fix(tui): keep closed tabs closed

Co-authored-by: Aiden Cline <rekram1-node@users.noreply.github.com>

.opencode/plugins/tui/discovery-smoke.ts

@@ -0,0 +1,13 @@
+import type { Context } from "../../../packages/plugin/src/tui/context"
+
+export default {
+  id: "test.tui-discovery-smoke",
+  setup(_context: Context) {
+    // context.ui.toast.show({
+    //   title: "TUI plugin discovery works",
+    //   message: "Loaded .opencode/plugins/tui/discovery-smoke.ts",
+    //   variant: "success",
+    //   duration: 30_000,
+    // })
+  },
+}

AGENTS.md

@@ -4,6 +4,13 @@
 - The default branch in this repo is `dev`.
 - Local `main` ref may not exist; use `dev` or `origin/dev` for diffs.
 
+## Live V2 TUI Testing
+
+- Run `bun run dev:live` from a development worktree to test its TUI against the currently elected `opencode2` background server and live sessions.
+- Pass a directory after the script when needed, for example `bun run dev:live /path/to/project`.
+- The script discovers the server with `opencode2 service status`, injects its private local credential from `opencode2 service get password`, and uses the `next` TUI storage channel so tabs and other client-local state match the installed client.
+- Prefer `dev:live` over plain `bun run dev` for this workflow. An implicit managed-service connection may replace the live server when the worktree client version differs; explicit `--server` warns and continues without replacing it.
+
 ## Branch Names
 
 Use a short branch name of at most three words, separated by hyphens. Do not use slashes or type prefixes such as `feat/` or `fix/`.
@@ -60,6 +67,7 @@ const { a, b } = obj
 ### Imports
 
 - Never alias imports. Do not use `import { foo as bar } from "..."` or renamed imports like `resolve as pathResolve`.
+- Never use type-position `import("...")` references such as `Schema.declare<import("@opencode-ai/plugin/effect/plugin").Plugin["effect"]>`. Only when two imports genuinely collide on a name and no other option exists, an aliased type import (`import type { Plugin as PluginDefinition } from "..."`) is permitted as a last resort — still strongly preferred not to.
 - Never use star imports. Do not use `import * as Foo from "..."` or `import type * as Foo from "..."`.
 - If a namespace-style value is needed, import the module's own exported namespace by name, for example `import { Project } from "@opencode-ai/core/project"`, then reference `Project.ID`.
 - Prefer dynamic imports for heavy modules that are only needed in selected code paths, especially in startup-sensitive entrypoints. Destructure dynamic import bindings near the top of the narrowest scope that needs them so they read like normal imports. Avoid inline chains such as `await import("./module").then((mod) => mod.value())` or `(await import("./module")).value()`. Keep branch-specific imports inside the branch that needs them to preserve lazy loading.

docs/design/v1-v2-database-migration.md

@@ -0,0 +1,118 @@
+# V1 to V2 Database Migration
+
+## Approach
+
+- Use the `dev` branch database schema and migration registry as the V1 baseline.
+- Remove migrations that exist only on the V2 branch.
+- Generate one canonical migration from the `dev` schema to the final V2 schema.
+- Add explicit data operations to that migration where generated DDL is insufficient.
+- Test the migration against a populated database at the exact `dev` schema.
+
+## Preserve
+
+The canonical V1 data remains in its existing tables. In particular, preserve `session`, `message`, and `part` rows.
+
+Preserve `workspace` rows and existing `session.workspace_id` values unchanged. The migration must not clear or rebuild
+workspace relationships.
+
+Keep the `todo` table and its data unchanged. V2 does not currently migrate todos into another representation, and the
+generated migration must not drop the table.
+
+## Truncate
+
+Truncate these pre-launch V2 tables before applying schema changes:
+
+- `event`
+- `event_sequence`
+- `session_message`
+
+These rows are not canonical V1 data. Truncating `event` before adding the required `event.created` column means the
+column needs neither a backfill nor a default. After truncation, rebuild `session_message` from canonical V1 `message`
+and `part` rows rather than retaining its pre-launch V2 contents.
+
+## Message Backfill
+
+Backfill canonical V1 history from `message` and `part` into `session_message`. This is the main data transformation in
+the migration. Preserving the V1 tables alone keeps the data safe but does not make existing history visible through the
+V2 session APIs, which read `session_message`.
+
+Reuse each V1 `message.id` as the corresponding `session_message.id`. Stable IDs keep the migration deterministic and
+avoid rewriting other persisted state that may refer to a message.
+
+Within each session, order V1 messages by `time_created` and then `id`, matching the existing V1 message index. Assign
+contiguous `session_message.seq` values starting at `0`.
+
+Map ordinary V1 messages one-to-one by role. Each ordinary V1 user message becomes one V2 `user` row, and each ordinary
+V1 assistant message becomes one V2 `assistant` row. Fold the source message's ordered V1 parts into that row's V2
+payload.
+
+Handle semantic marker parts before applying the ordinary mapping. In particular, a V1 user message containing a
+`compaction` part and its paired assistant summary represent one compaction operation, not two ordinary messages. Special
+part mappings must be decided explicitly before implementing the backfill.
+
+V1 synthetic content is represented by user text parts with `synthetic: true`, not by a separate message role. A V1 user
+message whose visible text parts are all synthetic should become a V2 `synthetic` message. If a V1 user message mixes
+ordinary and synthetic content, preserve the ordinary content in the V2 `user` row and emit the synthetic content as an
+adjacent V2 `synthetic` row. Ignore text parts marked `ignored`, matching V1 model-history behavior.
+
+Use the V1 compaction user message ID as the ID of the collapsed V2 compaction message. This matches V2's use of the
+admitted compaction input ID and preserves references to the initiating message.
+
+For a completed compaction, create one V2 `compaction` row with `status: "completed"`. Set `reason` from the V1
+compaction part's `auto` flag, join the paired summary assistant's nonempty text parts with blank lines for `summary`, and
+serialize the retained V1 tail beginning at `tail_start_id` for `recent`. Use an empty `recent` value when no tail was
+retained, and use the compaction user message creation time. Do not emit the paired summary assistant as a separate V2
+assistant row.
+
+After rebuilding `session_message`, seed `event_sequence` with one row per migrated session. Set its watermark to that
+session's maximum backfilled `session_message.seq`. This prevents new V2 events from reusing sequence numbers or sorting
+before migrated history. The `event` table remains empty.
+
+## Drop
+
+Drop these pre-launch V2 tables without preserving or transforming their rows:
+
+- `session_input`
+- `session_context_epoch`
+
+Do not transfer `session_input` rows into `session_pending`.
+
+## Create Empty
+
+Let the generated migration create these tables empty:
+
+- `instruction_blob`
+- `instruction_entry`
+- `instruction_state`
+- `session_pending`
+- `kv`
+
+V1 has no canonical data to backfill into these tables. V2 initializes their state as it runs.
+
+## Fork Storage
+
+V1 has no fork-boundary state to backfill. New V2 forks use a required message boundary and persist it in
+`session.fork_boundary`. The durable fork event contains no parent sequence. Its resolved boundary is one of:
+
+- `before`: copy messages before the identified message.
+- `through`: copy messages through the identified message.
+
+Forking an empty session is not supported. `session.fork_seq` and `session.fork_message_id` are not part of the final V2
+schema.
+
+New nullable session columns, including `fork_session_id`, `fork_boundary`, and `time_suspended`, require no explicit
+backfill. Existing rows naturally receive `NULL` when the generated migration adds the columns.
+
+## Verification
+
+The canonical migration test should seed representative V1 sessions, messages, parts, todos, projects, accounts,
+credentials, permissions, shares, and workspaces. After migration, it should verify:
+
+- Preserved rows and encoded values remain unchanged.
+- Todo rows remain available in the unchanged `todo` table.
+- `event` is empty, and stale pre-launch rows are absent from the rebuilt projections.
+- Backfilled `session_message` rows represent the canonical V1 `message` and `part` history.
+- Each migrated session's `event_sequence` watermark matches its maximum backfilled message sequence.
+- Dropped tables no longer exist.
+- New tables exist and are empty.
+- The final schema has no ungenerated changes.

nix/hashes.json

@@ -1,8 +1,8 @@
 {
   "nodeModules": {
-    "x86_64-linux": "sha256-0kcwV34P2C3yKg2eG9W2nW+OedrSBb+1TdpuUeYtauY=",
-    "aarch64-linux": "sha256-yHVygApQchAB34wrtFR4GU0CkmZOlLsl3wsp15u0xzs=",
-    "aarch64-darwin": "sha256-DyalcwyK2Wn5R6249keFcNVECbgtjYNjscOFqTi88FI=",
-    "x86_64-darwin": "sha256-BkGw0GWN9W9q+/g4FYR0MqxUuFP80BPoERO+ypz/arQ="
+    "x86_64-linux": "sha256-RFek0QoEEjsgbqmTE/SxQAmPtYyzs0IPR2ugFn5Okrs=",
+    "aarch64-linux": "sha256-BmAxapY1YrAFn7mVq3/6A9+6Au5UIvSqBboHMkyJH3I=",
+    "aarch64-darwin": "sha256-Sx3bGWQqLlgoa/RudJxanjSzhFRNklckT2ffnO2I5F4=",
+    "x86_64-darwin": "sha256-CMOhiisHNowg06qadvgg4K+60zrynglwiT0qKYQ4NiA="
   }
 }

package.json

@@ -8,6 +8,7 @@
   "packageManager": "bun@1.3.14",
   "scripts": {
     "dev": "bun run --cwd packages/cli --conditions=browser src/index.ts",
+    "dev:live": "OPENCODE_TUI_CHANNEL=next OPENCODE_PASSWORD=\"$(opencode2 service get password)\" bun run dev --server \"$(opencode2 service status)\"",
     "dev:desktop": "bun --cwd packages/desktop dev",
     "dev:web": "bun --cwd packages/app dev",
     "dev:console": "ulimit -n 10240 2>/dev/null; bun run --cwd packages/console/app dev",

packages/ai/AGENTS.md

@@ -10,7 +10,9 @@
 
 ## Conventions
 
-Per-type constructors live on the type, not as top-level re-exports. Use `Message.system(...)`, `Message.user(...)`, `Message.assistant(...)`, `Message.tool(...)`, `Model.make(...)`, `ToolDefinition.make(...)`, `ToolCallPart.make(...)`, `ToolResultPart.make(...)`, `ToolChoice.make(...)`, `ToolChoice.named(...)`, `SystemPart.make(...)`, and `GenerationOptions.make(...)` directly. The top-level `LLM` namespace is reserved for request-shaped call APIs: `LLM.request`, `LLM.generate`, `LLM.stream`, `LLM.updateRequest`, and `LLM.generateObject`. Two ways to construct the same thing is one too many.
+Per-type constructors live on the type, not as top-level re-exports. Use `Message.system(...)`, `Message.user(...)`, `Message.assistant(...)`, `Message.tool(...)`, `LanguageModel.make(...)`, `ToolDefinition.make(...)`, `ToolCallPart.make(...)`, `ToolResultPart.make(...)`, `ToolChoice.make(...)`, `ToolChoice.named(...)`, `SystemPart.make(...)`, and `GenerationOptions.make(...)` directly. The top-level `LLM` namespace is reserved for request-shaped call APIs: `LLM.request`, `LLM.generate`, `LLM.stream`, `LLM.updateRequest`, and `LLM.generateObject`. Two ways to construct the same thing is one too many.
+
+- Keep provider-defined string enums forward-compatible. Expose known values for autocomplete while accepting future values with `Known | (string & {})`; use `Schema.String` at runtime unless rejecting unknown values is required for correctness.
 
 ## Tests
 
@@ -46,7 +48,7 @@ const response = yield * LLMClient.generate(request)
 
 `LLM.request(...)` builds an `LLMRequest`. `LLMClient.generate(...)` reads the executable route carried by `request.model.route`, builds the provider-native body, asks the route's transport for a real `HttpClientRequest.HttpClientRequest`, sends it through `RequestExecutor.Service`, parses the provider stream into common `LLMEvent`s, and finally returns an `LLMResponse`.
 
-Use `LLMClient.stream(request)` when callers want incremental `LLMEvent`s. Use `LLMClient.generate(request)` when callers want those same events collected into an `LLMResponse`. Use `LLMClient.prepare<Body>(request)` to compile a request through the route pipeline without sending it — the optional `Body` type argument narrows `.body` to the route's native shape (e.g. `prepare<OpenAIChatBody>(...)` returns a `PreparedRequestOf<OpenAIChatBody>`). The runtime body is identical; the generic is a type-level assertion.
+Use `LLMClient.stream(request)` when callers want incremental `LLMEvent`s. Use `LLMClient.generate(request)` when callers want those same events collected into an `LLMResponse`.
 
 Filter or narrow `LLMEvent` streams with `LLMEvent.is.*` (camelCase guards, e.g. `events.filter(LLMEvent.is.toolCall)`). The kebab-case `LLMEvent.guards["tool-call"]` form also works but prefer `is.*` in new code.
 
@@ -74,7 +76,7 @@ export const route = Route.make({
 })
 ```
 
-Route defaults are request-shaping defaults such as `headers`, `limits`, `generation`, `providerOptions`, and `http`. Endpoint host/query belongs on the route endpoint. Selected `Model` values carry only model id, provider id, and the configured route value. Model capability/catalog metadata lives outside this package; protocol support is enforced by request lowering and typed `LLMError`s.
+Route defaults are request-shaping defaults such as `headers`, `limits`, `generation`, `providerOptions`, and `http`. Endpoint host/query belongs on the route endpoint. Selected `LanguageModel` values carry only model id, provider id, and the configured route value. Model capability/catalog metadata lives outside this package; protocol support is enforced by request lowering and typed `AIError`s.
 
 The four-axis decomposition is the reason DeepSeek, TogetherAI, Cerebras, Baseten, Fireworks, and DeepInfra all reuse `OpenAIChat.protocol` verbatim — each provider deployment is a 5-15 line `Route.make(...)` call instead of a 300-400 line route clone. Bug fixes in one protocol propagate to every consumer of that protocol in a single commit.
 
@@ -126,7 +128,7 @@ const selected = model("gpt-5", {
 })
 ```
 
-Keep semantic APIs as separate entrypoints, such as OpenAI `chat` and `responses`. Keep transport choices inside the semantic entrypoint settings, so OpenAI Responses HTTP and WebSocket share one entrypoint. Provider facades may still expose named selectors such as `responsesWebSocket` for direct typed call sites; the package-like contract maps its settings to those selectors before returning an executable `Model`.
+Keep semantic APIs as separate entrypoints, such as OpenAI `chat` and `responses`. Keep transport choices inside the semantic entrypoint settings, so OpenAI Responses HTTP and WebSocket share one entrypoint. Provider facades may still expose named selectors such as `responsesWebSocket` for direct typed call sites; the package-like contract maps its settings to those selectors before returning an executable `LanguageModel`.
 
 Do not expose `Route` in provider package settings. Route composition stays an implementation detail behind `model(...)`.
 
@@ -136,15 +138,15 @@ Do not expose `Route` in provider package settings. Route composition stays an i
 packages/ai/src/
   schema/                   canonical Schema model, split by concern
     ids.ts                  branded IDs, literal types, ProviderMetadata
-    options.ts              Generation/Provider/Http options, Limits, Model, cache policy
+    options.ts              Generation/Provider/Http options, Limits, LanguageModel, cache policy
     messages.ts             content parts, Message, ToolDefinition, LLMRequest
-    events.ts               Usage, individual events, LLMEvent, PreparedRequest, LLMResponse
-    errors.ts               error reasons, LLMError, ToolFailure
+    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.prepare/stream/generate
+    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

packages/ai/DESIGN.md

@@ -96,7 +96,7 @@ contains identity, capabilities, pricing metadata, provider-specific option
 types, reusable request-behavior defaults, and hidden execution behavior.
 
 Normal users do not need to learn the current `Route` composite. Protocol,
-endpoint, auth, transport, and hooks are bound behind `Model`.
+endpoint, auth, transport, and hooks are bound behind `LanguageModel`.
 
 ### Request
 
@@ -539,7 +539,7 @@ Hosted tools do not pretend to have local handlers, and callers do not inspect a
 
 ### Run stream
 
-`LLM.stream` returns an Effect `Stream<RunEvent, LLMError, Requirements>`.
+`LLM.stream` returns an Effect `Stream<RunEvent, AIError, Requirements>`.
 Run events explicitly expose orchestration boundaries:
 
 ```ts
@@ -828,11 +828,11 @@ portable semantic guarantee.
 
 ## Error Model
 
-The Effect error channel is a tagged domain union rather than one `LLMError`
+The Effect error channel is a tagged domain union rather than one `AIError`
 wrapper with nested reasons. Illustrative categories:
 
 ```ts
-type LLMError =
+type AIError =
   | AuthenticationError
   | InvalidRequestError
   | UnsupportedCapabilityError
@@ -1079,7 +1079,7 @@ The redesign intentionally removes or changes these current concepts:
 | `LLM.generate` means one turn           | `LLM.generate` means complete run                           |
 | `LLMClient.generate/stream`             | `LLM.generateTurn/streamTurn` for one turn                  |
 | `LLMClient.layer` requirement           | Standard Effect requirements exposed directly               |
-| Public `Route` mental model             | Hidden behind executable `Model`                            |
+| Public `Route` mental model             | Hidden behind executable `LanguageModel`                    |
 | `Provider.make` structural helper       | Experimental declarative `Provider.define`                  |
 | Schema classes as canonical values      | Plain immutable values plus schema subpath                  |
 | `LLM.updateRequest`                     | Object spread                                               |
@@ -1089,7 +1089,7 @@ The redesign intentionally removes or changes these current concepts:
 | `generateObject`                        | Typed `output` option on `generate`                         |
 | One event union for provider output     | Separate `TurnEvent` and `RunEvent` unions                  |
 | `providerExecuted` dispatch check       | Distinct hosted-tool constructors                           |
-| One wrapped `LLMError`                  | Tagged domain error union                                   |
+| One wrapped `AIError`                   | Tagged domain error union                                   |
 
 OpenCode should migrate to `generateTurn` / `streamTurn`, preserving its durable
 prompt admission, persistence, permission, tool settlement, and continuation

packages/ai/README.md

@@ -195,8 +195,7 @@ The hosted result is represented as a provider-executed tool call and tool resul
 - **`LLM.request({...})`** — build a provider-neutral `LLMRequest`. Accepts ergonomic inputs (`system: string`, `prompt: string`) that normalize into the canonical Schema classes.
 - **`LLM.generate` / `LLM.stream`** — re-exported from `LLMClient` for one-import use.
 - **`Message.user(...)` / `Message.assistant(...)` / `Message.tool(...)`** — message constructors from the canonical schema model.
-- **`Model.make(...)` / `ToolCallPart.make(...)` / `ToolResultPart.make(...)` / `ToolDefinition.make(...)`** — model and tool-related constructors from the canonical schema model.
-- **`LLMClient.prepare(request)`** — compile a request through protocol body construction, validation, and HTTP preparation without sending. Useful for inspection and testing.
+- **`LanguageModel.make(...)` / `ToolCallPart.make(...)` / `ToolResultPart.make(...)` / `ToolDefinition.make(...)`** — model and tool-related constructors from the canonical schema model.
 - **`LLMEvent.is.*`** — typed guards (`is.textDelta`, `is.toolCall`, `is.finish`, …) for filtering streams.
 - **`Image.generate({...})`** — generate images through a provider-neutral image request and response model.
 - **`ImageClient`** — Effect service and layer for image execution, parallel to `LLMClient`.

packages/ai/STATUS.md

@@ -60,7 +60,7 @@ Other `aisdk:` packages, including Google Vertex, Azure, and Bedrock, currently
 | `@ai-sdk/google-vertex/xai`       | Vertex Chat / Responses                                        | Partial / usable | Decide Chat/Responses selection for catalog models, add runner mapping and recorded coverage, and review xAI-specific request options.                                                                              |
 | `@ai-sdk/azure`                   | Azure OpenAI Chat/Responses facade                             | Partial          | Map runner/catalog metadata to native Azure, handle resourceName/baseURL/apiVersion variants, add AAD/token auth story, and verify Chat vs Responses deployment selection.                                          |
 | `@ai-sdk/amazon-bedrock`          | Bedrock Converse                                               | Partial          | Add default AWS credential chain/profile support, region/inference-profile model ID handling, provider option parity via `additionalModelRequestFields`, guardrails/performance config, and runner/catalog mapping. |
-| `@ai-sdk/amazon-bedrock/mantle`   | Bedrock Mantle OpenAI-compatible Chat/Responses namespace      | Missing          | Decide native Mantle shape, likely separate from Converse because it uses OpenAI-compatible Chat/Responses semantics over Bedrock. Add package mapping and tests.                                                   |
+| `@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
 
@@ -71,30 +71,31 @@ Other `aisdk:` packages, including Google Vertex, Azure, and Bedrock, currently
 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; the missing native boundary is Bedrock Mantle.
-9. Recorded coverage is uneven. OpenAI, Anthropic, Gemini, Bedrock Converse, Cloudflare, OpenRouter, and several OpenAI-compatible Chat providers have cassettes. Azure, Vertex, and Mantle need first-class recorded scenarios before switching defaults.
+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/WebSocket selected through settings. |
-| 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                | Missing                                                 | AWS Bedrock Mantle OpenAI-compatible APIs.                                   |
-| Azure OpenAI Chat             | `@opencode-ai/ai/providers/azure/chat`                  | Azure specialization of OpenAI Chat.                                         |
-| Azure OpenAI Responses        | `@opencode-ai/ai/providers/azure/responses`             | Azure specialization of OpenAI Responses.                                    |
+| API slice                     | Package-like entrypoint                                     | Purpose                                                                      |
+| ----------------------------- | ----------------------------------------------------------- | ---------------------------------------------------------------------------- |
+| OpenAI Chat                   | `@opencode-ai/ai/providers/openai/chat`                     | OpenAI `/chat/completions` semantics.                                        |
+| OpenAI Responses              | `@opencode-ai/ai/providers/openai/responses`                | OpenAI `/responses` semantics with HTTP/WebSocket selected through settings. |
+| OpenAI-compatible Chat        | `@opencode-ai/ai/providers/openai-compatible`               | Generic OpenAI-compatible `/chat/completions`.                               |
+| 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
 
@@ -103,6 +104,5 @@ These are implementation/API slices, not separate npm packages.
 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. Add Bedrock Mantle as a separate OpenAI-compatible Bedrock namespace after deciding whether it uses Chat, Responses, or both by model.
-7. Expand typed provider options from the existing V1 lowerer knowledge in `packages/core/src/v1/config/provider-options.ts` before adding more raw overlay examples.
-8. Add recorded provider tests for Azure, Vertex Gemini, Vertex Chat, Vertex Responses, Vertex Messages, Bedrock credential-chain behavior, and Mantle before making native runtime the default for those packages.
+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

@@ -33,7 +33,7 @@ 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 `Model` with a route value, protocol, transport, auth,
+- 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
@@ -137,7 +137,7 @@ starts hiding the real provider-specific config.
    - accepts model id only
    - returns executable models
    - does not accept endpoint/auth/deployment overrides
-4. **Model**
+4. **Language Model**
    - model id
    - route value
    - provider id
@@ -164,7 +164,7 @@ execution mechanism:
 ```ts
 type ProviderFacade<APIs, Config> = {
   readonly id: ProviderID
-  readonly model: (id: string) => Model
+  readonly model: (id: string) => LanguageModel
   readonly configure: (input?: Config) => ProviderFacade<APIs, Config>
 } & APIs
 ```
@@ -181,8 +181,8 @@ export const OpenAI = {
   configure: configureOpenAI,
 } satisfies ProviderFacade<
   {
-    responses: (id: string) => Model
-    chat: (id: string) => Model
+    responses: (id: string) => LanguageModel
+    chat: (id: string) => LanguageModel
   },
   OpenAIConfig
 >
@@ -528,7 +528,7 @@ The chosen split is:
 ```txt
 Route = execution mechanics
 Provider facade = configured route group
-Model = selected executable model carrying route value
+LanguageModel = selected executable model carrying route value
 App boundary = explicit durable-config -> typed-provider call
 ```
 
@@ -549,13 +549,13 @@ App boundary = explicit durable-config -> typed-provider call
   entrypoint maps its scoped `transport` setting before constructing the model.
 - No separate public `LLMClient.layerWithWebSocket`. The runtime should expose one
   client layer with the available transport capabilities.
-- No executable `ModelRef`. The executable handle is `Model`; durable model
+- 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 `Model`.
-- [x] Change `Model.route` to carry a route value, not a `RouteID` string.
+- [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.
@@ -566,9 +566,9 @@ App boundary = explicit durable-config -> typed-provider call
 - [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 `Model`; selected models now carry only
+- [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.prepare` / `stream` / `generate` to read
+- [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.

packages/ai/example/tutorial.ts

@@ -50,18 +50,6 @@ const request = LLM.request({
   },
 })
 
-// `http` is intentionally not needed for normal calls. This shows the shape for
-// newly released provider fields before they deserve a typed provider option.
-const rawOverlayExample = LLM.request({
-  model,
-  prompt: "Show the final HTTP overlay shape.",
-  http: {
-    body: { metadata: { example: "tutorial" } },
-    headers: { "x-opencode-tutorial": "1" },
-    query: { debug: "1" },
-  },
-})
-
 // 3. `generate` sends the request and collects the event stream into one
 // response object. `response.text` is the collected text output.
 const generateOnce = Effect.gen(function* () {
@@ -222,33 +210,15 @@ const FakeEcho = {
   }),
 }
 
