mkDerivation (finalAttrs: {
     description = "The open source coding agent";
     homepage = "https://opencode.ai";
     license = lib.licenses.mit;
-    mainProgram = "opencode";
+    mainProgram = "opencode2";
     inherit (node_modules.meta) platforms;
   };
 })

package.json

@@ -15,6 +15,7 @@
     "dev:stats": "bun sst shell --stage=production -- bun run --cwd packages/stats/app dev",
     "dev:www": "bun run --cwd packages/www dev",
     "dev:storybook": "bun --cwd packages/storybook storybook",
+    "bench:devex": "bun run --cwd packages/app test:bench:devex",
     "lint": "oxlint",
     "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",
@@ -37,10 +38,10 @@
       "packages/slack"
     ],
     "catalog": {
-      "@effect/opentelemetry": "4.0.0-beta.107",
+      "@effect/opentelemetry": "4.0.0-rc.110",
-      "@effect/platform-node": "4.0.0-beta.107",
+      "@effect/platform-node": "4.0.0-rc.110",
-      "@effect/platform-node-shared": "4.0.0-beta.107",
+      "@effect/platform-node-shared": "4.0.0-rc.110",
-      "@effect/sql-sqlite-bun": "4.0.0-beta.107",
+      "@effect/sql-sqlite-bun": "4.0.0-rc.110",
       "@npmcli/arborist": "9.4.0",
       "@types/bun": "1.3.13",
       "@types/cross-spawn": "6.0.6",
@@ -71,7 +72,7 @@
       "dompurify": "3.3.1",
       "drizzle-kit": "1.0.0-rc.2",
       "drizzle-orm": "1.0.0-rc.2",
-      "effect": "4.0.0-beta.107",
+      "effect": "4.0.0-rc.110",
       "ai": "6.0.168",
       "cross-spawn": "7.0.6",
       "hono": "4.10.7",

packages/ai/AGENTS.md

@@ -49,7 +49,7 @@ Filter or narrow `LLMEvent` streams with `LLMEvent.is.*` (camelCase guards, e.g.
 
 ### Routes
 
-A route is the registered, runnable composition of four orthogonal pieces:
+A route is the runnable composition of four orthogonal pieces:
 
 - **`Protocol`** (`src/route/protocol.ts`) — semantic API contract. Owns request body construction (`body.from`), the body schema (`body.schema`), the streaming-event schema (`stream.event`), and the event-to-`LLMEvent` state machine (`stream.step`). `Route.make(...)` validates and JSON-encodes the body from `body.schema` and decodes frames with `stream.event`. Examples: `OpenAIChat.protocol`, `OpenResponses.protocol`, `OpenAIResponses.protocol`, `AnthropicMessages.protocol`, `Gemini.protocol`, `BedrockConverse.protocol`.
 - **`Endpoint`** (`src/route/endpoint.ts`) — URL construction. The host, path, and route query live on the endpoint. `Endpoint.path("/chat/completions", { baseURL })` is the common case; pass a function for paths that embed the model id or a body field (e.g. `Endpoint.path(({ body }) => `/model/${body.modelId}/converse-stream`)`).
@@ -66,7 +66,7 @@ export const route = Route.make({
   endpoint: Endpoint.path("/chat/completions", {
     baseURL: "https://api.openai.com/v1",
   }),
-  auth: Auth.bearer(),
+  auth: Auth.bearer(Auth.config("OPENAI_API_KEY")),
   framing: Framing.sse,
 })
 ```
@@ -79,7 +79,7 @@ When a provider supports multiple physical transports, selection remains executi
 
 ### URL Construction
 
-`Endpoint` owns `{ baseURL, path, query }`. Each protocol route includes a canonical endpoint when the provider has one (e.g. `https://api.openai.com/v1`); provider helpers override endpoint fields by configuring the route before selecting a model. Routes that have no canonical URL (OpenAI-compatible Chat, GitHub Copilot) require configuration before execution.
+`Endpoint` owns `{ baseURL, path, query }`. Each protocol route includes a canonical endpoint when the provider has one (e.g. `https://api.openai.com/v1`); provider helpers override endpoint fields by configuring the route before selecting a model. Generic OpenAI-compatible routes have no canonical URL and require configuration before execution.
 
 For providers where the URL is derived from typed inputs (Azure resource name, Bedrock region), the provider helper configures the route endpoint before calling `.model(...)`. Use `AtLeastOne<T>` from `route/auth-options.ts` for inputs that accept either of two derivation paths (Azure: `resourceName` or `baseURL`).
 
@@ -126,54 +126,6 @@ Keep semantic APIs as separate entrypoints, such as OpenAI `chat` and `responses
 
 Do not expose `Route` in provider package settings. Route composition stays an implementation detail behind `model(...)`.
 
-### Folder layout
-
-```
-packages/ai/src/
-  schema/                   canonical Schema model, split by concern
-    ids.ts                  branded IDs, literal types, ProviderMetadata
-    options.ts              Generation/Provider/Http options, Limits, LanguageModel, cache policy
-    messages.ts             content parts, Message, ToolDefinition, LLMRequest
-    events.ts               Usage, individual events, LLMEvent, LLMResponse
-    errors.ts               error reasons, AIError, ToolFailure
-    index.ts                barrel
-  llm.ts                    request constructors and convenience helpers
-  route/
-    index.ts                @opencode-ai/ai/route advanced barrel
-    client.ts               Route.make + LLMClient.stream/generate
-    executor.ts             RequestExecutor service + transport error mapping
-    protocol.ts             Protocol type + Protocol.make
-    endpoint.ts             Endpoint type + Endpoint.path
-    auth.ts                 Auth type + Auth.bearer / Auth.apiKeyHeader / Auth.passthrough
-    auth-options.ts         ProviderAuthOption shape, AuthOptions.bearer, AtLeastOne helper
-    framing.ts              Framing type + Framing.sse
-    transport/              transport implementations
-      index.ts              Transport execution types + HttpTransport / WebSocketTransport namespaces
-      websocket-channel.ts  generic sequential channel executor/driver contract
-      http.ts               HttpTransport.httpJson — POST + framing
-      websocket.ts          direct one-request channel executor + raw socket adapter
-  protocols/
-    shared.ts               ProviderShared toolkit used inside protocol impls
-    openai-chat.ts          protocol + route (compose OpenAIChat.protocol)
-    open-responses.ts         provider-neutral Responses protocol baseline
-    open-responses-channel.ts provider-neutral Responses WebSocket transport factory
-    openai-responses.ts       OpenAI tools/events and channel policy composed over OpenResponses
-    anthropic-messages.ts
-    gemini.ts
-    bedrock-converse.ts
-    bedrock-event-stream.ts framing for AWS event-stream binary frames
-    openai-compatible-chat.ts route that reuses OpenAIChat.protocol, no canonical URL
-    openai-compatible-responses.ts deployment adapter that reuses OpenResponses.protocol, no canonical URL
-    utils/                  per-protocol helpers (auth, cache, media, tool-stream, ...)
-  providers/
-    openai-compatible.ts    generic Chat helper + family model helpers
-    openai-compatible-responses.ts generic Responses helper
-    openai-compatible-profile.ts family defaults (deepseek, togetherai, ...)
-    azure.ts / amazon-bedrock.ts / cloudflare.ts / github-copilot.ts / google.ts / xai.ts / openai.ts / anthropic.ts / openrouter.ts
-  tool.ts                   typed tool() helper
-  tool-runtime.ts           narrow one-call typed tool dispatcher
-```
-
 The dependency arrow points down: `providers/*.ts` files import protocol routes and auth-option utilities; protocol modules import `endpoint`, `auth`, `framing`, and transport pieces. Protocols do not import provider facades. Lower-level modules know nothing about provider catalog metadata. `OpenAIResponses` composes the provider-neutral `OpenResponses` protocol; the baseline never imports the OpenAI extension.
 
 ### Shared protocol helpers
@@ -193,7 +145,7 @@ If you find yourself copying a 3-to-5-line snippet between two protocols, lift i
 
 `LLMRequest.system` is the initial privileged prompt that applies ahead of the conversation. `Message.system(...)` is a separate, provider-neutral chronological operator update inside `LLMRequest.messages`; it applies only from its position in history onward and accepts text content only.
 
-Native chronological system messages are route/model-specific. Anthropic Messages lowers them natively for Claude Opus 4.8 (`claude-opus-4-8`). Other routes and models intentionally lower the update in place into ordinary user-compatible text using this stable escaped representation:
+Native chronological system messages are route/model-specific. Open Responses lowers them to standard `developer` messages, while Anthropic Messages lowers them to native system messages for Claude Opus 4.8 (`claude-opus-4-8`). Other routes and models intentionally lower the update in place into ordinary user-compatible text using this stable escaped representation:
 
 ```text
 <system-update>
@@ -221,19 +173,17 @@ Routes lower these into provider-native assistant tool-call messages and tool-re
 
 ### Tool dispatch
 
-`LLM.stream(request)` and `LLM.generate(request)` each run exactly one provider turn. Add tool schemas to `request.tools` with `Tool.toDefinitions(tools)`. When a caller wants the package's typed one-call execution behavior, pass each canonical local `tool-call` event to `ToolRuntime.dispatch(tools, call)`.
+`LLM.stream(request)` and `LLM.generate(request)` each run exactly one model call. Add tool schemas to `request.tools` with `Tool.toDefinitions(tools)`. When a caller wants the package's typed one-call execution behavior, pass each canonical local `tool-call` event to `ToolRuntime.dispatch(tools, call)`.
 
 ```ts
-const get_weather = tool({
+const get_weather = Tool.make({
   description: "Get current weather for a city",
   parameters: Schema.Struct({ city: Schema.String }),
   success: Schema.Struct({ temperature: Schema.Number, condition: Schema.String }),
-  execute: ({ city }) =>
+  execute: (input) =>
     Effect.gen(function* () {
-      // city: string  — typed from parameters Schema
+      const data = yield* WeatherApi.fetch(input.city)
-      const data = yield* WeatherApi.fetch(city)
       return { temperature: data.temp, condition: data.cond }
-      // return type checked against success Schema
     }),
 })
 

packages/ai/README.md

@@ -237,11 +237,11 @@ Prompt caching is **on by default**. Every `LLMRequest` resolves to `cache: "aut
 
 ### Auto placement
 
-`"auto"` places up to four breakpoints — the last tool definition, the first system part, the last system part when distinct, and the final message boundary. These expose successively larger reusable prefixes for tools, the base agent, project instructions, and the active conversation. The rolling final-message boundary is the load-bearing detail in tool loops: it advances on every request so the previous cache entry stays within Anthropic's 20-block lookback.
+`"auto"` places up to four breakpoints — the last tool definition, the first system part, the last system part when distinct, and the final message boundary. These expose successively larger reusable prefixes for tools, the base agent, project instructions, and the active conversation. The rolling final-message boundary advances on every request so recent conversation prefixes remain reusable during tool loops.
 
 Tools precede every system and conversation block in the provider prefix, so tool definitions must remain byte-stable and deterministically ordered for downstream breakpoints to remain reusable.
 
-The math justifies the default: Anthropic's 5-minute cache write is 1.25× base, read is 0.1×, so a single reuse within 5 minutes already wins. One-shot completions below the per-model minimum-cacheable-token threshold silently no-op on the wire, so the worst case is harmless.
+Requests below a provider's minimum cacheable size simply do not produce a reusable cache entry.
 
 ### Opting out
 
@@ -285,6 +285,7 @@ LLM.request({
 | ----------------------- | ------------------------------------------------------------------------- |
 | Anthropic Messages      | emits up to 4 `cache_control` markers (4-breakpoint cap enforced)         |
 | Bedrock Converse        | emits up to 4 `cachePoint` blocks (4-breakpoint cap enforced)             |
+| OpenRouter              | emits up to 4 `cache_control` markers                                     |
 | OpenAI Chat / Responses | no-op (implicit caching above 1024 tokens)                                |
 | Gemini                  | no-op (implicit caching on 2.5+; explicit `CachedContent` is out-of-band) |
 
@@ -308,7 +309,7 @@ Included providers: OpenAI, Anthropic, Google (Gemini), Google Vertex Gemini and
 
 ### Package-like entrypoints
 
-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.
+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` and `body` overlays.
 
