refactor(ai): unify prompt cache configuration

refactor(ai): promote prompt cache key
refactor(core): remove unused helpers (#39943 )
2026-08-01 15:38:44 +00:00 · 2026-07-31 17:52:32 -05:00 · 2026-07-31 16:59:24 -05:00 · 2026-07-31 16:17:14 -05:00 · 2026-07-31 21:08:33 +00:00 · 2026-07-31 16:35:05 -04:00
380 changed files with 11557 additions and 6424 deletions
@@ -2,15 +2,12 @@ import type { Context } from "../../../packages/plugin/src/tui/context"

 export default {
  id: "test.tui-discovery-smoke",
-  setup(context: Context) {
-    const timer = setTimeout(() => {
-      context.ui.toast.show({
-        title: "TUI plugin discovery works",
-        message: "Loaded .opencode/plugins/tui/discovery-smoke.ts",
-        variant: "success",
-        duration: 30_000,
-      })
-    }, 1_000)
-    return () => clearTimeout(timer)
+  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,
+    // })
  },
 }
@@ -67,6 +67,7 @@
        "@opencode-ai/sdk": "file:vendor/opencode-ai-sdk-1.18.8-dev.tgz",
        "@opencode-ai/session-ui": "workspace:*",
        "@opencode-ai/ui": "workspace:*",
+        "@opencode-ai/util": "workspace:*",
        "@pierre/trees": "1.0.0-beta.4",
        "@sentry/solid": "catalog:",
        "@shikijs/transformers": "3.9.2",
@@ -143,7 +144,10 @@
        "open": "10.1.2",
        "semver": "catalog:",
        "solid-js": "catalog:",
+        "tree-sitter-bash": "0.25.0",
+        "tree-sitter-powershell": "0.25.10",
        "uqr": "0.1.3",
+        "web-tree-sitter": "0.25.10",
        "ws": "8.21.0",
      },
      "devDependencies": {
@@ -402,8 +406,11 @@
        "immer": "11.1.4",
        "jsonc-parser": "3.3.1",
        "semver": "^7.6.3",
+        "tree-sitter-bash": "0.25.0",
+        "tree-sitter-powershell": "0.25.10",
        "turndown": "7.2.0",
        "venice-ai-sdk-provider": "2.1.1",
+        "web-tree-sitter": "0.25.10",
        "which": "6.0.1",
        "zod": "catalog:",
      },
@@ -514,6 +521,7 @@
        "@opencode-ai/core": "workspace:*",
        "@opencode-ai/session-ui": "workspace:*",
        "@opencode-ai/ui": "workspace:*",
+        "@opencode-ai/util": "workspace:*",
        "@pierre/diffs": "catalog:",
        "@solidjs/meta": "catalog:",
        "@solidjs/router": "catalog:",
@@ -600,6 +608,7 @@
        "zod": "catalog:",
      },
      "devDependencies": {
+        "@opencode-ai/theme": "workspace:*",
        "@opentui/core": "catalog:",
        "@opentui/keymap": "catalog:",
        "@opentui/solid": "catalog:",
@@ -611,12 +620,14 @@
        "typescript": "catalog:",
      },
      "peerDependencies": {
+        "@opencode-ai/theme": "workspace:*",
        "@opentui/core": ">=0.4.5",
        "@opentui/keymap": ">=0.4.5",
        "@opentui/solid": ">=0.4.5",
        "solid-js": ">=1.9.0",
      },
      "optionalPeers": [
+        "@opencode-ai/theme",
        "@opentui/core",
        "@opentui/keymap",
        "@opentui/solid",
@@ -1070,9 +1081,11 @@
  },
  "trustedDependencies": [
    "esbuild",
+    "tree-sitter-powershell",
    "protobufjs",
    "electron",
    "web-tree-sitter",
+    "tree-sitter-bash",
  ],
  "patchedDependencies": {
    "@pierre/trees@1.0.0-beta.4": "patches/@pierre%2Ftrees@1.0.0-beta.4.patch",
@@ -4069,10 +4082,10 @@