-// `LLMClient.prepare` is the lower-level inspection hook: it compiles through
-// body conversion, validation, endpoint, auth, and HTTP construction without
-// sending anything over the network.
-const inspectFakeProvider = Effect.gen(function* () {
-  const prepared = yield* LLMClient.prepare(
-    LLM.request({
-      model: FakeEcho.configure().model("tiny-echo"),
-      prompt: "Show me the provider pipeline.",
-    }),
-  )
-
-  console.log("\n== fake provider prepare ==")
-  console.log("route:", prepared.route)
-  console.log("body:", Formatter.formatJson(prepared.body, { space: 2 }))
-})
-
 // Provide the LLM runtime and the HTTP request executor once. Keep one path
-// enabled at a time so the tutorial can demonstrate generate, prepare, stream,
-// or tool-loop behavior without spending tokens on every example.
+// enabled at a time so the tutorial can demonstrate generate, stream, or
+// tool-loop behavior without spending tokens on every example.
 const requestExecutorLayer = RequestExecutor.fetchLayer
 const llmDeps = Layer.mergeAll(requestExecutorLayer, WebSocketExecutor.layer)
 const llmClientLayer = LLMClient.layer.pipe(Layer.provide(llmDeps))
 
 const program = Effect.gen(function* () {
   // yield* generateOnce
-  // yield* inspectFakeProvider
-  // yield* LLMClient.prepare(rawOverlayExample).pipe(Effect.andThen((prepared) => Effect.sync(() => console.log(prepared.body))))
   // yield* streamText
   // yield* generateStructuredObject
   // yield* generateDynamicObject.pipe(Effect.andThen((response) => Effect.sync(() => console.log(response.object))))

packages/ai/package.json

@@ -15,6 +15,7 @@
   ],
   "exports": {
     ".": "./src/index.ts",
+    "./testing": "./src/testing.ts",
     "./*": "./src/*.ts"
   },
   "devDependencies": {

packages/ai/src/cache-policy.ts

@@ -38,7 +38,7 @@ const resolve = (policy: CachePolicy | undefined): CachePolicyObject => {
 // Protocols whose wire format ignores inline cache markers (OpenAI's implicit
 // prefix caching, Gemini's implicit + out-of-band CachedContent). Skip the
 // whole policy pass for these — emitting hints would be harmless but pointless.
-const RESPECTS_INLINE_HINTS = new Set(["anthropic-messages", "bedrock-converse"])
+const RESPECTS_INLINE_HINTS = new Set(["anthropic-messages", "bedrock-converse", "openrouter"])
 
 const makeHint = (ttlSeconds: number | undefined): CacheHint =>
   ttlSeconds !== undefined ? new CacheHint({ type: "ephemeral", ttlSeconds }) : new CacheHint({ type: "ephemeral" })
@@ -133,6 +133,7 @@ const countHints = (request: LLMRequest) =>
 
 export const applyCachePolicy = (request: LLMRequest): LLMRequest => {
   if (!RESPECTS_INLINE_HINTS.has(request.model.route.id)) return request
+  if (request.model.route.id === "openrouter" && (request.cache === undefined || request.cache === "auto")) return request
   const policy = resolve(request.cache)
   if (!policy.tools && !policy.system && !policy.messages) return request
 

packages/ai/src/image-client.ts

@@ -1,25 +1,25 @@
 import { Context, Effect, Layer } from "effect"
 import { RequestExecutor } from "./route/executor"
 import type { ImageOptions, ImageRequest, ImageRequestFor, ImageResponse } from "./image"
-import type { LLMError } from "./schema"
+import type { AIError } from "./schema"
 
 export type Execute = RequestExecutor.Interface["execute"]
 
 export interface Interface {
   readonly generate: <Options extends ImageOptions>(
     request: ImageRequestFor<Options>,
-  ) => Effect.Effect<ImageResponse, LLMError>
+  ) => Effect.Effect<ImageResponse, AIError>
 }
 
 export class Service extends Context.Service<Service, Interface>()("@opencode/ImageClient") {}
 
 export const generate = <Options extends ImageOptions>(
   request: ImageRequestFor<Options>,
-): Effect.Effect<ImageResponse, LLMError> =>
+): Effect.Effect<ImageResponse, AIError> =>
   Effect.gen(function* () {
     const client = yield* Service
     return yield* client.generate(request)
-  }) as Effect.Effect<ImageResponse, LLMError>
+  }) as Effect.Effect<ImageResponse, AIError>
 
 export const layer: Layer.Layer<Service, never, RequestExecutor.Service> = Layer.effect(
   Service,

packages/ai/src/image.ts

@@ -1,13 +1,10 @@
 import { Effect, Schema } from "effect"
-import { HttpOptions, InvalidRequestReason, LLMError, ModelID, ProviderID, ProviderMetadata, Usage } from "./schema"
+import { HttpOptions, InvalidRequestReason, AIError, ModelID, ProviderID, ProviderMetadata, Usage } from "./schema"
 import { ImageClient, Service, type Execute as ImageExecute } from "./image-client"
 
 export interface ImageRoute<Options extends ImageOptions = ImageOptions> {
   readonly id: string
-  readonly generate: (
-    request: ImageRequestFor<Options>,
-    execute: ImageExecute,
-  ) => Effect.Effect<ImageResponse, LLMError>
+  readonly generate: (request: ImageRequestFor<Options>, execute: ImageExecute) => Effect.Effect<ImageResponse, AIError>
 }
 
 export type ImageOptions = Record<string, unknown>
@@ -146,13 +143,13 @@ export function request(input: ImageRequest | ImageRequestInput) {
 
 export function generate<const Model extends object>(
   input: ImageRequestInput<Model>,
-): Effect.Effect<ImageResponse, LLMError, Service>
-export function generate(input: ImageRequest): Effect.Effect<ImageResponse, LLMError, Service>
+): Effect.Effect<ImageResponse, AIError, Service>
+export function generate(input: ImageRequest): Effect.Effect<ImageResponse, AIError, Service>
 export function generate(input: ImageRequest | ImageRequestInput) {
   return Effect.try({
     try: () => (input instanceof ImageRequest ? input : request(input)),
     catch: (error) =>
-      new LLMError({
+      new AIError({
         module: "Image",
         method: "generate",
         reason: new InvalidRequestReason({ message: error instanceof Error ? error.message : String(error) }),

packages/ai/src/index.ts

@@ -5,8 +5,8 @@ export { Provider } from "./provider"
 export { ProviderPackage } from "./provider-package"
 export { isContextOverflow, isContextOverflowFailure } from "./provider-error"
 export type {
-  RouteModelInput,
-  RouteRoutedModelInput,
+  RouteLanguageModelInput,
+  RouteRoutedLanguageModelInput,
   Interface as LLMClientShape,
   Service as LLMClientService,
 } from "./route/client"
@@ -33,7 +33,7 @@ export type {
 export * as LLM from "./llm"
 export type {
   Definition as ProviderDefinition,
-  ModelFactory as ProviderModelFactory,
-  ModelOptions as ProviderModelOptions,
+  LanguageModelFactory as ProviderLanguageModelFactory,
+  LanguageModelOptions as ProviderLanguageModelOptions,
 } from "./provider"
 export type { Definition as ProviderPackageDefinition, Settings as ProviderPackageSettings } from "./provider-package"

packages/ai/src/llm.ts

@@ -4,30 +4,33 @@ import {
   GenerationOptions,
   HttpOptions,
   InvalidProviderOutputReason,
-  LLMError,
+  AIError,
   LLMEvent,
   LLMRequest,
   LLMResponse,
   Message,
+  LanguageModel,
   SystemPart,
   ToolChoice,
   ToolDefinition,
   type ContentPart,
+  type LanguageModelProviderOptions,
 } from "./schema"
 import { make as makeTool, toDefinitions, type ToolSchema } from "./tool"
 
 /** Input accepted by `LLM.request`, normalized into the canonical `LLMRequest` class. */
-export type RequestInput = Omit<
+export type RequestInput<SelectedLanguageModel extends LanguageModel = LanguageModel> = Omit<
   ConstructorParameters<typeof LLMRequest>[0],
-  "system" | "messages" | "tools" | "toolChoice" | "generation" | "http" | "providerOptions"
+  "model" | "system" | "messages" | "tools" | "toolChoice" | "generation" | "http" | "providerOptions"
 > & {
+  readonly model: SelectedLanguageModel
   readonly system?: string | SystemPart | ReadonlyArray<SystemPart>
   readonly prompt?: string | ContentPart | ReadonlyArray<ContentPart>
   readonly messages?: ReadonlyArray<Message | Message.Input>
   readonly tools?: ReadonlyArray<ToolDefinition.Input>
   readonly toolChoice?: ToolChoice.Input
   readonly generation?: GenerationOptions.Input
-  readonly providerOptions?: ConstructorParameters<typeof LLMRequest>[0]["providerOptions"]
+  readonly providerOptions?: NoInfer<LanguageModelProviderOptions<SelectedLanguageModel>>
   readonly http?: HttpOptions.Input
 }
 
@@ -35,7 +38,9 @@ export const generate = LLMClient.generate
 
 export const stream = LLMClient.stream
 
-export const request = (input: RequestInput) => {
+export const request = <const SelectedLanguageModel extends LanguageModel>(
+  input: RequestInput<SelectedLanguageModel>,
+) => {
   const {
     system: requestSystem,
     prompt,
@@ -63,7 +68,10 @@ const GENERATE_OBJECT_TOOL_NAME = "generate_object"
 
 const GENERATE_OBJECT_TOOL_DESCRIPTION = "Return the structured result by calling this tool."
 
-type GenerateObjectBase = Omit<RequestInput, "tools" | "toolChoice">
+type GenerateObjectBase<SelectedLanguageModel extends LanguageModel = LanguageModel> = Omit<
+  RequestInput<SelectedLanguageModel>,
+  "tools" | "toolChoice"
+>
 
 export class GenerateObjectResponse<T> {
   constructor(
@@ -80,11 +88,15 @@ export class GenerateObjectResponse<T> {
   }
 }
 
-export interface GenerateObjectOptions<S extends ToolSchema<any>> extends GenerateObjectBase {
+export interface GenerateObjectOptions<
+  S extends ToolSchema<any>,
+  SelectedLanguageModel extends LanguageModel = LanguageModel,
+> extends GenerateObjectBase<SelectedLanguageModel> {
   readonly schema: S
 }
 
-export interface GenerateObjectDynamicOptions extends GenerateObjectBase {
+export interface GenerateObjectDynamicOptions<SelectedLanguageModel extends LanguageModel = LanguageModel>
+  extends GenerateObjectBase<SelectedLanguageModel> {
   /** Raw JSON Schema object describing the expected output shape. */
   readonly jsonSchema: JsonSchema.JsonSchema
 }
@@ -103,7 +115,7 @@ const runGenerateObject = Effect.fn("LLM.generateObject")(function* (
     (event) => LLMEvent.is.toolCall(event) && event.name === GENERATE_OBJECT_TOOL_NAME,
   )
   if (!call || !LLMEvent.is.toolCall(call))
-    return yield* new LLMError({
+    return yield* new AIError({
       module: "LLM",
       method: "generateObject",
       reason: new InvalidProviderOutputReason({
@@ -113,7 +125,7 @@ const runGenerateObject = Effect.fn("LLM.generateObject")(function* (
   const object = yield* tool._decode(call.input).pipe(
     Effect.mapError(
       (error) =>
-        new LLMError({
+        new AIError({
           module: "LLM",
           method: "generateObject",
           reason: new InvalidProviderOutputReason({
@@ -133,16 +145,16 @@ const runGenerateObject = Effect.fn("LLM.generateObject")(function* (
  * Two input modes:
  *
  * 1. `schema: EffectSchema<T>` — `.object` is decoded and typed as `T`.
- *    Decode failures surface as `LLMError`.
+ *    Decode failures surface as `AIError`.
  * 2. `jsonSchema: JsonSchema.JsonSchema` — `.object` is `unknown`. Use when
  *    the schema is only available at runtime (MCP, plugin manifests). Caller validates.
  */
-export function generateObject<S extends ToolSchema<any>>(
-  options: GenerateObjectOptions<S>,
-): Effect.Effect<GenerateObjectResponse<Schema.Schema.Type<S>>, LLMError>
-export function generateObject(
-  options: GenerateObjectDynamicOptions,
-): Effect.Effect<GenerateObjectResponse<unknown>, LLMError>
+export function generateObject<const SelectedLanguageModel extends LanguageModel, S extends ToolSchema<any>>(
+  options: GenerateObjectOptions<S, SelectedLanguageModel>,
+): Effect.Effect<GenerateObjectResponse<Schema.Schema.Type<S>>, AIError>
+export function generateObject<const SelectedLanguageModel extends LanguageModel>(
+  options: GenerateObjectDynamicOptions<SelectedLanguageModel>,
+): Effect.Effect<GenerateObjectResponse<unknown>, AIError>
 export function generateObject(options: GenerateObjectOptions<ToolSchema<any>> | GenerateObjectDynamicOptions) {
   if ("schema" in options) {
     const { schema, ...rest } = options

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

@@ -6,7 +6,7 @@ import { Endpoint } from "../route/endpoint"
 import { Framing } from "../route/framing"
 import { Protocol } from "../route/protocol"
 import {
-  LLMError,
+  AIError,
   LLMEvent,
   mergeJsonRecords,
   Usage,
@@ -242,16 +242,14 @@ const AnthropicUsage = Schema.StructWithRest(
     cache_creation_input_tokens: optionalNull(Schema.Number),
     cache_read_input_tokens: optionalNull(Schema.Number),
     server_tool_use: optionalNull(
-      Schema.StructWithRest(
-        Schema.Struct({ web_search_requests: Schema.optional(Schema.Number) }),
-        [Schema.Record(Schema.String, Schema.Unknown)],
-      ),
+      Schema.StructWithRest(Schema.Struct({ web_search_requests: Schema.optional(Schema.Number) }), [
+        Schema.Record(Schema.String, Schema.Unknown),
+      ]),
     ),
     output_tokens_details: optionalNull(
-      Schema.StructWithRest(
-        Schema.Struct({ thinking_tokens: Schema.optional(Schema.Number) }),
-        [Schema.Record(Schema.String, Schema.Unknown)],
-      ),
+      Schema.StructWithRest(Schema.Struct({ thinking_tokens: Schema.optional(Schema.Number) }), [
+        Schema.Record(Schema.String, Schema.Unknown),
+      ]),
     ),
   }),
   [Schema.Record(Schema.String, Schema.Unknown)],
@@ -725,8 +723,7 @@ const mergeUsage = (left: Usage | undefined, right: Usage | undefined) => {
     reasoningTokens,
     totalTokens: ProviderShared.totalTokens(inputTokens, outputTokens, undefined),
     providerMetadata: {
-      anthropic:
-        mergeJsonRecords(left.providerMetadata?.["anthropic"], right.providerMetadata?.["anthropic"]) ?? {},
+      anthropic: mergeJsonRecords(left.providerMetadata?.["anthropic"], right.providerMetadata?.["anthropic"]) ?? {},
     },
   })
 }
@@ -816,7 +813,8 @@ const onContentBlockStart = (state: ParserState, event: AnthropicEvent): StepRes
   if (block.type === "thinking" && block.thinking !== undefined) {
     const events: LLMEvent[] = []
     const id = `reasoning-${event.index ?? 0}`
-    const providerMetadata = block.signature === undefined ? undefined : anthropicMetadata({ signature: block.signature })
+    const providerMetadata =
+      block.signature === undefined ? undefined : anthropicMetadata({ signature: block.signature })
     const lifecycle = Lifecycle.reasoningStart(state.lifecycle, events, id, providerMetadata)
     return [
       {
@@ -980,7 +978,7 @@ const providerErrorMessage = (event: AnthropicEvent): string => {
 }
 
 const onError = (event: AnthropicEvent) =>
-  new LLMError({
+  new AIError({
     module: ADAPTER,
     method: "stream",
     reason: classifyProviderFailure({ message: providerErrorMessage(event), code: event.error?.type }),

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

@@ -3,7 +3,7 @@ import { Route } from "../route/client"
 import { Endpoint } from "../route/endpoint"
 import { Protocol } from "../route/protocol"
 import {
-  LLMError,
+  AIError,
   LLMEvent,
   Usage,
   type CacheHint,
@@ -11,7 +11,7 @@ import {
   type FinishReasonDetails,
   type JsonSchema,
   type LLMRequest,
-  type ModelToolSchemaCompatibility,
+  type LanguageModelToolSchemaCompatibility,
   type ProviderMetadata,
   type ReasoningPart,
   type ToolCallPart,
@@ -232,7 +232,7 @@ const lowerToolSpec = (tool: ToolDefinition, inputSchema: JsonSchema): BedrockTo
 })
 
 const lowerTools = (
-  compatibility: ModelToolSchemaCompatibility | undefined,
+  compatibility: LanguageModelToolSchemaCompatibility | undefined,
   breakpoints: BedrockCache.Breakpoints,
   tools: ReadonlyArray<ToolDefinition>,
 ): BedrockTool[] => {
@@ -274,9 +274,7 @@ const reasoningSignature = (part: ReasoningPart) => {
 
 const reasoningRedactedData = (part: ReasoningPart) => {
   const bedrock = part.providerMetadata?.bedrock
-  return ProviderShared.isRecord(bedrock) && typeof bedrock.redactedData === "string"
-    ? bedrock.redactedData
-    : undefined
+  return ProviderShared.isRecord(bedrock) && typeof bedrock.redactedData === "string" ? bedrock.redactedData : undefined
 }
 
 const lowerToolCall = (part: ToolCallPart): BedrockToolUseBlock => ({
@@ -656,7 +654,7 @@ const step = (state: ParserState, event: BedrockEvent) =>
       ] as const
     ).find((entry) => entry[1] !== undefined)
     if (exception) {
-      return yield* new LLMError({
+      return yield* new AIError({
         module: ADAPTER,
         method: "stream",
         reason: classifyProviderFailure({

packages/ai/src/protocols/gemini.ts

@@ -29,9 +29,30 @@ export const DEFAULT_BASE_URL = "https://generativelanguage.googleapis.com/v1bet
 
 export interface OptionsInput {
   readonly [key: string]: unknown
+  readonly cachedContent?: string
+  readonly safetySettings?: ReadonlyArray<{
+    readonly category:
+      | "HARM_CATEGORY_UNSPECIFIED"
+      | "HARM_CATEGORY_HATE_SPEECH"
+      | "HARM_CATEGORY_DANGEROUS_CONTENT"
+      | "HARM_CATEGORY_HARASSMENT"
+      | "HARM_CATEGORY_SEXUALLY_EXPLICIT"
+      | "HARM_CATEGORY_CIVIC_INTEGRITY"
+      | (string & {})
+    readonly threshold:
+      | "HARM_BLOCK_THRESHOLD_UNSPECIFIED"
+      | "BLOCK_LOW_AND_ABOVE"
+      | "BLOCK_MEDIUM_AND_ABOVE"
+      | "BLOCK_ONLY_HIGH"
+      | "BLOCK_NONE"
+      | "OFF"
+      | (string & {})
+  }>
+  readonly serviceTier?: "standard" | "flex" | "priority" | (string & {})
   readonly thinkingConfig?: {
     readonly thinkingBudget?: number
     readonly includeThoughts?: boolean
+    readonly thinkingLevel?: "minimal" | "low" | "medium" | "high" | (string & {})
   }
 }
 
@@ -111,6 +132,12 @@ const GeminiToolConfig = Schema.Struct({
 const GeminiThinkingConfig = Schema.Struct({
   thinkingBudget: Schema.optional(Schema.Number),
   includeThoughts: Schema.optional(Schema.Boolean),
+  thinkingLevel: Schema.optional(Schema.String),
+})
+
+const GeminiSafetySetting = Schema.Struct({
+  category: Schema.String,
+  threshold: Schema.String,
 })
 
 const GeminiGenerationConfig = Schema.Struct({
@@ -123,7 +150,10 @@ const GeminiGenerationConfig = Schema.Struct({
 })
 
 const GeminiBodyFields = {
+  cachedContent: Schema.optional(Schema.String),
   contents: Schema.Array(GeminiContent),
+  safetySettings: optionalArray(GeminiSafetySetting),
+  serviceTier: Schema.optional(Schema.String),
   systemInstruction: Schema.optional(GeminiSystemInstruction),
   tools: optionalArray(GeminiTool),
   toolConfig: Schema.optional(GeminiToolConfig),
@@ -316,17 +346,38 @@ const lowerMessages = Effect.fn("Gemini.lowerMessages")(function* (request: LLMR
 })
 
 const resolveOptions = (request: LLMRequest) => {
-  const value = request.providerOptions?.gemini?.thinkingConfig
-  if (!ProviderShared.isRecord(value)) return {}
+  const input = request.providerOptions?.gemini
+  const value = input?.thinkingConfig
   const thinkingConfig = {
-    thinkingBudget: typeof value.thinkingBudget === "number" ? value.thinkingBudget : undefined,
-    includeThoughts: typeof value.includeThoughts === "boolean" ? value.includeThoughts : undefined,
+    thinkingBudget:
+      ProviderShared.isRecord(value) && typeof value.thinkingBudget === "number" ? value.thinkingBudget : undefined,
+    includeThoughts:
+      ProviderShared.isRecord(value) && typeof value.includeThoughts === "boolean"
+        ? value.includeThoughts
+        : ProviderShared.isRecord(value)
+          ? true
+          : undefined,
+    thinkingLevel:
+      ProviderShared.isRecord(value) && typeof value.thinkingLevel === "string" ? value.thinkingLevel : undefined,
   }
   return {
+    cachedContent: typeof input?.cachedContent === "string" ? input.cachedContent : undefined,
+    safetySettings: mapSafetySettings(input?.safetySettings),
+    serviceTier: typeof input?.serviceTier === "string" ? input.serviceTier : undefined,
     thinkingConfig: Object.values(thinkingConfig).some((item) => item !== undefined) ? thinkingConfig : undefined,
   }
 }
 
+function mapSafetySettings(value: unknown) {
+  if (!Array.isArray(value)) return undefined
+  const settings = value.flatMap((item) =>
+    ProviderShared.isRecord(item) && typeof item.category === "string" && typeof item.threshold === "string"
+      ? [{ category: item.category, threshold: item.threshold }]
+      : [],
+  )
+  return settings
+}
+
 const fromRequest = Effect.fn("Gemini.fromRequest")(function* (request: LLMRequest) {
   const hasTools = request.tools.length > 0
   const generation = request.generation
@@ -342,7 +393,10 @@ const fromRequest = Effect.fn("Gemini.fromRequest")(function* (request: LLMReque
   }
 
   return {
+    cachedContent: options.cachedContent,
     contents: yield* lowerMessages(request),
+    safetySettings: options.safetySettings,
+    serviceTier: options.serviceTier,
     systemInstruction:
       request.system.length === 0 ? undefined : { parts: [{ text: ProviderShared.joinText(request.system) }] },
     tools: hasTools

packages/ai/src/protocols/google-images.ts

@@ -11,7 +11,7 @@ import {
 import { Auth, type Definition as AuthDefinition } from "../route/auth"
 import {
   InvalidProviderOutputReason,
-  LLMError,
+  AIError,
   Usage,
   mergeHttpOptions,
   mergeJsonRecords,
@@ -125,7 +125,7 @@ const nativeOptions = (options: GoogleImageOptions | undefined) => {
 }
 
 const invalidOutput = (message: string, providerMetadata?: ProviderMetadata) =>
-  new LLMError({
+  new AIError({
     module: ADAPTER,
     method: "generate",
     reason: new InvalidProviderOutputReason({ message, route: ADAPTER, providerMetadata }),
@@ -285,7 +285,7 @@ export const model = (input: ModelInput) => {
   return ImageModel.make<GoogleImageOptions>({ id: input.id, provider: "google", route, http: input.http })
 }
 
-const googleImagePart = (image: ImageInput): Effect.Effect<Record<string, unknown>, LLMError> => {
+const googleImagePart = (image: ImageInput): Effect.Effect<Record<string, unknown>, AIError> => {
   if (image.type === "bytes")
     return Effect.succeed({ inlineData: { mimeType: image.mediaType, data: Encoding.encodeBase64(image.data) } })
   if (image.type === "file-uri") return Effect.succeed({ fileData: { mimeType: image.mediaType, fileUri: image.uri } })

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

@@ -3,7 +3,7 @@ import type { Content } from "@opencode-ai/schema/tool"
 import { HttpTransport } from "../route/transport"
 import { Protocol } from "../route/protocol"
 import {
-  LLMError,
+  AIError,
   LLMEvent,
   Usage,
   type FinishReason,
@@ -434,10 +434,7 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
         const groups = content.reduce<Array<{ phase: MessagePhase | null | undefined; parts: TextPart[] }>>(
           (groups, part) => {
             const metadata = part.providerMetadata?.[providerMetadataKey]
-            const phase =
-              ProviderShared.isRecord(metadata)
-                ? messagePhase(metadata.phase, extension)
-                : undefined
+            const phase = ProviderShared.isRecord(metadata) ? messagePhase(metadata.phase, extension) : undefined
             const group = groups.at(-1)
             if (group && group.phase === phase) group.parts.push(part)
             else groups.push({ phase, parts: [part] })
@@ -646,10 +643,7 @@ const onOutputTextDelta = (state: ParserState, event: Event, id: string): StepRe
   const phase = state.messagePhases[id]
   const metadata = phase === undefined ? undefined : providerMetadata(state, { phase })
   const lifecycle = Lifecycle.textStart(state.lifecycle, events, id, metadata)
-  return [
-    { ...state, lifecycle: Lifecycle.textDelta(lifecycle, events, id, event.delta) },
-    events,
-  ]
+  return [{ ...state, lifecycle: Lifecycle.textDelta(lifecycle, events, id, event.delta) }, events]
 }
 
 const onOutputTextDone = (state: ParserState, event: Event, id: string): StepResult => {
@@ -975,7 +969,7 @@ const providerErrorMessage = (event: Event, fallback: string): string => {
 const providerError = (state: ParserState, event: Event, fallback: string) => {
   const code = event.code || event.error?.code || event.response?.error?.code || undefined
   const message = providerErrorMessage(event, fallback)
-  return new LLMError({
+  return new AIError({
     module: state.id,
     method: "stream",
     reason: classifyProviderFailure({ message, code }),

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

@@ -6,11 +6,12 @@ import { Endpoint } from "../route/endpoint"
 import { HttpTransport } from "../route/transport"
 import { Protocol } from "../route/protocol"
 import {
-  LLMError,
+  AIError,
   LLMEvent,
   Usage,
   type FinishReason,
   type FinishReasonDetails,
+  type CacheHint,
   type JsonSchema,
   type LLMRequest,
   type MediaPart,
@@ -38,6 +39,11 @@ export const PATH = "/chat/completions"
 // The body schema is the provider-native JSON body. `fromRequest` below builds
 // this shape from the common `LLMRequest`, then `Route.make` validates and
 // JSON-encodes it before transport.
+const OpenAIChatCacheControl = Schema.Struct({
+  type: Schema.Literal("ephemeral"),
+  ttl: Schema.optional(Schema.String),
+})
+
 const OpenAIChatFunction = Schema.Struct({
   name: Schema.String,
   description: Schema.String,
@@ -47,6 +53,7 @@ const OpenAIChatFunction = Schema.Struct({
 const OpenAIChatTool = Schema.Struct({
   type: Schema.tag("function"),
   function: OpenAIChatFunction,
+  cache_control: Schema.optional(OpenAIChatCacheControl),
 })
 type OpenAIChatTool = Schema.Schema.Type<typeof OpenAIChatTool>
 
@@ -61,7 +68,11 @@ const OpenAIChatAssistantToolCall = Schema.Struct({
 type OpenAIChatAssistantToolCall = Schema.Schema.Type<typeof OpenAIChatAssistantToolCall>
 
 const OpenAIChatUserContent = Schema.Union([
-  Schema.Struct({ type: Schema.Literal("text"), text: Schema.String }),
+  Schema.Struct({
+    type: Schema.Literal("text"),
+    text: Schema.String,
+    cache_control: Schema.optional(OpenAIChatCacheControl),
+  }),
   Schema.Struct({
     type: Schema.Literal("image_url"),
     image_url: Schema.Struct({ url: Schema.String }),
@@ -69,7 +80,10 @@ const OpenAIChatUserContent = Schema.Union([
 ])
 
 const OpenAIChatMessage = Schema.Union([
-  Schema.Struct({ role: Schema.Literal("system"), content: Schema.String }),
+  Schema.Struct({
+    role: Schema.Literal("system"),
+    content: Schema.Union([Schema.String, Schema.Array(OpenAIChatUserContent)]),
+  }),
   Schema.Struct({
     role: Schema.Literal("user"),
     content: Schema.Union([Schema.String, Schema.Array(OpenAIChatUserContent)]),
@@ -83,10 +97,16 @@ const OpenAIChatMessage = Schema.Union([
       reasoning: Schema.optional(Schema.String),
       reasoning_text: Schema.optional(Schema.String),
       reasoning_details: Schema.optional(Schema.Unknown),
+      cache_control: Schema.optional(OpenAIChatCacheControl),
     }),
     [Schema.Record(Schema.String, Schema.Unknown)],
   ),
-  Schema.Struct({ role: Schema.Literal("tool"), tool_call_id: Schema.String, content: Schema.String }),
+  Schema.Struct({
+    role: Schema.Literal("tool"),
+    tool_call_id: Schema.String,
+    content: Schema.String,
+    cache_control: Schema.optional(OpenAIChatCacheControl),
+  }),
 ]).pipe(Schema.toTaggedUnion("role"))
 type OpenAIChatMessage = Schema.Schema.Type<typeof OpenAIChatMessage>
 