 ```ts
 import { model } from "@opencode-ai/ai/providers/openai/responses"
@@ -316,7 +317,6 @@ import { model } from "@opencode-ai/ai/providers/openai/responses"
 const selected = model("gpt-5", {
   apiKey: process.env.OPENAI_API_KEY,
   headers: { "x-application": "opencode" },
-  limits: { context: 200_000, output: 64_000 },
 })
 ```
 
@@ -371,11 +371,23 @@ Request options in order of stability:
 
 1. **`generation`** — portable knobs (`maxTokens`, `temperature`, `topP`, `topK`, penalties, seed, stop).
 2. **`promptCacheKey`** — stable cache affinity lowered by every protocol that supports it.
-3. **`providerOptions: { <provider>: {...} }`** — typed-at-the-facade provider-specific knobs (OpenAI `store`, Anthropic `thinking`, Gemini `thinkingConfig`, OpenRouter routing).
+3. **`providerOptions: { ... }`** — flat options inferred from the selected model (OpenAI `store`, Anthropic `thinking`, Gemini `thinkingConfig`, OpenRouter routing).
 4. **`http: { body, headers, query }`** — last-resort serializable overlays merged into the final HTTP request. Reach for this only when a stable typed path doesn't yet exist.
 
 Route/provider defaults are overridden by request-level values for each axis.
 
+The selected model supplies the provider-specific option type, so per-request overrides stay flat while the canonical runtime request remains provider-neutral:
+
+```ts
+LLM.request({
+  model,
+  prompt,
+  providerOptions: {
+    reasoningEffort: "low",
+  },
+})
+```
+
 ## Routes
 
 Adding a new model or deployment is usually 5-15 lines using `Route.make({ protocol, endpoint, auth, framing, ... })`. The route owns endpoint/auth/framing and the protocol owns body construction plus stream parsing. Transports are reusable IO templates that receive route endpoint/auth at compile time. Capability/catalog metadata lives outside this low-level package; unsupported request shapes fail during protocol lowering. See `AGENTS.md` for the architectural detail.
@@ -387,6 +399,5 @@ This package is built on Effect. Public methods return `Effect` or `Stream`; pro
 ## See also
 
 - `AGENTS.md` — architecture, route construction, contributor guide
-- `STATUS.md` — native provider parity status and AI SDK migration gaps
 - `example/tutorial.ts` — runnable end-to-end walkthrough
 - `test/provider/*.test.ts` — fixture-first protocol tests; `*.recorded.test.ts` files cover live cassettes

packages/ai/STATUS.md

@@ -1,107 +0,0 @@
-# LLM Provider Parity Status
-
-Last reviewed: 2026-08-07
-
-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/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
-
-| Native slice                       | Source                                                                                                                            | Current state                                                                                                                                                            | Main gaps                                                                                                                                                                                                                                                 |
-| ---------------------------------- | --------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| OpenAI Chat                        | `src/protocols/openai-chat.ts`, `src/providers/openai.ts`                                                                         | Usable. Streams text, reasoning deltas, tool calls, usage, images, and common generation controls.                                                                       | No typed structured-output / `response_format` path. Limited typed OpenAI option surface compared with SDK escape hatches.                                                                                                                                |
-| OpenAI Responses                   | `src/protocols/open-responses.ts`, `src/protocols/openai-responses.ts`, `src/providers/openai.ts`                                 | Usable over HTTP by default, with optional per-call WebSocket channel execution on the same model and route identity.                                                    | No incremental `previous_response_id` path or persistent Session channel manager yet. Typed options cover only a subset of Responses fields. Structured output is still mostly synthetic-tool based.                                                      |
-| 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.                                                                                                                                                                                      |
-| Open Responses-compatible          | `src/protocols/open-responses.ts`, `src/protocols/openai-compatible-responses.ts`, `src/providers/openai-compatible-responses.ts` | Usable for deployments that implement the provider-neutral Open Responses protocol. The deployment adapter does not inherit OpenAI tools, events, metadata, or defaults. | 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; 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/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/open-responses.ts`, `src/providers/google-vertex-responses.ts`                                                     | Usable for Grok models through Open Responses with explicit OAuth tokens or ADC, project/location endpoint derivation, and an explicit `store: false` Vertex 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.                                                                                                                         |
-| OpenRouter                         | `src/providers/openrouter.ts`                                                                                                     | Present with OpenRouter-specific usage/reasoning/prompt-cache options over Chat.                                                                                         | Responses-style OpenRouter support is absent.                                                                                                                                                                                                             |
-| xAI                                | `src/providers/xai.ts`                                                                                                            | Present with Responses and Chat selectors.                                                                                                                               | Needs package-parity review against the AI SDK xAI provider.                                                                                                                                                                                              |
-| GitHub Copilot                     | `src/providers/github-copilot.ts`                                                                                                 | Present as explicit-base-URL OpenAI Chat/Responses facade.                                                                                                               | Runtime/catalog integration remains specialized and should stay separate from public OpenAI-compatible defaults.                                                                                                                                          |
-
-## V2 Runner Status
-
-`packages/core/src/session/runner/model.ts` currently resolves only this native subset from catalog `aisdk` metadata:
-
-| Catalog API                                         | Native route used today      |
-| --------------------------------------------------- | ---------------------------- |
-| `aisdk:@ai-sdk/openai`                              | `OpenAIResponses.route`      |
-| `aisdk:@ai-sdk/anthropic`                           | `AnthropicMessages.route`    |
-| `aisdk:@ai-sdk/openai-compatible` with explicit URL | `OpenAICompatibleChat.route` |
-
-Other `aisdk:` packages, including Google Vertex, Azure, and Bedrock, currently fall back through the AI SDK loader in the production runner. The dependency-free resolver seam rejects them with `SessionRunnerModel.UnsupportedPackageError`; they are not native route mappings yet.
-
-## AI SDK Package Parity Matrix
-
-| AI SDK package                    | Intended native target                                    | Status           | Biggest gaps                                                                                                                                                                                                        |
-| --------------------------------- | --------------------------------------------------------- | ---------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| `@ai-sdk/openai`                  | `OpenAI.chat`, `OpenAI.responses`                         | Partial / usable | Add complete typed option coverage, structured output strategy, explicit Responses continuation support, and runner execution policy for optional WebSocket channels.                                               |
-| `@ai-sdk/openai-compatible`       | Generic OpenAI-compatible Chat and Responses              | Partial / usable | Decide per-family namespace/profile behavior and runner API selection for providers that support Responses versus Chat only.                                                                                        |
-| `@ai-sdk/anthropic`               | `AnthropicMessages`                                       | Partial / usable | Finish Messages API parity for headers/betas/metadata/newer fields and document hosted-tool continuation expectations.                                                                                              |
-| `@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 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 | Partial / usable | Add default AWS credential chain/profile support; native catalog mapping currently requires bearer auth or explicit static credentials.                                                                             |
-
-## Highest-Risk Gaps
-
-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. The Open Responses adapter is available through 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, 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. Vertex xAI still needs catalog API selection.
-9. Recorded coverage is uneven. OpenAI, Anthropic, Gemini, Bedrock Converse, Bedrock Mantle, Cloudflare, OpenRouter, and several OpenAI-compatible Chat providers have cassettes. Azure and Vertex still 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/ai/providers/openai/chat`                     | OpenAI `/chat/completions` semantics.                                                      |
-| OpenAI Responses              | `@opencode-ai/ai/providers/openai/responses`                | OpenAI `/responses` semantics with HTTP default and optional per-call WebSocket execution. |
-| OpenAI-compatible Chat        | `@opencode-ai/ai/providers/openai-compatible`               | Generic OpenAI-compatible `/chat/completions`.                                             |
-| Open Responses-compatible     | `@opencode-ai/ai/providers/openai-compatible/responses`     | Generic provider-neutral `/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 Open 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 Chat           | `@opencode-ai/ai/providers/amazon-bedrock/mantle/chat`      | AWS Bedrock Mantle OpenAI-compatible Chat API.                                             |
-| Bedrock Mantle Responses      | `@opencode-ai/ai/providers/amazon-bedrock/mantle/responses` | AWS Bedrock Mantle OpenAI-compatible Responses API.                                        |
-| 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, Chat, Responses, and Messages entrypoints.
-5. Decide Chat/Responses selection for `@ai-sdk/google-vertex/xai` catalog models.
-6. 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.
-7. Add recorded provider tests for Azure, Vertex Gemini, Vertex Chat, Vertex Responses, Vertex Messages, and Bedrock credential-chain behavior before making native runtime the default for those packages.

packages/ai/example/call-sites.md

@@ -1,606 +0,0 @@
-# LLM Call Site Sketches
-
-Scratchpad for examples first, abstractions second. Current direction: routes
-execute, provider facades organize configured route sets, and models carry route
-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/ai` where possible. The goal is not a big
-generic transform layer; the goal is small composable route definitions backed by
-recorded golden tests.
-
-Things to keep testing against:
-
-- Cache placement: `cache: "auto"`, manual cache breakpoints, provider cache usage.
-- Images: golden image tests for providers/protocols that claim image support.
-- Reasoning: canonical reasoning parts/events versus provider-native knobs.
-- Auth: bearer, custom headers, multiple credentials, query auth, SigV4, OAuth, no auth.
-- OpenAI-compatible providers: DeepSeek, Together, Groq, Alibaba/DashScope, custom routers.
-- Provider switching: stale signatures, encrypted reasoning, provider metadata, incompatible parts.
-- Error quality: typed errors instead of generic SDK/server failures.
-
-## Final Guide: Routes Execute, Providers Organize
-
-Do not introduce a first-class `Deployment` abstraction unless it gains real
-semantics. Provider facades are ergonomic configured route groups, not execution
-registries. The executable/composable thing is still a route. Do not make route
-construction publish to a global registry; models should carry their route value
-directly.
-
-Keep durable identity separate from runtime capability:
-
-- Durable identity is small serializable data like `{ providerID, modelID }` for
-  config, sessions, logs, and catalogs.
-- Runtime capability is a `LanguageModel` with a route value, protocol, transport, auth,
-  and defaults. It is allowed to contain functions and schemas.
-- If persisted identity needs to become executable, resolve it through an app
-  boundary first. Do not make `LLMRequest` recover behavior from a global route
-  side table.
-
-Keep unconfigured behavior values as values, not factories. A transport like
-`HttpTransport.sseJson` should be a reusable immutable value. Use a function only
-when the caller supplies options or when construction needs fresh state.
-
-Use constants to remove repetition before inventing abstractions. Provider ids
-are branded once per provider facade and reused across routes; a plain exported
-object is enough for the provider-facing API unless a helper earns its keep by
-removing repeated route projection.
-
-Expose default configured provider instances, and put provider-specific setup on
-`.configure(...)`. Model selectors stay pure: `model(id)`, `responses(id)`,
-`chat(id)`, etc. Endpoint/auth/resource/api-version configuration happens before
-model selection, not as a second argument to model selection.
-
-Use provider/product facades consistently:
-
-- One coherent provider/product config surface gets one top-level facade.
-- APIs/model kinds that share that config are methods on the facade.
-- Different products with different required config get separate top-level
-  facades, not a shared namespace with unrelated children.
-- Default facades are exposed only when concrete defaults or lazy env/credential
-  defaults make the facade valid.
-
-Examples:
-
-```ts
-OpenAI.responses("gpt-4o")
-OpenAI.chat("gpt-4o")
-
-Azure.configure({ resourceName, apiKey }).responses("my-deployment")
-AmazonBedrock.configure({ region, credentials }).model("anthropic.claude-3-5-sonnet-20241022-v2:0")
-
-CloudflareAIGateway.configure({ accountId, gatewayId, gatewayApiKey, apiKey }).model("openai/gpt-4o")
-CloudflareWorkersAI.configure({ accountId, apiKey }).model("@cf/meta/llama-3.1-8b-instruct")
-
-OpenAICompatible.configure({
-  provider: "custom",
-  baseURL: "https://custom.example/v1",
-  auth: Auth.bearer(apiKey),
-}).model("custom-model")
-```
-
-Standardize the provider facade contract before abstracting construction. A
-plain object is enough at first; add a helper only if repeated route projection
-starts hiding the real provider-specific config.
-
-`Route.with(...)` patch semantics should be boring and explicit:
-
-- Omitted fields inherit from the original route.
-- `endpoint` patches merge with the existing endpoint, so overriding `baseURL`
-  keeps the existing `path`.
-- `endpoint.query` merges by default; later values win.
-- `auth` replaces.
-- `headers` merge by default; undefined values are omitted.
-- `id` is optional in patches. Route ids are diagnostic/provider API labels, not
-  global runtime registry keys.
-
-1. **Route**
-   - route id
-   - provider id
-   - protocol
-   - body schema
-   - body builder
-   - stream event schema
-   - parser/state machine
-   - transport
-     - method / IO shape
-     - framing
-     - request preparation
-     - constants when unconfigured; functions only when configured
-   - endpoint
-     - base URL
-     - static path
-     - body/model-derived path
-     - query params
-   - auth
-     - bearer
-     - custom header
-     - multiple credentials
-     - SigV4
-     - none
-   - defaults
-     - headers
-     - generation defaults
-     - provider options
-     - limits
-2. **Provider Facade**
-   - default configured provider instance
-   - provider-specific `.configure(...)`
-   - plain object/function facade over one or more routes
-   - top-level export only when it represents one coherent config surface
-   - no passive `Provider.make(...)` wrapper unless it gains runtime behavior
-3. **Model Selector**
-   - route/provider-owned selector
-   - accepts model id only
-   - returns executable models
-   - does not accept endpoint/auth/deployment overrides
-4. **Language Model**
-   - model id
-   - route value
-   - provider id
-   - configured route value at selection time
-5. **LLM Request**
-   - model
-   - messages/tools
-   - generation/cache/reasoning/response-format options
-   - request-level HTTP overlays for per-request headers/query/body additions,
-     not provider endpoint/auth reconfiguration
-6. **Compile**
-   - read route from model
-   - merge route defaults and request overrides
-   - build final URL from route endpoint
-   - apply auth from the configured route
-   - build body with protocol
-   - execute with transport and parse with protocol
-
-## Provider Facade Shape
-
-The provider abstraction is a facade over configured routes, not the runtime
-execution mechanism:
-
-```ts
-type ProviderFacade<APIs, Config> = {
-  readonly id: ProviderID
-  readonly model: (id: string) => LanguageModel
-  readonly configure: (input?: Config) => ProviderFacade<APIs, Config>
-} & APIs
-```
-
-Manual construction is fine and should be the default until duplication earns a
-helper:
-
-```ts
-export const OpenAI = {
-  id: openAIProvider,
-  model: openAIResponses.model,
-  responses: openAIResponses.model,
-  chat: openAIChat.model,
-  configure: configureOpenAI,
-} satisfies ProviderFacade<
-  {
-    responses: (id: string) => LanguageModel
-    chat: (id: string) => LanguageModel
-  },
-  OpenAIConfig
->
-```
-
-If several providers repeat the same projection from route values to model
-methods, the helper can stay deliberately tiny:
-
-```ts
-const configureOpenAI = (input: OpenAIConfig = {}) =>
-  Provider.define({
-    id: openAIProvider,
-    routes: {
-      responses: openAIResponses.with(openAIConfig(input)),
-      chat: openAIChat.with(openAIConfig(input)),
-    },
-    default: "responses",
-    configure: configureOpenAI,
-  })
-
-export const OpenAI = configureOpenAI()
-```
-
-`Provider.define(...)` would only project route methods and preserve types:
-
-```ts
-OpenAI.model("gpt-4o")
-OpenAI.responses("gpt-4o")
-OpenAI.chat("gpt-4o")
-OpenAI.configure({ apiKey }).responses("gpt-4o")
-```
-
-It must not register routes, select routes dynamically, or participate in
-execution. Execution still reads the route value carried by the model.
-
-## Ideal Call Sites
-
-Define concrete routes for a native provider, then project them through a
-provider facade:
-
-```ts
-const openAIProvider = ProviderID.make("openai")
-
-const openAIResponses = Route.make({
-  id: "openai-responses",
-  provider: openAIProvider,
-  protocol: OpenAIResponses.protocol,
-  transport: HttpTransport.sseJson,
-  endpoint: {
-    baseURL: "https://api.openai.com/v1",
-    path: "/responses",
-  },
-  auth: Auth.envBearer("OPENAI_API_KEY"),
-})
-
-const openAIChat = Route.make({
-  id: "openai-chat",
-  provider: openAIProvider,
-  protocol: OpenAIChat.protocol,
-  transport: HttpTransport.sseJson,
-  endpoint: {
-    baseURL: "https://api.openai.com/v1",
-    path: "/chat/completions",
-  },
-  auth: Auth.envBearer("OPENAI_API_KEY"),
-})
-
-const openAIConfig = (input: OpenAIConfig) => ({
-  endpoint: input.endpoint,
-  auth: input.auth ?? (input.apiKey ? Auth.bearer(input.apiKey) : undefined),
-  headers: {
-    "OpenAI-Organization": input.organization,
-    "OpenAI-Project": input.project,
-  },
-})
-
-const configureOpenAI = (input: OpenAIConfig = {}) => {
-  const responses = openAIResponses.with(openAIConfig(input))
-  const chat = openAIChat.with(openAIConfig(input))
-
-  return {
-    id: openAIProvider,
-    responses: responses.model,
-    chat: chat.model,
-    model: responses.model,
-    configure: configureOpenAI,
-  }
-}
-
-export const OpenAI = configureOpenAI()
-```
-
-Specialize it functionally for concrete providers:
-
-```ts
-const deepSeekProvider = ProviderID.make("deepseek")
-
-const deepseekChat = openAIChat.with({
-  id: "deepseek-chat",
-  provider: deepSeekProvider,
-  endpoint: {
-    baseURL: "https://api.deepseek.com/v1",
-  },
-  auth: Auth.envBearer("DEEPSEEK_API_KEY"),
-})
-
-const configureDeepSeek = (input: OpenAICompatibleConfig = {}) => {
-  const route = deepseekChat.with({
-    endpoint: input.endpoint,
-    auth: input.auth ?? (input.apiKey ? Auth.bearer(input.apiKey) : undefined),
-  })
-
-  return {
-    id: deepSeekProvider,
-    model: route.model,
-    configure: configureDeepSeek,
-  }
-}
-
-export const DeepSeek = {
-  id: deepSeekProvider,
-  model: deepseekChat.model,
-  configure: configureDeepSeek,
-}
-```
-
-Provider-specific configuration happens before model selection:
-
-```ts
-const deepseek = DeepSeek.configure({
-  endpoint: {
-    baseURL: "https://proxy.example.com/v1",
-  },
-  auth: Auth.bearer(apiKey),
-})
-
-const model = deepseek.model("deepseek-chat")
-```
-
-Final request call site stays boring:
-
-```ts
-const response =
-  yield *
-  LLM.generate(
-    LLM.request({
-      model: DeepSeek.model("deepseek-chat"),
-      prompt: "Hello.",
-    }),
-  )
-```
-
-For direct provider-facade calls, Responses has one semantic model and route:
-
-```ts
-OpenAI.responses("gpt-4o")
-```
-
-The package-like OpenAI Responses entrypoint has the same transport-neutral
-`model(...)` contract:
-
-```ts
-import { model } from "@opencode-ai/ai/providers/openai/responses"
-
-model("gpt-4o", { apiKey })
-```
-
-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/ai/providers/google-vertex/gemini"
-
-model("gemini-3.5-flash", { project, location: "global" })
-```
-
-```ts
-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" })
-```
-
-The client does not require a different public layer for WebSocket execution.
-Responses routes use HTTP by default, and callers may pass a channel executor per
-call. Routes without channel support simply ignore that execution capability.
-
-Azure is a route specialization with auth/path/default changes plus input
-mapping. The public API configures the Azure resource once, then selects
-deployment ids with pure model selectors:
-
-```ts
-const azureProvider = ProviderID.make("azure")
-
-const azureResponses = openAIResponses.with({
-  id: "azure-openai-responses",
-  provider: azureProvider,
-  auth: Auth.envHeader("api-key", "AZURE_OPENAI_API_KEY"),
-})
-
-const configureAzure = (input: AzureConfig = {}) => {
-  const route = azureResponses.with({
-    endpoint: {
-      baseURL:
-        input.baseURL ??
-        Endpoint.envBaseURL(
-          "AZURE_RESOURCE_NAME",
-          (resourceName) => `https://${resourceName}.openai.azure.com/openai/v1`,
-        ),
-      query: { "api-version": input.apiVersion ?? "v1" },
-    },
-    auth: input.apiKey ? Auth.header("api-key", input.apiKey) : Auth.envHeader("api-key", "AZURE_OPENAI_API_KEY"),
-  })
-
-  return {
-    id: azureProvider,
-    model: route.model,
-    responses: route.model,
-    configure: configureAzure,
-  }
-}
-
-export const Azure = configureAzure()
-
-const azure = Azure.configure({
-  resourceName: "my-resource",
-  apiVersion: "v1",
-})
-
-const model = azure.responses("my-deployment")
-```
-
-Default provider facades are only valid when required configuration has a lazy
-default source. `Azure.responses("my-deployment")` can be valid if endpoint
-resolution reads `AZURE_RESOURCE_NAME` lazily and fails with a typed
-configuration error when missing. If a provider has no sensible lazy default,
-do not expose a default model selector; expose only a configured entrypoint.
-
-Cloudflare AI Gateway and Workers AI are separate product facades because their
-configuration surfaces differ. Do not make a root `Cloudflare.configure(...)`
-pretend there is one coherent Cloudflare provider configuration:
-
-```ts
-const cloudflareProvider = ProviderID.make("cloudflare-ai-gateway")
-
-const cloudflareOpenAIChat = openAIChat.with({
-  id: "cloudflare-ai-gateway-openai-chat",
-  provider: cloudflareProvider,
-  auth: Auth.bearerHeader("cf-aig-authorization").andThen(Auth.bearer()),
-})
-
-const configureCloudflareAIGateway = (input: CloudflareAIGatewayConfig) => {
-  const route = cloudflareOpenAIChat.with({
-    endpoint: {
-      baseURL: `https://gateway.ai.cloudflare.com/v1/${input.accountId}/${input.gatewayId}/openai`,
-    },
-    auth: Auth.bearerHeader("cf-aig-authorization", input.gatewayApiKey).andThen(Auth.bearer(input.apiKey)),
-  })
-
-  return {
-    id: cloudflareProvider,
-    model: (modelID: string) => route.model({ id: modelID }),
-    configure: configureCloudflareAIGateway,
-  }
-}
-
-export const CloudflareAIGateway = {
-  id: cloudflareProvider,
-  configure: configureCloudflareAIGateway,
-}
-
-const gateway = CloudflareAIGateway.configure({
-  accountId: "account",
-  gatewayId: "gateway",
-  gatewayApiKey,
-  apiKey,
-})
-
-const model = gateway.model("openai/gpt-4o")
-```
-
-If a Cloudflare product gains a full lazy env default, it can expose a direct
-selector too. Until then, omitting `CloudflareAIGateway.model(...)` makes missing
-account/gateway configuration unrepresentable.
-
-opencode's dynamic runtime should construct executable models at its app
-boundary instead of exposing a giant unstructured public model constructor or a
-generic dynamic resolver:
-
-```ts
-const model =
-  providerID === "azure" ? Azure.configure(resolvedAzureConfig).responses(apiModelID) : OpenAI.responses(apiModelID)
-```
-
-That boundary can branch on durable config/catalog metadata and call typed
-provider APIs directly. Transport selection remains execution policy: a Session
-or other caller may pass a WebSocket channel executor per call without changing
-the model constructed by this boundary.
-
-## Competitive Shape
-
-This follows the strongest parts of adjacent libraries:
-
-- AI SDK: configured provider instances expose provider-specific model methods.
-- Effect AI: executable models carry provider requirements and can be resolved by
-  an app boundary.
-- LiteLLM/opencode config: dynamic `providerID/modelID` branching belongs at the
-  app boundary, not in the typed public provider API or a global runtime
-  resolver.
-- LangChain/LlamaIndex: constructor-style config plus model id is convenient,
-  but we avoid making model selection also configure endpoint/auth.
-
-The chosen split is:
-
-```txt
-Route = execution mechanics
-Provider facade = configured route group
-LanguageModel = selected executable model carrying route value
-App boundary = explicit durable-config -> typed-provider call
-```
-
-## What This Removes
-
-- No `Provider.make(...)` as a core abstraction.
-- No `Provider.make(...)` wrapper just to bind an id to model functions. Use a
-  branded provider id constant and a plain exported provider facade.
-- No `Deployment.define(...)` unless future examples force it.
-- No global route registry as the normal execution path.
-- No import side effects required before a model can execute.
-- No duplicate `provider.id` object when selected models already carry provider
-  id.
-- No `model(id, overrides)` escape hatch. Model selection takes the model id;
-  endpoint/auth/deployment customization happens by configuring the route first.
-- No transport setting on a provider or executable model. OpenAI Responses uses
-  HTTP by default and accepts an optional per-call channel executor as execution policy.
-- No separate public `LLMClient.layerWithWebSocket`. The runtime should expose one
-  client layer with the available transport capabilities.
-- No executable `ModelRef`. The executable handle is `LanguageModel`; durable model
-  identity stays separate and cannot execute on its own.
-
-## Implementation Todo
-
-- [x] Replace the current executable `ModelRef` with `LanguageModel`.
-- [x] Change `LanguageModel.route` to carry a route value, not a `RouteID` string.
-- [ ] Keep a separate durable model identity type for persisted/session/catalog
-      data, likely `{ providerID, modelID }`, and make it clear that it cannot
-      execute without resolver context.
-- [x] Change route model selectors so `route.model(id)` returns an executable
-      model with the route value attached, not a globally registered route id.
-- [x] Remove the standalone `Route.model(route, defaults, mapInput)` helper;
-      configured route instances own model selection.
-- [x] Remove endpoint/auth escape hatches from route model selection; callers must
-      configure endpoint/auth through `route.with(...)` or provider facades before
-      calling `.model(...)`.
-- [x] Remove request-shaping defaults from `LanguageModel`; selected models now carry only
-      id, provider, and configured route while defaults live on routes or requests.
-- [x] Rework `LLMClient.stream` / `generate` to read
-      `request.model.route` directly instead of calling `registeredRoute(...)`.
-- [x] Remove `Route.make(...)` global registration from the normal execution
-      path; keep route ids only as diagnostics/provider API labels.
-- [x] Model endpoint as `{ baseURL, path, query }` on routes, then remove the
-      current split where host/query live on the model and path lives in route
-      transport setup.
-- [x] Define `Route.with(...)` with explicit patch semantics for endpoint merge,
-      query merge, header merge, auth replacement, and optional diagnostic id.
-- [x] Make unconfigured transports reusable constants such as
-      `HttpTransport.sseJson`; keep transport functions only for configured/fresh
-      state construction.
-- [x] Collapse the public WebSocket runtime split so one `LLMClient.layer` accepts
-      optional per-call channel execution without changing route identity.
-- [x] Convert OpenAI provider APIs to provider-facade shape:
-      `OpenAI.configure(config).responses(id)` and `.chat(id)`.
-- [x] Convert Azure to a configured facade where resource/base URL/api version
-      setup happens before selecting deployment ids.
-- [x] Split Cloudflare products into separate facades such as
-      `CloudflareAIGateway` and `CloudflareWorkersAI`; do not expose a shared root
-      config surface unless one product actually exists.
-- [x] Migrate remaining built-in provider facades one at a time so configuration
-      happens before model selection and selectors accept only ids:
-      xAI, GitHub Copilot, OpenRouter, OpenAI-compatible families, Anthropic,
-      Google/Gemini, and Amazon Bedrock now use configured facades such as
-      `Provider.configure(options).model(id)` with named selectors where needed.
-- [ ] Decide whether a tiny `Provider.define(...)` helper is warranted after two
-      or three provider conversions; start with plain objects if duplication is not
-      yet painful.
-- [x] Keep executable model construction transport-neutral at the Session boundary;
-      Session-scoped execution policy supplies channel capability separately.
-- [ ] Update tests so direct route/provider tests assert route values are carried
-      by executable models, and opencode/native tests assert boundary-based route
-      selection.
-- [ ] Remove compatibility exports or stale docs only after internal call sites
-      are migrated; do not keep duplicate constructor paths without an external
-      compatibility need.
-
-## Open Questions
-
-- Default facades with required setup: should providers like Azure and Bedrock
-  expose default model selectors only when all required setup has lazy env or
-  credential-chain defaults? If not, omit the default selector so missing config
-  is impossible at the type/API level.
-- Lazy endpoint/auth values: should `Endpoint.envBaseURL(...)` and env-backed
-  auth produce typed configuration/authentication errors at compile/prepare time
-  or only when executing the transport?
-- `Route.with(...)` clearing semantics: endpoint/query/header patches merge by
-  default, but what is the explicit way to remove an inherited value?
-- Provider facade helper: keep plain objects until duplication hurts, or add a
-  tiny `Provider.define(...)` immediately to enforce shape and method projection?
-- Auth shape: should auth stay as today's composable `Auth`, or split into an
-  auth placement/strategy and credential sources?
-- Naming: is `baseURL` still the right endpoint field name, or should it be
-  `origin` / `urlPrefix` to clarify that route `path` is appended?

packages/ai/example/tutorial.ts

@@ -22,7 +22,7 @@ const model = OpenAI.configure({
   apiKey,
   generation: { maxTokens: 160 },
   providerOptions: {
-    openai: { store: false },
+    store: false,
   },
 }).model("gpt-4o-mini")
 
@@ -34,7 +34,7 @@ const model = OpenAI.configure({
 //   - `generation`: common controls such as max tokens, temperature, topP/topK,
 //     penalties, seed, and stop sequences.
 //   - `promptCacheKey`: stable cache affinity for protocols that support it.
-//   - `providerOptions`: namespaced provider-native behavior. For example,
+//   - `providerOptions`: model-typed provider-native behavior. For example,
 //     OpenAI store behavior, Anthropic thinking, Gemini thinking config, or
 //     OpenRouter routing/reasoning.
 //   - `http`: last-resort serializable overlays for final request body, headers,
@@ -188,7 +188,7 @@ const FakeProtocol = Protocol.make<FakeBody, string, string, void>({
   },
 })
 
-// An route is the runnable binding for that protocol. It adds the deployment
+// A route is the runnable binding for that protocol. It adds the deployment
 // axes that the protocol deliberately does not know: URL, auth, and framing.
 const FakeAdapter = Route.make({
   id: "fake-echo",

packages/ai/src/cache-policy.ts

@@ -5,8 +5,8 @@
 // The default `"auto"` shape places breakpoints at the last tool definition,
 // the first and last distinct system parts, and the conversation tail. This
 // exposes reusable tool, base-agent, project, and session prefixes while