    "fetch-blob": ["fetch-blob@3.2.0", "", { "dependencies": { "node-domexception": "^1.0.0", "web-streams-polyfill": "^3.0.3" } }, "sha512-7yAQpD2UMJzLi1Dqv7qFYnPbaPx7ZfFK6PiIxQ4PfkGPyNyl2Ugx+a/umUonmKqjhM4DnfbMvdX6otXq83soQQ=="],

-    "file-uri-to-path": ["file-uri-to-path@1.0.0", "", {}, "sha512-0Zt+s3L7Vf1biwWZ29aARiVYLx7iMGnEUl9x33fbB/j3jR81u/O2LbqK+Bm1CDSNDKVtJ/YjwY7TUd5SkeLQLw=="],
-
    "ffi-rs": ["ffi-rs@1.3.2", "", { "optionalDependencies": { "@yuuang/ffi-rs-android-arm64": "1.3.2", "@yuuang/ffi-rs-darwin-arm64": "1.3.2", "@yuuang/ffi-rs-darwin-x64": "1.3.2", "@yuuang/ffi-rs-linux-arm-gnueabihf": "1.3.2", "@yuuang/ffi-rs-linux-arm64-gnu": "1.3.2", "@yuuang/ffi-rs-linux-arm64-musl": "1.3.2", "@yuuang/ffi-rs-linux-x64-gnu": "1.3.2", "@yuuang/ffi-rs-linux-x64-musl": "1.3.2", "@yuuang/ffi-rs-win32-arm64-msvc": "1.3.2", "@yuuang/ffi-rs-win32-ia32-msvc": "1.3.2", "@yuuang/ffi-rs-win32-x64-msvc": "1.3.2" } }, "sha512-4s8dX9VbBw/jd5NOuE3EJRqXaIVdjMyiumeeDzrOhtjQRwp6Bz2za7iksWXTnvTQKV/tTdm1s1w7mObe92zPjQ=="],

+    "file-uri-to-path": ["file-uri-to-path@1.0.0", "", {}, "sha512-0Zt+s3L7Vf1biwWZ29aARiVYLx7iMGnEUl9x33fbB/j3jR81u/O2LbqK+Bm1CDSNDKVtJ/YjwY7TUd5SkeLQLw=="],
+
    "filelist": ["filelist@1.0.6", "", { "dependencies": { "minimatch": "^5.0.1" } }, "sha512-5giy2PkLYY1cP39p17Ech+2xlpTRL9HLspOfEgm0L6CwBXBTgsK5ou0JtzYuepxkaQ/tvhCFIJ5uXo0OrM2DxA=="],

    "fill-range": ["fill-range@7.1.1", "", { "dependencies": { "to-regex-range": "^5.0.1" } }, "sha512-YsGpe3WHLK8ZYi4tWDg2Jy3ebRz2rXowDxnld4bkQB00cc/1Zw9AWnC0i9ztDJitivtQvaI9KaLyKrc+hBW0yg=="],
@@ -5631,6 +5644,10 @@

    "traverse": ["traverse@0.3.9", "", {}, "sha512-iawgk0hLP3SxGKDfnDJf8wTz4p2qImnyihM5Hh/sGvQ3K37dPi/w8sRhdNIxYA1TwFwc5mDhIJq+O0RsvXBKdQ=="],

+    "tree-sitter-bash": ["tree-sitter-bash@0.25.0", "", { "dependencies": { "node-addon-api": "^8.2.1", "node-gyp-build": "^4.8.2" }, "peerDependencies": { "tree-sitter": "^0.25.0" }, "optionalPeers": ["tree-sitter"] }, "sha512-gZtlj9+qFS81qKxpLfD6H0UssQ3QBc/F0nKkPsiFDyfQF2YBqYvglFJUzchrPpVhZe9kLZTrJ9n2J6lmka69Vg=="],
+
+    "tree-sitter-powershell": ["tree-sitter-powershell@0.25.10", "", { "dependencies": { "node-addon-api": "^7.1.0", "node-gyp-build": "^4.8.0" }, "peerDependencies": { "tree-sitter": "^0.25.0" }, "optionalPeers": ["tree-sitter"] }, "sha512-bEt8QoySpGFnU3aa8WedQyNMaN6aTwy/WUbvIVt0JSKF+BbJoSHNHu+wCbhj7xLMsfB0AuffmiJm+B8gzva8Lg=="],
+
    "treeverse": ["treeverse@3.0.0", "", {}, "sha512-gcANaAnd2QDZFmHFEOF4k7uc1J/6a6z3DJMd/QwEyxLoKGiptJRwid582r7QIsFlFMIZ3SnxfS52S4hm2DHkuQ=="],