@@ -107,6 +127,7 @@ export const bodyFields = {
   stream_options: Schema.optional(Schema.Struct({ include_usage: Schema.Boolean })),
   store: Schema.optional(Schema.Boolean),
   reasoning_effort: Schema.optional(OpenAIOptions.OpenAIReasoningEffort),
+  max_completion_tokens: Schema.optional(Schema.Number),
   max_tokens: Schema.optional(Schema.Number),
   temperature: Schema.optional(Schema.Number),
   top_p: Schema.optional(Schema.Number),
@@ -209,13 +230,20 @@ export interface ParserState {
 // Lowering is the only place that knows how common LLM messages map onto the
 // OpenAI Chat wire format. Keep provider quirks here instead of leaking native
 // fields into `LLMRequest`.
-const lowerTool = (tool: ToolDefinition, inputSchema: JsonSchema): OpenAIChatTool => ({
+interface LoweringOptions {
+  readonly cacheControl?: (
+    cache: CacheHint | undefined,
+  ) => Schema.Schema.Type<typeof OpenAIChatCacheControl> | undefined
+}
+
+const lowerTool = (tool: ToolDefinition, inputSchema: JsonSchema, options: LoweringOptions): OpenAIChatTool => ({
   type: "function",
   function: {
     name: tool.name,
     description: tool.description,
     parameters: ToolSchemaProjection.openAI(inputSchema),
   },
+  cache_control: options.cacheControl?.(tool.cache),
 })
 
 const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>) =>
@@ -257,11 +285,14 @@ const reasoningDetails = (parts: ReadonlyArray<ReasoningPart>, native: unknown)
   if (isRecord(native) && Array.isArray(native.reasoning_details)) return native.reasoning_details
 }
 
-const lowerUserMessage = Effect.fn("OpenAIChat.lowerUserMessage")(function* (message: OpenAIChatRequestMessage) {
+const lowerUserMessage = Effect.fn("OpenAIChat.lowerUserMessage")(function* (
+  message: OpenAIChatRequestMessage,
+  options: LoweringOptions,
+) {
   const content: Array<Schema.Schema.Type<typeof OpenAIChatUserContent>> = []
   for (const part of message.content) {
     if (part.type === "text") {
-      content.push({ type: "text", text: part.text })
+      content.push({ type: "text", text: part.text, cache_control: options.cacheControl?.(part.cache) })
       continue
     }
     if (part.type === "media") {
@@ -270,14 +301,18 @@ const lowerUserMessage = Effect.fn("OpenAIChat.lowerUserMessage")(function* (mes
     }
     return yield* ProviderShared.unsupportedContent("OpenAI Chat", "user", ["text", "media"])
   }
-  if (content.every((part) => part.type === "text"))
-    return { role: "user" as const, content: content.map((part) => part.text).join("") }
+  if (content.every((part) => part.type === "text" && part.cache_control === undefined))
+    return {
+      role: "user" as const,
+      content: content.map((part) => (part.type === "text" ? part.text : "")).join(""),
+    }
   return { role: "user" as const, content }
 })
 
 const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(function* (
   message: OpenAIChatRequestMessage,
   configuredField?: string,
+  options: LoweringOptions = {},
 ) {
   const content: TextPart[] = []
   const reasoning: ReasoningPart[] = []
@@ -315,29 +350,44 @@ const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(func
     if (reasoning.length === 0) return nativeReasoning
     return text
   })()
+  const cached = message.content.findLast((part) => "cache" in part && part.cache !== undefined)
   const result = {
     role: "assistant" as const,
     content: content.length === 0 ? null : ProviderShared.joinText(content),
     tool_calls: toolCalls.length === 0 ? undefined : toolCalls,
     reasoning_details: details,
+    cache_control: options.cacheControl?.(cached && "cache" in cached ? cached.cache : undefined),
   }
   if (field === undefined || reasoningText === undefined) return result
   return { ...result, [field]: reasoningText }
 })
 
-const lowerToolMessages = Effect.fn("OpenAIChat.lowerToolMessages")(function* (message: OpenAIChatRequestMessage) {
+const lowerToolMessages = Effect.fn("OpenAIChat.lowerToolMessages")(function* (
+  message: OpenAIChatRequestMessage,
+  options: LoweringOptions,
+) {
   const messages: OpenAIChatMessage[] = []
   const images: Array<Schema.Schema.Type<typeof OpenAIChatUserContent>> = []
   for (const part of message.content) {
     if (!ProviderShared.supportsContent(part, ["tool-result"]))
       return yield* ProviderShared.unsupportedContent("OpenAI Chat", "tool", ["tool-result"])
     if (part.result.type !== "content") {
-      messages.push({ role: "tool", tool_call_id: part.id, content: ProviderShared.toolResultText(part) })
+      messages.push({
+        role: "tool",
+        tool_call_id: part.id,
+        content: ProviderShared.toolResultText(part),
+        cache_control: options.cacheControl?.(part.cache),
+      })
       continue
     }
     const content: ReadonlyArray<Tool.Content> = part.result.value
     const text = content.filter((item) => item.type === "text").map((item) => item.text)
-    messages.push({ role: "tool", tool_call_id: part.id, content: text.join("\n") })
+    messages.push({
+      role: "tool",
+      tool_call_id: part.id,
+      content: text.join("\n"),
+      cache_control: options.cacheControl?.(part.cache),
+    })
     const files = content.filter((item) => item.type === "file")
     images.push(
       ...(yield* Effect.forEach(files, (item) =>
@@ -351,15 +401,29 @@ const lowerToolMessages = Effect.fn("OpenAIChat.lowerToolMessages")(function* (m
 const lowerMessage = Effect.fn("OpenAIChat.lowerMessage")(function* (
   message: OpenAIChatRequestMessage,
   reasoningField?: string,
+  options: LoweringOptions = {},
 ) {
-  if (message.role === "user") return [yield* lowerUserMessage(message)]
-  if (message.role === "assistant") return [yield* lowerAssistantMessage(message, reasoningField)]
-  return (yield* lowerToolMessages(message)).messages
+  if (message.role === "user") return [yield* lowerUserMessage(message, options)]
+  if (message.role === "assistant") return [yield* lowerAssistantMessage(message, reasoningField, options)]
+  return (yield* lowerToolMessages(message, options)).messages
 })
 
-const lowerMessages = Effect.fn("OpenAIChat.lowerMessages")(function* (request: LLMRequest) {
+const lowerMessages = Effect.fn("OpenAIChat.lowerMessages")(function* (request: LLMRequest, options: LoweringOptions) {
   const system: OpenAIChatMessage[] =
-    request.system.length === 0 ? [] : [{ role: "system", content: ProviderShared.joinText(request.system) }]
+    request.system.length === 0
+      ? []
+      : request.system.some((part) => part.cache !== undefined) && options.cacheControl !== undefined
+        ? [
+            {
+              role: "system",
+              content: request.system.map((part) => ({
+                type: "text",
+                text: part.text,
+                cache_control: options.cacheControl?.(part.cache),
+              })),
+            },
+          ]
+        : [{ role: "system", content: ProviderShared.joinText(request.system) }]
   const messages = [...system]
   const pendingImages: Array<Schema.Schema.Type<typeof OpenAIChatUserContent>> = []
   const flushImages = () => {
@@ -370,28 +434,53 @@ const lowerMessages = Effect.fn("OpenAIChat.lowerMessages")(function* (request:
     if (message.role === "system") {
       const part = yield* ProviderShared.wrappedSystemUpdate("OpenAI Chat", message)
       if (pendingImages.length > 0) {
-        messages.push({ role: "user", content: [...pendingImages.splice(0), { type: "text", text: part.text }] })
+        messages.push({
+          role: "user",
+          content: [
+            ...pendingImages.splice(0),
+            { type: "text", text: part.text, cache_control: options.cacheControl?.(part.cache) },
+          ],
+        })
         continue
       }
       const previous = messages.at(-1)
       if (previous?.role === "user" && typeof previous.content === "string")
-        messages[messages.length - 1] = { role: "user", content: `${previous.content}\n${part.text}` }
+        messages[messages.length - 1] = options.cacheControl?.(part.cache)
+          ? {
+              role: "user",
+              content: [
+                { type: "text", text: previous.content },
+                { type: "text", text: part.text, cache_control: options.cacheControl(part.cache) },
+              ],
+            }
+          : { role: "user", content: `${previous.content}\n${part.text}` }
       else if (previous?.role === "user" && Array.isArray(previous.content))
         messages[messages.length - 1] = {
           role: "user",
-          content: [...previous.content, { type: "text", text: part.text }],
+          content: [
+            ...previous.content,
+            { type: "text", text: part.text, cache_control: options.cacheControl?.(part.cache) },
+          ],
         }
-      else messages.push({ role: "user", content: part.text })
+      else
+        messages.push(
+          options.cacheControl?.(part.cache)
+            ? {
+                role: "user",
+                content: [{ type: "text", text: part.text, cache_control: options.cacheControl(part.cache) }],
+              }
+            : { role: "user", content: part.text },
+        )
       continue
     }
     if (message.role === "tool") {
-      const lowered = yield* lowerToolMessages(message)
+      const lowered = yield* lowerToolMessages(message, options)
       messages.push(...lowered.messages)
       pendingImages.push(...lowered.images)
       continue
     }
     flushImages()
-    messages.push(...(yield* lowerMessage(message, request.model.compatibility?.reasoningField)))
+    messages.push(...(yield* lowerMessage(message, request.model.compatibility?.reasoningField, options)))
   }
   flushImages()
   return messages
@@ -405,7 +494,10 @@ const lowerOptions = (request: LLMRequest) => {
   }
 }
 
-const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (request: LLMRequest) {
+export const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (
+  request: LLMRequest,
+  options: LoweringOptions = {},
+) {
   // `fromRequest` returns the provider body only. Endpoint, auth, framing,
   // validation, and HTTP execution are composed by `Route.make`.
   const reasoningField = request.model.compatibility?.reasoningField
@@ -415,19 +507,26 @@ const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (request: LLMR
     )
   const generation = request.generation
   const toolSchemaCompatibility = request.model.compatibility?.toolSchema
+  const maxTokensField = request.model.compatibility?.maxTokensField ?? "max_tokens"
   return {
     model: request.model.id,
-    messages: yield* lowerMessages(request),
+    messages: yield* lowerMessages(request, options),
     tools:
       request.tools.length === 0
         ? undefined
         : request.tools.map((tool) =>
-            lowerTool(tool, ToolSchemaProjection.modelCompatibility(tool.inputSchema, toolSchemaCompatibility)),
+            lowerTool(
+              tool,
+              ToolSchemaProjection.modelCompatibility(tool.inputSchema, toolSchemaCompatibility),
+              options,
+            ),
           ),
     tool_choice: request.toolChoice ? yield* lowerToolChoice(request.toolChoice) : undefined,
     stream: true as const,
     stream_options: { include_usage: true },
-    max_tokens: generation?.maxTokens,
+    ...(maxTokensField === "max_completion_tokens"
+      ? { max_completion_tokens: generation?.maxTokens }
+      : { max_tokens: generation?.maxTokens }),
     temperature: generation?.temperature,
     top_p: generation?.topP,
     frequency_penalty: generation?.frequencyPenalty,
@@ -456,7 +555,7 @@ const mapFinishReason = (reason: string | null | undefined): FinishReason => {
 // OpenAI Chat reports `prompt_tokens` (inclusive total) with a
 // cached-read and cache-write subsets, and `completion_tokens` (inclusive
 // total) with a `reasoning_tokens` subset. We pass the inclusive totals
-// through and derive the non-cached breakdown so the `LLM.Usage` contract is
+// through and derive the non-cached breakdown so the `AI.Usage` contract is
 // satisfied on both sides.
 const mapUsage = (usage: OpenAIChatEvent["usage"]): Usage | undefined => {
   if (!usage) return undefined
@@ -546,7 +645,7 @@ const reasoningMetadata = (field: ParserState["reasoningField"], details?: Reado
 const step = (state: ParserState, event: OpenAIChatEvent) =>
   Effect.gen(function* () {
     if (event.error)
-      return yield* new LLMError({
+      return yield* new AIError({
         module: ADAPTER,
         method: "stream",
         reason: classifyProviderFailure({

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

@@ -1,11 +1,11 @@
-import { Route, type RouteRoutedModelInput } from "../route/client"
+import { Route, type RouteRoutedLanguageModelInput } from "../route/client"
 import { Endpoint } from "../route/endpoint"
 import { Framing } from "../route/framing"
 import * as OpenAIChat from "./openai-chat"
 
 const ADAPTER = "openai-compatible-chat"
 
-export type OpenAICompatibleChatModelInput = RouteRoutedModelInput
+export type OpenAICompatibleChatLanguageModelInput = RouteRoutedLanguageModelInput
 
 /**
  * Route for non-OpenAI providers that expose an OpenAI Chat-compatible

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

@@ -1,10 +1,10 @@
-import { Route, type RouteRoutedModelInput } from "../route/client"
+import { Route, type RouteRoutedLanguageModelInput } from "../route/client"
 import { Endpoint } from "../route/endpoint"
 import { OpenResponses } from "./open-responses"
 
 const ADAPTER = "openai-compatible-responses"
 
-export type OpenAICompatibleResponsesModelInput = RouteRoutedModelInput
+export type OpenAICompatibleResponsesLanguageModelInput = RouteRoutedLanguageModelInput
 
 /**
  * Deployment adapter for providers that expose an Open Responses-compatible

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

@@ -11,7 +11,7 @@ import {
 import { Auth, type Definition as AuthDefinition } from "../route/auth"
 import {
   InvalidProviderOutputReason,
-  LLMError,
+  AIError,
   Usage,
   mergeHttpOptions,
   mergeJsonRecords,
@@ -85,7 +85,7 @@ const nativeOptions = (options: OpenAIImageOptions | undefined) => {
 }
 
 const invalidOutput = (message: string) =>
-  new LLMError({
+  new AIError({
     module: ADAPTER,
     method: "generate",
     reason: new InvalidProviderOutputReason({ message, route: ADAPTER }),

packages/ai/src/protocols/shared.ts

@@ -6,7 +6,7 @@ import { Headers, HttpClientRequest } from "effect/unstable/http"
 import {
   InvalidProviderOutputReason,
   InvalidRequestReason,
-  LLMError,
+  AIError,
   type ContentPart,
   type LLMRequest,
   type MediaPart,
@@ -41,7 +41,7 @@ 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 `LLM.Usage` contract, `inputTokens` and `outputTokens`
+ * Under the additive `AI.Usage` contract, `inputTokens` and `outputTokens`
  * are the non-cached input and visible output only. The provider-supplied
  * `total` is the source of truth when present; the computed fallback
  * under-counts cache and reasoning by design and exists mainly so
@@ -88,7 +88,7 @@ export const sumTokens = (...values: ReadonlyArray<number | undefined>): number
 }
 
 export const eventError = (route: string, message: string, raw?: string) =>
-  new LLMError({
+  new AIError({
     module: "ProviderShared",
     method: "stream",
     reason: new InvalidProviderOutputReason({ route, message, raw }),
@@ -238,9 +238,9 @@ export const errorText = (error: unknown) => {
  * `decodeChunk` 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) so the public error channel stays
- * `LLMError`.
+ * `AIError`.
  */
-export const sseFraming = (bytes: Stream.Stream<Uint8Array, LLMError>): Stream.Stream<string, LLMError> =>
+export const sseFraming = (bytes: Stream.Stream<Uint8Array, AIError>): Stream.Stream<string, AIError> =>
   bytes.pipe(
     Stream.decodeText(),
     Stream.pipeThroughChannel(Sse.decode()),
@@ -257,7 +257,7 @@ export const sseFraming = (bytes: Stream.Stream<Uint8Array, LLMError>): Stream.S
  * lands here.
  */
 export const invalidRequest = (message: string) =>
-  new LLMError({
+  new AIError({
     module: "ProviderShared",
     method: "request",
     reason: new InvalidRequestReason({ message }),
@@ -304,7 +304,7 @@ export const unsupportedContent = (
  * Build a `validate` step from a Schema decoder. Replaces the per-route
  * lambda body `(payload) => decode(payload).pipe(Effect.mapError((e) =>
  * invalid(e.message)))`. Any decode error is translated into
- * `LLMError` carrying the original parse-error message.
+ * `AIError` carrying the original parse-error message.
  */
 export const validateWith =
   <A, I, E extends { readonly message: string }>(decode: (input: I) => Effect.Effect<A, E>) =>

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

@@ -22,6 +22,8 @@ const signRequest = (input: {
   readonly body: string
   readonly headers: Headers.Headers
   readonly credentials: Credentials
+  readonly service: string
+  readonly name: string
 }) =>
   Effect.tryPromise({
     try: async () => {
@@ -34,23 +36,26 @@ const signRequest = (input: {
         accessKeyId: input.credentials.accessKeyId,
         secretAccessKey: input.credentials.secretAccessKey,
         sessionToken: input.credentials.sessionToken,
-        service: "bedrock",
+        service: input.service,
       }).sign()
       return Object.fromEntries(signed.headers.entries())
     },
     catch: (error) =>
       ProviderShared.invalidRequest(
-        `Bedrock Converse SigV4 signing failed: ${error instanceof Error ? error.message : String(error)}`,
+        `${input.name} SigV4 signing failed: ${error instanceof Error ? error.message : String(error)}`,
       ),
   })
 
 /** Sign the exact JSON bytes with SigV4 using credentials configured on the route. */
-export const sigV4 = (credentials: Credentials | undefined) =>
+export const sigV4 = (
+  credentials: Credentials | undefined,
+  options: { readonly service?: string; readonly name?: string } = {},
+) =>
   Auth.custom((input: AuthInput) => {
     return Effect.gen(function* () {
       if (!credentials) {
         return yield* ProviderShared.invalidRequest(
-          "Bedrock Converse requires either route bearer auth or AWS credentials configured on the route",
+          `${options.name ?? "Bedrock Converse"} requires either route bearer auth or AWS credentials configured on the route`,
         )
       }
       const headersForSigning = Headers.set(input.headers, "content-type", "application/json")
@@ -59,6 +64,8 @@ export const sigV4 = (credentials: Credentials | undefined) =>
         body: input.body,
         headers: headersForSigning,
         credentials,
+        service: options.service ?? "bedrock",
+        name: options.name ?? "Bedrock Converse",
       })
       return Headers.setAll(headersForSigning, signed)
     })

packages/ai/src/protocols/utils/image-input.ts

@@ -1,9 +1,9 @@
 import { Effect, Encoding } from "effect"
 import type { ImageInput } from "../../image"
-import { InvalidRequestReason, LLMError } from "../../schema"
+import { InvalidRequestReason, AIError } from "../../schema"
 
 const invalid = (module: string, message: string) =>
-  new LLMError({
+  new AIError({
     module,
     method: "generate",
     reason: new InvalidRequestReason({ message }),
@@ -15,7 +15,7 @@ export const dataUrl = (input: Extract<ImageInput, { readonly type: "bytes" }>)
 export const decodeDataUrl = (
   url: string,
   module: string,
-): Effect.Effect<{ readonly mediaType: string; readonly data: Uint8Array } | undefined, LLMError> => {
+): Effect.Effect<{ readonly mediaType: string; readonly data: Uint8Array } | undefined, AIError> => {
   if (!url.startsWith("data:")) return Effect.succeed(undefined)
   const match = /^data:([^;,]+);base64,(.*)$/s.exec(url)
   if (!match) return Effect.fail(invalid(module, "Image data URLs must contain a MIME type and base64 data"))

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

@@ -1,4 +1,4 @@
-import type { JsonSchema, ModelToolSchemaCompatibility } from "../../schema"
+import type { JsonSchema, LanguageModelToolSchemaCompatibility } from "../../schema"
 import { isRecord } from "../../utils/record"
 import { GeminiToolSchema } from "./gemini-tool-schema"
 
@@ -69,7 +69,7 @@ const gemini = (schema: JsonSchema): JsonSchema => GeminiToolSchema.convert(sche
 
 const modelCompatibility = (
   schema: JsonSchema,
-  compatibility: ModelToolSchemaCompatibility | undefined,
+  compatibility: LanguageModelToolSchemaCompatibility | undefined,
 ): JsonSchema => {
   if (compatibility === undefined) return schema
   switch (compatibility) {

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

@@ -1,5 +1,5 @@
 import { Effect } from "effect"
-import { LLMError, LLMEvent, type ProviderMetadata, type ToolCall, type ToolInputError } from "../../schema"
+import { AIError, LLMEvent, type ProviderMetadata, type ToolCall, type ToolInputError } from "../../schema"
 import { eventError, parseToolInput, type ToolAccumulator } from "../shared"
 
 type StreamKey = string | number
@@ -112,8 +112,8 @@ const appendTool = <K extends StreamKey>(
   }
 }
 
-export const isError = <K extends StreamKey>(result: AppendOutcome<K> | LLMError): result is LLMError =>
-  result instanceof LLMError
+export const isError = <K extends StreamKey>(result: AppendOutcome<K> | AIError): result is AIError =>
+  result instanceof AIError
 
 /**
  * Register a tool call whose start event arrived before any argument deltas.
@@ -138,7 +138,7 @@ export const appendOrStart = <K extends StreamKey>(
   key: K,
   delta: { readonly id?: string; readonly name?: string; readonly text: string },
   missingToolMessage: string,
-): AppendOutcome<K> | LLMError => {
+): AppendOutcome<K> | AIError => {
   const current = tools[key]
   const id = current?.id ?? delta.id
   const name = current?.name ?? delta.name
@@ -167,7 +167,7 @@ export const appendExisting = <K extends StreamKey>(
   key: K,
   text: string,
   missingToolMessage: string,
-): AppendOutcome<K> | LLMError => {
+): AppendOutcome<K> | AIError => {
   const current = tools[key]
   if (!current) return eventError(route, missingToolMessage)
   if (text.length === 0) return { tools, tool: current, events: [] }

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

@@ -4,7 +4,7 @@ import { GeneratedImage, ImageModel, ImageResponse, type ImageRequestFor, type I
 import { Auth, type Definition as AuthDefinition } from "../route/auth"
 import {
   InvalidProviderOutputReason,
-  LLMError,
+  AIError,
   Usage,
   mergeHttpOptions,
   mergeJsonRecords,
@@ -95,7 +95,7 @@ const nativeOptions = (options: XAIImageOptions | undefined) => {
 }
 
 const invalidOutput = (message: string) =>
-  new LLMError({
+  new AIError({
     module: ADAPTER,
     method: "generate",
     reason: new InvalidProviderOutputReason({ message, route: ADAPTER }),

packages/ai/src/protocols/zai-images.ts

@@ -2,7 +2,7 @@ import { Effect, Schema } from "effect"
 import { Headers, HttpClientRequest } from "effect/unstable/http"
 import { GeneratedImage, ImageModel, ImageResponse, type ImageRequestFor, type ImageRoute } from "../image"
 import { Auth, type Definition as AuthDefinition } from "../route/auth"
-import { InvalidProviderOutputReason, LLMError, mergeHttpOptions, mergeJsonRecords, type HttpOptions } from "../schema"
+import { InvalidProviderOutputReason, AIError, mergeHttpOptions, mergeJsonRecords, type HttpOptions } from "../schema"
 import { ProviderShared } from "./shared"
 import { ImageInputs } from "./utils/image-input"
 
@@ -58,7 +58,7 @@ const nativeOptions = (options: ZAIImageOptions | undefined) => {
 }
 
 const invalidOutput = (message: string) =>
-  new LLMError({
+  new AIError({
     module: ADAPTER,
     method: "generate",
     reason: new InvalidProviderOutputReason({ message, route: ADAPTER }),

packages/ai/src/provider-error.ts

@@ -3,7 +3,7 @@ import {
   AuthenticationReason,
   ContentPolicyReason,
   InvalidRequestReason,
-  LLMError,
+  AIError,
   ProviderErrorEvent,
   ProviderInternalReason,
   QuotaExceededReason,
@@ -16,7 +16,6 @@ import {
 
 const patterns = [
   /prompt is too long/i,
-  /request_too_large/i,
   /input is too long for requested model/i,
   /exceeds the context window/i,
   /exceeds (?:the )?(?:model'?s )?maximum context length(?: of [\d,]+ tokens?|\s*\([\d,]+\))/i,
@@ -33,7 +32,6 @@ const patterns = [
   /context window exceeds limit/i,
   /exceeded model token limit/i,
   /context[_ ]length[_ ]exceeded/i,
-  /request entity too large/i,
   /context length is only \d+ tokens/i,
   /input length.*exceeds.*context length/i,
   /prompt too long; exceeded (?:max )?context length/i,
@@ -44,14 +42,18 @@ const patterns = [
   /token limit exceeded/i,
 ]
 
+const payloadPatterns = [/request_too_large/i, /request entity too large/i, /payload too large/i, /request too large/i]
+
 const exclusions = [/^(throttling error|service unavailable):/i, /rate limit/i, /too many requests/i]
 
 export const isContextOverflow = (message: string) =>
   !exclusions.some((pattern) => pattern.test(message)) &&
-  (patterns.some((pattern) => pattern.test(message)) || /^4(00|13)\s*(status code)?\s*\(no body\)/i.test(message))
+  (patterns.some((pattern) => pattern.test(message)) || /^400\s*(status code)?\s*\(no body\)/i.test(message))
+
+export const isPayloadTooLarge = (message: string) => payloadPatterns.some((pattern) => pattern.test(message))
 
 export const isContextOverflowFailure = (failure: unknown) =>
-  failure instanceof LLMError
+  failure instanceof AIError
     ? failure.reason._tag === "InvalidRequest" && failure.reason.classification === "context-overflow"
     : Schema.is(ProviderErrorEvent)(failure) && failure.classification === "context-overflow"
 