-// advancing the tail after each tool result keeps the previous cache entry
+// advancing the tail after each tool result keeps recent conversation prefixes
-// within Anthropic's 20-block lookback during long agent turns.
+// reusable during long agent runs.
 //
 // Manual `cache: CacheHint` placements on individual parts are preserved and
 // count against the four-breakpoint budget; auto only fills remaining slots.
@@ -23,9 +23,7 @@ const NONE: CachePolicyObject = {}
 const BREAKPOINT_CAP = 4
 
 // Resolution rules:
-//   - undefined   → "auto" — caching is on by default. The math favors it:
+//   - undefined   → "auto" — caching is on by default.
-//                   Anthropic 5m-cache write is 1.25x base, read is 0.1x,
-//                   so a single reuse within 5 minutes already wins.
 //   - "auto"      → tools + first/last system + final message boundary.
 //   - "none"      → no auto placement; manual `CacheHint`s still flow.
 //   - object form → exactly what the caller asked for.

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

@@ -16,7 +16,6 @@ import {
   type JsonSchema,
   type LLMRequest,
   type MediaPart,
-  type ProviderOptions,
   type ProviderMetadata,
   type ToolCallPart,
   type ToolDefinition,
@@ -32,6 +31,7 @@ import { ToolStream } from "./utils/tool-stream.js"
 const ADAPTER = "anthropic-messages"
 export const DEFAULT_BASE_URL = "https://api.anthropic.com/v1"
 export const PATH = "/messages"
+export const DEFAULT_MAX_TOKENS = 32_000
 
 export type ThinkingInput =
   | {
@@ -52,9 +52,7 @@ export interface OptionsInput {
   readonly effort?: string
 }
 
-export type ProviderOptionsInput = ProviderOptions & {
+export type ProviderOptionsInput = OptionsInput
-  readonly anthropic?: OptionsInput
-}
 
 // =============================================================================
 // Request Body Schema
@@ -593,7 +591,7 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
 })
 
 const resolveOptions = Effect.fn("AnthropicMessages.resolveOptions")(function* (request: LLMRequest) {
-  const input = request.providerOptions?.anthropic
+  const input = request.providerOptions
   return {
     thinking: yield* resolveThinking(input?.thinking),
     effort: typeof input?.effort === "string" ? input.effort : undefined,
@@ -627,7 +625,6 @@ const resolveThinking = Effect.fn("AnthropicMessages.resolveThinking")(function*
 const fromRequest = Effect.fn("AnthropicMessages.fromRequest")(function* (request: LLMRequest) {
   const generation = request.generation
   const toolSchemaCompatibility = request.model.compatibility?.toolSchema
-  const outputLimit = request.model.defaults?.limits?.output ?? request.model.route.defaults.limits?.output ?? 4096
   // Allocate the 4-breakpoint budget in invalidation order: tools → system →
   // messages. Tools live highest in the cache hierarchy, so when callers
   // over-mark we keep their tool hints and shed the message-tail ones first.
@@ -666,7 +663,7 @@ const fromRequest = Effect.fn("AnthropicMessages.fromRequest")(function* (reques
     tools,
     tool_choice: toolChoice,
     stream: true as const,
-    max_tokens: generation?.maxTokens ?? outputLimit,
+    max_tokens: generation?.maxTokens ?? DEFAULT_MAX_TOKENS,
     temperature: generation?.temperature,
     top_p: generation?.topP,
     top_k: generation?.topK,
@@ -1015,8 +1012,8 @@ const step = (state: ParserState, event: AnthropicEvent) => {
 // =============================================================================
 /**
  * The Anthropic Messages protocol — request body construction, body schema,
- * and the streaming-event state machine. Used by native Anthropic Cloud and
+ * and the streaming-event state machine shared by Anthropic-compatible and
- * (once registered) Vertex Anthropic / Bedrock-hosted Anthropic passthrough.
+ * Vertex-hosted Messages routes.
  */
 export const protocol = Protocol.make({
   id: ADAPTER,

packages/ai/src/protocols/gemini.ts

@@ -12,7 +12,6 @@ import {
   type JsonSchema,
   type LLMRequest,
   type MediaPart,
-  type ProviderOptions,
   type ProviderMetadata,
   type TextPart,
   type ToolCallPart,
@@ -67,9 +66,7 @@ export interface OptionsInput {
   }
 }
 
-export type ProviderOptionsInput = ProviderOptions & {
+export type ProviderOptionsInput = OptionsInput
-  readonly gemini?: OptionsInput
-}
 
 // =============================================================================
 // Request Body Schema
@@ -292,7 +289,9 @@ const lowerMessages = Effect.fn("Gemini.lowerMessages")(function* (request: LLMR
     if (message.role === "system") {
       const part = yield* ProviderShared.wrappedSystemUpdate("Gemini", message)
       const previous = contents.at(-1)
-      if (previous?.role === "user")
+      // Gemini rejects a continuation whose function-response turn carries extra
+      // parts, so an update after a tool result starts its own user turn.
+      if (previous?.role === "user" && !previous.parts.some((item) => "functionResponse" in item))
         contents[contents.length - 1] = { role: "user", parts: [...previous.parts, { text: part.text }] }
       else contents.push({ role: "user", parts: [{ text: part.text }] })
       continue
@@ -387,7 +386,7 @@ const lowerMessages = Effect.fn("Gemini.lowerMessages")(function* (request: LLMR
 })
 
 const resolveOptions = (request: LLMRequest) => {
-  const input = request.providerOptions?.gemini
+  const input = request.providerOptions
   const value = input?.thinkingConfig
   const thinkingConfig = {
     thinkingBudget:
@@ -630,8 +629,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
 // =============================================================================
 /**
  * The Gemini protocol — request body construction, body schema, and the
- * streaming-event state machine. Used by Google AI Studio Gemini and (once
+ * streaming-event state machine shared by Google AI Studio and Vertex Gemini.
- * registered) Vertex Gemini.
  */
 export const protocol = Protocol.make({
   id: ADAPTER,

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

@@ -90,8 +90,11 @@ const OpenResponsesFunctionCallOutput = Schema.Union([
 
 export const InputItem = Schema.Union([
   Schema.Struct({ role: Schema.tag("system"), content: Schema.String }),
+  Schema.Struct({ role: Schema.tag("developer"), content: Schema.String }),
   Schema.Struct({ role: Schema.tag("user"), content: Schema.Array(OpenResponsesInputContent) }),
   Schema.Struct({
+    type: Schema.tag("message"),
+    id: Schema.optionalKey(Schema.String),
     role: Schema.tag("assistant"),
     content: Schema.Array(OpenResponsesOutputText),
     phase: Schema.optionalKey(MessagePhase),
@@ -100,6 +103,7 @@ export const InputItem = Schema.Union([
   OpenResponsesItemReference,
   Schema.Struct({
     type: Schema.tag("function_call"),
+    id: Schema.optionalKey(Schema.String),
     call_id: Schema.String,
     name: Schema.String,
     arguments: Schema.String,
@@ -114,6 +118,8 @@ type OpenResponsesInputItem = Schema.Schema.Type<typeof InputItem>
 type LoweredInputItem =
   | OpenResponsesInputItem
   | {
+      readonly type: "message"
+      readonly id?: string
       readonly role: "assistant"
       readonly content: ReadonlyArray<{ readonly type: "output_text"; readonly text: string }>
       readonly phase?: MessagePhase | null
@@ -127,8 +133,6 @@ type OpenResponsesReasoningInput = {
   summary: Array<{ type: "summary_text"; text: string }>
   encrypted_content?: string | null
 }
-type OpenResponsesReasoningReplay = Omit<OpenResponsesReasoningInput, "id">
-
 export const Tool = Schema.Struct({
   type: Schema.tag("function"),
   name: Schema.String,
@@ -140,6 +144,11 @@ export const Tool = Schema.Struct({
 export const ToolChoice = Schema.Union([
   Schema.Literals(["auto", "none", "required"]),
   Schema.Struct({ type: Schema.tag("function"), name: Schema.String }),
+  Schema.Struct({
+    type: Schema.tag("allowed_tools"),
+    mode: Schema.Literals(["auto", "none", "required"]),
+    tools: Schema.Array(Schema.Struct({ type: Schema.tag("function"), name: Schema.String })),
+  }),
 ])
 
 // Fields shared between the HTTP body and the WebSocket `response.create`
@@ -153,6 +162,15 @@ export const coreFields = {
   tools: optionalArray(Tool),
   tool_choice: Schema.optional(ToolChoice),
   store: Schema.optional(Schema.Boolean),
+  metadata: Schema.optional(Schema.Record(Schema.String, Schema.String)),
+  safety_identifier: Schema.optional(Schema.String),
+  stream_options: Schema.optional(
+    Schema.Struct({
+      include_obfuscation: Schema.optional(Schema.Boolean),
+    }),
+  ),
+  top_logprobs: Schema.optional(Schema.Int.check(Schema.isBetween({ minimum: 0, maximum: 20 }))),
+  truncation: Schema.optional(OpenResponsesOptions.TruncationSchema),
   service_tier: Schema.optional(OpenResponsesOptions.ServiceTierSchema),
   prompt_cache_key: Schema.optional(Schema.String),
   include: optionalArray(OpenResponsesOptions.ResponseIncludableSchema),
@@ -168,8 +186,12 @@ export const coreFields = {
     }),
   ),
   max_output_tokens: Schema.optional(Schema.Number),
+  max_tool_calls: Schema.optional(Schema.Int),
+  parallel_tool_calls: Schema.optional(Schema.Boolean),
   temperature: Schema.optional(Schema.Number),
   top_p: Schema.optional(Schema.Number),
+  presence_penalty: Schema.optional(Schema.Number),
+  frequency_penalty: Schema.optional(Schema.Number),
 }
 
 const OpenResponsesBody = Schema.Struct({
@@ -279,6 +301,20 @@ export const Event = Schema.StructWithRest(
 )
 export type Event = Schema.Schema.Type<typeof Event>
 
+const RefusalEvent = Schema.Union([
+  Schema.Struct({
+    type: Schema.tag("response.refusal.delta"),
+    item_id: Schema.String,
+    delta: Schema.String,
+  }),
+  Schema.Struct({
+    type: Schema.tag("response.refusal.done"),
+    item_id: Schema.String,
+    refusal: Schema.String,
+  }),
+])
+const isRefusalEvent = Schema.is(RefusalEvent)
+
 export interface Extension {
   readonly id: string
   readonly name: string
@@ -331,7 +367,7 @@ export const lowerTool = Effect.fn("OpenResponses.lowerTool")(function* (
     name: tool.name,
     description: tool.description,
     parameters: ToolSchemaProjection.responses(inputSchema),
-    // TODO: Read this from Responses tool options so direct LLM callers can opt into strict schemas.
+    // The common tool definition does not currently express Responses strict-schema policy.
     strict: false,
   }
 })
@@ -344,34 +380,42 @@ export const lowerToolChoice = (protocolName: string, toolChoice: NonNullable<LL
     tool: (toolName) => ({ type: "function" as const, name: toolName }),
   })
 
-const lowerToolCall = (part: ToolCallPart): OpenResponsesInputItem => ({
+const itemID = (providerMetadata: ProviderMetadata | undefined, providerMetadataKey: string) => {
-  type: "function_call",
+  const metadata = providerMetadata?.[providerMetadataKey]
-  call_id: part.id,
+  return ProviderShared.isRecord(metadata) && typeof metadata.itemId === "string" && metadata.itemId.length > 0
-  name: part.name,
+    ? metadata.itemId
-  arguments: ProviderShared.encodeJson(part.input),
+    : undefined
-})
+}
+
+const lowerToolCall = (part: ToolCallPart, providerMetadataKey: string): OpenResponsesInputItem => {
+  const id = itemID(part.providerMetadata, providerMetadataKey)
+  return {
+    type: "function_call",
+    ...(id ? { id } : {}),
+    call_id: part.id,
+    name: part.name,
+    arguments: ProviderShared.encodeJson(part.input),
+  }
+}
 
 const lowerReasoning = (part: ReasoningPart, providerMetadataKey: string): OpenResponsesReasoningInput | undefined => {
   const metadata = part.providerMetadata?.[providerMetadataKey]
-  if (!ProviderShared.isRecord(metadata) || typeof metadata.itemId !== "string" || metadata.itemId.length === 0)
+  const id = itemID(part.providerMetadata, providerMetadataKey)
-    return undefined
+  if (!ProviderShared.isRecord(metadata) || !id) return undefined
   const encryptedContent =
     typeof metadata.reasoningEncryptedContent === "string" || metadata.reasoningEncryptedContent === null
       ? metadata.reasoningEncryptedContent
       : undefined
   return {
     type: "reasoning",
-    id: metadata.itemId,
+    id,
     summary: part.text.length > 0 ? [{ type: "summary_text", text: part.text }] : [],
     encrypted_content: encryptedContent,
   }
 }
 
 const hostedToolItemID = (part: ToolResultPart, providerMetadataKey: string) => {
-  const metadata = part.providerMetadata?.[providerMetadataKey]
+  return itemID(part.providerMetadata, providerMetadataKey)
-  return ProviderShared.isRecord(metadata) && typeof metadata.itemId === "string" && metadata.itemId.length > 0
-    ? metadata.itemId
-    : undefined
 }
 
 const lowerMedia = Effect.fn("OpenResponses.lowerMedia")(function* (
@@ -439,14 +483,10 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
 
   for (const message of request.messages) {
     if (message.role === "system") {
-      const part = yield* ProviderShared.wrappedSystemUpdate(extension.name, message)
+      input.push({
-      const previous = input.at(-1)
+        role: "developer",
-      if (previous && "role" in previous && previous.role === "user")
+        content: ProviderShared.joinText(yield* ProviderShared.systemUpdateText(extension.name, message)),
-        input[input.length - 1] = {
+      })
-          role: "user",
-          content: [...previous.content, { type: "input_text", text: part.text }],
-        }
-      else input.push({ role: "user", content: [{ type: "input_text", text: part.text }] })
       continue
     }
 
@@ -460,24 +500,26 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
 
     if (message.role === "assistant") {
       const content: TextPart[] = []
-      const reasoningItems: Record<string, OpenResponsesReasoningReplay> = {}
+      const reasoningItems: Record<string, OpenResponsesReasoningInput> = {}
       const reasoningReferences = new Set<string>()
       const hostedToolReferences = new Set<string>()
       const flushText = () => {
         if (content.length === 0) return
-        const groups = content.reduce<Array<{ phase: MessagePhase | null | undefined; parts: TextPart[] }>>(
+        const groups = content.reduce<
-          (groups, part) => {
+          Array<{ id: string | undefined; phase: MessagePhase | null | undefined; parts: TextPart[] }>
-            const metadata = part.providerMetadata?.[providerMetadataKey]
+        >((groups, part) => {
-            const phase = ProviderShared.isRecord(metadata) ? messagePhase(metadata.phase, extension) : undefined
+          const metadata = part.providerMetadata?.[providerMetadataKey]
-            const group = groups.at(-1)
+          const id = itemID(part.providerMetadata, providerMetadataKey)
-            if (group && group.phase === phase) group.parts.push(part)
+          const phase = ProviderShared.isRecord(metadata) ? messagePhase(metadata.phase, extension) : undefined
-            else groups.push({ phase, parts: [part] })
+          const group = groups.at(-1)
-            return groups
+          if (group && group.id === id && group.phase === phase) group.parts.push(part)
-          },
+          else groups.push({ id, phase, parts: [part] })
-          [],
+          return groups
-        )
+        }, [])
         input.push(
           ...groups.map((group) => ({
+            type: "message" as const,
+            ...(group.id === undefined ? {} : { id: group.id }),
             role: "assistant" as const,
             content: group.parts.map((part) => ({ type: "output_text" as const, text: part.text })),
             ...(group.phase === undefined ? {} : { phase: group.phase }),
@@ -506,19 +548,14 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
               existing.encrypted_content = reasoning.encrypted_content
             continue
           }
-          const replay = {
+          reasoningItems[reasoning.id] = reasoning
-            type: reasoning.type,
+          input.push(reasoning)
-            summary: reasoning.summary,
-            encrypted_content: reasoning.encrypted_content,
-          }
-          reasoningItems[reasoning.id] = replay
-          input.push(replay)
           continue
         }
         if (part.type === "tool-call") {
           flushText()
           if (part.providerExecuted === true) continue
-          input.push(lowerToolCall(part))
+          input.push(lowerToolCall(part, providerMetadataKey))
           continue
         }
         if (part.type === "tool-result" && part.providerExecuted === true) {
@@ -573,6 +610,12 @@ const lowerOptions = (request: LLMRequest) => {
   return {
     ...(options.instructions ? { instructions: options.instructions } : {}),
     ...(options.store !== undefined ? { store: options.store } : {}),
+    ...(options.metadata ? { metadata: options.metadata } : {}),
+    ...(options.safetyIdentifier ? { safety_identifier: options.safetyIdentifier } : {}),
+    ...(options.streamOptions?.includeObfuscation !== undefined
+      ? { stream_options: { include_obfuscation: options.streamOptions.includeObfuscation } }
+      : {}),
+    ...(options.topLogprobs !== undefined ? { top_logprobs: options.topLogprobs } : {}),
     ...(request.promptCacheKey ? { prompt_cache_key: request.promptCacheKey } : {}),
     ...(options.include ? { include: options.include } : {}),
     ...(options.reasoningEffort || options.reasoningSummary
@@ -580,6 +623,19 @@ const lowerOptions = (request: LLMRequest) => {
       : {}),
     ...(options.textVerbosity ? { text: { verbosity: options.textVerbosity } } : {}),
     ...(options.serviceTier ? { service_tier: options.serviceTier } : {}),
+    ...(options.maxToolCalls !== undefined ? { max_tool_calls: options.maxToolCalls } : {}),
+    ...(options.parallelToolCalls !== undefined ? { parallel_tool_calls: options.parallelToolCalls } : {}),
+    ...(options.truncation ? { truncation: options.truncation } : {}),
+  }
+}
+
+const allowedToolChoice = (request: LLMRequest) => {
+  const allowed = OpenResponsesOptions.resolve(request).allowedTools
+  if (!allowed) return undefined
+  return {
+    type: "allowed_tools" as const,
+    mode: allowed.mode,
+    tools: allowed.toolNames.map((name) => ({ type: "function" as const, name })),
   }
 }
 
@@ -602,11 +658,15 @@ export const fromRequestWithExtension = Effect.fn("OpenResponses.fromRequestWith
               ToolSchemaProjection.modelCompatibility(tool.inputSchema, toolSchemaCompatibility),
             ),
           ),
-    tool_choice: request.toolChoice ? yield* lowerToolChoice(extension.name, request.toolChoice) : undefined,
+    tool_choice:
+      allowedToolChoice(request) ??
+      (request.toolChoice ? yield* lowerToolChoice(extension.name, request.toolChoice) : undefined),
     stream: true as const,
     max_output_tokens: generation?.maxTokens,
     temperature: generation?.temperature,
     top_p: generation?.topP,
+    presence_penalty: generation?.presencePenalty,
+    frequency_penalty: generation?.frequencyPenalty,
     ...lowerOptions(request),
   }
 })
@@ -675,7 +735,7 @@ const onOutputTextDelta = (state: ParserState, event: Event, id: string): StepRe
   if (!event.delta) return [state, NO_EVENTS]
   const events: LLMEvent[] = []
   const phase = state.messagePhases[id]
-  const metadata = phase === undefined ? undefined : providerMetadata(state, { phase })
+  const metadata = providerMetadata(state, { itemId: id, ...(phase === undefined ? {} : { phase }) })
   const lifecycle = Lifecycle.textStart(state.lifecycle, events, id, metadata)
   return [{ ...state, lifecycle: Lifecycle.textDelta(lifecycle, events, id, event.delta) }, events]
 }
@@ -904,7 +964,7 @@ const onOutputItemDone = Effect.fn("OpenResponses.onOutputItemDone")(function* (
           state.lifecycle,
           events,
           item.id,
-          phase === undefined ? undefined : providerMetadata(state, { phase }),
+          providerMetadata(state, { itemId: item.id, ...(phase === undefined ? {} : { phase }) }),
         ),
         messageItems,
         messagePhases,
@@ -917,7 +977,11 @@ const onOutputItemDone = Effect.fn("OpenResponses.onOutputItemDone")(function* (
     if (!item.id || !item.call_id || !item.name) return [state, NO_EVENTS] satisfies StepResult
     const tools = state.tools[item.id]
       ? state.tools
-      : ToolStream.start(state.tools, item.id, { id: item.call_id, name: item.name })
+      : ToolStream.start(state.tools, item.id, {
+          id: item.call_id,
+          name: item.name,
+          providerMetadata: providerMetadata(state, { itemId: item.id }),
+        })
     const result =
       item.arguments === undefined
         ? yield* ToolStream.finish(state.id, tools, item.id)
@@ -968,11 +1032,19 @@ const onOutputItemDone = Effect.fn("OpenResponses.onOutputItemDone")(function* (
   return [state, NO_EVENTS] satisfies StepResult
 })
 
-const onResponseFinish = (state: ParserState, event: Event): StepResult => {
+const onResponseFinish = Effect.fn("OpenResponses.onResponseFinish")(function* (state: ParserState, event: Event) {
-  const events: LLMEvent[] = []
+  // Some compatible providers omit output_item.done even after completing the response.
+  const pending =
+    event.type === "response.completed"
+      ? yield* ToolStream.finishAll(state.id, state.tools)
+      : { tools: state.tools, events: NO_EVENTS }
+  const events: LLMEvent[] = [...pending.events]
+  const hasFunctionCall =
+    pending.events.some((event) => LLMEvent.is.toolCall(event) || LLMEvent.is.toolInputError(event)) ||
+    state.hasFunctionCall
   const lifecycle = Lifecycle.finish(state.lifecycle, events, {
     reason: {
-      normalized: mapFinishReason(event, state.hasFunctionCall),
+      normalized: mapFinishReason(event, hasFunctionCall),
       raw: event.response?.incomplete_details?.reason,
     },
     usage: mapUsage(event.response?.usage, state.providerMetadataKey),
@@ -984,8 +1056,8 @@ const onResponseFinish = (state: ParserState, event: Event): StepResult => {
           })
         : undefined,
   })
-  return [{ ...state, lifecycle }, events]
+  return [{ ...state, lifecycle, hasFunctionCall, tools: pending.tools }, events] satisfies StepResult
-}
+})
 
 // Build a single human-readable message from whatever the provider supplied.
 // When both code and message are present, prefix the code so consumers see
@@ -1027,6 +1099,14 @@ export const step = (state: ParserState, event: Event) => {
         : onOutputTextDone(state, event, event.item_id),
     )
   }
+  if (event.type === "response.refusal.delta" || event.type === "response.refusal.done") {
+    if (!isRefusalEvent(event)) return ProviderShared.eventError(state.id, `${event.type} is malformed`)
+    return Effect.succeed(
+      event.type === "response.refusal.delta"
+        ? onOutputTextDelta(state, event, event.item_id)
+        : onOutputTextDone(state, { ...event, text: event.refusal }, event.item_id),
+    )
+  }
   if (event.type === "response.reasoning.delta" || event.type === "response.reasoning_summary_text.delta") {
     if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
     return Effect.succeed(onReasoningDelta(state, event, event.item_id))
@@ -1054,8 +1134,7 @@ export const step = (state: ParserState, event: Event) => {
       return ProviderShared.eventError(state.id, `${event.type} message is missing id`)
     return onOutputItemDone(state, event)
   }
-  if (event.type === "response.completed" || event.type === "response.incomplete")
+  if (event.type === "response.completed" || event.type === "response.incomplete") return onResponseFinish(state, event)
-    return Effect.succeed(onResponseFinish(state, event))
   if (event.type === "response.failed") return providerError(state, event, `${state.name} response failed`)
   if (event.type === "error")
     return decodeKnownErrorEvent(event).pipe(

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

@@ -28,7 +28,7 @@ import { ToolSchemaProjection } from "./utils/tool-schema.js"
 import { ToolStream } from "./utils/tool-stream.js"
 
 const ADAPTER = "openai-chat"
-const RESERVED_REASONING_FIELDS = new Set(["role", "content", "tool_calls"])
+const RESERVED_REASONING_FIELDS = new Set(["role", "content", "refusal", "tool_calls"])
 export const DEFAULT_BASE_URL = "https://api.openai.com/v1"
 export const PATH = "/chat/completions"
 
@@ -194,6 +194,7 @@ type OpenAIChatToolCallDelta = Schema.Schema.Type<typeof OpenAIChatToolCallDelta
 const OpenAIChatDelta = Schema.StructWithRest(
   Schema.Struct({
     content: optionalNull(Schema.String),
+    refusal: optionalNull(Schema.String),
     reasoning_content: optionalNull(Schema.String),
     reasoning: optionalNull(Schema.String),
     reasoning_text: optionalNull(Schema.String),
@@ -709,6 +710,7 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
     const reasoning = reasoningDelta(delta, state.reasoningField)
     const hasLateContent =
       Boolean(delta?.content) ||
+      Boolean(delta?.refusal) ||
       reasoning !== undefined ||
       (Array.isArray(delta?.reasoning_details) && delta.reasoning_details.length > 0) ||
       toolDeltas.some((tool) => Boolean(tool.id) || Boolean(tool.function?.name) || Boolean(tool.function?.arguments))
@@ -728,7 +730,7 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
     else if (
       reasoningDetailsObserved &&
       !lifecycle.reasoning.has("reasoning-0") &&
-      (Boolean(delta?.content) || toolDeltas.length > 0)
+      (Boolean(delta?.content) || Boolean(delta?.refusal) || toolDeltas.length > 0)
     )
       lifecycle = Lifecycle.reasoningStart(lifecycle, events, "reasoning-0", deltaMetadata)
     const reasoningEmitted = state.reasoningEmitted || lifecycle.reasoning.has("reasoning-0")
@@ -743,6 +745,16 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
       lifecycle = Lifecycle.textDelta(lifecycle, events, "text-0", delta.content)
     }
 
+    if (delta?.refusal) {
+      lifecycle = Lifecycle.reasoningEnd(
+        lifecycle,
+        events,
+        "reasoning-0",
+        reasoningMetadata(reasoningField, reasoningDetailsObserved ? state.reasoningDetails : undefined),
+      )
+      lifecycle = Lifecycle.textDelta(lifecycle, events, "text-0", delta.refusal)
+    }
+
     // Compatible providers may omit indexes. Prefer durable identity, then use
     // batch position for parallel deltas or the latest call for sparse chunks.
     for (const [position, tool] of toolDeltas.entries()) {

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

@@ -42,6 +42,8 @@ const OpenAIResponsesToolChoice = Schema.Union([
 
 const OpenAIResponsesInputItem = Schema.Union([
   Schema.Struct({
+    type: Schema.tag("message"),
+    id: Schema.optionalKey(Schema.String),
     role: Schema.tag("assistant"),
     content: Schema.Array(Schema.Struct({ type: Schema.tag("output_text"), text: Schema.String })),
     phase: Schema.optionalKey(Schema.NullOr(OpenResponses.MessagePhase)),
@@ -121,7 +123,8 @@ const fromRequest = Effect.fn("OpenAIResponses.fromRequest")(function* (request:
         : yield* Effect.forEach(request.tools, (tool) =>
             lowerTool(tool, ToolSchemaProjection.modelCompatibility(tool.inputSchema, toolSchemaCompatibility)),
           ),
-    tool_choice: request.toolChoice ? yield* lowerToolChoice(request.toolChoice, request.tools) : undefined,
+    tool_choice:
+      body.tool_choice ?? (request.toolChoice ? yield* lowerToolChoice(request.toolChoice, request.tools) : undefined),
   } satisfies OpenAIResponsesBody
 })
 
@@ -260,7 +263,7 @@ export const route = Route.make({
   endpoint,
   auth,
   transport,
-  defaults: { providerOptions: { openai: { store: false } } },
+  defaults: { providerOptions: { store: false } },
 })
 
 export * as OpenAIResponses from "./openai-responses.js"

packages/ai/src/protocols/shared.ts

@@ -41,12 +41,10 @@ export interface ToolAccumulator {
  * when at least one is defined. Returns `undefined` when neither input nor
  * output is known so routes don't publish a misleading `0`.
  *
- * Under the additive `AI.Usage` contract, `inputTokens` and `outputTokens`
+ * Under the inclusive `AI.Usage` contract, `inputTokens` includes cached input
- * are the non-cached input and visible output only. The provider-supplied
+ * and `outputTokens` includes reasoning. Protocol mappers normalize those
- * `total` is the source of truth when present; the computed fallback
+ * inclusive values before calling this helper. The provider-supplied total is
- * under-counts cache and reasoning by design and exists mainly so
+ * the source of truth when present; otherwise their sum is the canonical total.
- * Anthropic-style providers (which don't surface a total) still get a
- * sensible aggregate on the input + output axes.
  */
 export const totalTokens = (
   inputTokens: number | undefined,
@@ -67,7 +65,7 @@ export const totalTokens = (
  *
  * If `total` is `undefined`, returns `undefined` (we don't fabricate
  * counts). If `subtrahend` is `undefined`, returns `total` unchanged. The
- * provider-native breakdown stays available on `Usage.native` for debugging.
+ * provider-native breakdown stays available on `Usage.providerMetadata` for debugging.
  */
 export const subtractTokens = (total: number | undefined, subtrahend: number | undefined): number | undefined => {
   if (total === undefined) return undefined
@@ -199,8 +197,8 @@ export const errorText = (error: unknown) => {
 
 /**
  * `framing` step for Server-Sent Events. Decodes UTF-8, runs the SSE channel
- * decoder, and drops empty / `[DONE]` keep-alive events so the downstream
+ * decoder, and drops empty / `[DONE]` keep-alive events so the protocol event
- * `decodeChunk` sees one JSON string per element. The SSE channel emits a
+ * schema sees one JSON string per element. The SSE channel emits a
  * `Retry` control event on its error channel; we drop it here (we don't
  * implement client-driven retries). Decoder failures become provider output
  * errors so the public error channel stays `AIError`.
@@ -216,11 +214,7 @@ export const sseFraming = (bytes: Stream.Stream<Uint8Array, AIError>): Stream.St
   )
 
 /**
- * Canonical invalid-request constructor. Lift one-line `const invalid =
+ * Canonical invalid-request constructor shared by protocol lowering.
- * (message) => invalidRequest(message)` aliases out of every
- * route so the error constructor lives in one place. If we ever extend
- * `InvalidRequestReason` with route context or trace metadata, the change
- * lands here.
  */
 export const invalidRequest = (message: string) =>
   new AIError({

packages/ai/src/protocols/utils/bedrock-cache.ts

@@ -4,7 +4,7 @@ import { newBreakpoints, ttlBucket, type Breakpoints } from "./cache.js"
 
 // Bedrock cache markers are positional: emit a `cachePoint` block immediately
 // after the content the caller wants treated as a cacheable prefix. Bedrock
-// accepts optional `ttl: "5m" | "1h"` on cachePoint, mirroring Anthropic.
+// accepts optional `ttl: "5m" | "1h"` on cachePoint.
 export const CachePointBlock = Schema.Struct({
   cachePoint: Schema.Struct({
     type: Schema.tag("default"),
@@ -13,9 +13,8 @@ export const CachePointBlock = Schema.Struct({
 })
 export type CachePointBlock = Schema.Schema.Type<typeof CachePointBlock>
 
-// Bedrock-Claude enforces the same 4-breakpoint cap as the Anthropic Messages
+// Callers pass a shared counter through every `block()` call site so the
-// API. Callers pass a shared counter through every `block()` call site so the
+// four-breakpoint budget is respected across `system`, `messages`, and `tools`.
-// budget is respected across `system`, `messages`, and `tools`.
 export const BEDROCK_BREAKPOINT_CAP = 4
 
 export type { Breakpoints } from "./cache.js"

packages/ai/src/protocols/utils/cache.ts

@@ -1,6 +1,4 @@
-// Shared helpers for provider cache-marker lowering. Anthropic and Bedrock
+// Shared counter and TTL mapping for provider cache-marker lowering.
-// both enforce a 4-breakpoint cap per request and accept the same `5m`/`1h`
-// TTL buckets, so the counter and TTL mapping live here.
 
 export interface Breakpoints {
   remaining: number
@@ -9,8 +7,7 @@ export interface Breakpoints {
 
 export const newBreakpoints = (cap: number): Breakpoints => ({ remaining: cap, dropped: 0 })
 
-// Returns `"1h"` for any `ttlSeconds >= 3600`, otherwise `undefined` (the
+// Requests of at least one hour use the explicit `"1h"` bucket; shorter
-// provider default 5m). Anthropic & Bedrock both treat anything shorter than
+// requests omit the wire TTL and use the provider default.
-// an hour as 5m.
 export const ttlBucket = (ttlSeconds: number | undefined): "1h" | undefined =>
   ttlSeconds !== undefined && ttlSeconds >= 3600 ? "1h" : undefined

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

@@ -1,5 +1,19 @@
-import { Schema } from "effect"
+import { Option, Schema } from "effect"
-import { TextVerbosity, type LLMRequest } from "../../schema/index.js"
+import type { LLMRequest } from "../../schema/index.js"
+
+export const ReasoningEfforts = ["none", "minimal", "low", "medium", "high", "xhigh", "max"] as const
+export type ReasoningEffort = (typeof ReasoningEfforts)[number] | (string & {})
+export const ReasoningEffort = Schema.declare<ReasoningEffort>(
+  (value): value is ReasoningEffort => typeof value === "string",
+  { title: "ReasoningEffort" },
+)
+
+export const TextVerbosities = ["low", "medium", "high"] as const
+export type TextVerbosity = (typeof TextVerbosities)[number] | (string & {})
+export const TextVerbosity = Schema.declare<TextVerbosity>(
+  (value): value is TextVerbosity => typeof value === "string",
+  { title: "TextVerbosity" },
+)
 
 export const ResponseIncludables = [
   "file_search_call.results",
@@ -11,52 +25,67 @@ export const ResponseIncludables = [
   "reasoning.encrypted_content",
   "message.output_text.logprobs",
 ] as const
-export type ResponseIncludable = (typeof ResponseIncludables)[number]
+export type ResponseIncludable = (typeof ResponseIncludables)[number] | (string & {})
 
 export const ServiceTiers = ["auto", "default", "flex", "priority"] as const
 export type ServiceTier = (typeof ServiceTiers)[number]
 
-const TEXT_VERBOSITY = new Set<string>(["low", "medium", "high"])
+export const Truncations = ["auto", "disabled"] as const
-const INCLUDABLES = new Set<string>(ResponseIncludables)
+export type Truncation = (typeof Truncations)[number]
-const SERVICE_TIERS = new Set<string>(ServiceTiers)
 
-const isTextVerbosity = (value: unknown): value is Schema.Schema.Type<typeof TextVerbosity> =>
-  typeof value === "string" && TEXT_VERBOSITY.has(value)
-
-const isServiceTier = (value: unknown): value is ServiceTier => typeof value === "string" && SERVICE_TIERS.has(value)
-
-export const ReasoningEffort = Schema.String
 export const TextVerbositySchema = TextVerbosity
-export const ResponseIncludableSchema = Schema.Literals(ResponseIncludables)
+export const ResponseIncludableSchema = Schema.declare<ResponseIncludable>(
+  (value): value is ResponseIncludable => typeof value === "string",
+  { title: "ResponseIncludable" },
+)
 export const ServiceTierSchema = Schema.Literals(ServiceTiers)
+export const TruncationSchema = Schema.Literals(Truncations)
 
-export interface Resolved {
+export const AllowedTools = Schema.Struct({
-  readonly instructions?: string
+  toolNames: Schema.Array(Schema.String),
-  readonly store?: boolean
+  mode: Schema.optional(Schema.Literals(["auto", "none", "required"])),
-  readonly reasoningEffort?: string
+})
-  readonly reasoningSummary?: "auto" | "concise" | "detailed"
+export type AllowedTools = typeof AllowedTools.Type
-  readonly include?: ReadonlyArray<ResponseIncludable>