    "trim-lines": ["trim-lines@3.0.1", "", {}, "sha512-kRj8B+YHZCc9kQYdWfJB2/oUl9rA99qbowYYBtr4ui4mZyAQ2JpvVBd/6U2YloATfqBhBTSMhTpgBHtU0Mf3Rg=="],
@@ -6853,6 +6870,8 @@

    "topojson-client/commander": ["commander@2.20.3", "", {}, "sha512-GpVkmM8vF2vQUkj2LvZmD35JxeJOLCwJ9cUkugyk2nuhbv3+mJvpLYYt+0+USMxE+oj+ey/lJEnhZw75x/OMcQ=="],

+    "tree-sitter-bash/node-addon-api": ["node-addon-api@8.9.0", "", {}, "sha512-ekZMeaaIzSQTSpr7X2X3iJM7lTzgnx8ahAG9pJfT/7+14mlEM8ZYQ9cgCDvSSRbReFK0oHli3WrZdCiRsgAT9Q=="],
+
    "tw-to-css/postcss": ["postcss@8.4.31", "", { "dependencies": { "nanoid": "^3.3.6", "picocolors": "^1.0.0", "source-map-js": "^1.0.2" } }, "sha512-PS08Iboia9mts/2ygV3eLpY5ghnUcfLV/EXTOW1E2qYxJKGGBUtNjN76FYHnMs36RmARn41bC0AZmn+rR0OVpQ=="],

    "tw-to-css/tailwindcss": ["tailwindcss@3.3.2", "", { "dependencies": { "@alloc/quick-lru": "^5.2.0", "arg": "^5.0.2", "chokidar": "^3.5.3", "didyoumean": "^1.2.2", "dlv": "^1.1.3", "fast-glob": "^3.2.12", "glob-parent": "^6.0.2", "is-glob": "^4.0.3", "jiti": "^1.18.2", "lilconfig": "^2.1.0", "micromatch": "^4.0.5", "normalize-path": "^3.0.0", "object-hash": "^3.0.0", "picocolors": "^1.0.0", "postcss": "^8.4.23", "postcss-import": "^15.1.0", "postcss-js": "^4.0.1", "postcss-load-config": "^4.0.1", "postcss-nested": "^6.0.1", "postcss-selector-parser": "^6.0.11", "postcss-value-parser": "^4.2.0", "resolve": "^1.22.2", "sucrase": "^3.32.0" }, "bin": { "tailwind": "lib/cli.js", "tailwindcss": "lib/cli.js" } }, "sha512-9jPkMiIBXvPc2KywkraqsUfbfj+dHDb+JPWtSJa9MLFdrPyazI7q6WX2sUrm7R9eVR7qqv3Pas7EvQFzxKnI6w=="],
@@ -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.
@@ -12,6 +12,8 @@

 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.

+- 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

 - Use `testEffect(...)` from `test/lib/effect.ts` for tests requiring Effect layers.
@@ -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.