@@ -84,7 +86,7 @@ export interface ProviderFailure {
 
 // Keep HTTP failures and provider-reported stream failures on one typed path so
 // session retry policy never needs provider-specific string matching.
-export function classifyProviderFailure(input: ProviderFailure): LLMError["reason"] {
+export function classifyProviderFailure(input: ProviderFailure): AIError["reason"] {
   const body = input.http?.body ?? ""
   const codes = [input.code, ...providerCodes(body), ...providerCodes(input.message)]
     .filter((code): code is string => code !== undefined)
@@ -100,6 +102,8 @@ export function classifyProviderFailure(input: ProviderFailure): LLMError["reaso
       isContextOverflow(text))
   )
     return new InvalidRequestReason({ ...common, classification: "context-overflow" })
+  if (input.status === 413 || isPayloadTooLarge(text))
+    return new InvalidRequestReason({ ...common, classification: "payload-too-large" })
   if (CONTENT_POLICY_TEXT.test(text)) return new ContentPolicyReason(common)
   if (codes.some((code) => QUOTA_CODES.has(code)) || (input.status === 429 && QUOTA_TEXT.test(text)))
     return new QuotaExceededReason(common)
@@ -135,20 +139,14 @@ export function classifyProviderFailure(input: ProviderFailure): LLMError["reaso
       rateLimit: input.rateLimit,
     })
   }
-  if (input.status !== undefined && input.status >= 500)
+  if (input.status === 408 || input.status === 409 || (input.status !== undefined && input.status >= 500))
     return new ProviderInternalReason({
       ...common,
       status: input.status,
       retryAfterMs: input.retryAfterMs,
     })
   if (codes.some((code) => INVALID_REQUEST_CODES.has(code))) return new InvalidRequestReason(common)
-  if (
-    input.status === 400 ||
-    input.status === 404 ||
-    input.status === 409 ||
-    input.status === 413 ||
-    input.status === 422
-  )
+  if (input.status === 400 || input.status === 404 || input.status === 413 || input.status === 422)
     return new InvalidRequestReason(common)
   return new UnknownProviderReason({ ...common, status: input.status })
 }

packages/ai/src/provider-package.ts

@@ -1,16 +1,21 @@
-import type { Model } from "./schema"
+import type { LanguageModel, ProviderOptions } from "./schema"
 
 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<ProviderSettings extends Settings = Settings> {
-  readonly model: (modelID: string, settings: ProviderSettings) => Model
+export interface Definition<
+  ProviderSettings extends Settings = Settings,
+  Options extends ProviderOptions = ProviderOptions,
+> {
+  readonly model: (modelID: string, settings: ProviderSettings) => LanguageModel<Options>
 }
 
 export * as ProviderPackage from "./provider-package"

packages/ai/src/provider.ts

@@ -1,6 +1,6 @@
-import type { Model, ModelID, ProviderID } from "./schema"
+import type { LanguageModel, ModelID, ProviderID } from "./schema"
 
-export type ModelOptions = Pick<Model.Input, "defaults" | "compatibility">
+export type LanguageModelOptions = Pick<LanguageModel.Input, "defaults" | "compatibility">
 
 /**
  * Advanced structural provider definition helper. Built-in providers should
@@ -8,23 +8,23 @@ export type ModelOptions = Pick<Model.Input, "defaults" | "compatibility">
  * chosen before model selection. The optional `apis` map remains for external
  * structural providers that expose multiple route selectors behind one provider.
  */
-export type ModelFactory<Options extends ModelOptions = ModelOptions> = (
+export type LanguageModelFactory<Options extends LanguageModelOptions = LanguageModelOptions> = (
   id: string | ModelID,
   options?: Options,
-) => Model
+) => LanguageModel
 
-type AnyModelFactory = (...args: never[]) => Model
+type AnyLanguageModelFactory = (...args: never[]) => LanguageModel
 
-export interface Definition<Factory extends AnyModelFactory = ModelFactory> {
+export interface Definition<Factory extends AnyLanguageModelFactory = LanguageModelFactory> {
   readonly id: ProviderID
   readonly model: Factory
-  readonly apis?: Record<string, AnyModelFactory>
+  readonly apis?: Record<string, AnyLanguageModelFactory>
 }
 
 type DefinitionShape = {
   readonly id: ProviderID
-  readonly model: (...args: never[]) => Model
-  readonly apis?: Record<string, (...args: never[]) => Model>
+  readonly model: (...args: never[]) => LanguageModel
+  readonly apis?: Record<string, (...args: never[]) => LanguageModel>
 }
 
 type NoExtraFields<Input, Shape> = Input & Record<Exclude<keyof Input, keyof Shape>, never>

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

@@ -0,0 +1,107 @@
+import { Auth } from "../route/auth"
+import type { Route as RouteDef, RouteDefaultsInput } from "../route/client"
+import type { ProviderPackage } from "../provider-package"
+import { OpenAIChat } from "../protocols/openai-chat"
+import { OpenAIResponses } from "../protocols/openai-responses"
+import { BedrockAuth, type Credentials } from "../protocols/utils/bedrock-auth"
+import { ProviderID, type ModelID } from "../schema"
+import { withOpenAIOptions, type OpenAIProviderOptionsInput } from "./openai-options"
+
+export const id = ProviderID.make("amazon-bedrock")
+
+export type Config = RouteDefaultsInput & {
+  readonly apiKey?: string
+  readonly baseURL?: string
+  readonly credentials?: Credentials
+  readonly region?: string
+  readonly providerOptions?: OpenAIProviderOptionsInput
+}
+
+export interface Settings extends ProviderPackage.Settings {
+  readonly apiKey?: string
+  readonly auth?: "bearer" | "sigv4"
+  readonly baseURL?: string
+  readonly credentials?: Credentials
+  readonly region?: string
+  readonly providerOptions?: OpenAIProviderOptionsInput
+}
+
+const responsesRoute = OpenAIResponses.route.with({
+  id: "bedrock-mantle-responses",
+  provider: id,
+})
+
+const chatRoute = OpenAIChat.route.with({
+  id: "bedrock-mantle-chat",
+  provider: id,
+})
+
+export const routes = [responsesRoute, chatRoute]
+
+const configuredRoute = <Body, Prepared>(route: RouteDef<Body, Prepared>, input: Config) => {
+  const region = input.region ?? input.credentials?.region ?? "us-east-1"
+  const credentials = input.credentials === undefined ? undefined : { ...input.credentials, region }
+  return route.with({
+    endpoint: { baseURL: input.baseURL ?? `https://bedrock-mantle.${region}.api.aws/v1` },
+    auth:
+      input.apiKey === undefined
+        ? BedrockAuth.sigV4(credentials, { service: "bedrock-mantle", name: "Bedrock Mantle" })
+        : Auth.bearer(input.apiKey),
+  })
+}
+
+const defaults = (input: Config) => {
+  const { apiKey: _, baseURL: _baseURL, credentials: _credentials, region: _region, ...rest } = input
+  return rest
+}
+
+export const configure = (input: Config = {}) => {
+  const configuredResponsesRoute = configuredRoute(responsesRoute, input)
+  const configuredChatRoute = configuredRoute(chatRoute, input)
+  const modelDefaults = defaults(input)
+  const responses = (modelID: string | ModelID) =>
+    configuredResponsesRoute
+      .with(withOpenAIOptions(modelID, modelDefaults))
+      .model<OpenAIProviderOptionsInput>({ id: modelID })
+  const chat = (modelID: string | ModelID) =>
+    configuredChatRoute
+      .with(withOpenAIOptions(modelID, modelDefaults))
+      .model<OpenAIProviderOptionsInput>({ id: modelID })
+
+  return {
+    id,
+    model: chat,
+    chat,
+    responses,
+    configure,
+  }
+}
+
+export const provider = configure()
+
+const config = (settings: Settings): Config => {
+  if (settings.auth === "bearer" && settings.apiKey === undefined)
+    throw new Error("Amazon Bedrock Mantle bearer auth requires apiKey")
+  if (settings.auth === "sigv4" && settings.apiKey !== undefined)
+    throw new Error("Amazon Bedrock Mantle SigV4 auth does not accept apiKey")
+  return {
+    apiKey: settings.auth === "sigv4" ? undefined : settings.apiKey,
+    baseURL: settings.baseURL,
+    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,
+  }
+}
+
+export const chatModel: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (
+  modelID,
+  settings,
+) => configure(config(settings)).chat(modelID)
+export const responsesModel: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (
+  modelID,
+  settings,
+) => configure(config(settings)).responses(modelID)
+export const model = chatModel

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

@@ -0,0 +1,2 @@
+export { chatModel as model } from "../amazon-bedrock-mantle"
+export type { Settings } from "../amazon-bedrock-mantle"

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

@@ -0,0 +1,2 @@
+export { chatModel as model } from "../../amazon-bedrock-mantle"
+export type { Settings } from "../../amazon-bedrock-mantle"

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

@@ -0,0 +1,2 @@
+export { responsesModel as model } from "../../amazon-bedrock-mantle"
+export type { Settings } from "../../amazon-bedrock-mantle"

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

@@ -47,7 +47,7 @@ export const configure = (input: Config) => {
   })
   return {
     id: ProviderID.make(provider),
-    model: (modelID: string | ModelID) => route.model({ id: modelID }),
+    model: (modelID: string | ModelID) => route.model<AnthropicMessages.ProviderOptionsInput>({ id: modelID }),
     configure,
   }
 }
@@ -57,7 +57,10 @@ export const provider = {
   configure,
 }
 
-export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) => {
+export const model: ProviderPackage.Definition<Settings, AnthropicMessages.ProviderOptionsInput>["model"] = (
+  modelID,
+  settings,
+) => {
   if (settings.apiKey !== undefined && settings.authToken !== undefined)
     throw new Error("Anthropic-compatible apiKey cannot be combined with authToken")
   return configure({

packages/ai/src/providers/anthropic.ts

@@ -52,7 +52,10 @@ export const configure = (input: Config = {}) => {
 }
 
 export const provider = configure()
-export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) => {
+export const model: ProviderPackage.Definition<Settings, AnthropicMessages.ProviderOptionsInput>["model"] = (
+  modelID,
+  settings,
+) => {
   if (settings.apiKey !== undefined && settings.authToken !== undefined)
     throw new Error("Anthropic apiKey cannot be combined with authToken")
   return configure({

packages/ai/src/providers/azure.ts

@@ -14,7 +14,7 @@ const routeAuth = Auth.remove("authorization")
 // (helper builds the URL) or `baseURL` directly.
 type AzureURL = AtLeastOne<{ readonly resourceName: string; readonly baseURL: string }>
 
-export type ModelOptions = AzureURL &
+export type LanguageModelOptions = AzureURL &
   RouteDefaultsInput &
   ProviderAuthOption<"optional"> & {
     readonly apiVersion?: string
@@ -22,7 +22,7 @@ export type ModelOptions = AzureURL &
     readonly useCompletionUrls?: boolean
     readonly providerOptions?: OpenAIProviderOptionsInput
   }
-export type Config = ModelOptions
+export type Config = LanguageModelOptions
 
 export type Settings = ProviderPackage.Settings &
   AzureURL & {
@@ -99,10 +99,14 @@ export const configure = (input: Config) => {
   const modelDefaults = defaults(input)
 
   const responses = (modelID: string | ModelID) =>
-    configuredResponsesRoute.with(withOpenAIOptions(modelID, modelDefaults)).model({ id: modelID })
+    configuredResponsesRoute
+      .with(withOpenAIOptions(modelID, modelDefaults))
+      .model<OpenAIProviderOptionsInput>({ id: modelID })
 
   const chat = (modelID: string | ModelID) =>
-    configuredChatRoute.with(withOpenAIOptions(modelID, modelDefaults)).model({ id: modelID })
+    configuredChatRoute
+      .with(withOpenAIOptions(modelID, modelDefaults))
+      .model<OpenAIProviderOptionsInput>({ id: modelID })
 
   return {
     id,
@@ -133,8 +137,12 @@ const config = (settings: Settings): Config => {
   throw new Error("Azure requires resourceName or baseURL")
 }
 
-export const responsesModel: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) =>
-  configure(config(settings)).responses(modelID)
-export const chatModel: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) =>
-  configure(config(settings)).chat(modelID)
+export const responsesModel: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (
+  modelID,
+  settings,
+) => configure(config(settings)).responses(modelID)
+export const chatModel: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (
+  modelID,
+  settings,
+) => configure(config(settings)).chat(modelID)
 export const model = responsesModel

packages/ai/src/providers/cloudflare.ts

@@ -4,6 +4,7 @@ import { Auth } from "../route/auth"
 import { AuthOptions, type AtLeastOne, type ProviderAuthOption } from "../route/auth-options"
 import type { RouteDefaultsInput } from "../route/client"
 import { ProviderID, type ModelID } from "../schema"
+import type { OpenAIProviderOptionsInput } from "./openai-options"
 
 export const aiGatewayID = ProviderID.make("cloudflare-ai-gateway")
 export const workersAIID = ProviderID.make("cloudflare-workers-ai")
@@ -20,10 +21,11 @@ type GatewayURL = AtLeastOne<{
 }
 
 export type AIGatewayOptions = GatewayURL &
-  RouteDefaultsInput &
+  Omit<RouteDefaultsInput, "providerOptions"> &
   ProviderAuthOption<"optional"> & {
     /** Cloudflare AI Gateway authentication token. Sent as `cf-aig-authorization`. */
     readonly gatewayApiKey?: CloudflareSecret
+    readonly providerOptions?: OpenAIProviderOptionsInput
   }
 
 type WorkersAIURL = AtLeastOne<{
@@ -31,7 +33,11 @@ type WorkersAIURL = AtLeastOne<{
   readonly baseURL: string
 }>
 
-export type WorkersAIOptions = WorkersAIURL & RouteDefaultsInput & ProviderAuthOption<"optional">
+export type WorkersAIOptions = WorkersAIURL &
+  Omit<RouteDefaultsInput, "providerOptions"> &
+  ProviderAuthOption<"optional"> & {
+    readonly providerOptions?: OpenAIProviderOptionsInput
+  }
 
 export const aiGatewayBaseURL = (input: GatewayURL) => {
   if (input.baseURL) return input.baseURL
@@ -98,7 +104,7 @@ const configureAIGateway = (options: AIGatewayOptions) => {
   })
   return {
     id: aiGatewayID,
-    model: (modelID: string | ModelID) => route.model({ id: modelID }),
+    model: (modelID: string | ModelID) => route.model<OpenAIProviderOptionsInput>({ id: modelID }),
     configure: configureAIGateway,
   }
 }
@@ -111,7 +117,7 @@ const configureWorkersAI = (options: WorkersAIOptions) => {
   })
   return {
     id: workersAIID,
-    model: (modelID: string | ModelID) => route.model({ id: modelID }),
+    model: (modelID: string | ModelID) => route.model<OpenAIProviderOptionsInput>({ id: modelID }),
     configure: configureWorkersAI,
   }
 }

packages/ai/src/providers/github-copilot.ts

@@ -9,14 +9,14 @@ export const id = ProviderID.make("github-copilot")
 
 // GitHub Copilot has no canonical public URL — callers (opencode, etc.) must
 // supply `baseURL` explicitly.
-export type ModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
+export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
   ProviderAuthOption<"optional"> & {
     readonly baseURL: string
     readonly endpoint?: "chat" | "responses"
     readonly providerOptions?: OpenAIProviderOptionsInput
   }
 
-export const shouldUseResponsesApi = (modelID: string | ModelID, endpoint?: ModelOptions["endpoint"]) => {
+export const shouldUseResponsesApi = (modelID: string | ModelID, endpoint?: LanguageModelOptions["endpoint"]) => {
   if (endpoint) return endpoint === "responses"
   const model = String(modelID)
   const match = /^gpt-(\d+)/.exec(model)
@@ -29,30 +29,32 @@ export const routes = [OpenAIResponses.route, OpenAIChat.route]
 const chatRoute = OpenAIChat.route.with({ provider: id })
 const responsesRoute = OpenAIResponses.route.with({ provider: id })
 
-const defaults = (options: ModelOptions) => {
+const defaults = (options: LanguageModelOptions) => {
   const { apiKey: _, auth: _auth, baseURL: _baseURL, endpoint: _endpoint, ...rest } = options
   return rest
 }
 
-const configuredResponsesRoute = (options: ModelOptions) =>
+const configuredResponsesRoute = (options: LanguageModelOptions) =>
   responsesRoute.with({
     endpoint: { baseURL: options.baseURL },
     auth: AuthOptions.bearer(options, []),
   })
 
-const configuredChatRoute = (options: ModelOptions) =>
+const configuredChatRoute = (options: LanguageModelOptions) =>
   chatRoute.with({
     endpoint: { baseURL: options.baseURL },
     auth: AuthOptions.bearer(options, []),
   })
 
-export const configure = (options: ModelOptions) => {
+export const configure = (options: LanguageModelOptions) => {
   const responsesRoute = configuredResponsesRoute(options)
   const chatRoute = configuredChatRoute(options)
   const responses = (modelID: string | ModelID) =>
-    responsesRoute.with(withOpenAIOptions(modelID, defaults(options))).model({ id: modelID })
+    responsesRoute
+      .with(withOpenAIOptions(modelID, defaults(options)))
+      .model<OpenAIProviderOptionsInput>({ id: modelID })
   const chat = (modelID: string | ModelID) =>
-    chatRoute.with(withOpenAIOptions(modelID, defaults(options))).model({ id: modelID })
+    chatRoute.with(withOpenAIOptions(modelID, defaults(options))).model<OpenAIProviderOptionsInput>({ id: modelID })
   return {
     id,
     model: (modelID: string | ModelID) =>

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

@@ -1,8 +1,9 @@
 import type { ProviderPackage } from "../provider-package"
 import { OpenAICompatibleChat } from "../protocols/openai-compatible-chat"
 import type { RouteDefaultsInput } from "../route/client"
-import { ProviderID, type ModelID, type ProviderOptions } from "../schema"
+import { ProviderID, type ModelID } from "../schema"
 import { GoogleVertexShared } from "./google-vertex-shared"
+import type { OpenAIProviderOptionsInput } from "./openai-options"
 
 export const id = ProviderID.make("google-vertex")
 
@@ -11,6 +12,7 @@ export type Config = RouteDefaultsInput &
     readonly baseURL?: string
     readonly location?: string
     readonly project?: string
+    readonly providerOptions?: OpenAIProviderOptionsInput
   }
 
 export interface Settings extends ProviderPackage.Settings {
@@ -19,7 +21,7 @@ export interface Settings extends ProviderPackage.Settings {
   readonly baseURL?: string
   readonly location?: string
   readonly project?: string
-  readonly providerOptions?: ProviderOptions
+  readonly providerOptions?: OpenAIProviderOptionsInput
 }
 
 const route = OpenAICompatibleChat.route.with({
@@ -56,7 +58,7 @@ export const configure = (input: Config = {}) => {
   const route = configuredRoute(input)
   return {
     id,
-    model: (modelID: string | ModelID) => route.model({ id: modelID }),
+    model: (modelID: string | ModelID) => route.model<OpenAIProviderOptionsInput>({ id: modelID }),
     configure,
   }
 }
@@ -66,7 +68,7 @@ export const provider = {
   configure,
 }
 
-export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) => {
+export const model: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (modelID, settings) => {
   if (settings.apiKey !== undefined) throw new Error("Google Vertex Chat does not support API keys")
   return configure({
     accessToken: settings.accessToken,

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

@@ -91,7 +91,7 @@ export const configure = (input: Config = {}) => {
   const route = configuredRoute(input)
   return {
     id,
-    model: (modelID: string | ModelID) => route.model({ id: modelID }),
+    model: (modelID: string | ModelID) => route.model<AnthropicMessages.ProviderOptionsInput>({ id: modelID }),
     configure,
   }
 }
@@ -101,7 +101,10 @@ export const provider = {
   configure,
 }
 
-export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) => {
+export const model: ProviderPackage.Definition<Settings, AnthropicMessages.ProviderOptionsInput>["model"] = (
+  modelID,
+  settings,
+) => {
   if (settings.apiKey !== undefined) throw new Error("Google Vertex Messages does not support API keys")
   return configure({
     accessToken: settings.accessToken,

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

@@ -1,8 +1,9 @@
 import type { ProviderPackage } from "../provider-package"
 import { OpenAICompatibleResponses } from "../protocols/openai-compatible-responses"
 import type { RouteDefaultsInput } from "../route/client"
-import { ProviderID, type ModelID, type ProviderOptions } from "../schema"
+import { ProviderID, type ModelID } from "../schema"
 import { GoogleVertexShared } from "./google-vertex-shared"
+import type { OpenResponsesProviderOptionsInput } from "./open-responses-options"
 
 export const id = ProviderID.make("google-vertex")
 
@@ -11,6 +12,7 @@ export type Config = RouteDefaultsInput &
     readonly baseURL?: string
     readonly location?: string
     readonly project?: string
+    readonly providerOptions?: OpenResponsesProviderOptionsInput
   }
 
 export interface Settings extends ProviderPackage.Settings {
@@ -19,7 +21,7 @@ export interface Settings extends ProviderPackage.Settings {
   readonly baseURL?: string
   readonly location?: string
   readonly project?: string
-  readonly providerOptions?: ProviderOptions
+  readonly providerOptions?: OpenResponsesProviderOptionsInput
 }
 
 const route = OpenAICompatibleResponses.route.with({
@@ -58,7 +60,7 @@ export const configure = (input: Config = {}) => {
   const route = configuredRoute(input)
   return {
     id,
-    model: (modelID: string | ModelID) => route.model({ id: modelID }),
+    model: (modelID: string | ModelID) => route.model<OpenResponsesProviderOptionsInput>({ id: modelID }),
     configure,
   }
 }
@@ -68,7 +70,10 @@ export const provider = {
   configure,
 }
 
-export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) => {
+export const model: ProviderPackage.Definition<Settings, OpenResponsesProviderOptionsInput>["model"] = (
+  modelID,
+  settings,
+) => {
   if (settings.apiKey !== undefined) throw new Error("Google Vertex Responses does not support API keys")
   return configure({
     accessToken: settings.accessToken,

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

@@ -77,7 +77,8 @@ const configuredRoute = (input: Config, modelID: string | ModelID) => {
 export const configure = (input: Config = {}) => {
   return {
     id,
-    model: (modelID: string | ModelID) => configuredRoute(input, modelID).model({ id: modelID }),
+    model: (modelID: string | ModelID) =>
+      configuredRoute(input, modelID).model<Gemini.ProviderOptionsInput>({ id: modelID }),
     configure,
   }
 }
@@ -86,7 +87,10 @@ export const provider = {
   id,
   configure,
 }
-export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) => {
+export const model: ProviderPackage.Definition<Settings, Gemini.ProviderOptionsInput>["model"] = (
+  modelID,
+  settings,
+) => {
   if (settings.apiKey !== undefined && settings.accessToken !== undefined)
     throw new Error("Google Vertex apiKey cannot be combined with accessToken or auth")
   return configure({

packages/ai/src/providers/google.ts

@@ -50,14 +50,14 @@ export const configure = (input: Config = {}) => {
     })
   return {
     id,
-    model: (modelID: string | ModelID) => route.model({ id: modelID }),
+    model: (modelID: string | ModelID) => route.model<Gemini.ProviderOptionsInput>({ id: modelID }),
     image,
     configure,
   }
 }
 
 export const provider = configure()
-export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) =>
+export const model: ProviderPackage.Definition<Settings, Gemini.ProviderOptionsInput>["model"] = (modelID, settings) =>
   configure({
     apiKey: settings.apiKey,
     baseURL: settings.baseURL,

packages/ai/src/providers/index.ts

@@ -1,6 +1,7 @@
 export * as Anthropic from "./anthropic"
 export * as AnthropicCompatible from "./anthropic-compatible"
 export * as AmazonBedrock from "./amazon-bedrock"
+export * as AmazonBedrockMantle from "./amazon-bedrock-mantle"
 export * as Azure from "./azure"
 export * as Cloudflare from "./cloudflare"
 export { CloudflareAIGateway, CloudflareWorkersAI } from "./cloudflare"

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

@@ -36,7 +36,7 @@ export const configure = (input: Config) => {
   })
   return {
     id: ProviderID.make(provider),
-    model: (modelID: string | ModelID) => route.model({ id: modelID }),
+    model: (modelID: string | ModelID) => route.model<OpenResponsesProviderOptionsInput>({ id: modelID }),
     configure,
   }
 }
@@ -46,7 +46,10 @@ export const provider = {
   configure,
 }
 
-export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) =>
+export const model: ProviderPackage.Definition<Settings, OpenResponsesProviderOptionsInput>["model"] = (
+  modelID,
+  settings,
+) =>
   configure({
     apiKey: settings.apiKey,
     baseURL: settings.baseURL,

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

@@ -4,13 +4,15 @@ import type { RouteDefaultsInput } from "../route/client"
 import { AuthOptions, type ProviderAuthOption } from "../route/auth-options"
 import type { ProviderPackage } from "../provider-package"
 import { profiles, type OpenAICompatibleProfile } from "./openai-compatible-profile"
+import type { OpenAIProviderOptionsInput } from "./openai-options"
 
 export const id = ProviderID.make("openai-compatible")
 
-type GenericModelOptions = RouteDefaultsInput &
+type GenericModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
   ProviderAuthOption<"optional"> & {
     readonly provider?: string
     readonly baseURL: string
+    readonly providerOptions?: OpenAIProviderOptionsInput
   }
 
 export interface Settings extends ProviderPackage.Settings {
@@ -19,9 +21,10 @@ export interface Settings extends ProviderPackage.Settings {
   readonly provider?: string
 }
 
-export type FamilyModelOptions = RouteDefaultsInput &
+export type FamilyModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
   ProviderAuthOption<"optional"> & {
     readonly baseURL?: string
+    readonly providerOptions?: OpenAIProviderOptionsInput
   }
 
 export const routes = [OpenAICompatibleChat.route]
@@ -37,7 +40,8 @@ export const configure = (input: GenericModelOptions) => {
   })
   return {
     id: ProviderID.make(provider),
-    model: (modelID: string | ModelID) => route.model({ id: modelID, provider: ProviderID.make(provider) }),
+    model: (modelID: string | ModelID) =>
+      route.model<OpenAIProviderOptionsInput>({ id: modelID, provider: ProviderID.make(provider) }),
     configure,
   }
 }
@@ -63,7 +67,7 @@ export const provider = {
   configure,
 }
 
-export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) =>
+export const model: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (modelID, settings) =>
   configure({
     apiKey: settings.apiKey,
     baseURL: settings.baseURL,

packages/ai/src/providers/openai.ts

@@ -86,10 +86,15 @@ export const configure = (input: Config = {}) => {
   const chatRoute = configuredRoute(OpenAIChat.route, input)
   const modelDefaults = defaults(input)
   const responses = (id: string | ModelID) =>
-    responsesRoute.with(withOpenAIOptions(id, modelDefaults, { textVerbosity: true })).model({ id })
+    responsesRoute
+      .with(withOpenAIOptions(id, modelDefaults, { textVerbosity: true }))
+      .model<OpenAIProviderOptionsInput>({ id })
   const responsesWebSocket = (id: string | ModelID) =>
-    responsesWebSocketRoute.with(withOpenAIOptions(id, modelDefaults, { textVerbosity: true })).model({ id })
-  const chat = (id: string | ModelID) => chatRoute.with(withOpenAIOptions(id, modelDefaults)).model({ id })
+    responsesWebSocketRoute
+      .with(withOpenAIOptions(id, modelDefaults, { textVerbosity: true }))
+      .model<OpenAIProviderOptionsInput>({ id })
+  const chat = (id: string | ModelID) =>
+    chatRoute.with(withOpenAIOptions(id, modelDefaults)).model<OpenAIProviderOptionsInput>({ id })
   const image = (modelID: string | ModelID) =>
     OpenAIImages.model({
       id: modelID,
@@ -132,15 +137,17 @@ const config = (settings: Settings): Config => {
   }
 }
 
-export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) => {
+export const model: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (modelID, settings) => {
   const configured = configure(config(settings))
   if (settings.transport === undefined || settings.transport === "http") return configured.responses(modelID)
   if (settings.transport === "websocket") return configured.responsesWebSocket(modelID)
   throw new Error(`Unsupported OpenAI Responses transport: ${String(settings.transport)}`)
 }
 