+
-  readonly textVerbosity?: Schema.Schema.Type<typeof TextVerbosity>
+export const StreamOptions = Schema.Struct({
-  readonly serviceTier?: ServiceTier
+  includeObfuscation: Schema.optional(Schema.Boolean),
+})
+
+export const Options = Schema.Struct({
+  instructions: Schema.optional(Schema.String),
+  store: Schema.optional(Schema.Boolean),
+  metadata: Schema.optional(Schema.Record(Schema.String, Schema.String)),
+  safetyIdentifier: Schema.optional(Schema.String),
+  streamOptions: Schema.optional(StreamOptions),
+  topLogprobs: Schema.optional(Schema.Int.check(Schema.isBetween({ minimum: 0, maximum: 20 }))),
+  reasoningEffort: Schema.optional(ReasoningEffort),
+  reasoningSummary: Schema.optional(Schema.Literals(["auto", "concise", "detailed"])),
+  include: Schema.optional(Schema.Array(ResponseIncludableSchema)),
+  textVerbosity: Schema.optional(TextVerbositySchema),
+  serviceTier: Schema.optional(ServiceTierSchema),
+  truncation: Schema.optional(TruncationSchema),
+  allowedTools: Schema.optional(AllowedTools),
+  maxToolCalls: Schema.optional(Schema.Int),
+  parallelToolCalls: Schema.optional(Schema.Boolean),
+})
+export type Options = typeof Options.Type
+
+export type Resolved = Omit<Options, "allowedTools"> & {
+  readonly allowedTools?: AllowedTools & { readonly mode: NonNullable<AllowedTools["mode"]> }
 }
 
+const decodeOptions = Schema.decodeUnknownOption(Options)
+
 export const resolve = (request: LLMRequest): Resolved => {
-  const input = request.providerOptions?.[request.model.route.providerMetadataKey ?? "openresponses"]
+  const input = Option.getOrUndefined(decodeOptions(request.providerOptions))
-  const include = Array.isArray(input?.include)
+  if (!input) return {}
-    ? input.include.filter((entry): entry is ResponseIncludable => INCLUDABLES.has(entry))
-    : []
-  const reasoningSummary = input?.reasoningSummary
   return {
-    instructions: typeof input?.instructions === "string" ? input.instructions : undefined,
+    ...input,
-    store: typeof input?.store === "boolean" ? input.store : undefined,
+    include: input.include?.length ? input.include : undefined,
-    reasoningEffort: typeof input?.reasoningEffort === "string" ? input.reasoningEffort : undefined,
+    allowedTools:
-    reasoningSummary:
+      input.allowedTools && input.allowedTools.toolNames.length > 0
-      reasoningSummary === "auto" || reasoningSummary === "concise" || reasoningSummary === "detailed"
+        ? { ...input.allowedTools, mode: input.allowedTools.mode ?? "auto" }
-        ? reasoningSummary
         : undefined,
-    include: include.length > 0 ? include : undefined,
-    textVerbosity: isTextVerbosity(input?.textVerbosity) ? input.textVerbosity : undefined,
-    serviceTier: isServiceTier(input?.serviceTier) ? input.serviceTier : undefined,
   }
 }
 

packages/ai/src/protocols/utils/openai-options.ts

@@ -1,8 +1,9 @@
-import { ReasoningEfforts } from "../../schema/index.js"
 import { OpenResponsesOptions } from "./open-responses-options.js"
 
-export const OpenAIReasoningEfforts = ReasoningEfforts
+export const OpenAIReasoningEfforts = OpenResponsesOptions.ReasoningEfforts
-export type OpenAIReasoningEffort = string
+export type OpenAIReasoningEffort = OpenResponsesOptions.ReasoningEffort
+export const OpenAITextVerbosities = OpenResponsesOptions.TextVerbosities
+export type OpenAITextVerbosity = OpenResponsesOptions.TextVerbosity
 
 // Mirrors OpenAI's `ResponseIncludable` union from the official SDK. Keep this
 // in lockstep with `openai-node/src/resources/responses/responses.ts`.
@@ -12,7 +13,7 @@ export const OpenAIServiceTiers = OpenResponsesOptions.ServiceTiers
 export type OpenAIServiceTier = OpenResponsesOptions.ServiceTier
 
 export const OpenAIReasoningEffort = OpenResponsesOptions.ReasoningEffort
-export const OpenAITextVerbosity = OpenResponsesOptions.TextVerbositySchema
+export const OpenAITextVerbosity = OpenResponsesOptions.TextVerbosity
 export const OpenAIResponseIncludable = OpenResponsesOptions.ResponseIncludableSchema
 export const OpenAIServiceTier = OpenResponsesOptions.ServiceTierSchema
 

packages/ai/src/provider-package.ts

@@ -4,11 +4,6 @@ export interface Settings extends Readonly<Record<string, unknown>> {
   readonly baseURL?: string
   readonly headers?: Readonly<Record<string, string>>
   readonly body?: Readonly<Record<string, unknown>>
-  readonly limits?: {
-    readonly context: number
-    readonly input?: number
-    readonly output: number
-  }
 }
 
 export interface Definition<

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

@@ -90,7 +90,6 @@ const config = (settings: Settings): Config => {
     credentials: settings.credentials,
     headers: settings.headers === undefined ? undefined : { ...settings.headers },
     http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    limits: settings.limits,
     providerOptions: settings.providerOptions,
     region: settings.region,
   }

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

@@ -62,7 +62,6 @@ export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, se
     generation: settings.topP === undefined ? undefined : { topP: settings.topP },
     headers: settings.headers === undefined ? undefined : { ...settings.headers },
     http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    limits: settings.limits,
     region: settings.region,
   }).model(modelID)
 }

packages/ai/src/providers/anthropic-compatible.ts

@@ -68,7 +68,6 @@ export const model: ProviderPackage.Definition<Settings, AnthropicMessages.Provi
     baseURL: settings.baseURL,
     headers: settings.headers === undefined ? undefined : { ...settings.headers },
     http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    limits: settings.limits,
     provider: settings.provider,
     providerOptions: settings.providerOptions,
   }).model(modelID)

packages/ai/src/providers/anthropic.ts

@@ -63,7 +63,6 @@ export const model: ProviderPackage.Definition<Settings, AnthropicMessages.Provi
     baseURL: settings.baseURL,
     headers: settings.headers === undefined ? undefined : { ...settings.headers },
     http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    limits: settings.limits,
     providerOptions: settings.providerOptions,
   }).model(modelID)
 }

packages/ai/src/providers/azure.ts

@@ -126,7 +126,6 @@ const config = (settings: Settings): Config => {
     apiVersion: settings.apiVersion,
     headers: settings.headers === undefined ? undefined : { ...settings.headers },
     http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    limits: settings.limits,
     providerOptions: settings.providerOptions,
     queryParams: settings.queryParams === undefined ? undefined : { ...settings.queryParams },
     useDeploymentBasedUrls: settings.useDeploymentBasedUrls,

packages/ai/src/providers/google-vertex-chat.ts

@@ -75,7 +75,6 @@ export const model: ProviderPackage.Definition<Settings, OpenAIProviderOptionsIn
     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,

packages/ai/src/providers/google-vertex-messages.ts

@@ -110,7 +110,6 @@ export const model: ProviderPackage.Definition<Settings, AnthropicMessages.Provi
     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,

packages/ai/src/providers/google-vertex-responses.ts

@@ -27,7 +27,7 @@ export interface Settings extends ProviderPackage.Settings {
 const route = OpenAICompatibleResponses.route.with({
   id: "google-vertex-responses",
   provider: id,
-  providerOptions: { openresponses: { store: false } },
+  providerOptions: { store: false },
 })
 
 export const routes = [route]
@@ -80,7 +80,6 @@ export const model: ProviderPackage.Definition<Settings, OpenResponsesProviderOp
     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,

packages/ai/src/providers/google-vertex.ts

@@ -6,16 +6,14 @@ import { Auth } from "../route/auth.js"
 import { Route, type RouteDefaultsInput } from "../route/client.js"
 import { Endpoint } from "../route/endpoint.js"
 import { Framing } from "../route/framing.js"
-import { ProviderID, type LLMRequest, type ModelID, type ProviderOptions } from "../schema/index.js"
+import { ProviderID, type LLMRequest, type ModelID } from "../schema/index.js"
 import { GoogleVertexShared } from "./google-vertex-shared.js"
 
 export interface GeminiOptionsInput extends Gemini.OptionsInput {
   readonly labels?: Readonly<Record<string, string>>
 }
 
-export type GeminiProviderOptionsInput = ProviderOptions & {
+export type GeminiProviderOptionsInput = GeminiOptionsInput
-  readonly gemini?: GeminiOptionsInput
-}
 
 export const id = ProviderID.make("google-vertex")
 
@@ -40,7 +38,7 @@ export type Settings = ProviderPackage.Settings &
 
 const fromRequest = Effect.fn("GoogleVertex.fromRequest")(function* (request: LLMRequest) {
   const body = yield* Gemini.protocol.body.from(request)
-  const value = request.providerOptions?.gemini?.labels
+  const value = request.providerOptions?.labels
   const labels = ProviderShared.isRecord(value)
     ? Object.fromEntries(
         Object.entries(value).filter((entry): entry is [string, string] => typeof entry[1] === "string"),
@@ -121,7 +119,6 @@ export const model: ProviderPackage.Definition<Settings, GeminiProviderOptionsIn
     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,

packages/ai/src/providers/google.ts

@@ -63,7 +63,6 @@ export const model: ProviderPackage.Definition<Settings, Gemini.ProviderOptionsI
     baseURL: settings.baseURL,
     headers: settings.headers === undefined ? undefined : { ...settings.headers },
     http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    limits: settings.limits,
     providerOptions: settings.providerOptions,
   }).model(modelID)
 

packages/ai/src/providers/open-responses-options.ts

@@ -1,19 +1,6 @@
-import type { ResponseIncludable, ServiceTier } from "../protocols/utils/open-responses-options.js"
+import type { Options } from "../protocols/utils/open-responses-options.js"
-import type { ProviderOptions, ReasoningEffort, TextVerbosity } from "../schema/index.js"
 
-export interface OpenResponsesOptionsInput {
+export type OpenResponsesOptionsInput = Options & { readonly [key: string]: unknown }
-  readonly [key: string]: unknown
+export type OpenResponsesProviderOptionsInput = OpenResponsesOptionsInput
-  readonly instructions?: string
-  readonly store?: boolean
-  readonly reasoningEffort?: ReasoningEffort
-  readonly reasoningSummary?: "auto" | "concise" | "detailed"
-  readonly include?: ReadonlyArray<ResponseIncludable>
-  readonly textVerbosity?: TextVerbosity
-  readonly serviceTier?: ServiceTier
-}
-
-export type OpenResponsesProviderOptionsInput = ProviderOptions & {
-  readonly openresponses?: OpenResponsesOptionsInput
-}
 
 export * as OpenResponsesProviderOptions from "./open-responses-options.js"

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

@@ -55,7 +55,6 @@ export const model: ProviderPackage.Definition<Settings, OpenResponsesProviderOp
     baseURL: settings.baseURL,
     headers: settings.headers === undefined ? undefined : { ...settings.headers },
     http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    limits: settings.limits,
     provider: settings.provider,
     providerOptions: settings.providerOptions,
   }).model(modelID)

packages/ai/src/providers/openai-compatible.ts

@@ -19,6 +19,7 @@ export interface Settings extends ProviderPackage.Settings {
   readonly apiKey?: string
   readonly baseURL: string
   readonly provider?: string
+  readonly providerOptions?: OpenAIProviderOptionsInput
 }
 
 export type FamilyModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
@@ -73,8 +74,8 @@ export const model: ProviderPackage.Definition<Settings, OpenAIProviderOptionsIn
     baseURL: settings.baseURL,
     headers: settings.headers === undefined ? undefined : { ...settings.headers },
     http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    limits: settings.limits,
     provider: settings.provider,
+    providerOptions: settings.providerOptions,
   }).model(modelID)
 
 export const baseten = define(profiles.baseten)

packages/ai/src/providers/openai-options.ts

@@ -1,20 +1,17 @@
-import type { ProviderOptions } from "../schema/index.js"
+import { mergeProviderOptions, type ProviderOptions } from "../schema/index.js"
-import { mergeProviderOptions } from "../schema/index.js"
 import type { OpenResponsesOptionsInput } from "./open-responses-options.js"
 
 export type { OpenAIResponseIncludable, OpenAIServiceTier } from "../protocols/utils/openai-options.js"
 
 export type OpenAIOptionsInput = OpenResponsesOptionsInput
 
-export type OpenAIProviderOptionsInput = ProviderOptions & {
+export type OpenAIProviderOptionsInput = OpenAIOptionsInput
-  readonly openai?: OpenAIOptionsInput
-}
 
 const definedEntries = (input: Record<string, unknown>) =>
   Object.entries(input).filter((entry) => entry[1] !== undefined)
 
 const openAIProviderOptions = (options: OpenAIOptionsInput | undefined): ProviderOptions | undefined => {
-  const openai = Object.fromEntries(
+  const result = Object.fromEntries(
     definedEntries({
       store: options?.store,
       reasoningEffort: options?.reasoningEffort,
@@ -24,8 +21,8 @@ const openAIProviderOptions = (options: OpenAIOptionsInput | undefined): Provide
       serviceTier: options?.serviceTier,
     }),
   )
-  if (Object.keys(openai).length === 0) return undefined
+  if (Object.keys(result).length === 0) return undefined
-  return { openai }
+  return result
 }
 
 export const gpt5DefaultOptions = (

packages/ai/src/providers/openai.ts

@@ -124,7 +124,6 @@ const config = (settings: Settings): Config => {
     baseURL: settings.baseURL,
     headers: Object.keys(headers).length === 0 ? undefined : headers,
     http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    limits: settings.limits,
     providerOptions: settings.providerOptions,
     queryParams: settings.queryParams === undefined ? undefined : { ...settings.queryParams },
   }

packages/ai/src/providers/openrouter.ts

@@ -4,7 +4,7 @@ import { Endpoint } from "../route/endpoint.js"
 import { Framing } from "../route/framing.js"
 import { Protocol } from "../route/protocol.js"
 import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
-import { ProviderID, type CacheHint, type ModelID, type ProviderOptions } from "../schema/index.js"
+import { ProviderID, type CacheHint, type ModelID } from "../schema/index.js"
 import type { ProviderPackage } from "../provider-package.js"
 import * as OpenAICompatibleProfiles from "./openai-compatible-profile.js"
 import * as OpenAIChat from "../protocols/openai-chat.js"
@@ -71,9 +71,7 @@ export interface OpenRouterOptions {
   }>
 }
 
-export type OpenRouterProviderOptionsInput = ProviderOptions & {
+export type OpenRouterProviderOptionsInput = OpenRouterOptions
-  readonly openrouter?: OpenRouterOptions
-}
 
 export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
   ProviderAuthOption<"optional"> & {
@@ -120,7 +118,7 @@ export const protocol = Protocol.make({
           return {
             ...body,
             messages,
-            ...bodyOptions(request.providerOptions?.openrouter),
+            ...bodyOptions(request.providerOptions),
             ...(request.promptCacheKey ? { prompt_cache_key: request.promptCacheKey } : {}),
           } as OpenRouterBody
         }),
@@ -202,6 +200,5 @@ export const model: ProviderPackage.Definition<Settings, OpenRouterProviderOptio
     baseURL: settings.baseURL,
     headers: settings.headers,
     http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    limits: settings.limits,
     providerOptions: settings.providerOptions,
   }).model(modelID)

packages/ai/src/providers/xai.ts

@@ -1,7 +1,7 @@
 import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
 import { Route, type RouteDefaultsInput } from "../route/client.js"
 import { Endpoint } from "../route/endpoint.js"
-import { HttpOptions, ProviderID, type ModelID, type ProviderOptions } from "../schema/index.js"
+import { HttpOptions, ProviderID, type ModelID } from "../schema/index.js"
 import * as OpenAICompatibleProfiles from "./openai-compatible-profile.js"
 import * as OpenAICompatibleChat from "../protocols/openai-compatible-chat.js"
 import * as OpenAIChat from "../protocols/openai-chat.js"
@@ -12,9 +12,7 @@ import type { ProviderPackage } from "../provider-package.js"
 
 export const id = ProviderID.make("xai")
 
-export type XAIProviderOptionsInput = ProviderOptions & {
+export type XAIProviderOptionsInput = OpenAIOptionsInput
-  readonly xai?: OpenAIOptionsInput
-}
 
 export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
   ProviderAuthOption<"optional"> & {
@@ -37,7 +35,7 @@ const responsesRoute = Route.make({
   protocol: OpenAIResponses.protocol,
   endpoint: Endpoint.path("/responses", { baseURL: OpenAICompatibleProfiles.profiles.xai.baseURL }),
   transport: OpenAIResponses.httpTransport,
-  defaults: { providerOptions: { xai: { store: false } } },
+  defaults: { providerOptions: { store: false } },
 })
 
 const chatRoute = Route.make({
@@ -103,7 +101,6 @@ export const model: ProviderPackage.Definition<Settings, XAIProviderOptionsInput
     baseURL: settings.baseURL,
     headers: settings.headers,
     http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    limits: settings.limits,
     providerOptions: settings.providerOptions,
   }).model(modelID)
 export const responses = provider.responses

packages/ai/src/route/client.ts

@@ -16,7 +16,6 @@ import {
   LLMRequest,
   LLMResponse,
   LanguageModel,
-  LanguageModelLimits,
   LLMEvent,
   InvalidProviderOutputReason,
   ProviderID,
@@ -74,7 +73,6 @@ export type RouteRoutedLanguageModelInput = Omit<LanguageModel.Input, "route">
 
 export interface RouteDefaults {
   readonly headers?: Record<string, string>
-  readonly limits?: LanguageModelLimits
   readonly generation?: GenerationOptions
   readonly providerOptions?: ProviderOptions
   readonly http?: HttpOptions
@@ -82,7 +80,6 @@ export interface RouteDefaults {
 
 export interface RouteDefaultsInput {
   readonly headers?: Record<string, string>
-  readonly limits?: LanguageModelLimits.Input
   readonly generation?: GenerationOptions.Input
   readonly providerOptions?: ProviderOptions
   readonly http?: HttpOptions.Input
@@ -119,7 +116,6 @@ const mergeRouteDefaults = (base: RouteDefaults | undefined, patch: RouteDefault
     ...base,
     ...patch,
     headers,
-    limits: patch.limits === undefined ? base?.limits : LanguageModelLimits.make(patch.limits),
     generation: mergeGenerationOptions(generationOptions(base?.generation), generationOptions(patch.generation)),
     providerOptions: mergeProviderOptions(base?.providerOptions, patch.providerOptions),
     http: mergeHttpOptions(

packages/ai/src/route/framing.ts

@@ -13,8 +13,8 @@ import type { AIError } from "../schema/index.js"
  * - AWS event stream — length-prefixed binary frames with CRC checksums.
  *   Each emitted frame is one parsed binary event record.
  *
- * The frame type is opaque to this layer; the protocol's `decode` step turns
+ * The frame type is opaque to this layer; the protocol's event schema decodes
- * a frame into a typed chunk.
+ * each frame before its state machine handles it.
  */
 export interface Definition<Frame> {
   readonly id: string

packages/ai/src/route/protocol.ts

@@ -73,8 +73,7 @@ export interface ProtocolStream<Frame, Event, State> {
  *
  * Provider implementations should usually call `Protocol.make({ ... })`
  * without explicit type arguments; the schemas and parser functions are the
- * source of truth. The constructor remains as the public seam for future
+ * source of truth.
- * cross-cutting concerns such as tracing or instrumentation.
  */
 export const make = <Body, Frame, Event, State>(
   input: Protocol<Body, Frame, Event, State>,

packages/ai/src/schema/errors.ts

@@ -1,6 +1,7 @@
 import { Schema } from "effect"
 import { Tool } from "@opencode-ai/schema/tool"
-import { ModelID, ProviderID, ProviderMetadata, RouteID } from "./ids.js"
+import { ModelID, ProviderID, RouteID } from "./ids.js"
+import { ProviderMetadata } from "./messages.js"
 
 export const ProviderFailureClassification = Schema.Literals(["context-overflow", "payload-too-large"])
 export type ProviderFailureClassification = typeof ProviderFailureClassification.Type

packages/ai/src/schema/events.ts

@@ -1,8 +1,21 @@
 import { Schema } from "effect"
-import { ContentBlockID, FinishReason, ProviderMetadata, ToolCallID } from "./ids.js"
+import { LLM } from "@opencode-ai/schema/llm"
-import { Message, ToolCallPart, ToolOutput, ToolResultPart, ToolResultValue, type ContentPart } from "./messages.js"
+import { ContentBlockID, ToolCallID } from "./ids.js"
+import {
+  Message,
+  ProviderMetadata,
+  ToolCallPart,
+  ToolOutput,
+  ToolResultPart,
+  ToolResultValue,
+  type ContentPart,
+} from "./messages.js"
 import { ProviderFailureClassification } from "./errors.js"
 
+export const FinishReason = LLM.FinishReason
+export type FinishReason = Schema.Schema.Type<typeof FinishReason>
+export { ProviderMetadata } from "./messages.js"
+
 /**
  * Token usage reported by an LLM provider.
  *

packages/ai/src/schema/ids.ts

@@ -1,8 +1,4 @@
 import { Schema } from "effect"
-import { ProviderMetadata } from "@opencode-ai/schema/ai"
-import { LLM } from "@opencode-ai/schema/llm"
-
-export { ProviderMetadata }
 
 /** Stable string identifier for a protocol implementation. */
 export const ProtocolID = Schema.String
@@ -26,19 +22,3 @@ export type ContentBlockID = Schema.Schema.Type<typeof ContentBlockID>
 
 export const ToolCallID = Schema.String
 export type ToolCallID = Schema.Schema.Type<typeof ToolCallID>
-
-export const ReasoningEfforts = ["none", "minimal", "low", "medium", "high", "xhigh", "max"] as const
-export const ReasoningEffort = Schema.String
-export type ReasoningEffort = Schema.Schema.Type<typeof ReasoningEffort>
-
-export const TextVerbosity = Schema.Literals(["low", "medium", "high"])
-export type TextVerbosity = Schema.Schema.Type<typeof TextVerbosity>
-
-export const MessageRole = Schema.Literals(["system", "user", "assistant", "tool"])
-export type MessageRole = Schema.Schema.Type<typeof MessageRole>
-
-export const FinishReason = LLM.FinishReason
-export type FinishReason = Schema.Schema.Type<typeof FinishReason>
-
-export const JsonSchema = Schema.Record(Schema.String, Schema.Unknown)
-export type JsonSchema = Schema.Schema.Type<typeof JsonSchema>

packages/ai/src/schema/messages.ts

@@ -1,16 +1,24 @@
 import { Schema } from "effect"
 import { Tool } from "@opencode-ai/schema/tool"
-import { JsonSchema, MessageRole, ProviderMetadata } from "./ids.js"
 import {
   CacheHint,
   CachePolicy,
   GenerationOptions,
   HttpOptions,
+  JsonSchema,
   LanguageModelSchema,
   ProviderOptions,
 } from "./options.js"
 import { isRecord } from "../utils/record.js"
 
+export const MessageRole = Schema.Literals(["system", "user", "assistant", "tool"])
+export type MessageRole = Schema.Schema.Type<typeof MessageRole>
+
+export const ProviderMetadata = Schema.Record(Schema.String, Schema.Record(Schema.String, Schema.Unknown)).annotate({
+  identifier: "LLM.ProviderMetadata",
+})
+export type ProviderMetadata = Schema.Schema.Type<typeof ProviderMetadata>
+
 const systemPartSchema = Schema.Struct({
   type: Schema.Literal("text"),
   text: Schema.String,

packages/ai/src/schema/options.ts

@@ -1,8 +1,11 @@
 import { Schema } from "effect"
-import { JsonSchema, ModelID, ProviderID } from "./ids.js"
+import { ModelID, ProviderID } from "./ids.js"
 import type { AnyRoute } from "../route/client.js"
 import { isRecord } from "../utils/record.js"
 
+export const JsonSchema = Schema.Record(Schema.String, Schema.Unknown)
+export type JsonSchema = Schema.Schema.Type<typeof JsonSchema>
+
 export const mergeJsonRecords = (
   ...items: ReadonlyArray<Record<string, unknown> | undefined>
 ): Record<string, unknown> | undefined => {
@@ -33,22 +36,12 @@ const mergeStringRecords = (
   return Object.keys(result).length === 0 ? undefined : result
 }
 
-export const ProviderOptions = Schema.Record(Schema.String, Schema.Record(Schema.String, Schema.Unknown))
+export const ProviderOptions = Schema.Record(Schema.String, Schema.Unknown)
 export type ProviderOptions = Schema.Schema.Type<typeof ProviderOptions>
 
 export const mergeProviderOptions = (
   ...items: ReadonlyArray<ProviderOptions | undefined>
-): ProviderOptions | undefined => {
+): ProviderOptions | undefined => mergeJsonRecords(...items)
-  const result: Record<string, Record<string, unknown>> = {}
-  for (const item of items) {
-    if (!item) continue
-    for (const [provider, options] of Object.entries(item)) {
-      const merged = mergeJsonRecords(result[provider], options)
-      if (merged) result[provider] = merged
-    }
-  }
-  return Object.keys(result).length === 0 ? undefined : result
-}
 
 export class HttpOptions extends Schema.Class<HttpOptions>("AI.HttpOptions")({
   body: Schema.optional(JsonSchema),
@@ -121,22 +114,7 @@ export const mergeGenerationOptions = (...items: ReadonlyArray<GenerationOptions
   return Object.values(result).some((value) => value !== undefined) ? result : undefined
 }
 
-export class LanguageModelLimits extends Schema.Class<LanguageModelLimits>("LLM.LanguageModelLimits")({
-  context: Schema.optional(Schema.Number),
-  input: Schema.optional(Schema.Number),
-  output: Schema.optional(Schema.Number),
-}) {}
-
-export namespace LanguageModelLimits {
-  export type Input = LanguageModelLimits | ConstructorParameters<typeof LanguageModelLimits>[0]
-
-  /** Normalize model limit input into the canonical `LanguageModelLimits` class. */
-  export const make = (input: Input | undefined) =>
-    input instanceof LanguageModelLimits ? input : new LanguageModelLimits(input ?? {})
-}
-
 export class LanguageModelDefaults extends Schema.Class<LanguageModelDefaults>("LLM.LanguageModelDefaults")({
-  limits: Schema.optional(LanguageModelLimits),
   generation: Schema.optional(GenerationOptions),
   providerOptions: Schema.optional(ProviderOptions),
   http: Schema.optional(HttpOptions),
@@ -146,7 +124,6 @@ export namespace LanguageModelDefaults {
   export type Input =
     | LanguageModelDefaults
     | {
-        readonly limits?: LanguageModelLimits.Input
         readonly generation?: GenerationOptions.Input
         readonly providerOptions?: ProviderOptions
         readonly http?: HttpOptions.Input
@@ -156,7 +133,6 @@ export namespace LanguageModelDefaults {
   export const make = (input: Input) => {
     if (input instanceof LanguageModelDefaults) return input
     return new LanguageModelDefaults({
-      limits: input.limits === undefined ? undefined : LanguageModelLimits.make(input.limits),
       generation: input.generation === undefined ? undefined : GenerationOptions.make(input.generation),
       providerOptions: input.providerOptions,
       http: input.http === undefined ? undefined : HttpOptions.make(input.http),
@@ -269,10 +245,9 @@ export class CacheHint extends Schema.Class<CacheHint>("LLM.CacheHint")({
 // Auto-placement policy for prompt caching. The protocol-neutral lowering step
 // reads this and injects `CacheHint`s at the configured boundaries; the
 // per-protocol body builders then translate those hints into wire markers as
-// usual. `"auto"` is the recommended default for agent loops — it places
+// usual. `"auto"` is the default for agent loops — it places
 // breakpoints at the last tool definition, the first and last distinct system
-// parts, and the conversation tail. The rolling message breakpoint keeps a
+// parts, and the conversation tail so recent prefixes remain reusable during
-// prior cache entry within Anthropic/Bedrock's 20-block lookback during long
 // tool loops.
 //
 // Pass `"none"` to opt out entirely (the legacy behavior). Pass the granular

packages/ai/src/testing.ts

@@ -7,6 +7,7 @@ import {
   type FinishReasonDetails,
   type AIError,
   type LLMRequest,
+  type ProviderMetadata,
   type UsageInput,
 } from "./schema/index.js"
 import { Context, Deferred, Effect, Latch, Layer, Queue, Scope, Stream } from "effect"
@@ -33,13 +34,22 @@ export interface LayerOptions {
 export class Service extends Context.Service<Service, Interface>()("@opencode/ai/TestLLM") {}
 
 export const complete = (
-  options: { readonly reason: FinishReasonDetails; readonly usage?: UsageInput },
+  options: {
+    readonly reason: FinishReasonDetails
+    readonly usage?: UsageInput
+    readonly providerMetadata?: ProviderMetadata
+  },
   ...events: readonly LLMEvent[]
 ) => [
   LLMEvent.stepStart({ index: 0 }),
   ...events,
-  LLMEvent.stepFinish({ index: 0, reason: options.reason, usage: options.usage }),
+  LLMEvent.stepFinish({
-  LLMEvent.finish({ reason: options.reason }),
+    index: 0,
+    reason: options.reason,
+    usage: options.usage,
+    providerMetadata: options.providerMetadata,
+  }),
+  LLMEvent.finish({ reason: options.reason, providerMetadata: options.providerMetadata }),
 ]
 
 export const stop = (...events: readonly LLMEvent[]) => complete({ reason: { normalized: "stop" } }, ...events)

packages/ai/test/auth-options.types.ts

@@ -81,7 +81,7 @@ OpenAI.configure({
 }).responses("gpt-4.1-mini")
 OpenAI.configure({
   generation: { maxTokens: 100 },
-  providerOptions: { openai: { store: false } },
+  providerOptions: { store: false },
 }).responses("gpt-4.1-mini")
 
 // @ts-expect-error OpenAI model selectors only accept model ids.
@@ -97,7 +97,7 @@ OpenAI.configure({ bogus: true })
 OpenAI.configure({ generation: { maxTokens: "many" } })
 
 // @ts-expect-error provider-native options remain typed.
-OpenAI.configure({ providerOptions: { openai: { store: "false" } } })
+OpenAI.configure({ providerOptions: { store: "false" } })
 
 // @ts-expect-error auth is an override, so OpenAI rejects apiKey with auth.
 OpenAI.configure({ apiKey: "sk-test", auth: Auth.bearer("oauth-token") })
@@ -139,7 +139,8 @@ Anthropic.configure({ apiKey: "anthropic-key" }).model("claude-haiku")
 Anthropic.configure({
   apiKey: "anthropic-key",
   providerOptions: {
-    anthropic: { thinking: { type: "enabled", budgetTokens: 1_024 }, effort: "high" },
+    thinking: { type: "enabled", budgetTokens: 1_024 },
+    effort: "high",
   },
 }).model("claude-haiku")
 // @ts-expect-error Anthropic model selectors only accept model ids.
@@ -147,15 +148,15 @@ Anthropic.configure({ apiKey: "anthropic-key" }).model("claude-haiku", {})
 // @ts-expect-error Anthropic package settings accept only one auth source.
 Anthropic.model("claude-sonnet-4-6", { apiKey: "anthropic-key", authToken: "anthropic-token" })
 // @ts-expect-error Enabled Anthropic thinking requires a token budget.
-Anthropic.configure({ providerOptions: { anthropic: { thinking: { type: "enabled" } } } })
+Anthropic.configure({ providerOptions: { thinking: { type: "enabled" } } })
 // @ts-expect-error Anthropic thinking budgets must be numbers.
-Anthropic.configure({ providerOptions: { anthropic: { thinking: { type: "enabled", budgetTokens: "large" } } } })
+Anthropic.configure({ providerOptions: { thinking: { type: "enabled", budgetTokens: "large" } } })
 
 AnthropicCompatible.configure({
   apiKey: "messages-key",
   baseURL: "https://messages.example.com/v1",
   provider: "example",
-  providerOptions: { anthropic: { thinking: { type: "disabled" } } },
+  providerOptions: { thinking: { type: "disabled" } },
 }).model("compatible-model")
 // @ts-expect-error Anthropic-compatible providers require a base URL.
 AnthropicCompatible.configure({ apiKey: "messages-key" })
@@ -171,16 +172,16 @@ AnthropicCompatible.model("compatible-model", {
 Google.configure({ apiKey: "google-key" }).model("gemini-2.5-flash")
 Google.configure({
   apiKey: "google-key",
-  providerOptions: { gemini: { thinkingConfig: { thinkingBudget: 0, includeThoughts: false } } },
+  providerOptions: { thinkingConfig: { thinkingBudget: 0, includeThoughts: false } },
 }).model("gemini-2.5-flash")
 // @ts-expect-error Google model selectors only accept model ids.
 Google.configure({ apiKey: "google-key" }).model("gemini-2.5-flash", {})
 // @ts-expect-error Gemini thinking budgets must be numbers.
-Google.configure({ providerOptions: { gemini: { thinkingConfig: { thinkingBudget: "large" } } } })
+Google.configure({ providerOptions: { thinkingConfig: { thinkingBudget: "large" } } })
 
 GoogleVertex.configure({
   apiKey: "vertex-key",
-  providerOptions: { gemini: { thinkingConfig: { thinkingBudget: 1_024 } } },
+  providerOptions: { thinkingConfig: { thinkingBudget: 1_024 } },
 }).model("gemini-3.5-flash")
 GoogleVertex.configure({ accessToken: "vertex-token", project: "project" }).model("gemini-3.5-flash")
 GoogleVertex.configure({ auth: Auth.bearer("vertex-token"), project: "project" }).model("gemini-3.5-flash")
@@ -230,7 +231,7 @@ GoogleVertexResponses.configure({
 GoogleVertexMessages.configure({
   accessToken: "vertex-token",
   project: "project",
-  providerOptions: { anthropic: { thinking: { type: "adaptive", display: "omitted" }, effort: "low" } },
+  providerOptions: { thinking: { type: "adaptive", display: "omitted" }, effort: "low" },
 }).model("claude-sonnet-4-6")
 // @ts-expect-error Vertex Messages package settings do not accept API keys.
 GoogleVertexMessages.model("claude-sonnet-4-6", { apiKey: "vertex-key", project: "project" })

packages/ai/test/compile.test.ts

@@ -17,31 +17,25 @@ describe("request option precedence", () => {
   test("deep-merges provider option records and replaces arrays, primitives, and null", () => {
     const merged = mergeProviderOptions(
       {
-        openai: {
+        include: ["route"],
-          include: ["route"],
+        metadata: { route: true, shared: "route" },
-          metadata: { route: true, shared: "route" },
+        nullable: "route",
-          nullable: "route",
+        primitive: "route",
-          primitive: "route",
-        },
       },
       {
-        openai: {
+        include: ["model"],
-          include: ["model"],
+        metadata: { model: true, shared: "model" },
-          metadata: { model: true, shared: "model" },
+        nullable: null,
-          nullable: null,
+        primitive: "model",
-          primitive: "model",
-        },
       },
-      { openai: { metadata: { request: true }, primitive: false } },
+      { metadata: { request: true }, primitive: false },
     )
 
     expect(merged).toEqual({
-      openai: {
+      include: ["model"],
-        include: ["model"],
+      metadata: { route: true, model: true, request: true, shared: "model" },
-        metadata: { route: true, model: true, request: true, shared: "model" },
+      nullable: null,
-        nullable: null,
+      primitive: false,
-        primitive: false,
-      },
     })
   })
 
@@ -51,13 +45,13 @@ describe("request option precedence", () => {
         endpoint: { baseURL: "https://api.openai.test/v1/" },
         auth: Auth.bearer("test"),
         generation: { maxTokens: 10, temperature: 1, stop: ["route"] },
-        providerOptions: { openai: { store: false, reasoningEffort: "low" } },
+        providerOptions: { store: false, reasoningEffort: "low" },
       })
       const model = route.model({
         id: "gpt-4o-mini",
         defaults: {
           generation: { maxTokens: 20, temperature: 0.5, frequencyPenalty: 0.25, stop: ["model"] },
-          providerOptions: { openai: { reasoningEffort: "medium" } },
+          providerOptions: { reasoningEffort: "medium" },
         },
       })
       const prepared = yield* compileRequest(
@@ -65,7 +59,7 @@ describe("request option precedence", () => {
           model,
           prompt: "Say hello.",