@@ -138,13 +140,13 @@ packages/ai/src/
    ids.ts                  branded IDs, literal types, ProviderMetadata
    options.ts              Generation/Provider/Http options, Limits, Model, cache policy
    messages.ts             content parts, Message, ToolDefinition, LLMRequest
-    events.ts               Usage, individual events, LLMEvent, PreparedRequest, LLMResponse
+    events.ts               Usage, individual events, LLMEvent, LLMResponse
    errors.ts               error reasons, LLMError, 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
@@ -196,7 +196,6 @@ The hosted result is represented as a provider-executed tool call and tool resul
 - **`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.
 - **`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`.
@@ -568,7 +568,7 @@ App boundary = explicit durable-config -> typed-provider call
      calling `.model(...)`.
 - [x] Remove request-shaping defaults from `Model`; 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.
@@ -33,9 +33,10 @@ const model = OpenAI.configure({
 //
 //   - `generation`: common controls such as max tokens, temperature, topP/topK,
 //     penalties, seed, and stop sequences.
+//   - `cache`: provider-neutral prompt caching behavior and cache affinity.
 //   - `providerOptions`: namespaced provider-native behavior. For example,
-//     OpenAI cache keys and store behavior, Anthropic thinking, Gemini thinking
-//     config, or OpenRouter routing/reasoning.
+//     OpenAI store behavior, Anthropic thinking, Gemini thinking config, or
+//     OpenRouter routing/reasoning.
 //   - `http`: last-resort serializable overlays for final request body, headers,
 //     and query params. Prefer typed `providerOptions` when a field is stable.
 //
@@ -45,21 +46,7 @@ const request = LLM.request({
  system: "You are concise and practical.",
  prompt: "Tell me a joke",
  generation: { maxTokens: 80, temperature: 0.7 },
-  providerOptions: {
-    openai: { promptCacheKey: "tutorial-joke" },
-  },
-})
-
-// `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" },
-  },
+  cache: { mode: "auto", key: "tutorial-joke" },
 })

 // 3. `generate` sends the request and collects the event stream into one
@@ -222,33 +209,15 @@ 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))))
@@ -10,16 +10,16 @@
 //
 // Manual `cache: CacheHint` placements on individual parts are preserved and
 // count against the four-breakpoint budget; auto only fills remaining slots.
-import { CacheHint, type CachePolicy, type CachePolicyObject } from "./schema/options"
+import { CacheHint, type CacheExplicit, type CachePolicy } from "./schema/options"
 import { LLMRequest, Message, ToolDefinition, type ContentPart } from "./schema/messages"

-const AUTO: CachePolicyObject = {
+const AUTO: Omit<CacheExplicit, "mode" | "key"> = {
  tools: true,
  system: true,
  messages: { tail: 1 },
 }

-const NONE: CachePolicyObject = {}
+const NONE: Omit<CacheExplicit, "mode" | "key"> = {}
 const BREAKPOINT_CAP = 4

 // Resolution rules:
@@ -29,8 +29,8 @@ const BREAKPOINT_CAP = 4
 //   - "auto"      → tools + first/last system + final message boundary.
 //   - "none"      → no auto placement; manual `CacheHint`s still flow.
 //   - object form → exactly what the caller asked for.
-const resolve = (policy: CachePolicy | undefined): CachePolicyObject => {
-  if (policy === undefined || policy === "auto") return AUTO
+const resolve = (policy: CachePolicy | undefined): Omit<CacheExplicit, "mode" | "key"> => {
+  if (policy === undefined || (policy !== "none" && policy.mode === "auto")) return AUTO
  if (policy === "none") return NONE
  return policy
 }
@@ -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" })
@@ -103,7 +103,7 @@ const markMessageAt = (

 const markMessages = (
  messages: ReadonlyArray<Message>,
-  strategy: NonNullable<CachePolicyObject["messages"]>,
+  strategy: NonNullable<CacheExplicit["messages"]>,
  hint: CacheHint,
  budget: Budget,
 ): ReadonlyArray<Message> => {
@@ -133,6 +133,11 @@ 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 !== "none" && request.cache.mode === "auto"))
+  )
+    return request
  const policy = resolve(request.cache)
  if (!policy.tools && !policy.system && !policy.messages) return request

@@ -9,25 +9,28 @@ import {
  LLMRequest,
  LLMResponse,
  Message,
+  Model,
  SystemPart,
  ToolChoice,
  ToolDefinition,
  type ContentPart,
+  type ModelProviderOptions,
 } 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<SelectedModel extends Model = Model> = Omit<
  ConstructorParameters<typeof LLMRequest>[0],
-  "system" | "messages" | "tools" | "toolChoice" | "generation" | "http" | "providerOptions"
+  "model" | "system" | "messages" | "tools" | "toolChoice" | "generation" | "http" | "providerOptions"
 > & {
+  readonly model: SelectedModel
  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<ModelProviderOptions<SelectedModel>>
  readonly http?: HttpOptions.Input
 }