-export const chatModel: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) =>
-  configure(config(settings)).chat(modelID)
+export const chatModel: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (
+  modelID,
+  settings,
+) => configure(config(settings)).chat(modelID)
 export const responses = provider.responses
 export const responsesWebSocket = provider.responsesWebSocket
 export const chat = provider.chat

packages/ai/src/providers/openrouter.ts

@@ -4,32 +4,90 @@ import { Endpoint } from "../route/endpoint"
 import { Framing } from "../route/framing"
 import { Protocol } from "../route/protocol"
 import { AuthOptions, type ProviderAuthOption } from "../route/auth-options"
-import { ProviderID, type ModelID, type ProviderOptions } from "../schema"
+import { ProviderID, type CacheHint, type ModelID, type ProviderOptions } from "../schema"
+import type { ProviderPackage } from "../provider-package"
 import * as OpenAICompatibleProfiles from "./openai-compatible-profile"
 import * as OpenAIChat from "../protocols/openai-chat"
+import { newBreakpoints, ttlBucket } from "../protocols/utils/cache"
 import { isRecord } from "../protocols/shared"
 
 export const profile = OpenAICompatibleProfiles.profiles.openrouter
 export const id = ProviderID.make(profile.provider)
 const ADAPTER = "openrouter"
 
+type OpenRouterString<Known extends string> = Known | (string & {})
+
+export interface OpenRouterProviderRouting {
+  readonly [key: string]: unknown
+  readonly order?: ReadonlyArray<string>
+  readonly allow_fallbacks?: boolean
+  readonly require_parameters?: boolean
+  readonly data_collection?: OpenRouterString<"allow" | "deny">
+  readonly only?: ReadonlyArray<string>
+  readonly ignore?: ReadonlyArray<string>
+  readonly quantizations?: ReadonlyArray<string>
+  readonly sort?: OpenRouterString<"price" | "throughput" | "latency">
+  readonly max_price?: Readonly<{
+    prompt?: number | string
+    completion?: number | string
+    image?: number | string
+    audio?: number | string
+    request?: number | string
+  }>
+  readonly zdr?: boolean
+}
+
+export type OpenRouterPlugin =
+  | Readonly<{
+      id: "web"
+      max_results?: number
+      search_prompt?: string
+      engine?: OpenRouterString<"native" | "exa">
+    }>
+  | Readonly<{ id: "file-parser"; max_files?: number; pdf?: { engine?: string } }>
+  | Readonly<{ id: "moderation" }>
+  | Readonly<{ id: "response-healing" }>
+  | Readonly<{ id: "auto-router"; allowed_models?: ReadonlyArray<string> }>
+  | Readonly<{ id: string & {}; [key: string]: unknown }>
+
 export interface OpenRouterOptions {
   readonly [key: string]: unknown
-  readonly usage?: boolean | Record<string, unknown>
-  readonly reasoning?: Record<string, unknown>
+  readonly debug?: Readonly<{ echo_upstream_body?: boolean }>
+  readonly models?: ReadonlyArray<string>
+  readonly plugins?: ReadonlyArray<OpenRouterPlugin>
   readonly promptCacheKey?: string
+  readonly provider?: OpenRouterProviderRouting
+  readonly reasoning?: Readonly<{
+    enabled?: boolean
+    exclude?: boolean
+    effort?: OpenRouterString<"none" | "minimal" | "low" | "medium" | "high" | "xhigh" | "max">
+    max_tokens?: number
+  }>
+  readonly usage?: boolean | Readonly<{ include: boolean }>
+  readonly user?: string
+  readonly web_search_options?: Readonly<{
+    max_results?: number
+    search_prompt?: string
+    engine?: OpenRouterString<"native" | "exa">
+  }>
 }
 
 export type OpenRouterProviderOptionsInput = ProviderOptions & {
   readonly openrouter?: OpenRouterOptions
 }
 
-export type ModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
+export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
   ProviderAuthOption<"optional"> & {
     readonly baseURL?: string
     readonly providerOptions?: OpenRouterProviderOptionsInput
   }
 
+export interface Settings extends ProviderPackage.Settings {
+  readonly apiKey?: string
+  readonly baseURL?: string
+  readonly providerOptions?: OpenRouterProviderOptionsInput
+}
+
 const OpenRouterBody = Schema.StructWithRest(Schema.Struct(OpenAIChat.bodyFields), [
   Schema.Record(Schema.String, Schema.Any),
 ])
@@ -40,7 +98,7 @@ export const protocol = Protocol.make({
   body: {
     schema: OpenRouterBody,
     from: (request) =>
-      OpenAIChat.protocol.body.from(request).pipe(
+      OpenAIChat.fromRequest(request, { cacheControl: cacheControl() }).pipe(
         Effect.map((body) => {
           const sourceAssistants = request.messages.filter((message) => message.role === "assistant")
           let assistantIndex = 0
@@ -71,16 +129,39 @@ export const protocol = Protocol.make({
   stream: OpenAIChat.protocol.stream,
 })
 
+const cacheControl = () => {
+  const breakpoints = newBreakpoints(4)
+  return (cache: CacheHint | undefined) => {
+    if (cache === undefined || breakpoints.remaining === 0) return undefined
+    breakpoints.remaining -= 1
+    return {
+      type: "ephemeral" as const,
+      ...(ttlBucket(cache.ttlSeconds) === "1h" ? { ttl: "1h" } : {}),
+    }
+  }
+}
+
 const bodyOptions = (input: unknown) => {
   const openrouter = isRecord(input) ? input : {}
+  const { usage, models, provider, plugins, web_search_options, debug, user, reasoning, promptCacheKey, ...options } =
+    openrouter
   return {
-    ...(openrouter.usage === true
+    ...options,
+    ...(usage === undefined || usage === true
       ? { usage: { include: true } }
-      : isRecord(openrouter.usage)
-        ? { usage: openrouter.usage }
-        : {}),
-    ...(isRecord(openrouter.reasoning) ? { reasoning: openrouter.reasoning } : {}),
-    ...(typeof openrouter.promptCacheKey === "string" ? { prompt_cache_key: openrouter.promptCacheKey } : {}),
+      : usage === false
+        ? { usage: { include: false } }
+        : isRecord(usage)
+          ? { usage }
+          : {}),
+    ...(Array.isArray(models) ? { models } : {}),
+    ...(isRecord(provider) ? { provider } : {}),
+    ...(Array.isArray(plugins) ? { plugins } : {}),
+    ...(isRecord(web_search_options) ? { web_search_options } : {}),
+    ...(isRecord(debug) ? { debug } : {}),
+    ...(typeof user === "string" ? { user } : {}),
+    ...(isRecord(reasoning) ? { reasoning } : {}),
+    ...(typeof promptCacheKey === "string" ? { prompt_cache_key: promptCacheKey } : {}),
   }
 }
 
@@ -94,7 +175,7 @@ export const route = Route.make({
 
 export const routes = [route]
 
-const configuredRoute = (input: ModelOptions) => {
+const configuredRoute = (input: LanguageModelOptions) => {
   const { apiKey: _, auth: _auth, baseURL, ...rest } = input
   return route.with({
     ...rest,
@@ -103,14 +184,25 @@ const configuredRoute = (input: ModelOptions) => {
   })
 }
 
-export const configure = (input: ModelOptions = {}) => {
+export const configure = (input: LanguageModelOptions = {}) => {
   const route = configuredRoute(input)
   return {
     id,
-    model: (modelID: string | ModelID) => route.model({ id: modelID }),
+    model: (modelID: string | ModelID) => route.model<OpenRouterProviderOptionsInput>({ id: modelID }),
     configure,
   }
 }
 
 export const provider = configure()
-export const model = provider.model
+export const model: ProviderPackage.Definition<Settings, OpenRouterProviderOptionsInput>["model"] = (
+  modelID,
+  settings,
+) =>
+  configure({
+    apiKey: settings.apiKey,
+    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

@@ -1,49 +1,81 @@
 import { AuthOptions, type ProviderAuthOption } from "../route/auth-options"
-import type { RouteDefaultsInput } from "../route/client"
-import { HttpOptions, ProviderID, type ModelID } from "../schema"
+import { Route, type RouteDefaultsInput } from "../route/client"
+import { Endpoint } from "../route/endpoint"
+import { HttpOptions, ProviderID, type ModelID, type ProviderOptions } from "../schema"
 import * as OpenAICompatibleProfiles from "./openai-compatible-profile"
 import * as OpenAICompatibleChat from "../protocols/openai-compatible-chat"
+import * as OpenAIChat from "../protocols/openai-chat"
 import * as OpenAIResponses from "../protocols/openai-responses"
 import { XAIImages } from "../protocols/xai-images"
+import type { OpenAIOptionsInput } from "./openai-options"
+import type { ProviderPackage } from "../provider-package"
 
 export const id = ProviderID.make("xai")
 
-export type ModelOptions = RouteDefaultsInput &
+export type XAIProviderOptionsInput = ProviderOptions & {
+  readonly xai?: OpenAIOptionsInput
+}
+
+export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
   ProviderAuthOption<"optional"> & {
     readonly baseURL?: string
+    readonly providerOptions?: XAIProviderOptionsInput
   }
 
+export interface Settings extends ProviderPackage.Settings {
+  readonly apiKey?: string
+  readonly baseURL?: string
+  readonly providerOptions?: XAIProviderOptionsInput
+}
+
 export type { XAIImageOptions } from "../protocols/xai-images"
 
-export const routes = [OpenAIResponses.route, OpenAICompatibleChat.route]
+const responsesRoute = Route.make({
+  id: "openai-responses",
+  provider: id,
+  providerMetadataKey: "xai",
+  protocol: OpenAIResponses.protocol,
+  endpoint: Endpoint.path("/responses", { baseURL: OpenAICompatibleProfiles.profiles.xai.baseURL }),
+  transport: OpenAIResponses.httpTransport,
+  defaults: { providerOptions: { xai: { store: false } } },
+})
+
+const chatRoute = Route.make({
+  id: "openai-compatible-chat",
+  provider: id,
+  providerMetadataKey: "xai",
+  protocol: OpenAIChat.protocol,
+  endpoint: Endpoint.path("/chat/completions", { baseURL: OpenAICompatibleProfiles.profiles.xai.baseURL }),
+  transport: OpenAICompatibleChat.route.transport,
+})
+
+export const routes = [responsesRoute, chatRoute]
 
 const auth = (options: ProviderAuthOption<"optional">) => AuthOptions.bearer(options, "XAI_API_KEY")
 
-const configuredResponsesRoute = (input: ModelOptions) => {
+const configuredResponsesRoute = (input: LanguageModelOptions) => {
   const { apiKey: _, auth: _auth, baseURL, ...rest } = input
-  return OpenAIResponses.route.with({
+  return responsesRoute.with({
     ...rest,
-    provider: id,
     endpoint: { baseURL: baseURL ?? OpenAICompatibleProfiles.profiles.xai.baseURL },
     auth: auth(input),
   })
 }
 
-const configuredChatRoute = (input: ModelOptions) => {
+const configuredChatRoute = (input: LanguageModelOptions) => {
   const { apiKey: _, auth: _auth, baseURL, ...rest } = input
-  return OpenAICompatibleChat.route.with({
+  return chatRoute.with({
     ...rest,
-    provider: id,
     endpoint: { baseURL: baseURL ?? OpenAICompatibleProfiles.profiles.xai.baseURL },
     auth: auth(input),
   })
 }
 
-export const configure = (input: ModelOptions = {}) => {
+export const configure = (input: LanguageModelOptions = {}) => {
   const responsesRoute = configuredResponsesRoute(input)
   const chatRoute = configuredChatRoute(input)
-  const responses = (modelID: string | ModelID) => responsesRoute.model({ id: modelID })
-  const chat = (modelID: string | ModelID) => chatRoute.model({ id: modelID })
+  const responses = (modelID: string | ModelID) => responsesRoute.model<XAIProviderOptionsInput>({ id: modelID })
+  const chat = (modelID: string | ModelID) => chatRoute.model<XAIProviderOptionsInput>({ id: modelID })
   const image = (modelID: string | ModelID) =>
     XAIImages.model({
       id: modelID,
@@ -63,7 +95,15 @@ export const configure = (input: ModelOptions = {}) => {
 }
 
 export const provider = configure()
-export const model = provider.model
+export const model: ProviderPackage.Definition<Settings, XAIProviderOptionsInput>["model"] = (modelID, settings) =>
+  configure({
+    apiKey: settings.apiKey,
+    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
 export const chat = provider.chat
 export const image = provider.image

packages/ai/src/route/auth-options.ts

@@ -22,13 +22,17 @@ export type ProviderAuthOption<Mode extends ApiKeyMode> =
   | AuthOverride
   | (Mode extends "optional" ? OptionalApiKeyAuth : RequiredApiKeyAuth)
 
-export type ModelOptions<Base, Mode extends ApiKeyMode> = Omit<Base, "apiKey" | "auth"> & ProviderAuthOption<Mode>
+export type LanguageModelOptions<Base, Mode extends ApiKeyMode> = Omit<Base, "apiKey" | "auth"> &
+  ProviderAuthOption<Mode>
 
-export type ModelArgs<Base, Mode extends ApiKeyMode> = Mode extends "optional"
-  ? readonly [options?: ModelOptions<Base, Mode>]
-  : readonly [options: ModelOptions<Base, Mode>]
+export type LanguageModelArgs<Base, Mode extends ApiKeyMode> = Mode extends "optional"
+  ? readonly [options?: LanguageModelOptions<Base, Mode>]
+  : readonly [options: LanguageModelOptions<Base, Mode>]
 
-export type ModelFactory<Base, Mode extends ApiKeyMode, Model> = (id: string, ...args: ModelArgs<Base, Mode>) => Model
+export type LanguageModelFactory<Base, Mode extends ApiKeyMode, LanguageModel> = (
+  id: string,
+  ...args: LanguageModelArgs<Base, Mode>
+) => LanguageModel
 
 /**
  * Require at least one of the keys in `T`. Use for option shapes where any

packages/ai/src/route/auth.ts

@@ -1,6 +1,6 @@
 import { Config, Effect, Redacted } from "effect"
 import { Headers } from "effect/unstable/http"
-import { AuthenticationReason, InvalidRequestReason, LLMError, type HttpOptions } from "../schema"
+import { AuthenticationReason, InvalidRequestReason, AIError, type HttpOptions } from "../schema"
 
 export class MissingCredentialError extends Error {
   readonly _tag = "MissingCredentialError"
@@ -11,7 +11,7 @@ export class MissingCredentialError extends Error {
 }
 
 export type CredentialError = MissingCredentialError | Config.ConfigError
-export type AuthError = CredentialError | LLMError
+export type AuthError = CredentialError | AIError
 type Secret = string | Redacted.Redacted | Config.Config<string | Redacted.Redacted>
 
 export interface AuthInput {
@@ -100,7 +100,7 @@ export const headers = (input: Headers.Input) =>
 
 export const remove = (name: string) => auth((input) => Effect.succeed(Headers.remove(input.headers, name)))
 
-export const custom = (apply: (input: AuthInput) => Effect.Effect<Headers.Headers, LLMError>) => auth(apply)
+export const custom = (apply: (input: AuthInput) => Effect.Effect<Headers.Headers, AIError>) => auth(apply)
 
 export const passthrough = none
 
@@ -134,9 +134,9 @@ export function bearerHeader(name: string, source?: Secret | Credential) {
   return render(source)
 }
 
-const toLLMError = (error: AuthError): LLMError => {
+const toAIError = (error: AuthError): AIError => {
   if (error instanceof MissingCredentialError || error instanceof Config.ConfigError) {
-    return new LLMError({
+    return new AIError({
       module: "Auth",
       method: "apply",
       reason:
@@ -150,7 +150,7 @@ const toLLMError = (error: AuthError): LLMError => {
 
 export const toEffect =
   (input: Definition) =>
-  (authInput: AuthInput): Effect.Effect<Headers.Headers, LLMError> =>
-    input.apply(authInput).pipe(Effect.mapError(toLLMError))
+  (authInput: AuthInput): Effect.Effect<Headers.Headers, AIError> =>
+    input.apply(authInput).pipe(Effect.mapError(toAIError))
 
 export * as Auth from "./auth"

packages/ai/src/route/client.ts

@@ -5,22 +5,21 @@ import { Endpoint, type EndpointPatch } from "./endpoint"
 import { RequestExecutor } from "./executor"
 import { Framing } from "./framing"
 import { HttpTransport } from "./transport"
-import type { Transport, TransportRuntime } from "./transport"
+import type { HttpRequestTransform, Transport, TransportRuntime } from "./transport"
 import { WebSocketExecutor } from "./transport"
 import type { Protocol } from "./protocol"
 import { applyCachePolicy } from "../cache-policy"
 import * as ProviderShared from "../protocols/shared"
-import type { LLMError, PreparedRequestOf, ProtocolID, ProviderOptions } from "../schema"
+import type { ProtocolID, ProviderOptions } from "../schema"
 import {
+  AIError,
   GenerationOptions,
   HttpOptions,
   LLMRequest,
   LLMResponse,
-  Model,
-  ModelLimits,
-  LLMError as LLMErrorClass,
+  LanguageModel,
+  LanguageModelLimits,
   LLMEvent,
-  PreparedRequest,
   ProviderID,
   mergeGenerationOptions,
   mergeHttpOptions,
@@ -31,7 +30,7 @@ export interface RouteBody<Body> {
   /** Schema for the validated provider-native body sent as the JSON request. */
   readonly schema: Schema.Codec<Body, unknown>
   /** Build the provider-native body from a common `LLMRequest`. */
-  readonly from: (request: LLMRequest) => Effect.Effect<Body, LLMError>
+  readonly from: (request: LLMRequest) => Effect.Effect<Body, AIError>
 }
 
 export interface Route<Body, Prepared = unknown> {
@@ -46,13 +45,19 @@ export interface Route<Body, Prepared = unknown> {
   readonly defaults: RouteDefaults
   readonly body: RouteBody<Body>
   readonly with: (patch: RoutePatch<Body, Prepared>) => Route<Body, Prepared>
-  readonly model: (input: RouteMappedModelInput) => Model
-  readonly prepareTransport: (body: Body, request: LLMRequest) => Effect.Effect<Prepared, LLMError>
+  readonly model: <Options extends ProviderOptions = ProviderOptions>(
+    input: RouteMappedLanguageModelInput,
+  ) => LanguageModel<Options>
+  readonly prepareTransport: (
+    body: Body,
+    request: LLMRequest,
+    options?: StreamOptions,
+  ) => Effect.Effect<Prepared, AIError>
   readonly streamPrepared: (
     prepared: Prepared,
     request: LLMRequest,
     runtime: TransportRuntime,
-  ) => Stream.Stream<LLMEvent, LLMError>
+  ) => Stream.Stream<LLMEvent, AIError>
 }
 
 // Route registries intentionally erase body generics after construction.
@@ -63,13 +68,13 @@ export type AnyRoute = Route<any, any>
 
 export type HttpOptionsInput = HttpOptions.Input
 
-export type RouteModelInput = Omit<Model.Input, "provider" | "route">
+export type RouteLanguageModelInput = Omit<LanguageModel.Input, "provider" | "route">
 
-export type RouteRoutedModelInput = Omit<Model.Input, "route">
+export type RouteRoutedLanguageModelInput = Omit<LanguageModel.Input, "route">
 
 export interface RouteDefaults {
   readonly headers?: Record<string, string>
-  readonly limits?: ModelLimits
+  readonly limits?: LanguageModelLimits
   readonly generation?: GenerationOptions
   readonly providerOptions?: ProviderOptions
   readonly http?: HttpOptions
@@ -77,7 +82,7 @@ export interface RouteDefaults {
 
 export interface RouteDefaultsInput {
   readonly headers?: Record<string, string>
-  readonly limits?: ModelLimits.Input
+  readonly limits?: LanguageModelLimits.Input
   readonly generation?: GenerationOptions.Input
   readonly providerOptions?: ProviderOptions
   readonly http?: HttpOptions.Input
@@ -91,14 +96,17 @@ export interface RoutePatch<Body, Prepared> extends RouteDefaultsInput {
   readonly endpoint?: EndpointPatch<Body>
 }
 
-type RouteMappedModelInput = RouteModelInput | RouteRoutedModelInput
+type RouteMappedLanguageModelInput = RouteLanguageModelInput | RouteRoutedLanguageModelInput
 
-const makeRouteModel = (route: AnyRoute, mapped: RouteMappedModelInput) => {
+const makeRouteLanguageModel = <Options extends ProviderOptions = ProviderOptions>(
+  route: AnyRoute,
+  mapped: RouteMappedLanguageModelInput,
+) => {
   const provider = route.provider ?? ("provider" in mapped ? mapped.provider : undefined)
   if (!provider) throw new Error(`Route.model(${route.id}) requires a provider`)
   if (!endpointBaseURL(route.endpoint))
     throw new Error(`Route.model(${route.id}) requires an endpoint baseURL — configure it on the route first`)
-  return Model.make({
+  return LanguageModel.make<Options>({
     ...mapped,
     provider,
     route,
@@ -111,7 +119,7 @@ const mergeRouteDefaults = (base: RouteDefaults | undefined, patch: RouteDefault
     ...base,
     ...patch,
     headers,
-    limits: patch.limits === undefined ? base?.limits : ModelLimits.make(patch.limits),
+    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(
@@ -142,27 +150,20 @@ export const httpOptions = (input: HttpOptionsInput | undefined) => {
 }
 
 export interface Interface {
-  /**
-   * Compile a request through protocol body construction, validation, and HTTP
-   * preparation without sending it. Returns the prepared request including the
-   * provider-native body.
-   *
-   * Pass a `Body` type argument to statically expose the route's body
-   * shape (e.g. `prepare<OpenAIChatBody>(...)`) — the runtime body is
-   * identical, so this is a type-level assertion the caller makes about which
-   * route the request will resolve to.
-   */
-  readonly prepare: <Body = unknown>(request: LLMRequest) => Effect.Effect<PreparedRequestOf<Body>, LLMError>
   readonly stream: StreamMethod
   readonly generate: GenerateMethod
 }
 
+export interface StreamOptions {
+  readonly transform?: HttpRequestTransform
+}
+
 export interface StreamMethod {
-  (request: LLMRequest): Stream.Stream<LLMEvent, LLMError>
+  (request: LLMRequest, options?: StreamOptions): Stream.Stream<LLMEvent, AIError>
 }
 
 export interface GenerateMethod {
-  (request: LLMRequest): Effect.Effect<LLMResponse, LLMError>
+  (request: LLMRequest, options?: StreamOptions): Effect.Effect<LLMResponse, AIError>
 }
 
 export class Service extends Context.Service<Service, Interface>()("@opencode/LLMClient") {}
@@ -226,11 +227,11 @@ export interface MakeTransportInput<Body, Prepared, Frame, Event, State> {
 
 const streamError = (route: string, message: string, cause: Cause.Cause<unknown>) => {
   const failed = cause.reasons.find(Cause.isFailReason)?.error
-  if (failed instanceof LLMErrorClass) return failed
+  if (failed instanceof AIError) return failed
   return ProviderShared.eventError(route, message, Cause.pretty(cause))
 }
 
-const requireTerminalEvent = (route: string) => (events: Stream.Stream<LLMEvent, LLMError>) =>
+const requireTerminalEvent = (route: string) => (events: Stream.Stream<LLMEvent, AIError>) =>
   Stream.suspend(() => {
     let terminal = false
     return events.pipe(
@@ -296,8 +297,9 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
           defaults: mergeRouteDefaults(route.defaults, defaults),
         })
       },
-      model: (input) => makeRouteModel(route, input),
-      prepareTransport: (body, request) =>
+      model: <Options extends ProviderOptions = ProviderOptions>(input: RouteMappedLanguageModelInput) =>
+        makeRouteLanguageModel<Options>(route, input),
+      prepareTransport: (body, request, options) =>
         routeInput.transport.prepare({
           body,
           request,
@@ -305,6 +307,7 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
           auth: routeInput.auth ?? Auth.none,
           encodeBody,
           headers: routeInput.headers,
+          transform: options?.transform,
         }),
       streamPrepared: (prepared: Prepared, request: LLMRequest, runtime: TransportRuntime) => {
         const route = `${request.model.provider}/${request.model.route.id}`
@@ -370,17 +373,14 @@ export function make<Body, Prepared, Frame, Event, State>(
   })
 }
 
-// `compile` is the important boundary: it turns a common `LLMRequest` into a
-// validated provider body plus transport-private prepared data, but does not
-// execute transport.
-const compile = Effect.fn("LLM.compile")(function* (request: LLMRequest) {
+const compile = Effect.fn("LLM.compile")(function* (request: LLMRequest, options?: StreamOptions) {
   const resolved = applyCachePolicy(resolveRequestOptions(request))
   const route = resolved.model.route
 
   const body = yield* route.body
     .from(resolved)
     .pipe(Effect.flatMap(ProviderShared.validateWith(Schema.decodeUnknownEffect(route.body.schema))))
-  const prepared = yield* route.prepareTransport(body, resolved)
+  const prepared = yield* route.prepareTransport(body, resolved, options)
 
   return {
     request: resolved,
@@ -390,30 +390,30 @@ const compile = Effect.fn("LLM.compile")(function* (request: LLMRequest) {
   }
 })
 
-const prepareWith = Effect.fn("LLMClient.prepare")(function* (request: LLMRequest) {
+/** @internal Test-only projection of the execution compiler; not exported from package barrels. */
+export const compileRequest = Effect.fn("LLM.compileRequest")(function* (request: LLMRequest) {
   const compiled = yield* compile(request)
-
-  return new PreparedRequest({
+  return {
     id: compiled.request.id ?? "request",
     route: compiled.route.id,
     protocol: compiled.route.protocol,
     model: compiled.request.model,
     body: compiled.body,
     metadata: { transport: compiled.route.transport.id },
-  })
+  }
 })
 
-const streamRequestWith = (runtime: TransportRuntime) => (request: LLMRequest) =>
+const streamRequestWith = (runtime: TransportRuntime) => (request: LLMRequest, options?: StreamOptions) =>
   Stream.unwrap(
     Effect.gen(function* () {
-      const compiled = yield* compile(request)
+      const compiled = yield* compile(request, options)
       return compiled.route.streamPrepared(compiled.prepared, compiled.request, runtime)
     }),
   )
 
 const generateWith = (stream: Interface["stream"]) =>
-  Effect.fn("LLM.generate")(function* (request: LLMRequest) {
-    const state = yield* stream(request).pipe(Stream.runFold(LLMResponse.empty, LLMResponse.reduce))
+  Effect.fn("LLM.generate")(function* (request: LLMRequest, options?: StreamOptions) {
+    const state = yield* stream(request, options).pipe(Stream.runFold(LLMResponse.empty, LLMResponse.reduce))
     const response = LLMResponse.complete(state)
     if (response) return response
     return yield* ProviderShared.eventError(
@@ -422,27 +422,24 @@ const generateWith = (stream: Interface["stream"]) =>
     )
   })
 