           generation: { maxTokens: 30, topP: 0.9, stop: ["request"] },
-          providerOptions: { openai: { store: true } },
+          providerOptions: { store: true },
         }),
       )
 
@@ -276,21 +270,20 @@ describe("request option precedence", () => {
     ),
   )
 
-  it.effect("uses model output limits after route limits and before call maxTokens", () =>
+  it.effect("uses the Anthropic default before call maxTokens", () =>
     Effect.gen(function* () {
       const route = AnthropicMessages.route.with({
         endpoint: { baseURL: "https://api.anthropic.test/v1/" },
         auth: Auth.header("x-api-key", "test"),
-        limits: { output: 128 },
       })
-      const model = route.model({ id: "claude-sonnet-4-5", defaults: { limits: { output: 64 } } })
+      const model = route.model({ id: "claude-sonnet-4-5" })
       const withoutMaxTokens = yield* compileRequest(LLM.request({ model, prompt: "Say hello.", cache: "none" }))
       const withMaxTokens = yield* compileRequest(
-        LLM.request({ model, prompt: "Say hello.", cache: "none", generation: { maxTokens: 32 } }),
+        LLM.request({ model, prompt: "Say hello.", cache: "none", generation: { maxTokens: 8_000 } }),
       )
 
-      expect(withoutMaxTokens.body.max_tokens).toBe(64)
+      expect(withoutMaxTokens.body.max_tokens).toBe(32_000)
-      expect(withMaxTokens.body.max_tokens).toBe(32)
+      expect(withMaxTokens.body.max_tokens).toBe(8_000)
     }),
   )
 })

packages/ai/test/continuation-scenarios.ts

@@ -105,7 +105,7 @@ export function continuationRequest(input: {
     tools: features.has("tool-call") ? [continuationTool] : [],
     cache: "none",
     providerOptions: features.has("encrypted-reasoning")
-      ? { openai: { store: false, include: ["reasoning.encrypted_content"], reasoningSummary: "auto" } }
+      ? { store: false, include: ["reasoning.encrypted_content"], reasoningSummary: "auto" }
       : undefined,
     generation: { maxTokens: 80, temperature: 0 },
   })

packages/ai/test/llm-option-types.types.ts

@@ -13,9 +13,7 @@ interface ExampleOptions {
   readonly mode?: "fast" | "thorough"
 }
 
-type ExampleProviderOptions = ProviderOptions & {
+type ExampleProviderOptions = ProviderOptions & ExampleOptions
-  readonly example?: ExampleOptions
-}
 
 const model = OpenAIChat.route
   .with({ endpoint: { baseURL: "https://example.com/v1" } })
@@ -26,7 +24,7 @@ type StreamRequirements<T> = T extends Stream.Stream<infer _A, infer _E, infer R
 type Equal<A, B> = [A, B] extends [B, A] ? true : false
 type Assert<T extends true> = T
 
-LLM.request({ model, prompt: "Hello", providerOptions: { example: { mode: "fast" } } })
+LLM.request({ model, prompt: "Hello", providerOptions: { mode: "fast" } })
 LLM.request({ model, prompt: "Hello", providerOptions: { future: { option: true } } })
 
 const generated = LLM.generate(LLM.request({ model, prompt: "Hello" }))
@@ -38,14 +36,14 @@ LLM.request({
   model,
   prompt: "Hello",
   // @ts-expect-error Known provider options preserve their value types.
-  providerOptions: { example: { mode: "slow" } },
+  providerOptions: { mode: "slow" },
 })
 
 const generatedObject = LLM.generateObject({
   model,
   prompt: "Hello",
   schema: Schema.Struct({ answer: Schema.String }),
-  providerOptions: { example: { mode: "thorough" } },
+  providerOptions: { mode: "thorough" },
 })
 type GenerateObjectRequirements = Assert<Equal<Requirements<typeof generatedObject>, LLMClientService>>
 
@@ -61,13 +59,13 @@ LLM.generateObject({
   prompt: "Hello",
   jsonSchema: { type: "object" },
   // @ts-expect-error Dynamic object generation uses the selected model's provider options.
-  providerOptions: { example: { mode: false } },
+  providerOptions: { mode: false },
 })
 
 declare const generic: LanguageModel
 LLM.request({ model: generic, prompt: "Hello", providerOptions: { arbitrary: { option: true } } })
 
-const options: LanguageModelProviderOptions<typeof model> = { example: { mode: "fast" } }
+const options: LanguageModelProviderOptions<typeof model> = { mode: "fast" }
 void (options satisfies LanguageModelProviderOptions<typeof model>)
 void (true satisfies GenerateRequirements)
 void (true satisfies StreamClientRequirements)

packages/ai/test/llm.test.ts

@@ -59,18 +59,18 @@ describe("llm constructors", () => {
         provider: "fake",
         route: chatRoute.with({
           generation: { maxTokens: 100, temperature: 1 },
-          providerOptions: { openai: { store: false, metadata: { model: true } } },
+          providerOptions: { store: false, metadata: { model: true } },
           http: { body: { metadata: { model: true } }, headers: { "x-shared": "model" }, query: { model: "1" } },
         }),
       }),
       prompt: "Say hello.",
       generation: { temperature: 0 },
-      providerOptions: { openai: { store: true, metadata: { request: true } } },
+      providerOptions: { store: true, metadata: { request: true } },
       http: { body: { metadata: { request: true } }, headers: { "x-shared": "request" }, query: { request: "1" } },
     })
 
     expect(request.generation).toEqual({ temperature: 0 })
-    expect(request.providerOptions).toEqual({ openai: { store: true, metadata: { request: true } } })
+    expect(request.providerOptions).toEqual({ store: true, metadata: { request: true } })
     expect(request.http).toEqual({
       body: { metadata: { request: true } },
       headers: { "x-shared": "request" },
@@ -121,18 +121,16 @@ describe("llm constructors", () => {
     const model = chatRoute.model({
       id: "kimi-k2",
       defaults: {
-        limits: { context: 128_000, output: 8_192 },
         generation: { maxTokens: 1_024, stop: ["END"] },
-        providerOptions: { openai: { parallelToolCalls: false } },
+        providerOptions: { parallelToolCalls: false },
         http: { body: { extra_body: true } },
       },
       compatibility: { toolSchema: "moonshot" },
     })
     const request = LLM.request({ model, prompt: "Say hello." })
 
-    expect(request.model.defaults?.limits).toEqual({ context: 128_000, output: 8_192 })
     expect(request.model.defaults?.generation).toEqual({ maxTokens: 1_024, stop: ["END"] })
-    expect(request.model.defaults?.providerOptions).toEqual({ openai: { parallelToolCalls: false } })
+    expect(request.model.defaults?.providerOptions).toEqual({ parallelToolCalls: false })
     expect(request.model.defaults?.http).toEqual({ body: { extra_body: true } })
     expect(request.model.compatibility).toEqual({ toolSchema: "moonshot" })
     expect(request.generation).toBeUndefined()

packages/ai/test/provider-options/anthropic-compatible.types.ts

@@ -3,11 +3,11 @@ import { AnthropicCompatible } from "../../src/providers.js"
 
 const model = AnthropicCompatible.configure({ baseURL: "https://example.com" }).model("claude")
 
-LLM.request({ model, prompt: "Hello", providerOptions: { anthropic: { effort: "high" } } })
+LLM.request({ model, prompt: "Hello", providerOptions: { effort: "high" } })
 
 LLM.request({
   model,
   prompt: "Hello",
   // @ts-expect-error Anthropic effort must be a string.
-  providerOptions: { anthropic: { effort: 1 } },
+  providerOptions: { effort: 1 },
 })

packages/ai/test/provider-options/anthropic.types.ts

@@ -3,11 +3,11 @@ import { Anthropic } from "../../src/providers.js"
 
 const model = Anthropic.provider.model("claude-sonnet-4-5")
 
-LLM.request({ model, prompt: "Hello", providerOptions: { anthropic: { thinking: { type: "adaptive" } } } })
+LLM.request({ model, prompt: "Hello", providerOptions: { thinking: { type: "adaptive" } } })
 
 LLM.request({
   model,
   prompt: "Hello",
   // @ts-expect-error Anthropic thinking modes are a fixed union.
-  providerOptions: { anthropic: { thinking: { type: "automatic" } } },
+  providerOptions: { thinking: { type: "automatic" } },
 })

packages/ai/test/provider-options/azure.types.ts

@@ -3,11 +3,11 @@ import { Azure } from "../../src/providers.js"
 
 const model = Azure.configure({ resourceName: "example" }).responses("deployment")
 
-LLM.request({ model, prompt: "Hello", providerOptions: { openai: { store: false } } })
+LLM.request({ model, prompt: "Hello", providerOptions: { store: false } })
 
 LLM.request({
   model,
   prompt: "Hello",
   // @ts-expect-error Azure OpenAI store must be boolean.
-  providerOptions: { openai: { store: "false" } },
+  providerOptions: { store: "false" },
 })

packages/ai/test/provider-options/google-vertex-chat.types.ts

@@ -3,11 +3,11 @@ import { GoogleVertexChat } from "../../src/providers.js"
 
 const model = GoogleVertexChat.configure({ accessToken: "test", project: "project" }).model("gemini")
 
-LLM.request({ model, prompt: "Hello", providerOptions: { openai: { serviceTier: "priority" } } })
+LLM.request({ model, prompt: "Hello", providerOptions: { serviceTier: "priority" } })
 
 LLM.request({
   model,
   prompt: "Hello",
   // @ts-expect-error Vertex OpenAI-compatible service tiers use the OpenAI union.
-  providerOptions: { openai: { serviceTier: "premium" } },
+  providerOptions: { serviceTier: "premium" },
 })

packages/ai/test/provider-options/google-vertex-messages.types.ts

@@ -3,11 +3,11 @@ import { GoogleVertexMessages } from "../../src/providers.js"
 
 const model = GoogleVertexMessages.configure({ accessToken: "test", project: "project" }).model("claude")
 
-LLM.request({ model, prompt: "Hello", providerOptions: { anthropic: { effort: "medium" } } })
+LLM.request({ model, prompt: "Hello", providerOptions: { effort: "medium" } })
 
 LLM.request({
   model,
   prompt: "Hello",
   // @ts-expect-error Vertex Anthropic effort must be a string.
-  providerOptions: { anthropic: { effort: false } },
+  providerOptions: { effort: false },
 })

packages/ai/test/provider-options/google-vertex-responses.types.ts

@@ -3,11 +3,12 @@ import { GoogleVertexResponses } from "../../src/providers.js"
 
 const model = GoogleVertexResponses.configure({ accessToken: "test", project: "project" }).model("gemini")
 
-LLM.request({ model, prompt: "Hello", providerOptions: { openresponses: { textVerbosity: "high" } } })
+LLM.request({ model, prompt: "Hello", providerOptions: { textVerbosity: "high" } })
+LLM.request({ model, prompt: "Hello", providerOptions: { textVerbosity: "verbose" } })
 
 LLM.request({
   model,
   prompt: "Hello",
-  // @ts-expect-error Vertex Responses verbosity uses the Open Responses union.
+  // @ts-expect-error Vertex Responses verbosity must be a string.
-  providerOptions: { openresponses: { textVerbosity: "verbose" } },
+  providerOptions: { textVerbosity: 1 },
 })

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

@@ -6,12 +6,12 @@ const model = GoogleVertex.provider.configure({ apiKey: "test" }).model("gemini-
 LLM.request({
   model,
   prompt: "Hello",
-  providerOptions: { gemini: { thinkingConfig: { includeThoughts: true } } },
+  providerOptions: { thinkingConfig: { includeThoughts: true } },
 })
 
 LLM.request({
   model,
   prompt: "Hello",
   // @ts-expect-error Vertex Gemini includeThoughts must be boolean.
-  providerOptions: { gemini: { thinkingConfig: { includeThoughts: "yes" } } },
+  providerOptions: { thinkingConfig: { includeThoughts: "yes" } },
 })

packages/ai/test/provider-options/google.types.ts

@@ -6,17 +6,15 @@ const model = Google.provider.model("gemini-2.5-pro")
 LLM.request({
   model,
   prompt: "Hello",
-  providerOptions: { gemini: { thinkingConfig: { thinkingBudget: 1024 } } },
+  providerOptions: { thinkingConfig: { thinkingBudget: 1024 } },
 })
 
 LLM.request({
   model,
   prompt: "Hello",
   providerOptions: {
-    gemini: {
+    // @ts-expect-error Gemini safety settings require a threshold for every category.
-      // @ts-expect-error Gemini safety settings require a threshold for every category.
+    safetySettings: [{ category: "HARM_CATEGORY_HATE_SPEECH" }],
-      safetySettings: [{ category: "HARM_CATEGORY_HATE_SPEECH" }],
-    },
   },
 })
 
@@ -24,12 +22,10 @@ LLM.request({
   model,
   prompt: "Hello",
   providerOptions: {
-    gemini: {
+    cachedContent: "cachedContents/example",
-      cachedContent: "cachedContents/example",
+    safetySettings: [{ category: "HARM_CATEGORY_HATE_SPEECH", threshold: "BLOCK_ONLY_HIGH" }],
-      safetySettings: [{ category: "HARM_CATEGORY_HATE_SPEECH", threshold: "BLOCK_ONLY_HIGH" }],
+    serviceTier: "future-tier",
-      serviceTier: "future-tier",
+    thinkingConfig: { thinkingLevel: "high", includeThoughts: true },
-      thinkingConfig: { thinkingLevel: "high", includeThoughts: true },
-    },
   },
 })
 
@@ -37,11 +33,11 @@ LLM.request({
   model,
   prompt: "Hello",
   // @ts-expect-error Gemini thinking budgets must be numeric.
-  providerOptions: { gemini: { thinkingConfig: { thinkingBudget: "large" } } },
+  providerOptions: { thinkingConfig: { thinkingBudget: "large" } },
 })
 
 LLM.request({
   model,
   prompt: "Hello",
-  providerOptions: { gemini: { thinkingConfig: { thinkingLevel: "maximum" } } },
+  providerOptions: { thinkingConfig: { thinkingLevel: "maximum" } },
 })

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

@@ -3,11 +3,15 @@ import { OpenAICompatibleResponses } from "../../src/providers.js"
 
 const model = OpenAICompatibleResponses.configure({ baseURL: "https://example.com" }).model("model")
 
-LLM.request({ model, prompt: "Hello", providerOptions: { openresponses: { reasoningSummary: "detailed" } } })
+LLM.request({ model, prompt: "Hello", providerOptions: { reasoningSummary: "detailed" } })
+LLM.request({ model, prompt: "Hello", providerOptions: { reasoningEffort: "high" } })
+LLM.request({ model, prompt: "Hello", providerOptions: { reasoningEffort: "experimental" } })
+LLM.request({ model, prompt: "Hello", providerOptions: { textVerbosity: "low" } })
+LLM.request({ model, prompt: "Hello", providerOptions: { textVerbosity: "verbose" } })
 
 LLM.request({
   model,
   prompt: "Hello",
   // @ts-expect-error Open Responses reasoning summaries use a fixed union.
-  providerOptions: { openresponses: { reasoningSummary: "full" } },
+  providerOptions: { reasoningSummary: "full" },
 })

packages/ai/test/provider-options/openai-compatible.types.ts

@@ -3,11 +3,11 @@ import { OpenAICompatible } from "../../src/providers.js"
 
 const model = OpenAICompatible.deepseek.model("deepseek-chat")
 
-LLM.request({ model, prompt: "Hello", providerOptions: { openai: { store: false } } })
+LLM.request({ model, prompt: "Hello", providerOptions: { store: false } })
 
 LLM.request({
   model,
   prompt: "Hello",
   // @ts-expect-error OpenAI-compatible store must be boolean.
-  providerOptions: { openai: { store: "false" } },
+  providerOptions: { store: "false" },
 })

packages/ai/test/provider-options/openai.types.ts

@@ -2,14 +2,20 @@ import { LLM } from "../../src/index.js"
 import { OpenAI } from "../../src/providers.js"
 
 const selected = OpenAI.responses("gpt-5")
+const chat = OpenAI.chat("gpt-4o-mini")
 
-LLM.request({ model: selected, prompt: "Hello", providerOptions: { openai: { reasoningEffort: "high" } } })
+LLM.request({ model: selected, prompt: "Hello", providerOptions: { reasoningEffort: "high" } })
+LLM.request({ model: selected, prompt: "Hello", providerOptions: { reasoningEffort: "experimental" } })
+LLM.request({ model: selected, prompt: "Hello", providerOptions: { textVerbosity: "low" } })
+LLM.request({ model: selected, prompt: "Hello", providerOptions: { textVerbosity: "verbose" } })
+LLM.request({ model: chat, prompt: "Hello", providerOptions: { reasoningEffort: "max" } })
+LLM.request({ model: chat, prompt: "Hello", providerOptions: { reasoningEffort: "experimental" } })
 
 LLM.request({
   model: selected,
   prompt: "Hello",
   // @ts-expect-error OpenAI reasoning effort must be a string.
-  providerOptions: { openai: { reasoningEffort: 1 } },
+  providerOptions: { reasoningEffort: 1 },
 })
 
 OpenAI.configure({

packages/ai/test/provider-options/openrouter.types.ts

@@ -3,27 +3,25 @@ import { OpenRouter } from "../../src/providers.js"
 
 const model = OpenRouter.provider.model("anthropic/claude-sonnet-4.5")
 
-LLM.request({ model, prompt: "Hello", providerOptions: { openrouter: { usage: true } } })
+LLM.request({ model, prompt: "Hello", providerOptions: { usage: true } })
 
 LLM.request({
   model,
   prompt: "Hello",
   providerOptions: {
-    openrouter: {
+    models: ["google/gemini-3.1-pro"],
-      models: ["google/gemini-3.1-pro"],
+    provider: {
-      provider: {
+      order: ["anthropic"],
-        order: ["anthropic"],
+      require_parameters: true,
-        require_parameters: true,
+      data_collection: "future-policy",
-        data_collection: "future-policy",
+      sort: "future-sort",
-        sort: "future-sort",
+      max_price: { prompt: "0.50" },
-        max_price: { prompt: "0.50" },
-      },
-      reasoning: { effort: "future-effort", exclude: false },
-      plugins: [{ id: "future-plugin", enabled: true }],
-      web_search_options: { engine: "future-engine" },
-      debug: { echo_upstream_body: true },
-      user: "user_123",
     },
+    reasoning: { effort: "future-effort", exclude: false },
+    plugins: [{ id: "future-plugin", enabled: true }],
+    web_search_options: { engine: "future-engine" },
+    debug: { echo_upstream_body: true },
+    user: "user_123",
   },
 })
 
@@ -31,5 +29,5 @@ LLM.request({
   model,
   prompt: "Hello",
   // @ts-expect-error OpenRouter usage must be boolean or an option record.
-  providerOptions: { openrouter: { usage: "yes" } },
+  providerOptions: { usage: "yes" },
 })

packages/ai/test/provider-options/xai.types.ts

@@ -3,11 +3,12 @@ import { XAI } from "../../src/providers.js"
 
 const model = XAI.provider.model("grok-4")
 
-LLM.request({ model, prompt: "Hello", providerOptions: { xai: { reasoningEffort: "high" } } })
+LLM.request({ model, prompt: "Hello", providerOptions: { reasoningEffort: "high" } })
+LLM.request({ model, prompt: "Hello", providerOptions: { reasoningEffort: "experimental" } })
 
 LLM.request({
   model,
   prompt: "Hello",
   // @ts-expect-error xAI's OpenAI-compatible reasoning effort must be a string.
-  providerOptions: { xai: { reasoningEffort: true } },
+  providerOptions: { reasoningEffort: true },
 })

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

@@ -43,25 +43,23 @@ describe("provider package entrypoints", () => {
       baseURL: "https://provider.example.test/v1",
       headers: { "x-application": "opencode" },
       body: { service_tier: "priority" },
-      limits: { context: 200_000, output: 64_000 },
     }
     const openrouter = OpenRouter.model("anthropic/claude-sonnet-4", {
       ...settings,
-      providerOptions: { openrouter: { usage: true } },
+      providerOptions: { usage: true },
     })
     const xai = XAI.model("grok-4", {
       ...settings,
-      providerOptions: { xai: { reasoningEffort: "high" } },
+      providerOptions: { reasoningEffort: "high" },
     })
 
     for (const selected of [openrouter, xai]) {
       expect(selected.route.endpoint.baseURL).toBe(settings.baseURL)
       expect(selected.route.defaults.headers).toEqual(settings.headers)
       expect(selected.route.defaults.http?.body).toEqual(settings.body)
-      expect(selected.route.defaults.limits).toEqual(settings.limits)
     }
-    expect(openrouter.route.defaults.providerOptions).toEqual({ openrouter: { usage: true } })
+    expect(openrouter.route.defaults.providerOptions).toEqual({ usage: true })
-    expect(xai.route.defaults.providerOptions).toMatchObject({ xai: { reasoningEffort: "high", store: false } })
+    expect(xai.route.defaults.providerOptions).toMatchObject({ reasoningEffort: "high", store: false })
   })
 
   test("maps package settings onto the executable model", () => {
@@ -70,14 +68,12 @@ describe("provider package entrypoints", () => {
       baseURL: "https://api.openai.test/v1",
       headers: { "x-application": "opencode" },
       body: { service_tier: "priority" },
-      limits: { context: 200_000, output: 64_000 },
       unrelatedInheritedSetting: true,
     })
 
     expect(selected.route.id).toBe("openai-responses")
     expect(selected.route.defaults.headers).toEqual({ "x-application": "opencode" })
     expect(selected.route.defaults.http?.body).toEqual({ service_tier: "priority" })
-    expect(selected.route.defaults.limits).toEqual({ context: 200_000, output: 64_000 })
   })
 
   test("maps OpenAI-compatible Responses settings onto the executable model", async () => {
@@ -88,8 +84,7 @@ describe("provider package entrypoints", () => {
       provider: "example",
       headers: { "x-application": "opencode" },
       body: { service_tier: "priority" },
-      limits: { context: 200_000, output: 64_000 },
+      providerOptions: { reasoningEffort: "low", store: true },
-      providerOptions: { openresponses: { reasoningEffort: "low", store: true } },
     })
 
     expect(String(selected.provider)).toBe("example")
@@ -100,10 +95,7 @@ describe("provider package entrypoints", () => {
     })
     expect(selected.route.defaults.headers).toEqual({ "x-application": "opencode" })
     expect(selected.route.defaults.http?.body).toEqual({ service_tier: "priority" })
-    expect(selected.route.defaults.limits).toEqual({ context: 200_000, output: 64_000 })
+    expect(selected.route.defaults.providerOptions).toEqual({ reasoningEffort: "low", store: true })
-    expect(selected.route.defaults.providerOptions).toEqual({
-      openresponses: { reasoningEffort: "low", store: true },
-    })
   })
 
   test("maps Anthropic-compatible settings onto the executable model", async () => {
@@ -114,8 +106,7 @@ describe("provider package entrypoints", () => {
       provider: "example",
       headers: { "x-application": "opencode" },
       body: { metadata: { user_id: "user_1" } },
-      limits: { context: 200_000, output: 64_000 },
+      providerOptions: { effort: "low" },
-      providerOptions: { anthropic: { effort: "low" } },
     })
 
     expect(String(selected.provider)).toBe("example")
@@ -126,20 +117,17 @@ describe("provider package entrypoints", () => {
     })
     expect(selected.route.defaults.headers).toEqual({ "x-application": "opencode" })
     expect(selected.route.defaults.http?.body).toEqual({ metadata: { user_id: "user_1" } })
-    expect(selected.route.defaults.limits).toEqual({ context: 200_000, output: 64_000 })
+    expect(selected.route.defaults.providerOptions).toEqual({ effort: "low" })
-    expect(selected.route.defaults.providerOptions).toEqual({ anthropic: { effort: "low" } })
   })
 
   test("maps Anthropic provider options onto the executable model", async () => {
     const Anthropic = await import("@opencode-ai/ai/providers/anthropic")
     const selected = Anthropic.model("claude-sonnet-4-6", {
       apiKey: "fixture",
-      providerOptions: { anthropic: { thinking: { type: "adaptive" } } },
+      providerOptions: { thinking: { type: "adaptive" } },
     })
 
-    expect(selected.route.defaults.providerOptions).toEqual({
+    expect(selected.route.defaults.providerOptions).toEqual({ thinking: { type: "adaptive" } })
-      anthropic: { thinking: { type: "adaptive" } },
-    })
   })
 
   test("requires an Anthropic-compatible base URL at runtime", async () => {
@@ -189,7 +177,6 @@ describe("provider package entrypoints", () => {
       resourceName: "opencode-test",
       headers: { "x-application": "opencode" },
       body: { service_tier: "priority" },
-      limits: { context: 200_000, output: 64_000 },
     }
 
     const responses = AzureResponses.model("deployment", settings)
@@ -200,7 +187,6 @@ describe("provider package entrypoints", () => {
     expect(responses.route.endpoint.baseURL).toBe("https://opencode-test.openai.azure.com/openai/v1")
     expect(responses.route.defaults.headers).toEqual({ "x-application": "opencode" })
     expect(responses.route.defaults.http?.body).toEqual({ service_tier: "priority" })
-    expect(responses.route.defaults.limits).toEqual({ context: 200_000, output: 64_000 })
     expect(chat.route.id).toBe("azure-openai-chat")
   })
 
@@ -232,18 +218,14 @@ describe("provider package entrypoints", () => {
       baseURL: "https://generativelanguage.test/v1beta",
       headers: { "x-application": "opencode" },
       body: { safetySettings: [] },
-      limits: { context: 1_000_000, output: 65_536 },
+      providerOptions: { thinkingConfig: { thinkingBudget: 1_024 } },
-      providerOptions: { gemini: { thinkingConfig: { thinkingBudget: 1_024 } } },
     })
 
     expect(selected.route.id).toBe("gemini")
     expect(selected.route.endpoint.baseURL).toBe("https://generativelanguage.test/v1beta")
     expect(selected.route.defaults.headers).toEqual({ "x-application": "opencode" })
     expect(selected.route.defaults.http?.body).toEqual({ safetySettings: [] })
-    expect(selected.route.defaults.limits).toEqual({ context: 1_000_000, output: 65_536 })
+    expect(selected.route.defaults.providerOptions).toEqual({ thinkingConfig: { thinkingBudget: 1_024 } })
-    expect(selected.route.defaults.providerOptions).toEqual({
-      gemini: { thinkingConfig: { thinkingBudget: 1_024 } },
-    })
   })
 
   test("selects Vertex entrypoints with the same model contract", async () => {
@@ -256,7 +238,6 @@ describe("provider package entrypoints", () => {
       apiKey: "fixture",
       headers: { "x-application": "opencode" },
       body: { safetySettings: [] },
-      limits: { context: 1_000_000, output: 65_536 },
     })
     const messages = GoogleVertexMessages.model("claude-sonnet-4-6", {
       accessToken: "fixture",
@@ -280,7 +261,6 @@ describe("provider package entrypoints", () => {
     expect(gemini.route.endpoint.baseURL).toBe("https://aiplatform.googleapis.com/v1/publishers/google")
     expect(gemini.route.defaults.headers).toEqual({ "x-application": "opencode" })
     expect(gemini.route.defaults.http?.body).toEqual({ safetySettings: [] })
-    expect(gemini.route.defaults.limits).toEqual({ context: 1_000_000, output: 65_536 })
     expect(
       GoogleVertex.model("gemini-3.5-flash", {
         accessToken: "fixture",
@@ -305,7 +285,7 @@ describe("provider package entrypoints", () => {
       baseURL: "https://aiplatform.googleapis.com/v1/projects/vertex-project/locations/global/endpoints/openapi",
       path: "/responses",
     })
-    expect(responses.route.defaults.providerOptions).toEqual({ openresponses: { store: false } })
+    expect(responses.route.defaults.providerOptions).toEqual({ store: false })
   })
 
   test("rejects conflicting Vertex auth settings at runtime", async () => {

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

@@ -63,7 +63,8 @@ describe("Anthropic Messages route", () => {
       const prepared = yield* compileRequest(
         LLMRequest.update(request, {
           providerOptions: {
-            anthropic: { thinking: { type: "adaptive", display: "summarized" }, effort: "low" },
+            thinking: { type: "adaptive", display: "summarized" },
+            effort: "low",
           },
         }),
       )
@@ -79,17 +80,17 @@ describe("Anthropic Messages route", () => {
     Effect.gen(function* () {
       const enabled = yield* compileRequest(
         LLMRequest.update(request, {
-          providerOptions: { anthropic: { thinking: { type: "enabled", budgetTokens: 1_024 } } },
+          providerOptions: { thinking: { type: "enabled", budgetTokens: 1_024 } },
         }),
       )
       const legacy = yield* compileRequest(
         LLMRequest.update(request, {
-          providerOptions: { anthropic: { thinking: { type: "enabled", budget_tokens: 2_048 } } },
+          providerOptions: { thinking: { type: "enabled", budget_tokens: 2_048 } },
         }),
       )
       const disabled = yield* compileRequest(
         LLMRequest.update(request, {
-          providerOptions: { anthropic: { thinking: { type: "disabled" } } },
+          providerOptions: { thinking: { type: "disabled" } },
         }),
       )
 
@@ -103,7 +104,7 @@ describe("Anthropic Messages route", () => {
     Effect.gen(function* () {
       const error = yield* compileRequest(
         LLMRequest.update(request, {
-          providerOptions: { anthropic: { thinking: { type: "enabled" } } },
+          providerOptions: { thinking: { type: "enabled" } },
         }),
       ).pipe(Effect.flip)
 
@@ -321,7 +322,7 @@ describe("Anthropic Messages route", () => {
           { role: "user", content: [{ type: "tool_result", tool_use_id: "call_1", content: '{"forecast":"sunny"}' }] },
         ],
         stream: true,
-        max_tokens: 4096,
+        max_tokens: 32_000,
       })
     }),
   )
@@ -1062,9 +1063,7 @@ describe("Anthropic Messages route", () => {
         ),
       )
 
-      expect(response.toolCalls).toMatchObject([
+      expect(response.toolCalls).toMatchObject([{ id: "call_1", name: "lookup", input: { query: "weather" } }])
-        { id: "call_1", name: "lookup", input: { query: "weather" } },
-      ])
       expect(response.finishReason).toEqual({ normalized: "tool-calls", raw: "tool_use" })
     }),
   )

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

@@ -48,28 +48,26 @@ describe("Gemini route", () => {
       const prepared = yield* compileRequest(
         LLMRequest.update(request, {
           providerOptions: {
-            gemini: {
+            cachedContent: "cachedContents/example",
-              cachedContent: "cachedContents/example",
+            safetySettings: [{ category: "HARM_CATEGORY_HATE_SPEECH", threshold: "BLOCK_ONLY_HIGH" }],
-              safetySettings: [{ category: "HARM_CATEGORY_HATE_SPEECH", threshold: "BLOCK_ONLY_HIGH" }],
+            serviceTier: "priority",
-              serviceTier: "priority",
+            thinkingConfig: { thinkingBudget: 0, includeThoughts: false, thinkingLevel: "high" },
-              thinkingConfig: { thinkingBudget: 0, includeThoughts: false, thinkingLevel: "high" },
-            },
           },
         }),
       )
       const filtered = yield* compileRequest(
         LLMRequest.update(request, {
-          providerOptions: { gemini: { thinkingConfig: { thinkingBudget: "invalid", includeThoughts: false } } },
+          providerOptions: { thinkingConfig: { thinkingBudget: "invalid", includeThoughts: false } },
         }),
       )
       const defaulted = yield* compileRequest(
         LLMRequest.update(request, {
-          providerOptions: { gemini: { thinkingConfig: { thinkingLevel: "high" } } },
+          providerOptions: { thinkingConfig: { thinkingLevel: "high" } },
         }),
       )
       const emptySafetySettings = yield* compileRequest(
         LLMRequest.update(request, {
-          providerOptions: { gemini: { safetySettings: [] } },
+          providerOptions: { safetySettings: [] },
         }),
       )
 
@@ -141,6 +139,48 @@ describe("Gemini route", () => {
     }),
   )
 
+  it.effect("keeps system updates separate from function responses", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model,
+          messages: [
+            Message.assistant([ToolCallPart.make({ id: "call_1", name: "lookup", input: { query: "weather" } })]),
+            Message.tool({ id: "call_1", name: "lookup", result: "done", resultType: "text" }),
+            Message.system("Update."),
+            Message.system("Later update."),
+          ],
+        }),
+      )
+
+      expect(prepared.body.contents).toEqual([
+        {
+          role: "model",
+          parts: [{ functionCall: { id: undefined, name: "lookup", args: { query: "weather" } } }],
+        },
+        {
+          role: "user",
+          parts: [
+            {
+              functionResponse: {
+                id: undefined,
+                name: "lookup",
+                response: { name: "lookup", content: "done" },
+              },
+            },
+          ],
+        },
+        {
+          role: "user",
+          parts: [
+            { text: "<system-update>\nUpdate.\n</system-update>" },
+            { text: "<system-update>\nLater update.\n</system-update>" },
+          ],
+        },
+      ])
+    }),
+  )
+
   it.effect("prepares multimodal user input and tool history", () =>
     Effect.gen(function* () {
       const prepared = yield* compileRequest(

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

@@ -62,7 +62,7 @@ describe("Google Vertex providers", () => {
             accessToken: "vertex-token",
             project: "vertex-project",
             providerOptions: {
-              gemini: { labels: { component: "opencode", environment: "test" } },
+              labels: { component: "opencode", environment: "test" },
             },
           }).model("gemini-3.5-flash"),
           prompt: "Say hello.",

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

@@ -15,7 +15,7 @@ const cases = [
     model: LanguageModel.update(
       OpenRouter.configure({
         apiKey: process.env.OPENROUTER_API_KEY ?? "fixture",
-        providerOptions: { openrouter: { reasoning: { max_tokens: 1024 } } },
+        providerOptions: { reasoning: { max_tokens: 1024 } },
       }).model("anthropic/claude-sonnet-4.6"),
       { compatibility: { reasoningField: "reasoning" } },
     ),

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

@@ -165,7 +165,7 @@ describe("OpenAI Chat route", () => {
         LLM.request({
           model: OpenAI.configure({ baseURL: "https://api.openai.test/v1/", apiKey: "test" }).chat("gpt-4o-mini"),
           prompt: "think",
-          providerOptions: { openai: { reasoningEffort: "max" } },
+          providerOptions: { reasoningEffort: "max" },
         }),
       )
 
@@ -221,7 +221,7 @@ describe("OpenAI Chat route", () => {
         LLM.request({
           model,
           prompt: "think",
-          providerOptions: { openai: { reasoningEffort: "experimental" } },
+          providerOptions: { reasoningEffort: "experimental" },
         }),
       )
 
@@ -255,7 +255,7 @@ describe("OpenAI Chat route", () => {
     LLMClient.generate(
       LLMRequest.update(request, {
         model: Azure.configure({
-          baseURL: "https://opencode-test.openai.azure.com/openai/v1/",
+          baseURL: "https://opencode-test.openai.azure.com/openai/",
           apiKey: "azure-key",
           headers: { authorization: "Bearer stale" },
         }).chat("gpt-4o-mini"),
@@ -664,6 +664,74 @@ describe("OpenAI Chat route", () => {
     }),
   )
 
+  it.effect("preserves streamed refusals as ordinary assistant text", () =>
+    Effect.gen(function* () {
+      const response = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              deltaChunk({ role: "assistant", refusal: "I can't" }),
+              deltaChunk({ refusal: " help with that." }),
+              deltaChunk({}, "stop"),
+            ),
+          ),
+        ),
+      )
+
+      expect(response.text).toBe("I can't help with that.")
+      expect(response.finishReason).toEqual({ normalized: "stop", raw: "stop" })
+      expect(response.message.content).toEqual([{ type: "text", text: "I can't help with that." }])
+
+      const replay = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
+      expect(replay.body.messages).toEqual([{ role: "assistant", content: "I can't help with that." }])
+    }),
+  )
+
+  it.effect("orders metadata-only reasoning before refusal output", () =>
+    Effect.gen(function* () {
+      const response = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              { choices: [{ delta: { reasoning_details: [] } }] },
+              deltaChunk({ refusal: "I can't help with that." }),
+              deltaChunk({}, "stop"),
+            ),
+          ),
+        ),
+      )
+
+      expect(response.message.content).toEqual([