@@ -35,7 +38,7 @@ export const generate = LLMClient.generate

 export const stream = LLMClient.stream

-export const request = (input: RequestInput) => {
+export const request = <const SelectedModel extends Model>(input: RequestInput<SelectedModel>) => {
  const {
    system: requestSystem,
    prompt,
@@ -63,7 +66,7 @@ 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<SelectedModel extends Model = Model> = Omit<RequestInput<SelectedModel>, "tools" | "toolChoice">

 export class GenerateObjectResponse<T> {
  constructor(
@@ -80,11 +83,13 @@ export class GenerateObjectResponse<T> {
  }
 }

-export interface GenerateObjectOptions<S extends ToolSchema<any>> extends GenerateObjectBase {
+export interface GenerateObjectOptions<S extends ToolSchema<any>, SelectedModel extends Model = Model>
+  extends GenerateObjectBase<SelectedModel> {
  readonly schema: S
 }

-export interface GenerateObjectDynamicOptions extends GenerateObjectBase {
+export interface GenerateObjectDynamicOptions<SelectedModel extends Model = Model>
+  extends GenerateObjectBase<SelectedModel> {
  /** Raw JSON Schema object describing the expected output shape. */
  readonly jsonSchema: JsonSchema.JsonSchema
 }
@@ -137,11 +142,11 @@ const runGenerateObject = Effect.fn("LLM.generateObject")(function* (
 * 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>,
+export function generateObject<const SelectedModel extends Model, S extends ToolSchema<any>>(
+  options: GenerateObjectOptions<S, SelectedModel>,
 ): Effect.Effect<GenerateObjectResponse<Schema.Schema.Type<S>>, LLMError>
-export function generateObject(
-  options: GenerateObjectDynamicOptions,
+export function generateObject<const SelectedModel extends Model>(
+  options: GenerateObjectDynamicOptions<SelectedModel>,
 ): Effect.Effect<GenerateObjectResponse<unknown>, LLMError>
 export function generateObject(options: GenerateObjectOptions<ToolSchema<any>> | GenerateObjectDynamicOptions) {
  if ("schema" in options) {
@@ -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
@@ -3,6 +3,7 @@ import type { Content } from "@opencode-ai/schema/tool"
 import { HttpTransport } from "../route/transport"
 import { Protocol } from "../route/protocol"
 import {
+  cacheKey,
  LLMError,
  LLMEvent,
  Usage,
@@ -542,7 +543,7 @@ const lowerOptions = (request: LLMRequest) => {
  return {
    ...(options.instructions ? { instructions: options.instructions } : {}),
    ...(options.store !== undefined ? { store: options.store } : {}),
-    ...(options.promptCacheKey ? { prompt_cache_key: options.promptCacheKey } : {}),
+    ...(cacheKey(request.cache) ? { prompt_cache_key: cacheKey(request.cache) } : {}),
    ...(options.include ? { include: options.include } : {}),
    ...(options.reasoningEffort || options.reasoningSummary
      ? { reasoning: { effort: options.reasoningEffort, summary: options.reasoningSummary } }
@@ -11,6 +11,7 @@ import {
  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,
@@ -33,7 +33,6 @@ export const ServiceTierSchema = Schema.Literals(ServiceTiers)
 export interface Resolved {
  readonly instructions?: string
  readonly store?: boolean
-  readonly promptCacheKey?: string
  readonly reasoningEffort?: string
  readonly reasoningSummary?: "auto" | "concise" | "detailed"
  readonly include?: ReadonlyArray<ResponseIncludable>
@@ -50,7 +49,6 @@ export const resolve = (request: LLMRequest): Resolved => {
  return {
    instructions: typeof input?.instructions === "string" ? input.instructions : undefined,
    store: typeof input?.store === "boolean" ? input.store : undefined,
-    promptCacheKey: typeof input?.promptCacheKey === "string" ? input.promptCacheKey : undefined,
    reasoningEffort: typeof input?.reasoningEffort === "string" ? input.reasoningEffort : undefined,
    reasoningSummary:
      reasoningSummary === "auto" || reasoningSummary === "concise" || reasoningSummary === "detailed"
@@ -1,16 +1,21 @@
-import type { Model } from "./schema"
+import type { Model, 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) => Model<Options>
 }