-export const prepare = <Body = unknown>(request: LLMRequest) =>
-  prepareWith(request) as Effect.Effect<PreparedRequestOf<Body>, LLMError>
-
-export function stream(request: LLMRequest): Stream.Stream<LLMEvent, LLMError> {
+export function stream(request: LLMRequest, options?: StreamOptions): Stream.Stream<LLMEvent, AIError> {
   return Stream.unwrap(
     Effect.gen(function* () {
-      return (yield* Service).stream(request)
+      return (yield* Service).stream(request, options)
     }),
-  ) as Stream.Stream<LLMEvent, LLMError>
+  ) as Stream.Stream<LLMEvent, AIError>
 }
 
-export function generate(request: LLMRequest): Effect.Effect<LLMResponse, LLMError> {
+export function generate(request: LLMRequest, options?: StreamOptions): Effect.Effect<LLMResponse, AIError> {
   return Effect.gen(function* () {
-    return yield* (yield* Service).generate(request)
-  }) as Effect.Effect<LLMResponse, LLMError>
+    return yield* (yield* Service).generate(request, options)
+  }) as Effect.Effect<LLMResponse, AIError>
 }
 
-export const streamRequest = (request: LLMRequest) =>
+export const streamRequest = (request: LLMRequest, options?: StreamOptions) =>
   Stream.unwrap(
     Effect.gen(function* () {
-      return (yield* Service).stream(request)
+      return (yield* Service).stream(request, options)
     }),
   )
 
@@ -453,7 +450,7 @@ export const layer: Layer.Layer<Service, never, RequestExecutor.Service> = Layer
       http: yield* RequestExecutor.Service,
       webSocket: Option.getOrUndefined(yield* Effect.serviceOption(WebSocketExecutor.Service)),
     })
-    return Service.of({ prepare: prepareWith as Interface["prepare"], stream, generate: generateWith(stream) })
+    return Service.of({ stream, generate: generateWith(stream) })
   }),
 )
 
@@ -462,7 +459,6 @@ export const Route = { make } as const
 export const LLMClient = {
   Service,
   layer,
-  prepare,
   stream,
   generate,
 } as const

packages/ai/src/route/executor.ts

@@ -12,7 +12,7 @@ import {
   HttpRateLimitDetails,
   HttpRequestDetails,
   HttpResponseDetails,
-  LLMError,
+  AIError,
   TransportReason,
 } from "../schema"
 import { classifyProviderFailure } from "../provider-error"
@@ -20,10 +20,10 @@ import { classifyProviderFailure } from "../provider-error"
 export interface Interface {
   readonly execute: (
     request: HttpClientRequest.HttpClientRequest,
-  ) => Effect.Effect<HttpClientResponse.HttpClientResponse, LLMError>
+  ) => Effect.Effect<HttpClientResponse.HttpClientResponse, AIError>
 }
 
-export class Service extends Context.Service<Service, Interface>()("@opencode/LLM/RequestExecutor") {}
+export class Service extends Context.Service<Service, Interface>()("@opencode/AI/RequestExecutor") {}
 
 const BODY_LIMIT = 16_384
 const REDACTED = "<redacted>"
@@ -220,7 +220,7 @@ const statusError =
       const retryAfter = retryAfterMs(headers)
       const rateLimit = rateLimitDetails(headers, retryAfter)
       const details = responseBody(body, request)
-      return yield* new LLMError({
+      return yield* new AIError({
         module: "RequestExecutor",
         method: "execute",
         reason: classifyProviderFailure({
@@ -246,7 +246,7 @@ const toHttpError = (redactedNames: ReadonlyArray<string | RegExp>) => (error: u
     readonly kind?: string | undefined
     readonly request?: HttpClientRequest.HttpClientRequest | undefined
   }) =>
-    new LLMError({
+    new AIError({
       module: "RequestExecutor",
       method: "execute",
       reason: new TransportReason({

packages/ai/src/route/framing.ts

@@ -1,6 +1,6 @@
 import type { Stream } from "effect"
 import * as ProviderShared from "../protocols/shared"
-import type { LLMError } from "../schema"
+import type { AIError } from "../schema"
 
 /**
  * Decode a streaming HTTP response body into provider-protocol frames.
@@ -18,7 +18,7 @@ import type { LLMError } from "../schema"
  */
 export interface Definition<Frame> {
   readonly id: string
-  readonly frame: (bytes: Stream.Stream<Uint8Array, LLMError>) => Stream.Stream<Frame, LLMError>
+  readonly frame: (bytes: Stream.Stream<Uint8Array, AIError>) => Stream.Stream<Frame, AIError>
 }
 
 /** Server-Sent Events framing. Used by every JSON-streaming HTTP provider. */

packages/ai/src/route/index.ts

@@ -1,13 +1,14 @@
 export { Route, LLMClient } from "./client"
 export type {
   Route as RouteShape,
-  RouteModelInput,
-  RouteRoutedModelInput,
+  RouteLanguageModelInput,
+  RouteRoutedLanguageModelInput,
   RouteDefaults,
   RouteDefaultsInput,
   AnyRoute,
   Interface as LLMClientShape,
   Service as LLMClientService,
+  StreamOptions,
 } from "./client"
 export * from "./executor"
 export { Auth } from "./auth"
@@ -22,4 +23,4 @@ export type { ApiKeyMode, AuthOverride, ProviderAuthOption } from "./auth-option
 export type { Definition as EndpointFn, EndpointInput } from "./endpoint"
 export type { Definition as FramingDef } from "./framing"
 export type { Protocol as ProtocolDef } from "./protocol"
-export type { Transport as TransportDef, TransportRuntime } from "./transport"
+export type { HttpRequest, HttpRequestTransform, Transport as TransportDef, TransportRuntime } from "./transport"

packages/ai/src/route/protocol.ts

@@ -1,5 +1,5 @@
 import { Schema, type Effect } from "effect"
-import type { LLMError, LLMEvent, LLMRequest, ProtocolID } from "../schema"
+import type { AIError, LLMEvent, LLMRequest, ProtocolID } from "../schema"
 
 /**
  * The semantic API contract of one model server family.
@@ -47,7 +47,7 @@ export interface ProtocolBody<Body> {
   /** Schema for the validated provider-native body sent as the JSON request. */
   readonly schema: Schema.Codec<Body, unknown>
   /** Build the provider-native body from a common `LLMRequest`. */
-  readonly from: (request: LLMRequest) => Effect.Effect<Body, LLMError>
+  readonly from: (request: LLMRequest) => Effect.Effect<Body, AIError>
 }
 
 export interface ProtocolStream<Frame, Event, State> {
@@ -56,7 +56,7 @@ export interface ProtocolStream<Frame, Event, State> {
   /** Initial parser state. Called once per response with the resolved request. */
   readonly initial: (request: LLMRequest) => State
   /** Translate one event into emitted `LLMEvent`s plus the next state. */
-  readonly step: (state: State, event: Event) => Effect.Effect<readonly [State, ReadonlyArray<LLMEvent>], LLMError>
+  readonly step: (state: State, event: Event) => Effect.Effect<readonly [State, ReadonlyArray<LLMEvent>], AIError>
   /** Optional request-completion signal for transports that do not end naturally. */
   readonly terminal?: (event: Event) => boolean
   /** Optional flush emitted when the framed stream ends. */

packages/ai/src/route/transport/http.ts

@@ -28,57 +28,9 @@ const applyQuery = (url: string, query: Record<string, string> | undefined) => {
   return next.toString()
 }
 
-const PROTOCOL_BODY_OVERLAY_DENYLIST = new Set([
-  "anthropic_version",
-  "content",
-  "contents",
-  "frequencyPenalty",
-  "frequency_penalty",
-  "generationConfig",
-  "inferenceConfig",
-  "input",
-  "maxTokens",
-  "max_tokens",
-  "messages",
-  "model",
-  "presencePenalty",
-  "presence_penalty",
-  "responseFormat",
-  "response_format",
-  "seed",
-  "stop",
-  "stopSequences",
-  "stop_sequences",
-  "stream",
-  "streamOptions",
-  "stream_options",
-  "system",
-  "systemInstruction",
-  "system_instruction",
-  "temperature",
-  "thinking",
-  "toolChoice",
-  "toolConfig",
-  "tool_choice",
-  "tool_config",
-  "tools",
-  "topK",
-  "topP",
-  "top_k",
-  "top_p",
-])
-
-const forbiddenBodyOverlayKeys = (body: Record<string, unknown>) =>
-  Object.keys(body).filter((key) => PROTOCOL_BODY_OVERLAY_DENYLIST.has(key))
-
 const bodyWithOverlay = <Body>(body: Body, request: LLMRequest, encodeBody: (body: Body) => string) =>
   Effect.gen(function* () {
     if (request.http?.body === undefined) return { jsonBody: body, bodyText: encodeBody(body) }
-    const forbiddenKeys = forbiddenBodyOverlayKeys(request.http.body)
-    if (forbiddenKeys.length > 0)
-      return yield* ProviderShared.invalidRequest(
-        `http.body cannot overlay protocol-owned field(s): ${forbiddenKeys.join(", ")}`,
-      )
     if (ProviderShared.isRecord(body)) {
       const overlaid = mergeJsonRecords(body, request.http.body) ?? {}
       return { jsonBody: overlaid, bodyText: ProviderShared.encodeJson(overlaid) }
@@ -120,14 +72,19 @@ export const httpJson = <Body, Frame>(input: HttpJsonInput<Body, Frame>): HttpJs
   id: "http-json",
   with: (patch) => httpJson({ ...input, ...patch }),
   prepare: (prepareInput) =>
-    jsonRequestParts({
-      ...prepareInput,
-    }).pipe(
-      Effect.map((parts) => ({
-        request: ProviderShared.jsonPost({ url: parts.url, body: parts.bodyText, headers: parts.headers }),
+    Effect.gen(function* () {
+      const parts = yield* jsonRequestParts({ ...prepareInput })
+      const request = { url: parts.url, method: "POST", headers: { ...parts.headers }, body: parts.bodyText }
+      yield* (prepareInput.transform?.(request) ?? Effect.void)
+      return {
+        request: ProviderShared.jsonPost({
+          url: request.url,
+          body: request.body ?? "",
+          headers: Headers.fromInput(request.headers),
+        }),
         framing: input.framing,
-      })),
-    ),
+      }
+    }),
   frames: (prepared, request, runtime) =>
     Stream.unwrap(
       runtime.http

packages/ai/src/route/transport/index.ts

@@ -3,21 +3,26 @@ import { Endpoint } from "../endpoint"
 import { Auth } from "../auth"
 import type { Interface as RequestExecutorInterface } from "../executor"
 import type { Interface as WebSocketExecutorInterface } from "./websocket"
-import type { LLMError, LLMRequest } from "../../schema"
+import type { AIError, LLMRequest } from "../../schema"
 
 export interface TransportRuntime {
   readonly http: RequestExecutorInterface
   readonly webSocket?: WebSocketExecutorInterface
 }
 
+export interface HttpRequest {
+  url: string
+  readonly method: string
+  headers: Record<string, string>
+  body: string | undefined
+}
+
+export type HttpRequestTransform = (request: HttpRequest) => Effect.Effect<void>
+
 export interface Transport<Body, Prepared, Frame> {
   readonly id: string
-  readonly prepare: (input: TransportPrepareInput<Body>) => Effect.Effect<Prepared, LLMError>
-  readonly frames: (
-    prepared: Prepared,
-    request: LLMRequest,
-    runtime: TransportRuntime,
-  ) => Stream.Stream<Frame, LLMError>
+  readonly prepare: (input: TransportPrepareInput<Body>) => Effect.Effect<Prepared, AIError>
+  readonly frames: (prepared: Prepared, request: LLMRequest, runtime: TransportRuntime) => Stream.Stream<Frame, AIError>
 }
 
 export interface TransportPrepareInput<Body> {
@@ -27,6 +32,7 @@ export interface TransportPrepareInput<Body> {
   readonly auth: Auth.Definition
   readonly encodeBody: (body: Body) => string
   readonly headers?: (input: { readonly request: LLMRequest }) => Record<string, string>
+  readonly transform?: HttpRequestTransform
 }
 
 export * as HttpTransport from "./http"

packages/ai/src/route/transport/websocket.ts

@@ -1,6 +1,6 @@
 import { Cause, Context, Effect, Layer, Queue, Stream } from "effect"
 import { Headers } from "effect/unstable/http"
-import { LLMError, TransportReason } from "../../schema"
+import { AIError, TransportReason } from "../../schema"
 import * as HttpTransport from "./http"
 import type { Transport } from "./index"
 
@@ -10,13 +10,13 @@ export interface WebSocketRequest {
 }
 
 export interface WebSocketConnection {
-  readonly sendText: (message: string) => Effect.Effect<void, LLMError>
-  readonly messages: Stream.Stream<string | Uint8Array, LLMError>
+  readonly sendText: (message: string) => Effect.Effect<void, AIError>
+  readonly messages: Stream.Stream<string | Uint8Array, AIError>
   readonly close: Effect.Effect<void, never>
 }
 
 export interface Interface {
-  readonly open: (input: WebSocketRequest) => Effect.Effect<WebSocketConnection, LLMError>
+  readonly open: (input: WebSocketRequest) => Effect.Effect<WebSocketConnection, AIError>
 }
 
 type WebSocketConstructorWithHeaders = new (
@@ -24,14 +24,14 @@ type WebSocketConstructorWithHeaders = new (
   options?: { readonly headers?: Headers.Headers },
 ) => globalThis.WebSocket
 
-export class Service extends Context.Service<Service, Interface>()("@opencode/LLM/WebSocketExecutor") {}
+export class Service extends Context.Service<Service, Interface>()("@opencode/AI/WebSocketExecutor") {}
 
 const transportError = (
   method: string,
   message: string,
   input: { readonly url?: string; readonly kind?: string } = {},
 ) =>
-  new LLMError({
+  new AIError({
     module: "WebSocketExecutor",
     method,
     reason: new TransportReason({ message, url: input.url, kind: input.kind }),
@@ -59,7 +59,7 @@ const waitOpen = (ws: globalThis.WebSocket, input: WebSocketRequest) => {
       }),
     )
   }
-  return Effect.callback<void, LLMError>((resume, signal) => {
+  return Effect.callback<void, AIError>((resume, signal) => {
     const cleanup = () => {
       ws.removeEventListener("open", onOpen)
       ws.removeEventListener("error", onError)
@@ -138,10 +138,10 @@ export const layer: Layer.Layer<Service> = Layer.succeed(Service, Service.of({ o
 export const fromWebSocket = (
   ws: globalThis.WebSocket,
   input: WebSocketRequest,
-): Effect.Effect<WebSocketConnection, LLMError> =>
+): Effect.Effect<WebSocketConnection, AIError> =>
   Effect.gen(function* () {
     yield* waitOpen(ws, input)
-    const messages = yield* Queue.bounded<string | Uint8Array, LLMError | Cause.Done<void>>(128)
+    const messages = yield* Queue.bounded<string | Uint8Array, AIError | Cause.Done<void>>(128)
 
     const onMessage = (event: MessageEvent) => {
       if (typeof event.data === "string") return Queue.offerUnsafe(messages, event.data)
@@ -213,7 +213,7 @@ export interface JsonPrepared {
 }
 
 export interface JsonInput<Body, Message> {
-  readonly toMessage: (body: Body | Record<string, unknown>) => Effect.Effect<Message, LLMError>
+  readonly toMessage: (body: Body | Record<string, unknown>) => Effect.Effect<Message, AIError>
   readonly encodeMessage: (message: Message) => string
 }
 

packages/ai/src/schema/errors.ts

@@ -2,28 +2,28 @@ import { Schema } from "effect"
 import { Tool } from "@opencode-ai/schema/tool"
 import { ModelID, ProviderID, ProviderMetadata, RouteID } from "./ids"
 
-export const ProviderFailureClassification = Schema.Literal("context-overflow")
+export const ProviderFailureClassification = Schema.Literals(["context-overflow", "payload-too-large"])
 export type ProviderFailureClassification = typeof ProviderFailureClassification.Type
 
-export class HttpRequestDetails extends Schema.Class<HttpRequestDetails>("LLM.HttpRequestDetails")({
+export class HttpRequestDetails extends Schema.Class<HttpRequestDetails>("AI.HttpRequestDetails")({
   method: Schema.String,
   url: Schema.String,
   headers: Schema.Record(Schema.String, Schema.String),
 }) {}
 
-export class HttpResponseDetails extends Schema.Class<HttpResponseDetails>("LLM.HttpResponseDetails")({
+export class HttpResponseDetails extends Schema.Class<HttpResponseDetails>("AI.HttpResponseDetails")({
   status: Schema.Number,
   headers: Schema.Record(Schema.String, Schema.String),
 }) {}
 
-export class HttpRateLimitDetails extends Schema.Class<HttpRateLimitDetails>("LLM.HttpRateLimitDetails")({
+export class HttpRateLimitDetails extends Schema.Class<HttpRateLimitDetails>("AI.HttpRateLimitDetails")({
   retryAfterMs: Schema.optional(Schema.Number),
   limit: Schema.optional(Schema.Record(Schema.String, Schema.String)),
   remaining: Schema.optional(Schema.Record(Schema.String, Schema.String)),
   reset: Schema.optional(Schema.Record(Schema.String, Schema.String)),
 }) {}
 
-export class HttpContext extends Schema.Class<HttpContext>("LLM.HttpContext")({
+export class HttpContext extends Schema.Class<HttpContext>("AI.HttpContext")({
   request: HttpRequestDetails,
   response: Schema.optional(HttpResponseDetails),
   body: Schema.optional(Schema.String),
@@ -32,7 +32,7 @@ export class HttpContext extends Schema.Class<HttpContext>("LLM.HttpContext")({
   rateLimit: Schema.optional(HttpRateLimitDetails),
 }) {}
 
-export class InvalidRequestReason extends Schema.Class<InvalidRequestReason>("LLM.Error.InvalidRequest")({
+export class InvalidRequestReason extends Schema.Class<InvalidRequestReason>("AI.Error.InvalidRequest")({
   _tag: Schema.tag("InvalidRequest"),
   message: Schema.String,
   parameter: Schema.optional(Schema.String),
@@ -41,18 +41,18 @@ export class InvalidRequestReason extends Schema.Class<InvalidRequestReason>("LL
   http: Schema.optional(HttpContext),
 }) {}
 
-export class NoRouteReason extends Schema.Class<NoRouteReason>("LLM.Error.NoRoute")({
+export class NoRouteReason extends Schema.Class<NoRouteReason>("AI.Error.NoRoute")({
   _tag: Schema.tag("NoRoute"),
   route: RouteID,
   provider: ProviderID,
   model: ModelID,
 }) {
   get message() {
-    return `No LLM route for ${this.provider}/${this.model} using ${this.route}`
+    return `No AI route for ${this.provider}/${this.model} using ${this.route}`
   }
 }
 
-export class AuthenticationReason extends Schema.Class<AuthenticationReason>("LLM.Error.Authentication")({
+export class AuthenticationReason extends Schema.Class<AuthenticationReason>("AI.Error.Authentication")({
   _tag: Schema.tag("Authentication"),
   message: Schema.String,
   kind: Schema.Literals(["missing", "invalid", "expired", "insufficient-permissions", "unknown"]),
@@ -60,7 +60,7 @@ export class AuthenticationReason extends Schema.Class<AuthenticationReason>("LL
   http: Schema.optional(HttpContext),
 }) {}
 
-export class RateLimitReason extends Schema.Class<RateLimitReason>("LLM.Error.RateLimit")({
+export class RateLimitReason extends Schema.Class<RateLimitReason>("AI.Error.RateLimit")({
   _tag: Schema.tag("RateLimit"),
   message: Schema.String,
   retryAfterMs: Schema.optional(Schema.Number),
@@ -69,21 +69,21 @@ export class RateLimitReason extends Schema.Class<RateLimitReason>("LLM.Error.Ra
   http: Schema.optional(HttpContext),
 }) {}
 
-export class QuotaExceededReason extends Schema.Class<QuotaExceededReason>("LLM.Error.QuotaExceeded")({
+export class QuotaExceededReason extends Schema.Class<QuotaExceededReason>("AI.Error.QuotaExceeded")({
   _tag: Schema.tag("QuotaExceeded"),
   message: Schema.String,
   providerMetadata: Schema.optional(ProviderMetadata),
   http: Schema.optional(HttpContext),
 }) {}
 
-export class ContentPolicyReason extends Schema.Class<ContentPolicyReason>("LLM.Error.ContentPolicy")({
+export class ContentPolicyReason extends Schema.Class<ContentPolicyReason>("AI.Error.ContentPolicy")({
   _tag: Schema.tag("ContentPolicy"),
   message: Schema.String,
   providerMetadata: Schema.optional(ProviderMetadata),
   http: Schema.optional(HttpContext),
 }) {}
 
-export class ProviderInternalReason extends Schema.Class<ProviderInternalReason>("LLM.Error.ProviderInternal")({
+export class ProviderInternalReason extends Schema.Class<ProviderInternalReason>("AI.Error.ProviderInternal")({
   _tag: Schema.tag("ProviderInternal"),
   message: Schema.String,
   status: Schema.optional(Schema.Number),
@@ -92,7 +92,7 @@ export class ProviderInternalReason extends Schema.Class<ProviderInternalReason>
   http: Schema.optional(HttpContext),
 }) {}
 
-export class TransportReason extends Schema.Class<TransportReason>("LLM.Error.Transport")({
+export class TransportReason extends Schema.Class<TransportReason>("AI.Error.Transport")({
   _tag: Schema.tag("Transport"),
   message: Schema.String,
   kind: Schema.optional(Schema.String),
@@ -101,7 +101,7 @@ export class TransportReason extends Schema.Class<TransportReason>("LLM.Error.Tr
 }) {}
 
 export class InvalidProviderOutputReason extends Schema.Class<InvalidProviderOutputReason>(
-  "LLM.Error.InvalidProviderOutput",
+  "AI.Error.InvalidProviderOutput",
 )({
   _tag: Schema.tag("InvalidProviderOutput"),
   message: Schema.String,
@@ -110,7 +110,7 @@ export class InvalidProviderOutputReason extends Schema.Class<InvalidProviderOut
   providerMetadata: Schema.optional(ProviderMetadata),
 }) {}
 
-export class UnknownProviderReason extends Schema.Class<UnknownProviderReason>("LLM.Error.UnknownProvider")({
+export class UnknownProviderReason extends Schema.Class<UnknownProviderReason>("AI.Error.UnknownProvider")({
   _tag: Schema.tag("UnknownProvider"),
   message: Schema.String,
   status: Schema.optional(Schema.Number),
@@ -118,7 +118,7 @@ export class UnknownProviderReason extends Schema.Class<UnknownProviderReason>("
   http: Schema.optional(HttpContext),
 }) {}
 
-export const LLMErrorReason = Schema.Union([
+export const AIErrorReason = Schema.Union([
   InvalidRequestReason,
   NoRouteReason,
   AuthenticationReason,
@@ -130,12 +130,12 @@ export const LLMErrorReason = Schema.Union([
   InvalidProviderOutputReason,
   UnknownProviderReason,
 ]).pipe(Schema.toTaggedUnion("_tag"))
-export type LLMErrorReason = Schema.Schema.Type<typeof LLMErrorReason>
+export type AIErrorReason = Schema.Schema.Type<typeof AIErrorReason>
 
-export class LLMError extends Schema.TaggedErrorClass<LLMError>()("LLM.Error", {
+export class AIError extends Schema.TaggedErrorClass<AIError>()("AI.Error", {
   module: Schema.String,
   method: Schema.String,
-  reason: LLMErrorReason,
+  reason: AIErrorReason,
 }) {
   override readonly cause = this.reason
 

packages/ai/src/schema/events.ts

@@ -1,6 +1,5 @@
 import { Schema } from "effect"
-import { ContentBlockID, FinishReason, ProtocolID, ProviderMetadata, RouteID, ToolCallID } from "./ids"
-import { ModelSchema } from "./options"
+import { ContentBlockID, FinishReason, ProviderMetadata, ToolCallID } from "./ids"
 import { Message, ToolCallPart, ToolOutput, ToolResultPart, ToolResultValue, type ContentPart } from "./messages"
 import { ProviderFailureClassification } from "./errors"
 
@@ -49,7 +48,7 @@ import { ProviderFailureClassification } from "./errors"
  * — for fields we don't normalize and for billing-level audit trails.
  * Matches the same escape-hatch field on `LLMEvent`.
  */
-export class Usage extends Schema.Class<Usage>("LLM.Usage")({
+export class Usage extends Schema.Class<Usage>("AI.Usage")({
   inputTokens: Schema.optional(Schema.Number),
   outputTokens: Schema.optional(Schema.Number),
   nonCachedInputTokens: Schema.optional(Schema.Number),
@@ -314,29 +313,6 @@ export const LLMEvent = Object.assign(llmEventTagged, {
 })
 export type LLMEvent = Schema.Schema.Type<typeof llmEventTagged>
 
-export class PreparedRequest extends Schema.Class<PreparedRequest>("LLM.PreparedRequest")({
-  id: Schema.String,
-  route: RouteID,
-  protocol: ProtocolID,
-  model: ModelSchema,
-  body: Schema.Unknown,
-  metadata: Schema.optional(Schema.Record(Schema.String, Schema.Unknown)),
-}) {}
-
-/**
- * A `PreparedRequest` whose `body` is typed as `Body`. Use with the generic
- * on `LLMClient.prepare<Body>(...)` when the caller knows which route their
- * request will resolve to and wants its native shape statically exposed
- * (debug UIs, request previews, plan rendering).
- *
- * The runtime body is identical — the route still emits `body: unknown` — so
- * this is a type-level assertion the caller makes about what they expect to
- * find. The prepare runtime does not validate the assertion.
- */
-export type PreparedRequestOf<Body> = Omit<PreparedRequest, "body"> & {
-  readonly body: Body
-}
-
 const responseText = (events: ReadonlyArray<LLMEvent>) =>
   events
     .filter(LLMEvent.is.textDelta)

packages/ai/src/schema/ids.ts

@@ -1,5 +1,6 @@
 import { Schema } from "effect"
-import { LLM, ProviderMetadata } from "@opencode-ai/schema/llm"
+import { ProviderMetadata } from "@opencode-ai/schema/ai"
+import { LLM } from "@opencode-ai/schema/llm"
 
 export { ProviderMetadata }
 
@@ -11,10 +12,10 @@ export type ProtocolID = Schema.Schema.Type<typeof ProtocolID>
 export const RouteID = Schema.String
 export type RouteID = Schema.Schema.Type<typeof RouteID>
 
-export const ModelID = Schema.String.pipe(Schema.brand("LLM.ModelID"))
+export const ModelID = Schema.String.pipe(Schema.brand("AI.ModelID"))
 export type ModelID = typeof ModelID.Type
 