+        { type: "reasoning", text: "", providerMetadata: { openai: { reasoningDetails: [] } } },
+        {
+          type: "text",
+          text: "I can't help with that.",
+        },
+      ])
+    }),
+  )
+
+  it.effect("joins content and refusal deltas into ordinary assistant text", () =>
+    Effect.gen(function* () {
+      const response = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              deltaChunk({ refusal: "No." }),
+              deltaChunk({ content: " Alternative." }),
+              deltaChunk({ refusal: " Still no." }),
+              deltaChunk({}, "stop"),
+            ),
+          ),
+        ),
+      )
+
+      expect(response.text).toBe("No. Alternative. Still no.")
+      expect(response.events.filter(LLMEvent.is.textStart).map((event) => event.id)).toEqual(["text-0"])
+      expect(response.events.filter(LLMEvent.is.textEnd).map((event) => event.id)).toEqual(["text-0"])
+    }),
+  )
+
   it.effect("parses and replays OpenAI-compatible reasoning fields", () =>
     Effect.gen(function* () {
       const fields = ["reasoning_content", "reasoning", "reasoning_text"] as const

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

@@ -1,6 +1,6 @@
 import { describe, expect } from "bun:test"
 import { Effect } from "effect"
-import { LLM, LLMEvent, Message } from "../../src/index.js"
+import { LLM, LLMEvent, Message, ToolDefinition } from "../../src/index.js"
 import { configure } from "../../src/providers/openai-compatible-responses.js"
 import { OpenAI } from "../../src/providers.js"
 import { OpenResponses } from "../../src/protocols/open-responses.js"
@@ -56,6 +56,28 @@ describe("Open Responses-compatible route", () => {
     }),
   )
 
+  it.effect("lowers chronological system updates as standard developer messages", () =>
+    Effect.gen(function* () {
+      const model = configure({
+        apiKey: "test-key",
+        baseURL: "https://responses.example.test/v1",
+        provider: "example",
+      }).model("example-model")
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model,
+          messages: [Message.user("Before."), Message.system("Operator update."), Message.assistant("After.")],
+        }),
+      )
+
+      expect(prepared.body.input).toEqual([
+        { role: "user", content: [{ type: "input_text", text: "Before." }] },
+        { role: "developer", content: "Operator update." },
+        { type: "message", role: "assistant", content: [{ type: "output_text", text: "After." }] },
+      ])
+    }),
+  )
+
   it.effect("rejects OpenAI-native tools", () =>
     Effect.gen(function* () {
       const model = configure({
@@ -91,23 +113,111 @@ describe("Open Responses-compatible route", () => {
       )
 
       expect(prepared.body).toMatchObject({
-        input: [{ role: "assistant", content: [{ type: "output_text", text: "Unclassified." }] }],
+        input: [{ type: "message", role: "assistant", content: [{ type: "output_text", text: "Unclassified." }] }],
       })
     }),
   )
 
-  it.effect("reads standard options from the Open Responses namespace", () =>
+  it.effect("preserves standard refusal content as ordinary assistant text", () =>
     Effect.gen(function* () {
       const model = configure({
         apiKey: "test-key",
         baseURL: "https://responses.example.test/v1",
-        providerOptions: { openresponses: { reasoningEffort: "low", store: true } },
+        provider: "example",
       }).model("example-model")
-      const prepared = yield* compileRequest(LLM.request({ model, prompt: "Think." }))
+      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Unsafe request" })).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              {
+                type: "response.output_item.added",
+                output_index: 0,
+                item: { type: "message", id: "msg_refusal", content: [] },
+              },
+              {
+                type: "response.refusal.done",
+                item_id: "msg_refusal",
+                refusal: "I can't help with that.",
+              },
+              {
+                type: "response.output_item.done",
+                output_index: 0,
+                item: {
+                  type: "message",
+                  id: "msg_refusal",
+                  content: [{ type: "refusal", refusal: "I can't help with that." }],
+                },
+              },
+              { type: "response.completed", response: { id: "resp_1" } },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.message.content).toEqual([
+        {
+          type: "text",
+          text: "I can't help with that.",
+          providerMetadata: { openresponses: { itemId: "msg_refusal" } },
+        },
+      ])
+
+      const prepared = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
+      expect(prepared.body.input).toEqual([
+        {
+          type: "message",
+          id: "msg_refusal",
+          role: "assistant",
+          content: [{ type: "output_text", text: "I can't help with that." }],
+        },
+      ])
+    }),
+  )
+
+  it.effect("reads standard Open Responses options", () =>
+    Effect.gen(function* () {
+      const model = configure({
+        apiKey: "test-key",
+        baseURL: "https://responses.example.test/v1",
+        providerOptions: {
+          reasoningEffort: "low",
+          store: true,
+          metadata: { environment: "test" },
+          safetyIdentifier: "user_123",
+          streamOptions: { includeObfuscation: false },
+          topLogprobs: 3,
+          truncation: "auto",
+          allowedTools: { toolNames: ["lookup"] },
+          maxToolCalls: 2,
+          parallelToolCalls: false,
+        },
+      }).model("example-model")
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model,
+          prompt: "Think.",
+          generation: { presencePenalty: 0.2, frequencyPenalty: -0.1 },
+          tools: [ToolDefinition.make({ name: "lookup", description: "Lookup data", inputSchema: { type: "object" } })],
+        }),
+      )
 
       expect(prepared.body).toMatchObject({
         reasoning: { effort: "low" },
         store: true,
+        metadata: { environment: "test" },
+        safety_identifier: "user_123",
+        stream_options: { include_obfuscation: false },
+        top_logprobs: 3,
+        presence_penalty: 0.2,
+        frequency_penalty: -0.1,
+        truncation: "auto",
+        tool_choice: {
+          type: "allowed_tools",
+          mode: "auto",
+          tools: [{ type: "function", name: "lookup" }],
+        },
+        max_tool_calls: 2,
+        parallel_tool_calls: false,
       })
     }),
   )

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

@@ -159,8 +159,8 @@ describe("OpenAI Responses route", () => {
 
   it.effect("lowers semantic service tier options", () =>
     Effect.gen(function* () {
-      const input = LLMRequest.update(request, { providerOptions: { openai: { serviceTier: "priority" } } })
+      const input = LLMRequest.update(request, { providerOptions: { serviceTier: "priority" } })
-      expect(input.providerOptions).toEqual({ openai: { serviceTier: "priority" } })
+      expect(input.providerOptions).toEqual({ serviceTier: "priority" })
       const prepared = yield* compileRequest(input)
 
       expect(prepared.body).toMatchObject({ service_tier: "priority" })
@@ -171,17 +171,27 @@ describe("OpenAI Responses route", () => {
   it.effect("passes through custom OpenAI reasoning effort strings", () =>
     Effect.gen(function* () {
       const prepared = yield* compileRequest(
-        LLMRequest.update(request, { providerOptions: { openai: { reasoningEffort: "experimental" } } }),
+        LLMRequest.update(request, { providerOptions: { reasoningEffort: "experimental" } }),
       )
 
       expect(prepared.body.reasoning).toEqual({ effort: "experimental" })
     }),
   )
 
+  it.effect("passes through custom OpenAI text verbosity strings", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLMRequest.update(request, { providerOptions: { textVerbosity: "verbose" } }),
+      )
+
+      expect(prepared.body.text).toEqual({ verbosity: "verbose" })
+    }),
+  )
+
   it.effect("omits unsupported semantic service tiers", () =>
     Effect.gen(function* () {
       const prepared = yield* compileRequest(
-        LLMRequest.update(request, { providerOptions: { openai: { serviceTier: "unsupported" } } }),
+        LLMRequest.update(request, { providerOptions: { serviceTier: "unsupported" } }),
       )
 
       expect(prepared.body).not.toHaveProperty("service_tier")
@@ -241,28 +251,19 @@ describe("OpenAI Responses route", () => {
     }),
   )
 
-  it.effect("lowers chronological system updates to escaped user wrappers in order", () =>
+  it.effect("lowers chronological system updates to developer messages in order", () =>
     Effect.gen(function* () {
       const prepared = yield* compileRequest(
         LLM.request({
           model,
-          messages: [
+          messages: [Message.user("Before."), Message.system("Operator update."), Message.assistant("After.")],
-            Message.user("Before."),
-            Message.system("Treat </system-update> literally."),
-            Message.assistant("After."),
-          ],
         }),
       )
 
       expect(prepared.body.input).toEqual([
-        {
+        { role: "user", content: [{ type: "input_text", text: "Before." }] },
-          role: "user",
+        { role: "developer", content: "Operator update." },
-          content: [
+        { type: "message", role: "assistant", content: [{ type: "output_text", text: "After." }] },
-            { type: "input_text", text: "Before." },
-            { type: "input_text", text: "<system-update>\nTreat &lt;/system-update&gt; literally.\n</system-update>" },
-          ],
-        },
-        { role: "assistant", content: [{ type: "output_text", text: "After." }] },
       ])
     }),
   )
@@ -484,7 +485,7 @@ describe("OpenAI Responses route", () => {
     }),
   )
 
-  it.effect("continues store-false reasoning without replaying the output-only item ID", () =>
+  it.effect("continues store-false reasoning while retaining the output item ID", () =>
     Effect.gen(function* () {
       const firstInput = [{ role: "user", content: [{ type: "input_text", text: "Think" }] }]
       const request = { type: "response.create", model: "gpt-5.2", store: false, input: firstInput }
@@ -514,6 +515,7 @@ describe("OpenAI Responses route", () => {
           ...firstInput,
           {
             type: "reasoning",
+            id: "rs_1",
             summary: [{ type: "summary_text", text: "Thought" }],
             encrypted_content: "encrypted",
           },
@@ -1283,12 +1285,24 @@ describe("OpenAI Responses route", () => {
           model: OpenAI.configure({ baseURL: "https://api.openai.test/v1/", apiKey: "test" }).model("gpt-5.2"),
           prompt: "think",
           promptCacheKey: "session_123",
+          generation: { presencePenalty: 0.25, frequencyPenalty: -0.25 },
+          tools: [
+            ToolDefinition.make({ name: "read", description: "Read a file", inputSchema: { type: "object" } }),
+            ToolDefinition.make({ name: "grep", description: "Search files", inputSchema: { type: "object" } }),
+          ],
+          toolChoice: "none",
           providerOptions: {
-            openai: {
+            reasoningEffort: "high",
-              reasoningEffort: "high",
+            reasoningSummary: "auto",
-              reasoningSummary: "auto",
+            include: ["reasoning.encrypted_content"],
-              include: ["reasoning.encrypted_content"],
+            metadata: { environment: "test", tenant: "acme" },
-            },
+            safetyIdentifier: "user_123",
+            streamOptions: { includeObfuscation: false },
+            topLogprobs: 5,
+            truncation: "disabled",
+            allowedTools: { toolNames: ["read", "grep"], mode: "required" },
+            maxToolCalls: 4,
+            parallelToolCalls: false,
           },
         }),
       )
@@ -1298,6 +1312,23 @@ describe("OpenAI Responses route", () => {
       expect(prepared.body.include).toEqual(["reasoning.encrypted_content"])
       expect(prepared.body.reasoning).toEqual({ effort: "high", summary: "auto" })
       expect(prepared.body.text).toEqual({ verbosity: "low" })
+      expect(prepared.body.metadata).toEqual({ environment: "test", tenant: "acme" })
+      expect(prepared.body.safety_identifier).toBe("user_123")
+      expect(prepared.body.stream_options).toEqual({ include_obfuscation: false })
+      expect(prepared.body.top_logprobs).toBe(5)
+      expect(prepared.body.presence_penalty).toBe(0.25)
+      expect(prepared.body.frequency_penalty).toBe(-0.25)
+      expect(prepared.body.truncation).toBe("disabled")
+      expect(prepared.body.tool_choice).toEqual({
+        type: "allowed_tools",
+        mode: "required",
+        tools: [
+          { type: "function", name: "read" },
+          { type: "function", name: "grep" },
+        ],
+      })
+      expect(prepared.body.max_tool_calls).toBe(4)
+      expect(prepared.body.parallel_tool_calls).toBe(false)
     }),
   )
 
@@ -1308,9 +1339,7 @@ describe("OpenAI Responses route", () => {
           model,
           prompt: "hi",
           providerOptions: {
-            openai: {
+            include: ["reasoning.encrypted_content", "code_interpreter_call.outputs", "web_search_call.results"],
-              include: ["reasoning.encrypted_content", "code_interpreter_call.outputs", "web_search_call.results"],
-            },
           },
         }),
       )
@@ -1323,48 +1352,41 @@ describe("OpenAI Responses route", () => {
     }),
   )
 
-  it.effect("filters unknown includable values out of the include array", () =>
+  it.effect("passes forward-compatible includable values through", () =>
     Effect.gen(function* () {
       const prepared = yield* compileRequest(
         LLM.request({
           model,
           prompt: "hi",
-          // The user passed one invalid entry alongside a valid one. Keep the
+          providerOptions: { include: ["reasoning.encrypted_content", "bogus.thing"] },
-          // valid one so the request still succeeds rather than failing on a
-          // typo from upstream config.
-          providerOptions: { openai: { include: ["reasoning.encrypted_content", "bogus.thing"] } },
         }),
       )
 
-      expect(prepared.body.include).toEqual(["reasoning.encrypted_content"])
+      expect(prepared.body.include).toEqual(["reasoning.encrypted_content", "bogus.thing"])
     }),
   )
 
   it.effect("treats an explicit empty include as no include at all", () =>
     Effect.gen(function* () {
-      const prepared = yield* compileRequest(
+      const prepared = yield* compileRequest(LLM.request({ model, prompt: "hi", providerOptions: { include: [] } }))
-        LLM.request({ model, prompt: "hi", providerOptions: { openai: { include: [] } } }),
-      )
 
       expect(prepared.body.include).toBeUndefined()
     }),
   )
 
-  it.effect("treats an all-invalid include as no include at all", () =>
+  it.effect("passes an unknown includable value through", () =>
     Effect.gen(function* () {
       const prepared = yield* compileRequest(
-        LLM.request({ model, prompt: "hi", providerOptions: { openai: { include: ["bogus.thing"] } } }),
+        LLM.request({ model, prompt: "hi", providerOptions: { include: ["bogus.thing"] } }),
       )
 
-      expect(prepared.body.include).toBeUndefined()
+      expect(prepared.body.include).toEqual(["bogus.thing"])
     }),
   )
 
   it.effect("omits include when no include is set", () =>
     Effect.gen(function* () {
-      const prepared = yield* compileRequest(
+      const prepared = yield* compileRequest(LLM.request({ model, prompt: "hi", providerOptions: { store: false } }))
-        LLM.request({ model, prompt: "hi", providerOptions: { openai: { store: false } } }),
-      )
 
       expect(prepared.body.include).toBeUndefined()
     }),
@@ -1395,7 +1417,7 @@ describe("OpenAI Responses route", () => {
         LLM.request({
           model: OpenAI.configure({ baseURL: "https://api.openai.test/v1/", apiKey: "test" }).responses("gpt-5.2"),
           prompt: "hi",
-          providerOptions: { openai: { include: [] } },
+          providerOptions: { include: [] },
         }),
       )
 
@@ -1463,7 +1485,7 @@ describe("OpenAI Responses route", () => {
       expect(response.text).toBe("Hello!")
       expect(response.events).toEqual([
         { type: "step-start", index: 0 },
-        { type: "text-start", id: "msg_1" },
+        { type: "text-start", id: "msg_1", providerMetadata: { openai: { itemId: "msg_1" } } },
         { type: "text-delta", id: "msg_1", text: "Hello" },
         { type: "text-delta", id: "msg_1", text: "!" },
         { type: "text-end", id: "msg_1" },
@@ -1484,6 +1506,108 @@ describe("OpenAI Responses route", () => {
     }),
   )
 
+  it.effect("preserves standard refusal content as ordinary assistant text", () =>
+    Effect.gen(function* () {
+      const response = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              {
+                type: "response.output_item.added",
+                output_index: 0,
+                item: { type: "message", id: "msg_refusal", content: [] },
+              },
+              {
+                type: "response.content_part.added",
+                item_id: "msg_refusal",
+                output_index: 0,
+                content_index: 0,
+                part: { type: "refusal", refusal: "" },
+              },
+              {
+                type: "response.refusal.delta",
+                item_id: "msg_refusal",
+                output_index: 0,
+                content_index: 0,
+                delta: "I can't",
+              },
+              {
+                type: "response.refusal.delta",
+                item_id: "msg_refusal",
+                output_index: 0,
+                content_index: 0,
+                delta: " help with that.",
+              },
+              {
+                type: "response.refusal.done",
+                item_id: "msg_refusal",
+                output_index: 0,
+                content_index: 0,
+                refusal: "I can't help with that.",
+              },
+              {
+                type: "response.content_part.done",
+                item_id: "msg_refusal",
+                output_index: 0,
+                content_index: 0,
+                part: { type: "refusal", refusal: "I can't help with that." },
+              },
+              {
+                type: "response.output_item.done",
+                output_index: 0,
+                item: {
+                  type: "message",
+                  id: "msg_refusal",
+                  phase: "final_answer",
+                  content: [{ type: "refusal", refusal: "I can't help with that." }],
+                },
+              },
+              { type: "response.completed", response: { id: "resp_1" } },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.text).toBe("I can't help with that.")
+      expect(response.finishReason).toEqual({ normalized: "stop", raw: undefined })
+      expect(response.message.content).toEqual([
+        {
+          type: "text",
+          text: "I can't help with that.",
+          providerMetadata: { openai: { itemId: "msg_refusal", phase: "final_answer" } },
+        },
+      ])
+
+      const prepared = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
+      expect(prepared.body.input).toEqual([
+        {
+          type: "message",
+          id: "msg_refusal",
+          role: "assistant",
+          content: [{ type: "output_text", text: "I can't help with that." }],
+          phase: "final_answer",
+        },
+      ])
+    }),
+  )
+
+  it.effect("rejects malformed refusal events", () =>
+    Effect.gen(function* () {
+      const events = [
+        { type: "response.refusal.delta", output_index: 0, content_index: 0, delta: "missing item" },
+        { type: "response.refusal.delta", item_id: "msg_1", output_index: 0, content_index: 0 },
+        { type: "response.refusal.done", item_id: "msg_1", output_index: 0, content_index: 0 },
+      ]
+      for (const event of events) {
+        const error = yield* LLMClient.generate(request).pipe(
+          Effect.provide(fixedResponse(sseEvents(event))),
+          Effect.flip,
+        )
+        expect(error.reason._tag).toBe("InvalidProviderOutput")
+      }
+    }),
+  )
+
   it.effect("preserves and replays assistant message phases", () =>
     Effect.gen(function* () {
       const response = yield* LLMClient.generate(request).pipe(
@@ -1522,33 +1646,39 @@ describe("OpenAI Responses route", () => {
         {
           type: "text",
           text: "Checking.",
-          providerMetadata: { openai: { phase: "commentary" } },
+          providerMetadata: { openai: { itemId: "msg_commentary", phase: "commentary" } },
         },
         {
           type: "text",
           text: "Finished.",
-          providerMetadata: { openai: { phase: "final_answer" } },
+          providerMetadata: { openai: { itemId: "msg_final", phase: "final_answer" } },
         },
         {
           type: "text",
           text: "Unclassified.",
-          providerMetadata: { openai: { phase: null } },
+          providerMetadata: { openai: { itemId: "msg_null", phase: null } },
         },
       ])
 
       const prepared = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
       expect(prepared.body.input).toEqual([
         {
+          type: "message",
+          id: "msg_commentary",
           role: "assistant",
           content: [{ type: "output_text", text: "Checking." }],
           phase: "commentary",
         },
         {
+          type: "message",
+          id: "msg_final",
           role: "assistant",
           content: [{ type: "output_text", text: "Finished." }],
           phase: "final_answer",
         },
         {
+          type: "message",
+          id: "msg_null",
           role: "assistant",
           content: [{ type: "output_text", text: "Unclassified." }],
           phase: null,
@@ -1642,12 +1772,12 @@ describe("OpenAI Responses route", () => {
       )
 
       expect(response.events.filter((event) => event.type.startsWith("text-"))).toEqual([
-        { type: "text-start", id: "msg_1" },
+        { type: "text-start", id: "msg_1", providerMetadata: { openai: { itemId: "msg_1" } } },
         { type: "text-delta", id: "msg_1", text: "First" },
-        { type: "text-end", id: "msg_1" },
+        { type: "text-end", id: "msg_1", providerMetadata: undefined },
-        { type: "text-start", id: "msg_2" },
+        { type: "text-start", id: "msg_2", providerMetadata: { openai: { itemId: "msg_2" } } },
         { type: "text-delta", id: "msg_2", text: "Second" },
-        { type: "text-end", id: "msg_2" },
+        { type: "text-end", id: "msg_2", providerMetadata: { openai: { itemId: "msg_2" } } },
       ])
     }),
   )
@@ -1679,7 +1809,7 @@ describe("OpenAI Responses route", () => {
       expect(response.events.filter((event) => event.type === "finish")).toHaveLength(1)
       expect(response.message.content).toEqual([
         { type: "reasoning", text: "thinking" },
-        { type: "text", text: "Hello" },
+        { type: "text", text: "Hello", providerMetadata: { openai: { itemId: "msg_1" } } },
       ])
     }),
   )
@@ -1719,7 +1849,7 @@ describe("OpenAI Responses route", () => {
   it.effect("streams each reasoning summary part as a separate block", () =>
     Effect.gen(function* () {
       const response = yield* LLMClient.generate(
-        LLMRequest.update(request, { providerOptions: { openai: { store: false } } }),
+        LLMRequest.update(request, { providerOptions: { store: false } }),
       ).pipe(
         Effect.provide(
           fixedResponse(
@@ -1773,9 +1903,7 @@ describe("OpenAI Responses route", () => {
 
   it.effect("closes reasoning summary parts when storage is not disabled", () =>
     Effect.gen(function* () {
-      const response = yield* LLMClient.generate(
+      const response = yield* LLMClient.generate(LLMRequest.update(request, { providerOptions: { store: true } })).pipe(
-        LLMRequest.update(request, { providerOptions: { openai: { store: true } } }),
-      ).pipe(
         Effect.provide(
           fixedResponse(
             sseEvents(
@@ -1829,7 +1957,7 @@ describe("OpenAI Responses route", () => {
             ]),
             Message.user("Summarize it."),
           ],
-          providerOptions: { openai: { store: false } },
+          providerOptions: { store: false },
         }),
       ).pipe(
         Effect.provide(
@@ -1842,6 +1970,7 @@ describe("OpenAI Responses route", () => {
                   { role: "user", content: [{ type: "input_text", text: "What changed?" }] },
                   {
                     type: "reasoning",
+                    id: "rs_1",
                     encrypted_content: "encrypted-state",
                     summary: [{ type: "summary_text", text: "Checked the previous diff." }],
                   },
@@ -1849,7 +1978,6 @@ describe("OpenAI Responses route", () => {
                   { role: "user", content: [{ type: "input_text", text: "Summarize it." }] },
                 ],
               })
-              expect(body.input[1]).not.toHaveProperty("id")
               return input.respond(
                 sseEvents(
                   { type: "response.output_text.delta", item_id: "msg_1", delta: "Parser now round-trips reasoning." },
@@ -1888,18 +2016,19 @@ describe("OpenAI Responses route", () => {
               { type: "text", text: "After." },
             ]),
           ],
-          providerOptions: { openai: { store: false } },
+          providerOptions: { store: false },
         }),
       )
 
       expect(prepared.body.input).toEqual([
-        { role: "assistant", content: [{ type: "output_text", text: "Before." }] },
+        { type: "message", role: "assistant", content: [{ type: "output_text", text: "Before." }] },
         {
           type: "reasoning",
+          id: "rs_1",
           encrypted_content: "encrypted-state",
           summary: [{ type: "summary_text", text: "Checked order." }],
         },
-        { role: "assistant", content: [{ type: "output_text", text: "After." }] },
+        { type: "message", role: "assistant", content: [{ type: "output_text", text: "After." }] },
       ])
     }),
   )
@@ -1918,7 +2047,7 @@ describe("OpenAI Responses route", () => {
               },
             ]),
           ],
-          providerOptions: { openai: { store: true } },
+          providerOptions: { store: true },
         }),
       )
 
@@ -1951,7 +2080,7 @@ describe("OpenAI Responses route", () => {
             ]),
             Message.user("Continue."),
           ],
-          providerOptions: { openai: { store: true } },
+          providerOptions: { store: true },
         }),
       )
 
@@ -2024,13 +2153,14 @@ describe("OpenAI Responses route", () => {
               },
             ]),
           ],
-          providerOptions: { openai: { store: false } },
+          providerOptions: { store: false },
         }),
       )
 
       expect(prepared.body.input).toEqual([
         {
           type: "reasoning",
+          id: "rs_1",
           encrypted_content: "encrypted-state",
           summary: [
             { type: "summary_text", text: "First" },
@@ -2064,7 +2194,7 @@ describe("OpenAI Responses route", () => {
             ]),
             Message.user("Summarize it."),
           ],
-          providerOptions: { openai: { store: false } },
+          providerOptions: { store: false },
         }),
       )
 
@@ -2163,6 +2293,50 @@ describe("OpenAI Responses route", () => {
           usage,
         },
       ])
+
+      const prepared = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
+      expect(prepared.body.input).toEqual([
+        {
+          type: "function_call",
+          id: "item_1",
+          call_id: "call_1",
+          name: "lookup",
+          arguments: '{"query":"weather"}',
+        },
+      ])
+    }),
+  )
+
+  it.effect("finalizes a pending function call at response completion", () =>
+    Effect.gen(function* () {
+      const body = sseEvents(
+        {
+          type: "response.output_item.added",
+          item: { type: "function_call", id: "item_1", call_id: "call_1", name: "lookup", arguments: "" },
+        },
+        { type: "response.completed", response: { usage: { input_tokens: 5, output_tokens: 1 } } },
+      )
+      const response = yield* LLMClient.generate(request).pipe(Effect.provide(fixedResponse(body)))
+
+      expect(response.events.filter((event) => LLMEvent.is.toolInputEnd(event) || LLMEvent.is.toolCall(event))).toEqual(
+        [
+          {
+            type: "tool-input-end",
+            id: "call_1",
+            name: "lookup",
+            providerMetadata: { openai: { itemId: "item_1" } },
+          },
+          {
+            type: "tool-call",
+            id: "call_1",
+            name: "lookup",
+            input: {},
+            providerExecuted: undefined,
+            providerMetadata: { openai: { itemId: "item_1" } },
+          },
+        ],
+      )
+      expect(response.finishReason.normalized).toBe("tool-calls")
     }),
   )
 
@@ -2229,6 +2403,35 @@ describe("OpenAI Responses route", () => {
     }),
   )
 
+  it.effect("retains function call item metadata when output_item.added is absent", () =>
+    Effect.gen(function* () {
+      const response = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              {
+                type: "response.output_item.done",
+                item: {
+                  type: "function_call",
+                  id: "item_1",
+                  call_id: "call_1",
+                  name: "lookup",
+                  arguments: '{"query":"weather"}',
+                },
+              },
+              { type: "response.completed", response: { id: "resp_1" } },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.events.find(LLMEvent.is.toolCall)).toMatchObject({
+        id: "call_1",
+        providerMetadata: { openai: { itemId: "item_1" } },
+      })
+    }),
+  )
+
   it.effect("decodes web_search_call as provider-executed tool-call + tool-result", () =>
     Effect.gen(function* () {
       const item = {

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

@@ -141,7 +141,7 @@ describe("OpenRouter", () => {
         LLM.request({
           model: OpenRouter.configure({
             apiKey: "test-key",
-            providerOptions: { openrouter: { usage: false } },
+            providerOptions: { usage: false },
           }).model("openai/gpt-4o-mini"),
           cache: "none",
           prompt: "Hello",
@@ -159,17 +159,15 @@ describe("OpenRouter", () => {
           model: OpenRouter.configure({
             apiKey: "test-key",
             providerOptions: {
-              openrouter: {
+              usage: true,
-                usage: true,
+              reasoning: { effort: "high" },
-                reasoning: { effort: "high" },
+              models: ["anthropic/claude-sonnet-4.6", "google/gemini-3.1-pro"],
-                models: ["anthropic/claude-sonnet-4.6", "google/gemini-3.1-pro"],
+              provider: { order: ["anthropic", "google"], require_parameters: true },
-                provider: { order: ["anthropic", "google"], require_parameters: true },
+              plugins: [{ id: "response-healing" }],
-                plugins: [{ id: "response-healing" }],
+              web_search_options: { engine: "native", max_results: 3 },
-                web_search_options: { engine: "native", max_results: 3 },
+              debug: { echo_upstream_body: true },
-                debug: { echo_upstream_body: true },
+              user: "user_123",
-                user: "user_123",
+              future_option: { enabled: true },
-                future_option: { enabled: true },
-              },
             },
           }).model("anthropic/claude-3.7-sonnet:thinking"),
           prompt: "Think briefly.",
@@ -210,7 +208,7 @@ describe("OpenRouter", () => {
         LLM.request({
           model: OpenRouter.configure({
             apiKey: "test-key",
-            providerOptions: { openrouter: invalid },
+            providerOptions: invalid,
           }).model("openai/gpt-4o-mini"),
           prompt: "Hello",
         }),

packages/ai/test/recorded-scenarios.ts

@@ -181,12 +181,10 @@ const normalizeImageText = (value: string) =>
     .trim()
 
 const encryptedReasoningOptions = {
-  openai: {
+  store: false,
-    store: false,
+  include: ["reasoning.encrypted_content"],
-    include: ["reasoning.encrypted_content"],
+  reasoningEffort: "low",
-    reasoningEffort: "low",
+  reasoningSummary: "auto",
-    reasoningSummary: "auto",
-  },
 } as const
 
 type AssistantTextExpectation = string | RegExp
@@ -304,8 +302,7 @@ const runTextScenario = (context: GoldenScenarioContext) =>
     assistant.expectText(/^Hello!?$/, {
       system: "You are concise.",
       maxTokens: context.maxTokens ?? 40,
-      providerOptions:
+      providerOptions: context.model.route.id === "gemini" ? { thinkingConfig: { thinkingBudget: 0 } } : undefined,
-        context.model.route.id === "gemini" ? { gemini: { thinkingConfig: { thinkingBudget: 0 } } } : undefined,
     }),
   ])
 
@@ -388,7 +385,7 @@ const runReasoningScenario = (context: GoldenScenarioContext) =>
     user("Think briefly, then reply exactly with: Hello!"),
     assistant.expectText(/^Hello!?$/, {
       system: "Show concise reasoning when the provider supports visible reasoning summaries.",
-      providerOptions: { openai: { reasoningEffort: "low", reasoningSummary: "auto" } },
+      providerOptions: { reasoningEffort: "low", reasoningSummary: "auto" },
       maxTokens: context.maxTokens ?? 120,
       assert: (response) => expect(response.usage?.reasoningTokens ?? 0).toBeGreaterThan(0),
     }),

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

@@ -55,7 +55,7 @@ const schema_only_weather = Tool.make({
 })
 
 describe("LLMClient tools", () => {
-  it.effect("uses the registered model route when adding runtime tools", () =>
+  it.effect("uses the selected model route when adding runtime tools", () =>
     Effect.gen(function* () {
       const layer = scriptedResponses([
         sseEvents(deltaChunk({ role: "assistant", content: "Done." }), finishChunk("stop")),
@@ -636,7 +636,7 @@ describe("LLMClient tools", () => {
             .with({ endpoint: { baseURL: "https://api.openai.test/v1/" }, auth: Auth.bearer("test") })
             .model({ id: "gpt-5.5" }),
           prompt: "Use the tool.",
-          providerOptions: { openai: { store: false, include: ["reasoning.encrypted_content"] } },
+          providerOptions: { store: false, include: ["reasoning.encrypted_content"] },
         }),
         tools: { get_weather },
       }).pipe(Stream.runCollect, Effect.provide(layer))

packages/app/e2e/performance/README.md

@@ -2,12 +2,34 @@
 
 The app's high-volume performance diagnostics live under `packages/app/e2e/performance` and are excluded from normal local and CI Playwright discovery. The benchmark config builds the app and serves the production bundle before running scenarios serially.
 
+The `devex` category is the explicit exception to the production-build rule. It measures development commands from submission through a user-visible ready state and has its own Playwright configuration.
+
 Run the suite explicitly from `packages/app`:
 