 export * as ProviderPackage from "./provider-package"
@@ -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({
@@ -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({
@@ -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
@@ -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,
  }
 }
@@ -50,9 +50,11 @@ export const configure = (options: ModelOptions) => {
  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) =>
@@ -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,
@@ -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,
@@ -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,
@@ -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({
@@ -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,
@@ -5,7 +5,6 @@ export interface OpenResponsesOptionsInput {
  readonly [key: string]: unknown
  readonly instructions?: string
  readonly store?: boolean
-  readonly promptCacheKey?: string
  readonly reasoningEffort?: ReasoningEffort
  readonly reasoningSummary?: "auto" | "concise" | "detailed"
  readonly include?: ReadonlyArray<ResponseIncludable>
@@ -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,
@@ -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,
@@ -17,7 +17,6 @@ const openAIProviderOptions = (options: OpenAIOptionsInput | undefined): Provide
  const openai = Object.fromEntries(
    definedEntries({
      store: options?.store,
-      promptCacheKey: options?.promptCacheKey,
      reasoningEffort: options?.reasoningEffort,
      reasoningSummary: options?.reasoningSummary,
      include: options?.include,
@@ -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
@@ -4,20 +4,71 @@ 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 { cacheKey, 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 promptCacheKey?: string
+  readonly debug?: Readonly<{ echo_upstream_body?: boolean }>
+  readonly models?: ReadonlyArray<string>
+  readonly plugins?: ReadonlyArray<OpenRouterPlugin>
+  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 & {
@@ -30,6 +81,12 @@ export type ModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
    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 +97,7 @@ 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
@@ -64,6 +121,7 @@ export const protocol = Protocol.make({
            ...body,
            messages,
            ...bodyOptions(request.providerOptions?.openrouter),
+            ...(cacheKey(request.cache) ? { prompt_cache_key: cacheKey(request.cache) } : {}),
          } as OpenRouterBody
        }),
      ),
@@ -71,16 +129,38 @@ 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 } : {}),
  }
 }

@@ -107,10 +187,21 @@ export const configure = (input: ModelOptions = {}) => {
  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)
@@ -1,29 +1,62 @@
 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 ModelOptions = 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 { 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),
  })
@@ -31,9 +64,8 @@ const configuredResponsesRoute = (input: ModelOptions) => {

 const configuredChatRoute = (input: ModelOptions) => {
  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),
  })
@@ -42,8 +74,8 @@ const configuredChatRoute = (input: ModelOptions) => {
 export const configure = (input: ModelOptions = {}) => {
  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
@@ -5,12 +5,12 @@ 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 { LLMError, ProtocolID, ProviderOptions } from "../schema"
 import {
  GenerationOptions,
  HttpOptions,
@@ -20,7 +20,6 @@ import {
  ModelLimits,
  LLMError as LLMErrorClass,
  LLMEvent,
-  PreparedRequest,
  ProviderID,
  mergeGenerationOptions,
  mergeHttpOptions,
@@ -46,8 +45,12 @@ 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: RouteMappedModelInput) => Model<Options>
+  readonly prepareTransport: (
+    body: Body,
+    request: LLMRequest,
+    options?: StreamOptions,
+  ) => Effect.Effect<Prepared, LLMError>
  readonly streamPrepared: (
    prepared: Prepared,
    request: LLMRequest,
@@ -93,12 +96,12 @@ export interface RoutePatch<Body, Prepared> extends RouteDefaultsInput {

 type RouteMappedModelInput = RouteModelInput | RouteRoutedModelInput

-const makeRouteModel = (route: AnyRoute, mapped: RouteMappedModelInput) => {
+const makeRouteModel = <Options extends ProviderOptions = ProviderOptions>(route: AnyRoute, mapped: RouteMappedModelInput) => {
  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 Model.make<Options>({
    ...mapped,
    provider,
    route,
@@ -142,27 +145,20 @@ 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, LLMError>
 }