-export const ProviderID = Schema.String.pipe(Schema.brand("LLM.ProviderID"))
+export const ProviderID = Schema.String.pipe(Schema.brand("AI.ProviderID"))
 export type ProviderID = typeof ProviderID.Type
 
 export const ResponseID = Schema.String

packages/ai/src/schema/messages.ts

@@ -1,7 +1,7 @@
 import { Schema } from "effect"
 import { Tool } from "@opencode-ai/schema/tool"
 import { JsonSchema, MessageRole, ProviderMetadata } from "./ids"
-import { CacheHint, CachePolicy, GenerationOptions, HttpOptions, ModelSchema, ProviderOptions } from "./options"
+import { CacheHint, CachePolicy, GenerationOptions, HttpOptions, LanguageModelSchema, ProviderOptions } from "./options"
 import { isRecord } from "../utils/record"
 
 const systemPartSchema = Schema.Struct({
@@ -124,6 +124,7 @@ export const ToolCallPart = Object.assign(
     name: Schema.String,
     input: Schema.Unknown,
     providerExecuted: Schema.optional(Schema.Boolean),
+    cache: Schema.optional(CacheHint),
     metadata: Schema.optional(Schema.Record(Schema.String, Schema.Unknown)),
     providerMetadata: Schema.optional(ProviderMetadata),
   }).annotate({ identifier: "LLM.Content.ToolCall" }),
@@ -168,6 +169,7 @@ export const ReasoningPart = Schema.Struct({
   type: Schema.Literal("reasoning"),
   text: Schema.String,
   encrypted: Schema.optional(Schema.String),
+  cache: Schema.optional(CacheHint),
   metadata: Schema.optional(Schema.Record(Schema.String, Schema.Unknown)),
   providerMetadata: Schema.optional(ProviderMetadata),
 }).annotate({ identifier: "LLM.Content.Reasoning" })
@@ -261,7 +263,7 @@ export namespace ToolChoice {
 
 export class LLMRequest extends Schema.Class<LLMRequest>("LLM.Request")({
   id: Schema.optional(Schema.String),
-  model: ModelSchema,
+  model: LanguageModelSchema,
   system: Schema.Array(SystemPart),
   messages: Schema.Array(Message),
   tools: Schema.Array(ToolDefinition),

packages/ai/src/schema/options.ts

@@ -50,7 +50,7 @@ export const mergeProviderOptions = (
   return Object.keys(result).length === 0 ? undefined : result
 }
 
-export class HttpOptions extends Schema.Class<HttpOptions>("LLM.HttpOptions")({
+export class HttpOptions extends Schema.Class<HttpOptions>("AI.HttpOptions")({
   body: Schema.optional(JsonSchema),
   headers: Schema.optional(Schema.Record(Schema.String, Schema.String)),
   query: Schema.optional(Schema.Record(Schema.String, Schema.String)),
@@ -121,31 +121,32 @@ export const mergeGenerationOptions = (...items: ReadonlyArray<GenerationOptions
   return Object.values(result).some((value) => value !== undefined) ? result : undefined
 }
 
-export class ModelLimits extends Schema.Class<ModelLimits>("LLM.ModelLimits")({
+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 ModelLimits {
-  export type Input = ModelLimits | ConstructorParameters<typeof ModelLimits>[0]
+export namespace LanguageModelLimits {
+  export type Input = LanguageModelLimits | ConstructorParameters<typeof LanguageModelLimits>[0]
 
-  /** Normalize model limit input into the canonical `ModelLimits` class. */
+  /** Normalize model limit input into the canonical `LanguageModelLimits` class. */
   export const make = (input: Input | undefined) =>
-    input instanceof ModelLimits ? input : new ModelLimits(input ?? {})
+    input instanceof LanguageModelLimits ? input : new LanguageModelLimits(input ?? {})
 }
 
-export class ModelDefaults extends Schema.Class<ModelDefaults>("LLM.ModelDefaults")({
-  limits: Schema.optional(ModelLimits),
+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),
 }) {}
 
-export namespace ModelDefaults {
+export namespace LanguageModelDefaults {
   export type Input =
-    | ModelDefaults
+    | LanguageModelDefaults
     | {
-        readonly limits?: ModelLimits.Input
+        readonly limits?: LanguageModelLimits.Input
         readonly generation?: GenerationOptions.Input
         readonly providerOptions?: ProviderOptions
         readonly http?: HttpOptions.Input
@@ -153,9 +154,9 @@ export namespace ModelDefaults {
 
   /** Normalize selected-model request defaults without applying precedence. */
   export const make = (input: Input) => {
-    if (input instanceof ModelDefaults) return input
-    return new ModelDefaults({
-      limits: input.limits === undefined ? undefined : ModelLimits.make(input.limits),
+    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),
@@ -163,29 +164,39 @@ export namespace ModelDefaults {
   }
 }
 
-export const ModelToolSchemaCompatibility = Schema.Literals(["gemini", "moonshot"])
-export type ModelToolSchemaCompatibility = Schema.Schema.Type<typeof ModelToolSchemaCompatibility>
+export const LanguageModelToolSchemaCompatibility = Schema.Literals(["gemini", "moonshot"])
+export type LanguageModelToolSchemaCompatibility = Schema.Schema.Type<typeof LanguageModelToolSchemaCompatibility>
 
-export class ModelCompatibility extends Schema.Class<ModelCompatibility>("LLM.ModelCompatibility")({
-  toolSchema: Schema.optional(ModelToolSchemaCompatibility),
+export const LanguageModelMaxTokensFieldCompatibility = Schema.Literals(["max_completion_tokens", "max_tokens"])
+export type LanguageModelMaxTokensFieldCompatibility = Schema.Schema.Type<
+  typeof LanguageModelMaxTokensFieldCompatibility
+>
+
+export class LanguageModelCompatibility extends Schema.Class<LanguageModelCompatibility>(
+  "LLM.LanguageModelCompatibility",
+)({
+  toolSchema: Schema.optional(LanguageModelToolSchemaCompatibility),
   reasoningField: Schema.optional(Schema.String),
+  maxTokensField: Schema.optional(LanguageModelMaxTokensFieldCompatibility),
 }) {}
 
-export namespace ModelCompatibility {
-  export type Input = ModelCompatibility | ConstructorParameters<typeof ModelCompatibility>[0]
+export namespace LanguageModelCompatibility {
+  export type Input = LanguageModelCompatibility | ConstructorParameters<typeof LanguageModelCompatibility>[0]
 
   /** Normalize model/upstream compatibility metadata without projecting requests. */
-  export const make = (input: Input) => (input instanceof ModelCompatibility ? input : new ModelCompatibility(input))
+  export const make = (input: Input) =>
+    input instanceof LanguageModelCompatibility ? input : new LanguageModelCompatibility(input)
 }
 
-export class Model {
+export class LanguageModel<Options extends ProviderOptions = ProviderOptions> {
+  declare protected readonly _ProviderOptions: Options
   readonly id: ModelID
   readonly provider: ProviderID
   readonly route: AnyRoute
-  readonly defaults?: ModelDefaults
-  readonly compatibility?: ModelCompatibility
+  readonly defaults?: LanguageModelDefaults
+  readonly compatibility?: LanguageModelCompatibility
 
-  constructor(input: Model.ConstructorInput) {
+  constructor(input: LanguageModel.ConstructorInput) {
     this.id = input.id
     this.provider = input.provider
     this.route = input.route
@@ -193,17 +204,18 @@ export class Model {
     this.compatibility = input.compatibility
   }
 
-  static make(input: Model.Input) {
-    return new Model({
+  static make<Options extends ProviderOptions = ProviderOptions>(input: LanguageModel.Input) {
+    return new LanguageModel<Options>({
       id: ModelID.make(input.id),
       provider: ProviderID.make(input.provider),
       route: input.route,
-      defaults: input.defaults === undefined ? undefined : ModelDefaults.make(input.defaults),
-      compatibility: input.compatibility === undefined ? undefined : ModelCompatibility.make(input.compatibility),
+      defaults: input.defaults === undefined ? undefined : LanguageModelDefaults.make(input.defaults),
+      compatibility:
+        input.compatibility === undefined ? undefined : LanguageModelCompatibility.make(input.compatibility),
     })
   }
 
-  static input(model: Model): Model.ConstructorInput {
+  static input<Options extends ProviderOptions>(model: LanguageModel<Options>): LanguageModel.ConstructorInput {
     return {
       id: model.id,
       provider: model.provider,
@@ -213,35 +225,40 @@ export class Model {
     }
   }
 
-  static update(model: Model, patch: Partial<Model.Input>) {
+  static update<Options extends ProviderOptions>(model: LanguageModel<Options>, patch: Partial<LanguageModel.Input>) {
     if (Object.keys(patch).length === 0) return model
-    return Model.make({
-      ...Model.input(model),
+    return LanguageModel.make<Options>({
+      ...LanguageModel.input(model),
       ...patch,
     })
   }
 }
 
-export namespace Model {
+export namespace LanguageModel {
   export type ConstructorInput = {
     readonly id: ModelID
     readonly provider: ProviderID
     readonly route: AnyRoute
-    readonly defaults?: ModelDefaults
-    readonly compatibility?: ModelCompatibility
+    readonly defaults?: LanguageModelDefaults
+    readonly compatibility?: LanguageModelCompatibility
   }
 
   export type Input = Omit<ConstructorInput, "id" | "provider" | "defaults" | "compatibility"> & {
     readonly id: string | ModelID
     readonly provider: string | ProviderID
-    readonly defaults?: ModelDefaults.Input
-    readonly compatibility?: ModelCompatibility.Input
+    readonly defaults?: LanguageModelDefaults.Input
+    readonly compatibility?: LanguageModelCompatibility.Input
   }
 }
 
-export type ModelInput = Model.Input
+export type LanguageModelInput = LanguageModel.Input
 
-export const ModelSchema = Schema.declare((value): value is Model => value instanceof Model, { expected: "LLM.Model" })
+export type LanguageModelProviderOptions<SelectedModel> =
+  SelectedModel extends LanguageModel<infer Options> ? Options : never
+
+export const LanguageModelSchema = Schema.declare((value): value is LanguageModel => value instanceof LanguageModel, {
+  expected: "LLM.LanguageModel",
+})
 
 export class CacheHint extends Schema.Class<CacheHint>("LLM.CacheHint")({
   type: Schema.Literals(["ephemeral", "persistent"]),

packages/ai/src/testing.ts

@@ -0,0 +1,156 @@
+export * as TestLLM from "./testing"
+
+import { LLMClient, type Interface as LLMClientShape } from "./route/client"
+import {
+  LLMEvent,
+  LLMResponse,
+  type FinishReasonDetails,
+  type AIError,
+  type LLMRequest,
+  type UsageInput,
+} from "./schema"
+import { Context, Deferred, Effect, Latch, Layer, Queue, Scope, Stream } from "effect"
+
+export type Response = readonly LLMEvent[] | Stream.Stream<LLMEvent, AIError>
+
+export type Gate = Readonly<{ started: Effect.Effect<void>; release: Effect.Effect<void> }>
+
+export interface Interface {
+  readonly requests: LLMRequest[]
+  readonly push: (...responses: readonly Response[]) => Effect.Effect<void>
+  readonly always: (response: Response) => Effect.Effect<void>
+  readonly wait: (count: number) => Effect.Effect<void>
+  readonly gate: Effect.Effect<Gate, never, Scope.Scope>
+  readonly client: LLMClientShape
+}
+
+export interface LayerOptions {
+  readonly transformRequest?: (request: LLMRequest) => LLMRequest
+  /** Used after the one-shot response queue is exhausted. Omit to defect on unexpected requests. */
+  readonly fallback?: Response
+}
+
+export class Service extends Context.Service<Service, Interface>()("@opencode/ai/TestLLM") {}
+
+export const complete = (
+  options: { readonly reason: FinishReasonDetails; readonly usage?: UsageInput },
+  ...events: readonly LLMEvent[]
+) => [
+  LLMEvent.stepStart({ index: 0 }),
+  ...events,
+  LLMEvent.stepFinish({ index: 0, reason: options.reason, usage: options.usage }),
+  LLMEvent.finish({ reason: options.reason }),
+]
+
+export const stop = (...events: readonly LLMEvent[]) => complete({ reason: { normalized: "stop" } }, ...events)
+
+export const toolCalls = (...events: readonly LLMEvent[]) =>
+  complete({ reason: { normalized: "tool-calls" } }, ...events)
+
+const textEvents = (value: string, id: string) => [
+  LLMEvent.textStart({ id }),
+  LLMEvent.textDelta({ id, text: value }),
+  LLMEvent.textEnd({ id }),
+]
+
+export const text = (value: string, id: string) => stop(...textEvents(value, id))
+
+export const textWithUsage = (value: string, id: string, inputTokens: number) =>
+  complete(
+    { reason: { normalized: "stop" }, usage: { inputTokens, nonCachedInputTokens: inputTokens } },
+    ...textEvents(value, id),
+  )
+
+export const tool = (id: string, name: string, input: unknown) => toolCalls(LLMEvent.toolCall({ id, name, input }))
+
+export const failAfter = (error: AIError, ...events: readonly LLMEvent[]) =>
+  Stream.fromIterable(events).pipe(Stream.concat(Stream.fail(error)))
+
+export const hangAfter = (...events: readonly LLMEvent[]) => Stream.concat(Stream.fromIterable(events), Stream.never)
+
+const toStream = (response: Response) => (Stream.isStream(response) ? response : Stream.fromIterable(response))
+
+export const layer = (options: LayerOptions = {}) =>
+  Layer.effect(
+    Service,
+    Effect.gen(function* () {
+      const requests: LLMRequest[] = []
+      const responses: Response[] = []
+      let started = Deferred.makeUnsafe<void>()
+      let fallback = options.fallback
+      let activeGate: { readonly started: Queue.Queue<void>; readonly release: Latch.Latch } | undefined
+      const wait = (count: number): Effect.Effect<void> =>
+        Effect.suspend(() =>
+          requests.length >= count ? Effect.void : Deferred.await(started).pipe(Effect.andThen(wait(count))),
+        )
+
+      const stream = ((request: LLMRequest) => {
+        requests.push(options.transformRequest?.(request) ?? request)
+        const waiting = started
+        started = Deferred.makeUnsafe()
+        Deferred.doneUnsafe(waiting, Effect.void)
+        const response = responses.shift() ?? fallback
+        if (!response) return Stream.die(new Error(`TestLLM has no response for request ${requests.length}`))
+        const streamed = toStream(response)
+        const gate = activeGate
+        if (!gate) return streamed
+        return Stream.unwrap(
+          Queue.offer(gate.started, undefined).pipe(Effect.andThen(gate.release.await), Effect.as(streamed)),
+        )
+      }) as LLMClientShape["stream"]
+      const client = LLMClient.Service.of({
+        stream,
+        generate: (request) =>
+          stream(request).pipe(
+            Stream.runFold(LLMResponse.empty, LLMResponse.reduce),
+            Effect.flatMap((state) => {
+              const response = LLMResponse.complete(state)
+              if (response) return Effect.succeed(response)
+              return Effect.die("TestLLM response ended without a terminal finish event")
+            }),
+          ),
+      })
+
+      return Service.of({
+        requests,
+        push: (...input) =>
+          Effect.sync(() => {
+            responses.push(...input)
+          }),
+        always: (response) =>
+          Effect.sync(() => {
+            fallback = response
+          }),
+        wait,
+        gate: Effect.gen(function* () {
+          const gate = {
+            started: yield* Effect.acquireRelease(Queue.unbounded<void>(), Queue.shutdown),
+            release: yield* Latch.make(),
+          }
+          activeGate = gate
+          const release = Effect.sync(() => {
+            if (activeGate === gate) activeGate = undefined
+          }).pipe(Effect.andThen(gate.release.open), Effect.asVoid)
+          yield* Effect.addFinalizer(() => release)
+          return {
+            started: Queue.take(gate.started),
+            release,
+          }
+        }),
+        client,
+      })
+    }),
+  )
+
+export const clientLayer = Layer.effect(
+  LLMClient.Service,
+  Effect.map(Service, (service) => service.client),
+)
+
+export const push = (...responses: readonly Response[]) => Service.use((service) => service.push(...responses))
+
+export const always = (response: Response) => Service.use((service) => service.always(response))
+
+export const wait = (count: number) => Service.use((service) => service.wait(count))
+
+export const gate = Service.use((service) => service.gate)

packages/ai/test/adapter.test.ts

@@ -2,11 +2,12 @@ import { describe, expect } from "bun:test"
 import { Effect, Schema, Stream } from "effect"
 import { LLM, LLMRequest, LLMResponse } from "../src"
 import { Route, Endpoint, LLMClient, Protocol, type FramingDef } from "../src/route"
-import { Model } from "../src/schema"
+import { compileRequest } from "../src/route/client"
+import { LanguageModel } from "../src/schema"
 import { testEffect } from "./lib/effect"
 import { dynamicResponse } from "./lib/http"
 
-const updateModel = (model: Model, patch: Partial<Model.Input>) => Model.update(model, patch)
+const updateModel = (model: LanguageModel, patch: Partial<LanguageModel.Input>) => LanguageModel.update(model, patch)
 
 const Json = Schema.fromJsonString(Schema.Unknown)
 const encodeJson = Schema.encodeSync(Json)
@@ -85,7 +86,7 @@ const configuredGemini = gemini.with({ endpoint: { baseURL: "https://fake.local"
 
 const request = LLM.request({
   id: "req_1",
-  model: Model.make({
+  model: LanguageModel.make({
     id: "fake-model",
     provider: "fake-provider",
     route: configuredFake,
@@ -139,8 +140,7 @@ describe("llm route", () => {
 
   it.effect("selects routes by model route value", () =>
     Effect.gen(function* () {
-      const llm = yield* LLMClient.Service
-      const prepared = yield* llm.prepare(
+      const prepared = yield* compileRequest(
         LLMRequest.update(request, { model: updateModel(request.model, { route: configuredGemini }) }),
       )
 
@@ -173,7 +173,7 @@ describe("llm route", () => {
         framing: fakeFraming,
       })
 
-      const prepared = yield* (yield* LLMClient.Service).prepare(
+      const prepared = yield* compileRequest(
         LLMRequest.update(request, { model: updateModel(request.model, { route: duplicate }) }),
       )
 

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

@@ -1,6 +1,6 @@
 import { Config } from "effect"
 import { Auth } from "../src/route"
-import type { ModelFactory } from "../src/route/auth-options"
+import type { LanguageModelFactory } from "../src/route/auth-options"
 import * as OpenAIChat from "../src/protocols/openai-chat"
 import * as AmazonBedrock from "../src/providers/amazon-bedrock"
 import * as Anthropic from "../src/providers/anthropic"
@@ -23,13 +23,13 @@ type BaseOptions = {
   readonly headers?: Record<string, string>
 }
 
-type Model = {
+type LanguageModel = {
   readonly id: string
 }
 
 declare const auth: Auth.Definition
-declare const optionalAuthModel: ModelFactory<BaseOptions, "optional", Model>
-declare const requiredAuthModel: ModelFactory<BaseOptions, "required", Model>
+declare const optionalAuthModel: LanguageModelFactory<BaseOptions, "optional", LanguageModel>
+declare const requiredAuthModel: LanguageModelFactory<BaseOptions, "required", LanguageModel>
 const configApiKey = Config.redacted("OPENAI_API_KEY")
 
 OpenAIChat.route.model({ id: "gpt-4.1-mini" })

packages/ai/test/auth.test.ts

@@ -4,12 +4,12 @@ import { Headers } from "effect/unstable/http"
 import { LLM } from "../src"
 import { Auth } from "../src/route/auth"
 import * as OpenAIChat from "../src/protocols/openai-chat"
-import { Model } from "../src/schema"
+import { LanguageModel } from "../src/schema"
 import { it } from "./lib/effect"
 
 const request = LLM.request({
   id: "req_auth",
-  model: Model.make({ id: "fake-model", provider: "fake", route: OpenAIChat.route }),
+  model: LanguageModel.make({ id: "fake-model", provider: "fake", route: OpenAIChat.route }),
   prompt: "hello",
 })
 

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

@@ -1,7 +1,8 @@
 import { describe, expect, test } from "bun:test"
 import { Effect } from "effect"
 import { CacheHint, LLM, Message } from "../src"
-import { Auth, LLMClient } from "../src/route"
+import { Auth } from "../src/route"
+import { compileRequest } from "../src/route/client"
 import { AmazonBedrock } from "../src/providers"
 import * as AnthropicMessages from "../src/protocols/anthropic-messages"
 import * as Gemini from "../src/protocols/gemini"
@@ -31,7 +32,7 @@ const geminiModel = Gemini.route
 describe("applyCachePolicy", () => {
   it.effect("undefined cache resolves to 'auto' (the recommended default)", () =>
     Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
         LLM.request({
           model: anthropicModel,
           system: "You are concise.",
@@ -50,7 +51,7 @@ describe("applyCachePolicy", () => {
 
   it.effect("'auto' marks the last tool, first and last system parts, and final message boundary on Anthropic", () =>
     Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
         LLM.request({
           model: anthropicModel,
           system: [
@@ -87,7 +88,7 @@ describe("applyCachePolicy", () => {
 
   it.effect("'auto' is a no-op on OpenAI (implicit caching protocol)", () =>
     Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
         LLM.request({
           model: openaiModel,
           system: "Sys",
@@ -106,7 +107,7 @@ describe("applyCachePolicy", () => {
 
   it.effect("'auto' is a no-op on Gemini (out-of-band caching protocol)", () =>
     Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
         LLM.request({
           model: geminiModel,
           system: "Sys",
@@ -123,7 +124,7 @@ describe("applyCachePolicy", () => {
 
   it.effect("'auto' on Bedrock emits cachePoint markers in the right places", () =>
     Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
         LLM.request({
           model: bedrockModel,
           system: [
@@ -157,7 +158,7 @@ describe("applyCachePolicy", () => {
 
   it.effect("'none' disables auto placement even when manual hints exist", () =>
     Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
         LLM.request({
           model: anthropicModel,
           system: "Sys",
@@ -176,7 +177,7 @@ describe("applyCachePolicy", () => {
 
   it.effect("granular object form: tools-only marks just tools", () =>
     Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
         LLM.request({
           model: anthropicModel,
           system: "Sys",
@@ -195,7 +196,7 @@ describe("applyCachePolicy", () => {
 
   it.effect("auto policy preserves manual CacheHints on other parts", () =>
     Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
         LLM.request({
           model: anthropicModel,
           system: [
@@ -241,7 +242,7 @@ describe("applyCachePolicy", () => {
       expect("cache" in tail ? tail.cache : undefined).toBeUndefined()
       expect(applyCachePolicy(applied)).toBe(applied)
 
-      const prepared = yield* LLMClient.prepare(request)
+      const prepared = yield* compileRequest(request)
 
       const body = prepared.body as {
         tools: Array<{ cache_control?: unknown }>
@@ -261,7 +262,7 @@ describe("applyCachePolicy", () => {
 
   it.effect("ttlSeconds in the policy flows through to wire markers", () =>
     Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
         LLM.request({
           model: anthropicModel,
           system: "Sys",
@@ -278,7 +279,7 @@ describe("applyCachePolicy", () => {
 
   it.effect("messages: { tail: 2 } marks the last 2 message boundaries", () =>
     Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
         LLM.request({
           model: anthropicModel,
           messages: [Message.user("u1"), Message.assistant("a1"), Message.user("u2"), Message.assistant("a2")],
@@ -296,7 +297,7 @@ describe("applyCachePolicy", () => {
 
   it.effect("'latest-assistant' marks the last assistant message", () =>
     Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
         LLM.request({
           model: anthropicModel,
           messages: [Message.user("u1"), Message.assistant("a1"), Message.user("u2")],

packages/ai/test/compile.test.ts

@@ -4,6 +4,7 @@ import { HttpClientRequest } from "effect/unstable/http"
 import { LLM, mergeProviderOptions } from "../src"
 import { AnthropicMessages, OpenAIChat } from "../src/protocols"
 import { Auth, LLMClient } from "../src/route"
+import { compileRequest } from "../src/route/client"
 import { it } from "./lib/effect"
 import { dynamicResponse } from "./lib/http"
 import { deltaChunk } from "./lib/openai-chunks"
@@ -44,7 +45,7 @@ describe("request option precedence", () => {
     })
   })
 
-  it.effect("prepares bodies with route defaults, model defaults, and call options in order", () =>
+  it.effect("compiles bodies with route defaults, model defaults, and call options in order", () =>
     Effect.gen(function* () {
       const route = OpenAIChat.route.with({
         endpoint: { baseURL: "https://api.openai.test/v1/" },
@@ -59,7 +60,7 @@ describe("request option precedence", () => {
           providerOptions: { openai: { reasoningEffort: "medium" } },
         },
       })
-      const prepared = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
+      const prepared = yield* compileRequest(
         LLM.request({
           model,
           prompt: "Say hello.",
@@ -136,24 +137,61 @@ describe("request option precedence", () => {
     ),
   )
 
-  it.effect("rejects raw body overlays for protocol-owned roots", () =>
-    Effect.gen(function* () {
-      const model = OpenAIChat.route
-        .with({ endpoint: { baseURL: "https://api.openai.test/v1/" }, auth: Auth.bearer("test") })
-        .model({ id: "gpt-4o-mini" })
-      const error = yield* LLMClient.prepare(
-        LLM.request({
-          model,
-          prompt: "Say hello.",
-          http: { body: { model: "gpt-5", messages: [], tools: [] } },
-        }),
-      ).pipe(Effect.flip)
+  it.effect("transforms the final HTTP request after serialization and authentication", () =>
+    LLMClient.generate(
+      LLM.request({
+        model: OpenAIChat.route
+          .with({ endpoint: { baseURL: "https://api.openai.test/v1/" }, auth: Auth.bearer("fresh-key") })
+          .model({ id: "gpt-4o-mini" }),
+        prompt: "Say hello.",
+      }),
+      {
+        transform: (request) =>
+          Effect.sync(() => {
+            expect(request.headers.authorization).toBe("Bearer fresh-key")
+            request.url = "https://proxy.test/v1/chat/completions"
+            request.headers["x-plugin"] = "transformed"
+            request.body = JSON.stringify({ transformed: true })
+          }),
+      },
+    ).pipe(
+      Effect.provide(
+        dynamicResponse((input) =>
+          Effect.gen(function* () {
+            const web = yield* HttpClientRequest.toWeb(input.request).pipe(Effect.orDie)
+            expect(web.url).toBe("https://proxy.test/v1/chat/completions")
+            expect(web.headers.get("x-plugin")).toBe("transformed")
+            expect(decodeJson(input.text)).toEqual({ transformed: true })
+            return input.respond(sseEvents(deltaChunk({}, "stop")), {
+              headers: { "content-type": "text/event-stream" },
+            })
+          }),
+        ),
+      ),
+    ),
+  )
 