 ```sh
 bun run test:bench
 ```
 
+Run the desktop development startup benchmark from the repository root:
+
+```sh
+bun run bench:devex
+```
+
+It runs five serial samples of the exact `bun dev:desktop` command. Each sample uses a fresh desktop profile, database, service configuration, service registration, and service process; the desktop selects an isolated ephemeral loopback endpoint. It removes desktop build output and the desktop Vite cache before every run; dependencies, Bun's package cache, and Electron remain installed. The harness stops only that sample's service; it does not stop or change the elected global OpenCode service. The measured endpoint is a visible Home page whose empty-state controls pass Playwright actionability checks. The command's Electron installation check remains inside the measured interval.
+
+Set `DESKTOP_STARTUP_RUNS` only for focused diagnostics:
+
+```sh
+DESKTOP_STARTUP_RUNS=1 bun run bench:devex
+```
+
+Set `OPENCODE_PERFORMANCE_TRACE_DIR` to capture the renderer's CDP trace from attachment through actionable Home:
+
+```sh
+DESKTOP_STARTUP_RUNS=1 OPENCODE_PERFORMANCE_TRACE_DIR=/tmp/opencode-desktop-traces bun run bench:devex
+```
+
 PowerShell:
 
 ```powershell

packages/app/e2e/performance/chrome-trace.ts

@@ -18,9 +18,9 @@ const categories = [
   "disabled-by-default-v8.cpu_profiler",
 ]
 
-export async function startChromeTrace(page: Page, name: string) {
+export async function startChromeTrace(page: Page, name: string): Promise<undefined | (() => Promise<string>)> {
   const directory = process.env.OPENCODE_PERFORMANCE_TRACE_DIR
-  if (!directory) return
+  if (!directory) return undefined
 
   const selectors = process.env.OPENCODE_PERFORMANCE_SELECTOR_TRACE === "1"
   const file = await prepareChromeTrace(directory, name, selectors)

packages/app/e2e/performance/devex/desktop-startup-benchmark.spec.ts

@@ -0,0 +1,26 @@
+import { benchmark } from "../benchmark"
+import {
+  desktopBenchmarkContext,
+  runDesktopStartup,
+  summarizeDesktopStartup,
+  type DesktopStartupSample,
+} from "./desktop-startup"
+
+benchmark.describe("devex: desktop startup", () => {
+  benchmark("opens a cold desktop on Home", async ({ report }, testInfo) => {
+    benchmark.setTimeout(15 * 60_000)
+    const runs = Number(process.env.DESKTOP_STARTUP_RUNS ?? 5)
+    if (!Number.isSafeInteger(runs) || runs < 1) throw new Error("DESKTOP_STARTUP_RUNS must be a positive integer")
+
+    const samples: DesktopStartupSample[] = []
+    const context = await desktopBenchmarkContext(runs)
+    for (let run = 1; run <= runs; run++) {
+      const sample = await runDesktopStartup(run, testInfo).catch((error) => {
+        report(samples.length ? { samples, summary: summarizeDesktopStartup(samples) } : { samples }, context)
+        throw error
+      })
+      samples.push(sample)
+    }
+    report({ samples, summary: summarizeDesktopStartup(samples) }, context)
+  })
+})

packages/app/e2e/performance/devex/desktop-startup.ts

@@ -0,0 +1,526 @@
+import { Service } from "@opencode-ai/client/service"
+import { chromium, expect, type Browser, type Page, type TestInfo } from "@playwright/test"
+import { spawn, spawnSync, type ChildProcess } from "node:child_process"
+import { mkdir, mkdtemp, readFile, rm, writeFile } from "node:fs/promises"
+import { tmpdir } from "node:os"
+import { join, resolve } from "node:path"
+import { startChromeTrace } from "../chrome-trace"
+
+const repository = resolve(import.meta.dirname, "../../../../..")
+const milestones = [
+  "bunRootScript",
+  "bunDesktopScript",
+  "desktopPrepared",
+  "mainBundleReady",
+  "preloadBundleReady",
+  "rendererDevServerReady",
+  "electronSpawnStarted",
+  "debugEndpointReady",
+  "electronStarted",
+  "serviceEnsureStarted",
+  "serviceSpawnRequested",
+  "serviceReady",
+  "backgroundLoadingReady",
+  "rendererViteConnected",
+  "rendererInitializationStarted",
+  "rendererInitializationReady",
+  "windowVisible",
+  "homeReady",
+] as const
+const phases = [
+  "desktopPreparation",
+  "viteMainBundle",
+  "vitePreloadBundle",
+  "rendererServerStartup",
+  "electronStartup",
+  "serviceSpawnWait",
+  "serviceProcessStartup",
+  "rendererStartup",
+  "visibleWindowToHome",
+] as const
+
+type Milestone = (typeof milestones)[number]
+type Phase = (typeof phases)[number]
+type ServiceInfo = { id: string; version: string; url: string; pid: number }
+
+export type DesktopStartupSample = {
+  run: number
+  commandToHomeReadyMs: number
+  milestonesMs: Record<Milestone, number>
+  phasesMs: Record<Phase, number>
+  service: Omit<ServiceInfo, "id">
+}
+
+export async function runDesktopStartup(run: number, testInfo: TestInfo) {
+  const profile = await createColdProfile()
+  const desktop = await Promise.resolve()
+    .then(() => startDesktop(profile))
+    .catch(async (error) => {
+      await rm(profile.root, { recursive: true, force: true, maxRetries: 5, retryDelay: 100 })
+      throw error
+    })
+  try {
+    const page = await desktop.open()
+    const stopTrace = await startChromeTrace(page, `desktop-startup-${run}`)
+    try {
+      await startThemeObservation(page)
+      await waitForHome(page, desktop.mark)
+      await requireStableTheme(page)
+      return await desktop.result(run)
+    } finally {
+      await stopTrace?.()
+    }
+  } finally {
+    await desktop.close(testInfo, run)
+  }
+}
+
+export async function desktopBenchmarkContext(runs: number) {
+  const pkg = JSON.parse(await readFile(join(repository, "packages/desktop/package.json"), "utf8"))
+  const revision = spawnSync("git", ["rev-parse", "HEAD"], { cwd: repository })
+  if (revision.status !== 0) throw new Error("Failed to read the benchmark Git revision")
+  const status = spawnSync("git", ["status", "--porcelain"], { cwd: repository })
+  if (status.status !== 0) throw new Error("Failed to read the benchmark Git status")
+  const bun = spawnSync("bun", ["--version"], { cwd: repository })
+  if (bun.status !== 0) throw new Error("Failed to read the benchmark Bun version")
+  return {
+    arch: process.arch,
+    command: "bun dev:desktop",
+    runs,
+    profile: "fresh",
+    service: "isolated-cold",
+    install: "complete",
+    viteCache: "cold",
+    electronInstall: "present",
+    bunVersion: bun.stdout.toString().trim(),
+    electronVersion: pkg.devDependencies.electron,
+    electronViteVersionRange: pkg.devDependencies["electron-vite"],
+    gitCommit: revision.stdout.toString().trim(),
+    gitDirty: status.stdout.length > 0,
+    trace: Boolean(process.env.OPENCODE_PERFORMANCE_TRACE_DIR),
+  }
+}
+
+export function summarizeDesktopStartup(samples: DesktopStartupSample[]) {
+  return {
+    commandToHomeReadyMs: statistics(samples.map((sample) => sample.commandToHomeReadyMs)),
+    milestonesMs: Object.fromEntries(
+      milestones.map((name) => [name, statistics(samples.map((sample) => sample.milestonesMs[name]))]),
+    ),
+    phasesMs: Object.fromEntries(
+      phases.map((name) => [name, statistics(samples.map((sample) => sample.phasesMs[name]))]),
+    ),
+  }
+}
+
+export function milestoneForLine(line: string): Milestone | undefined {
+  const text = stripAnsi(line)
+  return milestonePatterns.find((item) => text.includes(item.text))?.name
+}
+
+const milestonePatterns: ReadonlyArray<{ name: Milestone; text: string }> = [
+  { name: "bunRootScript", text: "$ bun --cwd packages/desktop dev" },
+  { name: "bunDesktopScript", text: "$ bun ./scripts/dev.ts" },
+  { name: "desktopPrepared", text: "Copied dev icons from" },
+  { name: "mainBundleReady", text: "electron main process built successfully" },
+  { name: "preloadBundleReady", text: "electron preload scripts built successfully" },
+  { name: "rendererDevServerReady", text: "dev server running for the electron renderer process at:" },
+  { name: "electronSpawnStarted", text: "starting electron app..." },
+  { name: "debugEndpointReady", text: "DevTools listening on ws://" },
+  { name: "electronStarted", text: "app starting" },
+  { name: "serviceEnsureStarted", text: "starting v2 background service" },
+  { name: "serviceSpawnRequested", text: "v2 CLI background service starting" },
+  { name: "serviceReady", text: "v2 CLI background service ready" },
+  { name: "backgroundLoadingReady", text: "loading task finished" },
+  { name: "rendererViteConnected", text: "[vite] connected." },
+  { name: "rendererInitializationStarted", text: "awaiting server ready" },
+  { name: "rendererInitializationReady", text: "server ready" },
+  { name: "windowVisible", text: "main window visible" },
+]
+
+async function createColdProfile() {
+  await Promise.all(
+    ["packages/desktop/node_modules/.vite", "packages/desktop/out"].map((path) =>
+      rm(join(repository, path), { recursive: true, force: true }),
+    ),
+  )
+  const root = await mkdtemp(join(tmpdir(), "opencode-desktop-startup-"))
+  return initializeColdProfile(root).catch(async (error) => {
+    await rm(root, { recursive: true, force: true, maxRetries: 5, retryDelay: 100 })
+    throw error
+  })
+}
+
+async function initializeColdProfile(root: string) {
+  await Promise.all(
+    ["data", "config", "cache", "state", "desktop", "session", "home"].map((dir) =>
+      mkdir(join(root, dir), { recursive: true }),
+    ),
+  )
+  await Promise.all([
+    writeFile(
+      join(root, "desktop", "opencode.settings"),
+      JSON.stringify({ firstLaunchOnboardingComplete: true }),
+    ),
+    writeFile(join(root, "desktop", "opencode.global.dat"), JSON.stringify({ language: '{"locale":"en"}' })),
+  ])
+  const registration = join(root, "desktop", "opencode", "service-local.json")
+  await Service.stop({ file: registration })
+  return { root, registration }
+}
+
+function startDesktop(profile: Awaited<ReturnType<typeof createColdProfile>>) {
+  const started = performance.now()
+  const child = spawn("bun", ["dev:desktop"], {
+    cwd: repository,
+    detached: process.platform !== "win32",
+    env: {
+      ...process.env,
+      OPENCODE_CONFIG_DIR: join(profile.root, "config"),
+      OPENCODE_DB: join(profile.root, "data", "opencode.db"),
+      OPENCODE_TEST_HOME: join(profile.root, "home"),
+      OPENCODE_TEST_ONBOARDING: "0",
+      OPENCODE_DESKTOP_TEST_ROOT: profile.root,
+      OPENCODE_DESKTOP_REMOTE_DEBUGGING_PORT: "0",
+      OPENCODE_DESKTOP_DISABLE_PROTOCOL_REGISTRATION: "1",
+    },
+    stdio: ["ignore", "pipe", "pipe"],
+  })
+  if (!child.pid || !child.stdout || !child.stderr) throw new Error("Failed to start the desktop command")
+  const exited = childExit(child)
+  const observed: Partial<Record<Milestone, number>> = {}
+  const endpoint = Promise.withResolvers<string>()
+  const pageErrors: string[] = []
+  let browser: Browser | undefined
+  let service: ServiceInfo | undefined
+  const mark = (name: Milestone) => {
+    observed[name] ??= elapsed(started)
+  }
+  const record = (line: string) => {
+    const milestone = milestoneForLine(line)
+    if (milestone) mark(milestone)
+    const match = stripAnsi(line).match(/DevTools listening on (ws:\/\/\S+)/)
+    if (match?.[1]) endpoint.resolve(match[1])
+  }
+  const stdout = observeOutput(child.stdout, record)
+  const stderr = observeOutput(child.stderr, record)
+
+  return {
+    mark,
+    async open() {
+      const url = await Promise.race([
+        endpoint.promise,
+        exited.then((code) => {
+          throw new Error(`Desktop command exited with code ${code} before opening its debug endpoint`)
+        }),
+        sleep(120_000).then(() => {
+          throw new Error("Timed out waiting for the desktop debug endpoint")
+        }),
+      ])
+      browser = await chromium.connectOverCDP(url, { timeout: 120_000 })
+      const context = browser.contexts()[0]
+      if (!context) throw new Error("Electron did not expose a browser context")
+      await expect.poll(() => context.pages().length, { timeout: 120_000 }).toBeGreaterThan(0)
+      const page = context.pages()[0]
+      if (!page) throw new Error("Electron did not expose a renderer page")
+      page.on("pageerror", (error) => pageErrors.push(error.stack ?? error.message))
+      return page
+    },
+    async result(run: number): Promise<DesktopStartupSample> {
+      if (pageErrors.length) throw new Error(`Desktop renderer reported errors:\n\n${pageErrors.join("\n\n")}`)
+      service = await readService(profile)
+      const milestonesMs = requireMilestones(observed)
+      return {
+        run,
+        commandToHomeReadyMs: milestonesMs.homeReady,
+        milestonesMs,
+        phasesMs: calculatePhases(milestonesMs),
+        service: {
+          version: service.version,
+          url: service.url,
+          pid: service.pid,
+        },
+      }
+    },
+    async close(testInfo: TestInfo, run: number) {
+      const errors: unknown[] = []
+      await browser?.close().catch(() => undefined)
+      await stopProcessTree(child, exited).catch((error) => {
+        errors.push(error)
+        child.stdout?.destroy()
+        child.stderr?.destroy()
+      })
+      const [stdoutText, stderrText] = await Promise.all([stdout, stderr]).catch((error) => {
+        errors.push(error)
+        return ["", ""]
+      })
+      await Promise.all([
+        testInfo.attach(`desktop-startup-${run}-stdout`, { body: stdoutText, contentType: "text/plain" }),
+        testInfo.attach(`desktop-startup-${run}-stderr`, { body: stderrText, contentType: "text/plain" }),
+        pageErrors.length
+          ? testInfo.attach(`desktop-startup-${run}-page-errors`, {
+              body: pageErrors.join("\n\n"),
+              contentType: "text/plain",
+            })
+          : Promise.resolve(),
+      ]).catch((error) => errors.push(error))
+      await Service.stop({ file: profile.registration }).catch((error) => errors.push(error))
+      if (service && processAlive(service.pid))
+        errors.push(new Error(`Desktop service process ${service.pid} did not stop`))
+      await rm(profile.root, { recursive: true, force: true, maxRetries: 5, retryDelay: 100 }).catch((error) =>
+        errors.push(error),
+      )
+      if (errors.length) throw new AggregateError(errors, "Desktop benchmark cleanup failed")
+    },
+  }
+}
+
+async function waitForHome(page: Page, mark: (name: Milestone) => void) {
+  await expect.poll(() => page.evaluate(() => document.visibilityState), { timeout: 120_000 }).toBe("visible")
+
+  const projects = page.getByRole("complementary", { name: "Projects", exact: true })
+  const sessions = page.getByRole("region", { name: "Recent sessions", exact: true })
+  const search = page.getByRole("textbox", { name: "Search sessions", exact: true })
+  const addProject = projects.locator('button[data-action="home-add-project-row"]')
+  await expect(projects).toBeVisible({ timeout: 120_000 })
+  await expect(sessions).toBeVisible()
+  await expect(search).toBeEditable()
+  await expect(sessions.getByText("Nothing here yet", { exact: true })).toBeVisible()
+  await expect(addProject).toHaveCount(1)
+  await addProject.click({ trial: true })
+  mark("homeReady")
+}
+
+type ThemeWindow = Window & {
+  __OPENCODE_THEME_STATES__?: string[]
+  __OPENCODE_THEME_OBSERVER__?: MutationObserver
+}
+
+async function startThemeObservation(page: Page) {
+  await page.addInitScript(installThemeObservation)
+  await page.evaluate(installThemeObservation)
+}
+
+async function requireStableTheme(page: Page) {
+  const states = await page.evaluate(() => {
+    const target = window as ThemeWindow
+    target.__OPENCODE_THEME_OBSERVER__?.disconnect()
+    return target.__OPENCODE_THEME_STATES__ ?? []
+  })
+  if (states.length !== 1) throw new Error(`Desktop theme changed during startup: ${states.join(" -> ")}`)
+}
+
+function installThemeObservation() {
+  const target = window as ThemeWindow
+  const observeRoot = () => {
+    const root = document.documentElement
+    if (!root) return false
+    const state = () => {
+      const theme = root.dataset.theme
+      const scheme = root.dataset.colorScheme
+      return theme && scheme ? `${theme}:${scheme}` : undefined
+    }
+    const initial = state()
+    target.__OPENCODE_THEME_STATES__ = initial ? [initial] : []
+    target.__OPENCODE_THEME_OBSERVER__ = new MutationObserver(() => {
+      const next = state()
+      if (!next) return
+      if (target.__OPENCODE_THEME_STATES__?.at(-1) !== next) target.__OPENCODE_THEME_STATES__?.push(next)
+    })
+    target.__OPENCODE_THEME_OBSERVER__.observe(root, {
+      attributes: true,
+      attributeFilter: ["data-theme", "data-color-scheme"],
+    })
+    return true
+  }
+  if (observeRoot()) return
+  const documentObserver = new MutationObserver(() => {
+    if (!observeRoot()) return
+    documentObserver.disconnect()
+  })
+  target.__OPENCODE_THEME_OBSERVER__ = documentObserver
+  documentObserver.observe(document, { childList: true })
+}
+
+async function observeOutput(stream: NodeJS.ReadableStream, record: (line: string) => void) {
+  const decoder = new TextDecoder()
+  const output: string[] = []
+  let pending = ""
+  for await (const chunk of stream) {
+    const text = typeof chunk === "string" ? chunk : decoder.decode(chunk, { stream: true })
+    output.push(text)
+    pending += text
+    const lines = pending.split(/\r?\n/)
+    pending = lines.pop() ?? ""
+    lines.forEach(record)
+  }
+  const final = decoder.decode()
+  output.push(final)
+  pending += final
+  if (pending) record(pending)
+  return output.join("")
+}
+
+async function readService(profile: Awaited<ReturnType<typeof createColdProfile>>) {
+  const value: unknown = JSON.parse(await readFile(profile.registration, "utf8"))
+  if (!isServiceInfo(value)) throw new Error("Desktop service registration is invalid")
+  const url = new URL(value.url)
+  const port = Number(url.port)
+  if (url.hostname !== "127.0.0.1" || !Number.isInteger(port) || port <= 0)
+    throw new Error(`Desktop service used unexpected endpoint ${value.url}`)
+  if (!value.version.startsWith("2.0.0-local-"))
+    throw new Error(`Desktop service used unexpected version ${value.version}`)
+  return value
+}
+
+function isServiceInfo(value: unknown): value is ServiceInfo {
+  return (
+    typeof value === "object" &&
+    value !== null &&
+    "id" in value &&
+    typeof value.id === "string" &&
+    "version" in value &&
+    typeof value.version === "string" &&
+    "url" in value &&
+    typeof value.url === "string" &&
+    "pid" in value &&
+    typeof value.pid === "number"
+  )
+}
+
+function requireMilestones(observed: Partial<Record<Milestone, number>>) {
+  const get = (name: Milestone) => {
+    const value = observed[name]
+    if (value === undefined) throw new Error(`Desktop startup did not report milestone: ${name}`)
+    return round(value)
+  }
+  return {
+    bunRootScript: get("bunRootScript"),
+    bunDesktopScript: get("bunDesktopScript"),
+    desktopPrepared: get("desktopPrepared"),
+    mainBundleReady: get("mainBundleReady"),
+    preloadBundleReady: get("preloadBundleReady"),
+    rendererDevServerReady: get("rendererDevServerReady"),
+    electronSpawnStarted: get("electronSpawnStarted"),
+    debugEndpointReady: get("debugEndpointReady"),
+    electronStarted: get("electronStarted"),
+    serviceEnsureStarted: get("serviceEnsureStarted"),
+    serviceSpawnRequested: get("serviceSpawnRequested"),
+    serviceReady: get("serviceReady"),
+    backgroundLoadingReady: get("backgroundLoadingReady"),
+    rendererViteConnected: get("rendererViteConnected"),
+    rendererInitializationStarted: get("rendererInitializationStarted"),
+    rendererInitializationReady: get("rendererInitializationReady"),
+    windowVisible: get("windowVisible"),
+    homeReady: get("homeReady"),
+  }
+}
+
+function calculatePhases(value: Record<Milestone, number>): Record<Phase, number> {
+  return {
+    desktopPreparation: value.desktopPrepared,
+    viteMainBundle: round(value.mainBundleReady - value.desktopPrepared),
+    vitePreloadBundle: round(value.preloadBundleReady - value.mainBundleReady),
+    rendererServerStartup: round(value.rendererDevServerReady - value.preloadBundleReady),
+    electronStartup: round(value.electronStarted - value.electronSpawnStarted),
+    serviceSpawnWait: round(value.serviceSpawnRequested - value.serviceEnsureStarted),
+    serviceProcessStartup: round(value.serviceReady - value.serviceSpawnRequested),
+    rendererStartup: round(value.homeReady - value.rendererViteConnected),
+    visibleWindowToHome: round(value.homeReady - value.windowVisible),
+  }
+}
+
+function statistics(values: number[]) {
+  if (!values.length) throw new Error("Cannot summarize an empty benchmark")
+  const sorted = values.toSorted((left, right) => left - right)
+  const median = medianOf(sorted)
+  return {
+    min: round(sorted[0]),
+    median: round(median),
+    max: round(sorted.at(-1)!),
+    medianAbsoluteDeviation: round(medianOf(sorted.map((value) => Math.abs(value - median)).toSorted((a, b) => a - b))),
+  }
+}
+
+function medianOf(sorted: number[]) {
+  const middle = Math.floor(sorted.length / 2)
+  if (sorted.length % 2) return sorted[middle]
+  return (sorted[middle - 1] + sorted[middle]) / 2
+}
+
+async function stopProcessTree(child: ChildProcess, exited: Promise<number | null>) {
+  if (!child.pid) throw new Error("Desktop command has no process ID")
+  if (process.platform !== "win32") return stopProcessGroup(child.pid, exited)
+  if (child.exitCode !== null || (await exitsWithin(child, exited, 2_000))) return
+  const kill = spawn("taskkill.exe", ["/PID", String(child.pid), "/T", "/F"], {
+    stdio: "ignore",
+  })
+  await childExit(kill)
+  if (await exitsWithin(child, exited, 10_000)) return
+  if (!(await exitsWithin(child, exited, 5_000))) throw new Error(`Desktop command process ${child.pid} did not stop`)
+}
+
+async function stopProcessGroup(pid: number, exited: Promise<number | null>) {
+  await Promise.race([exited, sleep(2_000)])
+  if (!processGroupAlive(pid)) return
+  process.kill(-pid, "SIGTERM")
+  if (await processGroupStopsWithin(pid, 10_000)) return
+  process.kill(-pid, "SIGKILL")
+  if (!(await processGroupStopsWithin(pid, 5_000))) throw new Error(`Desktop command process group ${pid} did not stop`)
+}
+
+async function processGroupStopsWithin(pid: number, timeout: number) {
+  const deadline = Date.now() + timeout
+  while (Date.now() < deadline) {
+    if (!processGroupAlive(pid)) return true
+    await sleep(50)
+  }
+  return !processGroupAlive(pid)
+}
+
+function processGroupAlive(pid: number) {
+  try {
+    process.kill(-pid, 0)
+    return true
+  } catch {
+    return false
+  }
+}
+
+async function exitsWithin(child: ChildProcess, exited: Promise<number | null>, timeout: number) {
+  if (child.exitCode !== null) return true
+  const result = await Promise.race([exited.then(() => true), sleep(timeout).then(() => false)])
+  return result
+}
+
+function childExit(child: ChildProcess) {
+  return new Promise<number | null>((resolve, reject) => {
+    child.once("error", reject)
+    child.once("exit", (code) => resolve(code))
+  })
+}
+
+function sleep(milliseconds: number) {
+  return new Promise<void>((resolve) => setTimeout(resolve, milliseconds))
+}
+
+function processAlive(pid: number) {
+  try {
+    process.kill(pid, 0)
+    return true
+  } catch {
+    return false
+  }
+}
+
+function stripAnsi(value: string) {
+  return value.replace(/\x1b\[[0-?]*[ -/]*[@-~]/g, "")
+}
+
+function elapsed(started: number) {
+  return round(performance.now() - started)
+}
+
+function round(value: number) {
+  return Math.round(value * 100) / 100
+}