-      expect(error.reason).toMatchObject({
-        _tag: "InvalidRequest",
-        message: "http.body cannot overlay protocol-owned field(s): model, messages, tools",
-      })
-    }),
+  it.effect("applies raw body overlays after protocol lowering", () =>
+    LLMClient.generate(
+      LLM.request({
+        model: OpenAIChat.route
+          .with({ endpoint: { baseURL: "https://api.openai.test/v1/" }, auth: Auth.bearer("test") })
+          .model({ id: "gpt-4o-mini" }),
+        prompt: "Say hello.",
+        http: { body: { model: "gpt-5", messages: [], tools: [] } },
+      }),
+    ).pipe(
+      Effect.provide(
+        dynamicResponse((input) =>
+          Effect.gen(function* () {
+            expect(decodeJson(input.text)).toMatchObject({ model: "gpt-5", messages: [], tools: [] })
+            return input.respond(sseEvents(deltaChunk({}, "stop")), {
+              headers: { "content-type": "text/event-stream" },
+            })
+          }),
+        ),
+      ),
+    ),
   )
 
   it.effect("uses model output limits after route limits and before call maxTokens", () =>
@@ -164,10 +202,8 @@ describe("request option precedence", () => {
         limits: { output: 128 },
       })
       const model = route.model({ id: "claude-sonnet-4-5", defaults: { limits: { output: 64 } } })
-      const withoutMaxTokens = yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
-        LLM.request({ model, prompt: "Say hello.", cache: "none" }),
-      )
-      const withMaxTokens = yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
+      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 } }),
       )
 

packages/ai/test/continuation-scenarios.ts

@@ -1,4 +1,12 @@
-import { LLM, Message, ToolCallPart, ToolDefinition, ToolResultPart, type ContentPart, type Model } from "../src"
+import {
+  LLM,
+  Message,
+  ToolCallPart,
+  ToolDefinition,
+  ToolResultPart,
+  type ContentPart,
+  type LanguageModel,
+} from "../src"
 
 export const basicContinuation = ["system", "user-text", "assistant-text", "user-follow-up"] as const
 export const toolContinuation = ["tool-call", "tool-result"] as const
@@ -40,7 +48,7 @@ export const continuationTool = ToolDefinition.make({
 
 export function continuationRequest(input: {
   readonly id: string
-  readonly model: Model
+  readonly model: LanguageModel
   readonly features: ReadonlyArray<ContinuationFeature>
   readonly image?: string
 }) {

packages/ai/test/endpoint.test.ts

@@ -2,11 +2,11 @@ import { describe, expect, test } from "bun:test"
 import { LLM } from "../src"
 import * as OpenAIChat from "../src/protocols/openai-chat"
 import { Endpoint } from "../src/route"
-import { Model } from "../src/schema"
+import { LanguageModel } from "../src/schema"
 
 const request = () =>
   LLM.request({
-    model: Model.make({
+    model: LanguageModel.make({
       id: "model-1",
       provider: "test",
       route: OpenAIChat.route,

packages/ai/test/executor.test.ts

@@ -1,7 +1,7 @@
 import { describe, expect } from "bun:test"
 import { Effect, Layer, Ref } from "effect"
 import { Headers, HttpClient, HttpClientRequest, HttpClientResponse } from "effect/unstable/http"
-import { LLM, LLMError } from "../src"
+import { LLM, AIError } from "../src"
 import { LLMClient, RequestExecutor } from "../src/route"
 import * as OpenAIChat from "../src/protocols/openai-chat"
 import { dynamicResponse } from "./lib/http"
@@ -58,13 +58,13 @@ const countedResponsesLayer = (attempts: Ref.Ref<number>, responses: ReadonlyArr
     ),
   )
 
-const expectLLMError = (error: unknown) => {
-  expect(error).toBeInstanceOf(LLMError)
-  if (!(error instanceof LLMError)) throw new Error("expected LLMError")
+const expectAIError = (error: unknown) => {
+  expect(error).toBeInstanceOf(AIError)
+  if (!(error instanceof AIError)) throw new Error("expected AIError")
   return error
 }
 
-const errorHttp = (error: LLMError) => ("http" in error.reason ? error.reason.http : undefined)
+const errorHttp = (error: AIError) => ("http" in error.reason ? error.reason.http : undefined)
 
 describe("RequestExecutor", () => {
   it.effect("classifies context overflow responses", () =>
@@ -72,7 +72,7 @@ describe("RequestExecutor", () => {
       const executor = yield* RequestExecutor.Service
       const error = yield* executor.execute(request).pipe(Effect.flip)
 
-      expectLLMError(error)
+      expectAIError(error)
       expect(error.reason).toMatchObject({ _tag: "InvalidRequest", classification: "context-overflow" })
     }).pipe(
       Effect.provide(
@@ -85,14 +85,17 @@ describe("RequestExecutor", () => {
     ),
   )
 
-  it.effect("does not classify generic HTTP 413 payload errors as context overflow", () =>
+  it.effect("classifies generic HTTP 413 payload errors", () =>
     Effect.gen(function* () {
       const executor = yield* RequestExecutor.Service
       const error = yield* executor.execute(request).pipe(Effect.flip)
 
-      expectLLMError(error)
-      expect(error.reason).toMatchObject({ _tag: "InvalidRequest" })
-      expect("classification" in error.reason ? error.reason.classification : undefined).toBeUndefined()
+      expectAIError(error)
+      expect(error.reason).toMatchObject({
+        _tag: "InvalidRequest",
+        classification: "payload-too-large",
+        http: { response: { status: 413 } },
+      })
     }).pipe(Effect.provide(responsesLayer([new Response("request too large", { status: 413 })]))),
   )
 
@@ -101,7 +104,7 @@ describe("RequestExecutor", () => {
       const executor = yield* RequestExecutor.Service
       const error = yield* executor.execute(request).pipe(Effect.flip)
 
-      expectLLMError(error)
+      expectAIError(error)
       expect(error.reason).toMatchObject({ _tag: "InvalidRequest" })
       expect("classification" in error.reason ? error.reason.classification : undefined).toBeUndefined()
     }).pipe(Effect.provide(responsesLayer([new Response("invalid parameter", { status: 400 })]))),
@@ -114,7 +117,7 @@ describe("RequestExecutor", () => {
           const executor = yield* RequestExecutor.Service
           const error = yield* executor.execute(request).pipe(Effect.flip)
 
-          expectLLMError(error)
+          expectAIError(error)
           expect(error.reason).toMatchObject({ _tag: "RateLimit" })
         }).pipe(Effect.provide(responsesLayer([new Response(body, { status: 400 })])))
 
@@ -131,7 +134,7 @@ describe("RequestExecutor", () => {
           const executor = yield* RequestExecutor.Service
           const error = yield* executor.execute(request).pipe(Effect.flip)
 
-          expectLLMError(error)
+          expectAIError(error)
           expect(error.reason).toMatchObject({ _tag: "ProviderInternal" })
         }).pipe(Effect.provide(responsesLayer([new Response(body, { status: 400 })])))
 
@@ -145,7 +148,7 @@ describe("RequestExecutor", () => {
       const executor = yield* RequestExecutor.Service
       const error = yield* executor.execute(request).pipe(Effect.flip)
 
-      expectLLMError(error)
+      expectAIError(error)
       expect(error).toMatchObject({
         reason: {
           _tag: "RateLimit",
@@ -187,7 +190,7 @@ describe("RequestExecutor", () => {
       const executor = yield* RequestExecutor.Service
       const error = yield* executor.execute(request).pipe(Effect.flip)
 
-      expectLLMError(error)
+      expectAIError(error)
       expect(errorHttp(error)?.request.headers["x-safe"]).toBe("<redacted>")
       expect(errorHttp(error)?.response?.headers["x-safe"]).toBe("<redacted>")
     }).pipe(
@@ -201,7 +204,7 @@ describe("RequestExecutor", () => {
       const executor = yield* RequestExecutor.Service
       const error = yield* executor.execute(request).pipe(Effect.flip)
 
-      expectLLMError(error)
+      expectAIError(error)
       expect(error.reason).toMatchObject({ _tag: "RateLimit" })
       expect(error.reason._tag === "RateLimit" ? error.reason.rateLimit : undefined).toEqual({
         retryAfterMs: 0,
@@ -234,7 +237,7 @@ describe("RequestExecutor", () => {
       const executor = yield* RequestExecutor.Service
       const error = yield* executor.execute(request).pipe(Effect.flip)
 
-      expectLLMError(error)
+      expectAIError(error)
       expect(error.reason).toMatchObject({ _tag: "ProviderInternal" })
       expect(errorHttp(error)?.rateLimit).toEqual({
         retryAfterMs: 0,
@@ -277,7 +280,7 @@ describe("RequestExecutor", () => {
         ),
       )
 
-      expectLLMError(error)
+      expectAIError(error)
       expect(error.reason).toMatchObject({ _tag: "ProviderInternal", status: 503 })
       expect(yield* Ref.get(attempts)).toBe(1)
     }),
@@ -290,7 +293,7 @@ describe("RequestExecutor", () => {
           const executor = yield* RequestExecutor.Service
           const error = yield* executor.execute(request).pipe(Effect.flip)
 
-          expectLLMError(error)
+          expectAIError(error)
           expect(error.reason).toMatchObject({ _tag: "ProviderInternal", status })
         }).pipe(
           Effect.provide(
@@ -313,7 +316,7 @@ describe("RequestExecutor", () => {
       const executor = yield* RequestExecutor.Service
       const error = yield* executor.execute(request).pipe(Effect.flip)
 
-      expectLLMError(error)
+      expectAIError(error)
       expect(error.reason).toMatchObject({ _tag: "Authentication" })
       expect(errorHttp(error)?.bodyTruncated).toBe(true)
       expect(errorHttp(error)?.body).toHaveLength(16_384)
@@ -332,7 +335,7 @@ describe("RequestExecutor", () => {
       const executor = yield* RequestExecutor.Service
       const error = yield* executor.execute(request).pipe(Effect.flip)
 
-      expectLLMError(error)
+      expectAIError(error)
       expect(errorHttp(error)?.body).toContain('"key":"<redacted>"')
       expect(errorHttp(error)?.body).toContain("api_key=<redacted>")
       expect(errorHttp(error)?.body).not.toContain("body-secret")
@@ -353,7 +356,7 @@ describe("RequestExecutor", () => {
       const executor = yield* RequestExecutor.Service
       const error = yield* executor.execute(secretRequest).pipe(Effect.flip)
 
-      expectLLMError(error)
+      expectAIError(error)
       expect(errorHttp(error)?.body).toContain("provider echoed <redacted>")
       expect(errorHttp(error)?.body).toContain("authorization <redacted>")
       expect(errorHttp(error)?.body).not.toContain("query-secret-123")
@@ -392,7 +395,7 @@ describe("RequestExecutor", () => {
         Effect.flip,
       )
 
-      expectLLMError(error)
+      expectAIError(error)
       expect(error.reason).toMatchObject({ _tag: "InvalidProviderOutput" })
       expect(yield* Ref.get(attempts)).toBe(1)
     }),

packages/ai/test/exports.test.ts

@@ -1,5 +1,5 @@
 import { describe, expect, test } from "bun:test"
-import { ImageInput, LLM, LLMClient, Provider } from "@opencode-ai/ai"
+import { AIError, ImageInput, LanguageModel, LLM, LLMClient, Provider } from "@opencode-ai/ai"
 import { Route, Protocol } from "@opencode-ai/ai/route"
 import { Provider as ProviderSubpath } from "@opencode-ai/ai/provider"
 import {
@@ -19,15 +19,19 @@ import {
   OpenResponses,
 } from "@opencode-ai/ai/protocols"
 import * as AnthropicMessages from "@opencode-ai/ai/protocols/anthropic-messages"
+import { TestLLM } from "@opencode-ai/ai/testing"
 
 describe("public exports", () => {
   test("root exposes app-facing runtime APIs", () => {
     expect(LLM.request).toBeFunction()
     expect(LLMClient.Service).toBeFunction()
     expect(LLMClient.layer).toBeDefined()
+    expect(AIError).toBeFunction()
+    expect(LanguageModel.make).toBeFunction()
     expect(ImageInput.bytes).toBeFunction()
     expect(Provider.make).toBeFunction()
     expect(ProviderSubpath.make).toBe(Provider.make)
+    expect(TestLLM.layer).toBeFunction()
   })
 
   test("route barrel exposes route-authoring APIs", () => {
@@ -51,9 +55,7 @@ describe("public exports", () => {
     expect(CloudflareWorkersAI.configure).toBeFunction()
     expect(CloudflareWorkersAI.configure({ accountId: "fixture", apiKey: "fixture" }).model).toBeFunction()
     expect(OpenRouter.model).toBeFunction()
-    expect(OpenRouter.provider.model).toBe(OpenRouter.model)
     expect(XAI.model).toBeFunction()
-    expect(XAI.provider.model).toBe(XAI.model)
     expect(XAI.provider.responses).toBe(XAI.responses)
     expect(XAI.provider.chat).toBe(XAI.chat)
     expect(XAI.configure({ apiKey: "fixture" }).responses("grok-4.3").route.id).toBe("openai-responses")

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

@@ -0,0 +1,47 @@
+import { Schema } from "effect"
+import { LLM, type LanguageModel, type LanguageModelProviderOptions, type ProviderOptions } from "../src"
+import { OpenAIChat } from "../src/protocols"
+
+interface ExampleOptions {
+  readonly [key: string]: unknown
+  readonly mode?: "fast" | "thorough"
+}
+
+type ExampleProviderOptions = ProviderOptions & {
+  readonly example?: ExampleOptions
+}
+
+const model = OpenAIChat.route
+  .with({ endpoint: { baseURL: "https://example.com/v1" } })
+  .model<ExampleProviderOptions>({ id: "example" })
+
+LLM.request({ model, prompt: "Hello", providerOptions: { example: { mode: "fast" } } })
+LLM.request({ model, prompt: "Hello", providerOptions: { future: { option: true } } })
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  // @ts-expect-error Known provider options preserve their value types.
+  providerOptions: { example: { mode: "slow" } },
+})
+
+LLM.generateObject({
+  model,
+  prompt: "Hello",
+  schema: Schema.Struct({ answer: Schema.String }),
+  providerOptions: { example: { mode: "thorough" } },
+})
+
+LLM.generateObject({
+  model,
+  prompt: "Hello",
+  jsonSchema: { type: "object" },
+  // @ts-expect-error Dynamic object generation uses the selected model's provider options.
+  providerOptions: { example: { mode: false } },
+})
+
+declare const generic: LanguageModel
+LLM.request({ model: generic, prompt: "Hello", providerOptions: { arbitrary: { option: true } } })
+
+const options: LanguageModelProviderOptions<typeof model> = { example: { mode: "fast" } }
+void options

packages/ai/test/llm.test.ts

@@ -6,7 +6,7 @@ import {
   GenerationOptions,
   LLMRequest,
   Message,
-  Model,
+  LanguageModel,
   ToolCallPart,
   ToolChoice,
   ToolDefinition,
@@ -20,13 +20,13 @@ describe("llm constructors", () => {
   test("builds canonical schema classes from ergonomic input", () => {
     const request = LLM.request({
       id: "req_1",
-      model: Model.make({ id: "fake-model", provider: "fake", route: chatRoute }),
+      model: LanguageModel.make({ id: "fake-model", provider: "fake", route: chatRoute }),
       system: "You are concise.",
       prompt: "Say hello.",
     })
 
     expect(request).toBeInstanceOf(LLMRequest)
-    expect(request.model).toBeInstanceOf(Model)
+    expect(request.model).toBeInstanceOf(LanguageModel)
     expect(request.messages[0]).toBeInstanceOf(Message)
     expect(request.system).toEqual([{ type: "text", text: "You are concise." }])
     expect(request.messages[0]?.content).toEqual([{ type: "text", text: "Say hello." }])
@@ -37,7 +37,7 @@ describe("llm constructors", () => {
   test("updates requests without spreading schema class instances", () => {
     const base = LLM.request({
       id: "req_1",
-      model: Model.make({ id: "fake-model", provider: "fake", route: chatRoute }),
+      model: LanguageModel.make({ id: "fake-model", provider: "fake", route: chatRoute }),
       prompt: "Say hello.",
     })
     const updated = LLMRequest.update(base, {
@@ -54,7 +54,7 @@ describe("llm constructors", () => {
 
   test("keeps request options separate from route defaults", () => {
     const request = LLM.request({
-      model: Model.make({
+      model: LanguageModel.make({
         id: "fake-model",
         provider: "fake",
         route: chatRoute.with({
@@ -81,7 +81,7 @@ describe("llm constructors", () => {
   test("updates canonical requests from the request datatype", () => {
     const base = LLM.request({
       id: "req_1",
-      model: Model.make({ id: "fake-model", provider: "fake", route: chatRoute }),
+      model: LanguageModel.make({ id: "fake-model", provider: "fake", route: chatRoute }),
       prompt: "Say hello.",
     })
     const updated = LLMRequest.update(base, { messages: [...base.messages, Message.assistant("Hi.")] })
@@ -94,19 +94,19 @@ describe("llm constructors", () => {
   })
 
   test("updates canonical models from the model datatype", () => {
-    const base = Model.make({
+    const base = LanguageModel.make({
       id: "fake-model",
       provider: "fake",
       route: chatRoute,
     })
-    const updated = Model.update(base, {
+    const updated = LanguageModel.update(base, {
       route: responsesRoute,
       defaults: { generation: { maxTokens: 20 } },
       compatibility: { toolSchema: "gemini" },
     })
-    const updatedInput = Model.input(updated)
+    const updatedInput = LanguageModel.input(updated)
 
-    expect(updated).toBeInstanceOf(Model)
+    expect(updated).toBeInstanceOf(LanguageModel)
     expect(String(updated.id)).toBe("fake-model")
     expect(updated.route).toBe(responsesRoute)
     expect(updated.defaults?.generation).toEqual({ maxTokens: 20 })
@@ -114,7 +114,7 @@ describe("llm constructors", () => {
     expect(updatedInput.defaults).toBe(updated.defaults)
     expect(updatedInput.compatibility).toBe(updated.compatibility)
     expect(String(updatedInput.provider)).toBe("fake")
-    expect(Model.update(updated, {})).toBe(updated)
+    expect(LanguageModel.update(updated, {})).toBe(updated)
   })
 
   test("carries model defaults and compatibility through route model selection", () => {
@@ -155,7 +155,7 @@ describe("llm constructors", () => {
     expect(ToolChoice.make("required")).toEqual(new ToolChoice({ type: "required" }))
     expect(
       LLM.request({
-        model: Model.make({
+        model: LanguageModel.make({
           id: "fake-model",
           provider: "fake",
           route: chatRoute,
@@ -181,7 +181,7 @@ describe("llm constructors", () => {
       { type: "text", text: "Use parameterized SQL.", cache: new CacheHint({ type: "ephemeral" }) },
     ])
     const request = LLM.request({
-      model: Model.make({ id: "fake-model", provider: "fake", route: chatRoute }),
+      model: LanguageModel.make({ id: "fake-model", provider: "fake", route: chatRoute }),
       system: "Initial operator prompt.",
       messages: [Message.user("Review this."), update],
     })

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

@@ -6,7 +6,6 @@ describe("provider error classification", () => {
   test("classifies provider token limit messages as context overflow", () => {
     const messages = [
       "tokens in request more than max tokens allowed",
-      '{"error":{"type":"request_too_large","message":"Request exceeds the maximum size"}}',
       "Requested token count exceeds the model's maximum context length of 131072 tokens.",
       "Input length (265330) exceeds model's maximum context length (262144).",
       "Input length 131393 exceeds the maximum allowed input length of 131040 tokens.",
@@ -19,6 +18,24 @@ describe("provider error classification", () => {
     expect(messages.every(isContextOverflow)).toBe(true)
   })
 
+  test("classifies request size failures separately from context overflow", () => {
+    const failures = [
+      classifyProviderFailure({ message: "request too large", status: 413 }),
+      classifyProviderFailure({
+        message: '{"error":{"type":"request_too_large","message":"Request exceeds the maximum size"}}',
+        status: 400,
+      }),
+      classifyProviderFailure({ message: "upstream request entity too large", status: 502 }),
+    ]
+
+    expect(failures).toEqual(
+      failures.map((failure) =>
+        expect.objectContaining({ _tag: "InvalidRequest", classification: "payload-too-large" }),
+      ),
+    )
+    expect(isContextOverflow("413 status code (no body)")).toBe(false)
+  })
+
   test("does not classify rate limits as context overflow", () => {
     const messages = [
       "Throttling error: Too many tokens, please wait before trying again.",
@@ -58,6 +75,13 @@ describe("provider error classification", () => {
     ).toEqual(["ProviderInternal", "ProviderInternal"])
   })
 
+  test("classifies transient client statuses as provider internal", () => {
+    expect([408, 409].map((status) => classifyProviderFailure({ message: `HTTP ${status}`, status })._tag)).toEqual([
+      "ProviderInternal",
+      "ProviderInternal",
+    ])
+  })
+
   test("classifies nested provider codes when a top-level code is also present", () => {
     expect(
       [

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

@@ -0,0 +1,13 @@
+import { LLM } from "../../src"
+import { AnthropicCompatible } from "../../src/providers"
+
+const model = AnthropicCompatible.configure({ baseURL: "https://example.com" }).model("claude")
+
+LLM.request({ model, prompt: "Hello", providerOptions: { anthropic: { effort: "high" } } })
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  // @ts-expect-error Anthropic effort must be a string.
+  providerOptions: { anthropic: { effort: 1 } },
+})

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

@@ -0,0 +1,13 @@
+import { LLM } from "../../src"
+import { Anthropic } from "../../src/providers"
+
+const model = Anthropic.provider.model("claude-sonnet-4-5")
+
+LLM.request({ model, prompt: "Hello", providerOptions: { anthropic: { thinking: { type: "adaptive" } } } })
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  // @ts-expect-error Anthropic thinking modes are a fixed union.
+  providerOptions: { anthropic: { thinking: { type: "automatic" } } },
+})

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

@@ -0,0 +1,13 @@
+import { LLM } from "../../src"
+import { Azure } from "../../src/providers"
+
+const model = Azure.configure({ resourceName: "example" }).responses("deployment")
+
+LLM.request({ model, prompt: "Hello", providerOptions: { openai: { store: false } } })
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  // @ts-expect-error Azure OpenAI store must be boolean.
+  providerOptions: { openai: { store: "false" } },
+})

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

@@ -0,0 +1,13 @@
+import { LLM } from "../../src"
+import { CloudflareWorkersAI } from "../../src/providers"
+
+const model = CloudflareWorkersAI.configure({ accountId: "account", apiKey: "test" }).model("model")
+
+LLM.request({ model, prompt: "Hello", providerOptions: { openai: { promptCacheKey: "cache" } } })
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  // @ts-expect-error Cloudflare's OpenAI-compatible prompt cache key must be a string.
+  providerOptions: { openai: { promptCacheKey: 1 } },
+})

packages/ai/test/provider-options/github-copilot.types.ts

@@ -0,0 +1,13 @@
+import { LLM } from "../../src"
+import { GitHubCopilot } from "../../src/providers"
+
+const model = GitHubCopilot.configure({ baseURL: "https://example.com" }).model("gpt-5")
+
+LLM.request({ model, prompt: "Hello", providerOptions: { openai: { reasoningSummary: "auto" } } })
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  // @ts-expect-error Copilot reasoning summaries use the OpenAI union.
+  providerOptions: { openai: { reasoningSummary: "full" } },
+})

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

@@ -0,0 +1,13 @@
+import { LLM } from "../../src"
+import { GoogleVertexChat } from "../../src/providers"
+
+const model = GoogleVertexChat.configure({ accessToken: "test", project: "project" }).model("gemini")
+
+LLM.request({ model, prompt: "Hello", providerOptions: { openai: { serviceTier: "priority" } } })
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  // @ts-expect-error Vertex OpenAI-compatible service tiers use the OpenAI union.
+  providerOptions: { openai: { serviceTier: "premium" } },
+})

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

@@ -0,0 +1,13 @@
+import { LLM } from "../../src"
+import { GoogleVertexMessages } from "../../src/providers"
+
+const model = GoogleVertexMessages.configure({ accessToken: "test", project: "project" }).model("claude")
+
+LLM.request({ model, prompt: "Hello", providerOptions: { anthropic: { effort: "medium" } } })
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  // @ts-expect-error Vertex Anthropic effort must be a string.
+  providerOptions: { anthropic: { effort: false } },
+})

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

@@ -0,0 +1,13 @@
+import { LLM } from "../../src"
+import { GoogleVertexResponses } from "../../src/providers"
+
+const model = GoogleVertexResponses.configure({ accessToken: "test", project: "project" }).model("gemini")
+
+LLM.request({ model, prompt: "Hello", providerOptions: { openresponses: { textVerbosity: "high" } } })
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  // @ts-expect-error Vertex Responses verbosity uses the Open Responses union.
+  providerOptions: { openresponses: { textVerbosity: "verbose" } },
+})

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

@@ -0,0 +1,17 @@
+import { LLM } from "../../src"
+import { GoogleVertex } from "../../src/providers"
+
+const model = GoogleVertex.provider.configure({ apiKey: "test" }).model("gemini-2.5-pro")
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  providerOptions: { gemini: { thinkingConfig: { includeThoughts: true } } },
+})
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  // @ts-expect-error Vertex Gemini includeThoughts must be boolean.
+  providerOptions: { gemini: { thinkingConfig: { includeThoughts: "yes" } } },
+})

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

@@ -0,0 +1,47 @@
+import { LLM } from "../../src"
+import { Google } from "../../src/providers"
+
+const model = Google.provider.model("gemini-2.5-pro")
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  providerOptions: { gemini: { thinkingConfig: { thinkingBudget: 1024 } } },
+})
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  providerOptions: {
+    gemini: {
+      // @ts-expect-error Gemini safety settings require a threshold for every category.
+      safetySettings: [{ category: "HARM_CATEGORY_HATE_SPEECH" }],
+    },
+  },
+})
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  providerOptions: {
+    gemini: {
+      cachedContent: "cachedContents/example",
+      safetySettings: [{ category: "HARM_CATEGORY_HATE_SPEECH", threshold: "BLOCK_ONLY_HIGH" }],
+      serviceTier: "future-tier",
+      thinkingConfig: { thinkingLevel: "high", includeThoughts: true },
+    },
+  },
+})
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  // @ts-expect-error Gemini thinking budgets must be numeric.
+  providerOptions: { gemini: { thinkingConfig: { thinkingBudget: "large" } } },
+})
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  providerOptions: { gemini: { thinkingConfig: { thinkingLevel: "maximum" } } },
+})

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

@@ -0,0 +1,13 @@
+import { LLM } from "../../src"
+import { OpenAICompatibleResponses } from "../../src/providers"
+
+const model = OpenAICompatibleResponses.configure({ baseURL: "https://example.com" }).model("model")
+
+LLM.request({ model, prompt: "Hello", providerOptions: { openresponses: { reasoningSummary: "detailed" } } })
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  // @ts-expect-error Open Responses reasoning summaries use a fixed union.
+  providerOptions: { openresponses: { reasoningSummary: "full" } },
+})

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

@@ -0,0 +1,13 @@
+import { LLM } from "../../src"
+import { OpenAICompatible } from "../../src/providers"
+
+const model = OpenAICompatible.deepseek.model("deepseek-chat")
+
+LLM.request({ model, prompt: "Hello", providerOptions: { openai: { store: false } } })
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  // @ts-expect-error OpenAI-compatible store must be boolean.
+  providerOptions: { openai: { store: "false" } },
+})

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

@@ -0,0 +1,13 @@
+import { LLM } from "../../src"
+import { OpenAI } from "../../src/providers"
+
+const model = OpenAI.responses("gpt-5")
+
+LLM.request({ model, prompt: "Hello", providerOptions: { openai: { reasoningEffort: "high" } } })
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  // @ts-expect-error OpenAI reasoning effort must be a string.
+  providerOptions: { openai: { reasoningEffort: 1 } },
+})

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

@@ -0,0 +1,35 @@
+import { LLM } from "../../src"
+import { OpenRouter } from "../../src/providers"
+
+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: {
+    openrouter: {
+      models: ["google/gemini-3.1-pro"],
+      provider: {
+        order: ["anthropic"],
+        require_parameters: true,
+        data_collection: "future-policy",
+        sort: "future-sort",
+        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",
+    },
+  },
+})
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  // @ts-expect-error OpenRouter usage must be boolean or an option record.
+  providerOptions: { openrouter: { usage: "yes" } },
+})