packages/app/e2e/performance/devex/playwright.config.ts

@@ -0,0 +1,17 @@
+import { defineConfig } from "@playwright/test"
+
+process.env.OPENCODE_PERFORMANCE_RUN_ID ??= `${new Date().toISOString().replace(/[:.]/g, "-")}-${process.pid}`
+
+export default defineConfig({
+  testDir: ".",
+  testMatch: "desktop-startup-benchmark.spec.ts",
+  outputDir: "../../test-results/performance-devex",
+  timeout: 15 * 60_000,
+  expect: {
+    timeout: 120_000,
+  },
+  fullyParallel: false,
+  workers: 1,
+  reporter: [["html", { outputFolder: "../../playwright-report/performance-devex", open: "never" }], ["line"]],
+  projects: [{ name: "desktop" }],
+})

packages/app/e2e/performance/playwright.config.ts

@@ -7,7 +7,7 @@ process.env.OPENCODE_PERFORMANCE_RUN_ID ??= `${new Date().toISOString().replace(
 export default {
   ...config,
   testDir: ".",
-  testIgnore: "unit/**",
+  testIgnore: ["unit/**", "devex/**"],
   outputDir: "../test-results/performance",
   fullyParallel: false,
   workers: 1,

packages/app/e2e/performance/timeline-stability/adverse.spec.ts

@@ -179,7 +179,7 @@ test.describe("timeline adverse visual stability", () => {
         userMessage(),
         assistantMessage([
           shell(shellID, "completed", wideLines(15)),
-          toolPart(contextIDs[0]!, "read", "completed", { filePath: "src/a.ts" }),
+          toolPart(contextIDs[0]!, "read", "completed", { path: "src/a.ts" }),
           toolPart(contextIDs[1]!, "glob", "completed", { path: ".", pattern: "**/*.ts" }),
           textPart(followingID, "Following responsive timeline content that wraps on narrow screens."),
         ]),

packages/app/e2e/performance/timeline-stability/file-matrix.spec.ts

@@ -1,4 +1,5 @@
 import { test } from "@playwright/test"
+import { createTwoFilesPatch } from "diff"
 import {
   defineVisualRegions,
   reportVisualStability,
@@ -18,16 +19,20 @@ import {
 } from "./fixture"
 
 const profiles = [
-  { name: "edit", tool: "edit", input: { filePath: "src/edit.ts" } },
+  {
+    name: "edit",
+    tool: "edit",
+    input: { path: "src/edit.ts", oldString: "export const value = 1", newString: "export const value = 2" },
+  },
   {
     name: "multi patch",
-    tool: "apply_patch",
+    tool: "patch",
-    input: { files: ["src/a.ts", "src/b.ts", "src/old.ts", "src/moved.ts"] },
+    input: { patchText: "Update generated files" },
   },
 ] as const
 
 for (const profile of profiles) {
-  test(`stabilizes ${profile.name} pending to completed`, async ({ page }, testInfo) => {
+  test(`stabilizes ${profile.name} streaming to completed`, async ({ page }, testInfo) => {
     const partID = `prt_file_matrix_${profiles.indexOf(profile)}`
     const followingID = `prt_file_matrix_following_${profiles.indexOf(profile)}`
     const timeline = await setupTimeline(page, {
@@ -35,7 +40,7 @@ for (const profile of profiles) {
         userMessage(),
         assistantMessage(
           [
-            toolPart(partID, profile.tool, "pending", profile.input),
+            toolPart(partID, profile.tool, "streaming", profile.input),
             textPart(followingID, `Following ${profile.name}`),
           ],
           { completed: false },
@@ -89,34 +94,27 @@ function completedPart(partID: string, profile: (typeof profiles)[number]) {
   if (profile.tool === "edit") {
     return toolPart(partID, profile.tool, "completed", profile.input, {
       metadata: {
-        filediff: {
+        files: [patchFile("src/edit.ts", "modified", 50)],
-          file: "src/edit.ts",
-          additions: 50,
-          deletions: 50,
-          before: source(50, false),
-          after: source(50, true),
-        },
       },
     })
   }
   const files = [
-    patchFile("src/a.ts", "update"),
+    patchFile("src/a.ts", "modified", 20),
-    patchFile("src/b.ts", "add"),
+    patchFile("src/b.ts", "added", 20),
-    patchFile("src/old.ts", "delete"),
+    patchFile("src/old.ts", "deleted", 20),
-    { ...patchFile("src/moved.ts", "move"), move: "src/new-place.ts" },
   ]
   return toolPart(partID, profile.tool, "completed", profile.input, { metadata: { files } })
 }
 
-function patchFile(filePath: string, type: "add" | "update" | "delete" | "move") {
+function patchFile(file: string, status: "added" | "modified" | "deleted", lines: number) {
+  const before = status === "added" ? "" : source(lines, false)
+  const after = status === "deleted" ? "" : source(lines, true)
   return {
-    filePath,
+    file,
-    relativePath: filePath,
+    status,
-    type,
+    patch: createTwoFilesPatch(`a/${file}`, `b/${file}`, before, after),
-    additions: type === "delete" ? 0 : 20,
+    additions: status === "deleted" ? 0 : lines,
-    deletions: type === "add" ? 0 : 20,
+    deletions: status === "added" ? 0 : lines,
-    before: type === "add" ? undefined : source(20, false),
-    after: type === "delete" ? undefined : source(20, true),
   }
 }
 

packages/app/e2e/performance/timeline-stability/file-mutation.spec.ts

@@ -1,4 +1,5 @@
 import { expect, test } from "@playwright/test"
+import { createTwoFilesPatch } from "diff"
 import {
   defineVisualRegions,
   reportVisualStability,
@@ -20,13 +21,13 @@ import {
 test("adds patch files incrementally without resetting outer expansion", async ({ page }, testInfo) => {
   const patchID = "prt_incremental_01_patch"
   const followingID = "prt_incremental_02_following"
-  const first = patchFile("src/a.ts", "update")
+  const first = patchFile("src/a.ts", "modified")
   const timeline = await setupTimeline(page, {
     messages: [
       userMessage(),
       assistantMessage(
         [
-          toolPart(patchID, "apply_patch", "running", { files: [first.filePath] }, { metadata: { files: [first] } }),
+          toolPart(patchID, "patch", "running", { patchText: "Update files" }, { metadata: { files: [first] } }),
           textPart(followingID, "Following incremental patch"),
         ],
         { completed: false },
@@ -55,15 +56,15 @@ test("adds patch files incrementally without resetting outer expansion", async (
     },
   })
   await startVisualProbe(page, regions)
-  const second = patchFile("src/b.ts", "add")
+  const second = patchFile("src/b.ts", "added")
-  const third = patchFile("src/old.ts", "delete")
+  const third = patchFile("src/old.ts", "deleted")
   await timeline.send(
     partUpdated(
       toolPart(
         patchID,
-        "apply_patch",
+        "patch",
         "running",
-        { files: [first.filePath, second.filePath] },
+        { patchText: "Update files" },
         { metadata: { files: [first, second] } },
       ),
     ),
@@ -73,9 +74,9 @@ test("adds patch files incrementally without resetting outer expansion", async (
     partUpdated(
       toolPart(
         patchID,
-        "apply_patch",
+        "patch",
         "completed",
-        { files: [first.filePath, second.filePath, third.filePath] },
+        { patchText: "Update files" },
         { metadata: { files: [first, second, third] } },
       ),
     ),
@@ -106,15 +107,15 @@ test("adds patch files incrementally without resetting outer expansion", async (
   await expect(page.locator('[data-scope="apply-patch"] [data-type="delete"]')).toBeVisible()
 })
 
-function patchFile(filePath: string, type: "add" | "update" | "delete") {
+function patchFile(file: string, status: "added" | "modified" | "deleted") {
+  const before = status === "added" ? "" : source(false)
+  const after = status === "deleted" ? "" : source(true)
   return {
-    filePath,
+    file,
-    relativePath: filePath,
+    status,
-    type,
+    patch: createTwoFilesPatch(`a/${file}`, `b/${file}`, before, after),
-    additions: type === "delete" ? 0 : 4,
+    additions: status === "deleted" ? 0 : 4,
-    deletions: type === "add" ? 0 : 3,
+    deletions: status === "added" ? 0 : 3,
-    before: type === "add" ? undefined : source(false),
-    after: type === "delete" ? undefined : source(true),
   }
 }
 

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

@@ -35,7 +35,7 @@ describe("timeline fixture validation", () => {
         userMessage(),
         {
           ...assistantMessage(),
-          content: [{ type: "tool", id: "call_invalid", name: "bash", state: { status: "completed" } }],
+          content: [{ type: "tool", id: "call_invalid", name: "shell", state: { status: "completed" } }],
         } as never,
       ]),
     ).toThrow()
@@ -60,12 +60,11 @@ if (false) {
   const userSeed = { id: "prt_type_user", type: "text", text: "typed" } satisfies PartSeed<"user">
   userMessage([userSeed])
 
-  // @ts-expect-error Tool completion fields are not valid while pending.
+  // @ts-expect-error Tool completion fields are not valid while streaming.
-  toolPart("prt_invalid_pending", "bash", "pending", {}, { output: "impossible" })
+  toolPart("prt_invalid_streaming", "shell", "streaming", {}, { output: "impossible" })
-  // @ts-expect-error Tool completion fields are not valid while running.
+  toolPart("prt_valid_running", "shell", "running", {}, { output: "progressive output" })
-  toolPart("prt_invalid_running", "bash", "running", {}, { output: "impossible" })
   // @ts-expect-error Tool error fields are not valid after completion.
-  toolPart("prt_invalid_completed", "bash", "completed", {}, { error: "impossible" })
+  toolPart("prt_invalid_completed", "shell", "completed", {}, { error: "impossible" })
 
   assistantMessage([
     // @ts-expect-error Agent references belong to user messages, not assistant messages.

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

@@ -60,17 +60,17 @@ type ReasoningSeed = {
 type ToolSeed = {
   id: string
   type: "tool"
-  callID: string
+  name: string
-  tool: string
   messageID?: string
   executed?: boolean
   providerState?: Record<string, unknown>
   providerResultState?: Record<string, unknown>
   state:
-    | { status: "pending"; input: Record<string, unknown>; raw: string }
+    | { status: "streaming"; input: Record<string, unknown>; raw: string }
     | {
         status: "running"
         input: Record<string, unknown>
+        output?: string
         title?: string
         metadata: Record<string, unknown>
         time: { start: number }
@@ -100,10 +100,10 @@ export type PartSeed<Owner extends "user" | "assistant"> = Owner extends "user"
   ? TextSeed | FileSeed | AgentSeed
   : TextSeed | ReasoningSeed | ToolSeed
 
-type ToolOptions<State extends ToolStatus> = State extends "pending"
+type ToolOptions<State extends ToolStatus> = State extends "streaming"
   ? { output?: never; title?: never; metadata?: never; error?: never }
   : State extends "running"
-    ? { title?: string; metadata?: Record<string, unknown>; output?: never; error?: never }
+    ? { title?: string; metadata?: Record<string, unknown>; output?: string; error?: never }
     : State extends "error"
       ? { error?: string; metadata?: Record<string, unknown>; output?: never; title?: never }
       : { output?: string; title?: string; metadata?: Record<string, unknown>; error?: never }
@@ -371,6 +371,15 @@ export function partUpdated(part: PartSeed<"assistant">): readonly OpenCodeEvent
   }
   if (part.type === "reasoning") {
     startedParts.add(part.id)
+    if (!started && !part.text)
+      return [
+        makeEvent("session.reasoning.started", {
+          sessionID,
+          assistantMessageID: messageID,
+          ordinal: ref.ordinal!,
+          state: jsonRecord(part.metadata),
+        }),
+      ]
     return [
       ...(started
         ? []
@@ -542,9 +551,9 @@ export function reasoningPart(id: string, text: string): ReasoningSeed {
 export function toolPart(
   id: string,
   tool: string,
-  state: "pending",
+  state: "streaming",
   input: Record<string, unknown>,
-  options?: ToolOptions<"pending">,
+  options?: ToolOptions<"streaming">,
 ): ToolSeed
 export function toolPart(
   id: string,
@@ -574,14 +583,15 @@ export function toolPart(
   input: Record<string, unknown>,
   options: ToolOptions<ToolStatus> = {},
 ): ToolSeed {
-  const base = { id, type: "tool" as const, callID: id, tool }
+  const base = { id, type: "tool" as const, name: tool }
-  if (state === "pending") return { ...base, state: { status: state, input, raw: "" } }
+  if (state === "streaming") return { ...base, state: { status: state, input, raw: "" } }
   if (state === "running")
     return {
       ...base,
       state: {
         status: state,
         input,
+        ...(options.output === undefined ? {} : { output: options.output }),
         title: options.title,
         metadata: options.metadata ?? {},
         time: { start: 1700000001000 },
@@ -612,12 +622,10 @@ export function toolPart(
 }
 
 export function shell(id: string, state: ToolStatus, output = "", command = `echo ${id}`): ToolSeed {
-  if (state === "pending") return toolPart(id, "bash", state, { command })
+  if (state === "streaming") return toolPart(id, "shell", state, { command })
-  if (state === "running")
+  if (state === "running") return toolPart(id, "shell", state, { command }, { title: command, output })
-    return toolPart(id, "bash", state, { command }, { title: command, metadata: { command, output } })
+  if (state === "error") return toolPart(id, "shell", state, { command }, { error: output || undefined })
-  if (state === "error")
+  return toolPart(id, "shell", state, { command }, { title: command, output })
-    return toolPart(id, "bash", state, { command }, { error: output || undefined, metadata: { command, output } })
-  return toolPart(id, "bash", state, { command }, { title: command, output, metadata: { command, output } })
 }
 
 export function completedAssistantInfo(info: SessionMessageAssistant): SessionMessageAssistant {
@@ -655,7 +663,7 @@ function messageContent(
 ): SessionMessageAssistant["content"][number] {
   if (part.type === "tool") {
     partRefs.set(part.id, { messageID, type: part.type })
-    toolStates.set(part.callID, part.state.status)
+    toolStates.set(part.id, part.state.status)
   } else {
     partRefs.set(part.id, { messageID, type: part.type, ordinal: ordinals[part.type]++ })
     startedParts.add(part.id)
@@ -675,8 +683,8 @@ function messageContent(
   const completed = state.status === "completed" || state.status === "error" ? state.time.end : undefined
   const base = {
     type: "tool" as const,
-    id: part.callID,
+    id: part.id,
-    name: part.tool,
+    name: part.name,
     time: {
       created: time?.start ?? 1700000001000,
       ...(time?.start === undefined ? {} : { ran: time.start }),
@@ -686,11 +694,18 @@ function messageContent(
     ...(part.providerState ? { providerState: jsonRecord(part.providerState) } : {}),
     ...(part.providerResultState ? { providerResultState: jsonRecord(part.providerResultState) } : {}),
   }
-  if (state.status === "pending") return { ...base, state: { status: "streaming", input: state.raw } }
+  if (state.status === "streaming") return { ...base, state: { status: "streaming", input: state.raw } }
   if (state.status === "running")
     return {
       ...base,
-      state: { status: "running", input: jsonRecord(state.input), metadata: jsonRecord(state.metadata) },
+      state: {
+        status: "running",
+        input: jsonRecord(state.input),
+        metadata: jsonRecord({
+          ...state.metadata,
+          ...(state.output === undefined ? {} : { output: state.output }),
+        }),
+      },
     }
   if (state.status === "error")
     return {
@@ -714,7 +729,7 @@ function messageContent(
 }
 
 function toolEvents(part: ToolSeed, messageID: string): readonly OpenCodeEvent[] {
-  const previous = toolStates.get(part.callID)
+  const previous = toolStates.get(part.id)
   if (previous === "completed" || previous === "error") return []
 
   const events: OpenCodeEvent[] = []
@@ -723,27 +738,27 @@ function toolEvents(part: ToolSeed, messageID: string): readonly OpenCodeEvent[]
       makeEvent("session.tool.input.started", {
         sessionID,
         assistantMessageID: messageID,
-        id: part.callID,
+        id: part.id,
-        name: part.tool,
+        name: part.name,
       }),
     )
   }
-  if (part.state.status === "pending") {
+  if (part.state.status === "streaming") {
-    toolStates.set(part.callID, part.state.status)
+    toolStates.set(part.id, part.state.status)
     return events
   }
-  if (!previous || previous === "pending") {
+  if (!previous || previous === "streaming") {
     events.push(
       makeEvent("session.tool.input.ended", {
         sessionID,
         assistantMessageID: messageID,
-        id: part.callID,
+        id: part.id,
         text: JSON.stringify(part.state.input),
       }),
       makeEvent("session.tool.called", {
         sessionID,
         assistantMessageID: messageID,
-        id: part.callID,
+        id: part.id,
         input: part.state.input,
         executed: part.executed ?? true,
         state: jsonRecord(part.providerState),
@@ -751,16 +766,20 @@ function toolEvents(part: ToolSeed, messageID: string): readonly OpenCodeEvent[]
     )
   }
   if (part.state.status === "running") {
-    if (previous === "running" || Object.keys(part.state.metadata).length)
+    const metadata = {
+      ...part.state.metadata,
+      ...(part.state.output === undefined ? {} : { output: part.state.output }),
+    }
+    if (previous === "running" || Object.keys(metadata).length)
       events.push(
         makeEvent("session.tool.progress", {
           sessionID,
           assistantMessageID: messageID,
-          id: part.callID,
+          id: part.id,
-          metadata: jsonRecord(part.state.metadata),
+          metadata: jsonRecord(metadata),
         }),
       )
-    toolStates.set(part.callID, part.state.status)
+    toolStates.set(part.id, part.state.status)
     return events
   }
   if (part.state.status === "error") {
@@ -768,28 +787,28 @@ function toolEvents(part: ToolSeed, messageID: string): readonly OpenCodeEvent[]
       makeEvent("session.tool.failed", {
         sessionID,
         assistantMessageID: messageID,
-        id: part.callID,
+        id: part.id,
         error: { type: "ToolError", message: part.state.error },
         metadata: jsonRecord(part.state.metadata),
         executed: part.executed ?? true,
         resultState: jsonRecord(part.providerResultState),
       }),
     )
-    toolStates.set(part.callID, part.state.status)
+    toolStates.set(part.id, part.state.status)
     return events
   }
   events.push(
     makeEvent("session.tool.success", {
       sessionID,
       assistantMessageID: messageID,
-      id: part.callID,
+      id: part.id,
       content: [{ type: "text", text: part.state.output }],
       metadata: jsonRecord(part.state.metadata),
       executed: part.executed ?? true,
       resultState: jsonRecord(part.providerResultState),
     }),
   )
-  toolStates.set(part.callID, part.state.status)
+  toolStates.set(part.id, part.state.status)
   return events
 }
 

packages/app/e2e/performance/timeline-stability/interaction.spec.ts

@@ -1,4 +1,5 @@
 import { expect, test } from "@playwright/test"
+import { createTwoFilesPatch } from "diff"
 import {
   defineVisualRegions,
   reportVisualStability,
@@ -58,14 +59,14 @@ test("expands and collapses a long completed shell without overlap", async ({ pa
   await startVisualProbe(page, regions)
   await trigger.click()
   await expect(trigger).toHaveAttribute("aria-expanded", "true")
-  await page.waitForTimeout(500)
+  await waitForVisualSettle(page, [regions.shell.selector, regions.following.selector])
   const expanded = await stopVisualProbe<keyof typeof regions>(page)
   await reportVisualStability(testInfo, "shell-expand", expanded, plan)
 
   await startVisualProbe(page, regions)
   await trigger.click()
   await expect(trigger).toHaveAttribute("aria-expanded", "false")
-  await page.waitForTimeout(500)
+  await waitForVisualSettle(page, [regions.shell.selector, regions.following.selector])
   const collapsed = await stopVisualProbe<keyof typeof regions>(page)
   await reportVisualStability(testInfo, "shell-collapse", collapsed, plan)
 })
@@ -83,7 +84,7 @@ test("expands and collapses a completed context group without overlap", async ({
     messages: [
       userMessage(),
       assistantMessage([
-        toolPart(ids[0]!, "read", "completed", { filePath: "src/a.ts" }),
+        toolPart(ids[0]!, "read", "completed", { path: "src/a.ts" }),
         toolPart(ids[1]!, "glob", "completed", { path: ".", pattern: "**/*.ts" }),
         toolPart(ids[2]!, "grep", "completed", { path: ".", pattern: "stable" }),
         toolPart(ids[3]!, "list", "completed", { path: "src" }),
@@ -110,7 +111,7 @@ test("expands and collapses a completed context group without overlap", async ({
     await startVisualProbe(page, regions)
     await trigger.click()
     await expect(trigger).toHaveAttribute("aria-expanded", String(expanded))
-    await page.waitForTimeout(500)
+    await waitForVisualSettle(page, [regions.context.selector, regions.following.selector])
     const trace = await stopVisualProbe<keyof typeof regions>(page)
     await reportVisualStability(
       testInfo,
@@ -142,16 +143,23 @@ test("expands and collapses an edit diff without moving twice", async ({ page },
           editID,
           "edit",
           "completed",
-          { filePath: "src/edit.ts" },
+          { path: "src/edit.ts", oldString: "export const value = 1", newString: "export const value = 2" },
           {
             metadata: {
-              filediff: {
+              files: [
-                file: "src/edit.ts",
+                {
-                additions: 40,
+                  file: "src/edit.ts",
-                deletions: 40,
+                  patch: createTwoFilesPatch(
-                before: source(40, false),
+                    "a/src/edit.ts",
-                after: source(40, true),
+                    "b/src/edit.ts",
-              },
+                    source(40, false),
+                    source(40, true),
+                  ),
+                  additions: 40,
+                  deletions: 40,
+                  status: "modified",
+                },
+              ],
             },
           },
         ),
@@ -182,7 +190,7 @@ test("expands and collapses an edit diff without moving twice", async ({ page },
   await startVisualProbe(page, regions)
   await trigger.click()
   await expect(trigger).toHaveAttribute("aria-expanded", "true")
-  await page.waitForTimeout(900)
+  await waitForVisualSettle(page, [regions.edit.selector, regions.following.selector])
   const trace = await stopVisualProbe<keyof typeof regions>(page)
   await reportVisualStability(
     testInfo,

packages/app/e2e/performance/timeline-stability/tool-mutation.spec.ts

@@ -17,32 +17,32 @@ import {
   userMessage,
 } from "./fixture"
 
-test("adds a task child-session link without replacing the task row", async ({ page }, testInfo) => {
+test("adds a subagent child-session link without replacing the row", async ({ page }, testInfo) => {
-  const taskID = "prt_task_link"
+  const taskID = "prt_subagent_link"
-  const childID = "ses_task_child"
+  const childID = "ses_subagent_child"
-  const input = { description: "Inspect child", subagent_type: "explore" }
+  const input = { description: "Inspect child", agent: "explore", prompt: "Inspect the child Session." }
   const timeline = await setupTimeline(page, {
-    messages: [userMessage(), assistantMessage([toolPart(taskID, "task", "running", input)], { completed: false })],
+    messages: [userMessage(), assistantMessage([toolPart(taskID, "subagent", "running", input)], { completed: false })],
     sessions: [session(), session({ id: childID, parentID: sessionID, title: "Inspect child" })],
     cpuRate: 4,
   })
   const regions = defineVisualRegions({
-    task: { selector: `[data-timeline-part-id="${renderedPartID(taskID)}"] [data-slot="collapsible-trigger"]` },
+    subagent: { selector: `[data-timeline-part-id="${renderedPartID(taskID)}"] [data-slot="collapsible-trigger"]` },
   })
   await startVisualProbe(page, regions)
   await timeline.send(
-    partUpdated(toolPart(taskID, "task", "completed", input, { metadata: { sessionId: childID } })),
+    partUpdated(toolPart(taskID, "subagent", "completed", input, { metadata: { sessionID: childID } })),
     500,
   )
   const trace = await stopVisualProbe<keyof typeof regions>(page)
   await reportVisualStability(
     testInfo,
-    "task-link",
+    "subagent-link",
     trace,
     visualPlan(regions, [
-      { type: "required", regions: ["task"] },
+      { type: "required", regions: ["subagent"] },
-      { type: "unique", regions: ["task"] },
+      { type: "unique", regions: ["subagent"] },
-      { type: "stable", regions: ["task"] },
+      { type: "stable", regions: ["subagent"] },
       { type: "opacity", regions: "all" },
       { type: "continuity", regions: "all" },
       { type: "motion", regions: "all", maxPositionReversals: 0 },