 export interface GenerateMethod {
-  (request: LLMRequest): Effect.Effect<LLMResponse, LLMError>
+  (request: LLMRequest, options?: StreamOptions): Effect.Effect<LLMResponse, LLMError>
 }

 export class Service extends Context.Service<Service, Interface>()("@opencode/LLMClient") {}
@@ -296,8 +292,9 @@ 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: RouteMappedModelInput) =>
+        makeRouteModel<Options>(route, input),
+      prepareTransport: (body, request, options) =>
        routeInput.transport.prepare({
          body,
          request,
@@ -305,6 +302,7 @@ 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 +368,14 @@ 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 +385,30 @@ 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 +417,24 @@ 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, LLMError> {
  return Stream.unwrap(
    Effect.gen(function* () {
-      return (yield* Service).stream(request)
+      return (yield* Service).stream(request, options)
    }),
  ) as Stream.Stream<LLMEvent, LLMError>
 }

-export function generate(request: LLMRequest): Effect.Effect<LLMResponse, LLMError> {
+export function generate(request: LLMRequest, options?: StreamOptions): Effect.Effect<LLMResponse, LLMError> {
  return Effect.gen(function* () {
-    return yield* (yield* Service).generate(request)
+    return yield* (yield* Service).generate(request, options)
  }) as Effect.Effect<LLMResponse, LLMError>
 }

-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 +445,7 @@ 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 +454,6 @@ export const Route = { make } as const
 export const LLMClient = {
  Service,
  layer,
-  prepare,
  stream,
  generate,
 } as const
@@ -8,6 +8,7 @@ export type {
  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"
@@ -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
@@ -10,6 +10,15 @@ export interface TransportRuntime {
  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>
@@ -27,6 +36,7 @@ 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"
@@ -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"

@@ -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)
@@ -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" })
@@ -123,6 +123,7 @@ export const mergeGenerationOptions = (...items: ReadonlyArray<GenerationOptions

 export class ModelLimits extends Schema.Class<ModelLimits>("LLM.ModelLimits")({
  context: Schema.optional(Schema.Number),
+  input: Schema.optional(Schema.Number),
  output: Schema.optional(Schema.Number),
 }) {}

@@ -166,9 +167,13 @@ export namespace ModelDefaults {
 export const ModelToolSchemaCompatibility = Schema.Literals(["gemini", "moonshot"])
 export type ModelToolSchemaCompatibility = Schema.Schema.Type<typeof ModelToolSchemaCompatibility>

+export const ModelMaxTokensFieldCompatibility = Schema.Literals(["max_completion_tokens", "max_tokens"])
+export type ModelMaxTokensFieldCompatibility = Schema.Schema.Type<typeof ModelMaxTokensFieldCompatibility>
+
 export class ModelCompatibility extends Schema.Class<ModelCompatibility>("LLM.ModelCompatibility")({
  toolSchema: Schema.optional(ModelToolSchemaCompatibility),
  reasoningField: Schema.optional(Schema.String),
+  maxTokensField: Schema.optional(ModelMaxTokensFieldCompatibility),
 }) {}

 export namespace ModelCompatibility {
@@ -178,7 +183,8 @@ export namespace ModelCompatibility {
  export const make = (input: Input) => (input instanceof ModelCompatibility ? input : new ModelCompatibility(input))
 }

-export class Model {
+export class Model<Options extends ProviderOptions = ProviderOptions> {
+  declare protected readonly _ProviderOptions: Options
  readonly id: ModelID
  readonly provider: ProviderID
  readonly route: AnyRoute
@@ -193,8 +199,8 @@ export class Model {
    this.compatibility = input.compatibility
  }

-  static make(input: Model.Input) {
-    return new Model({
+  static make<Options extends ProviderOptions = ProviderOptions>(input: Model.Input) {
+    return new Model<Options>({
      id: ModelID.make(input.id),
      provider: ProviderID.make(input.provider),
      route: input.route,
@@ -203,7 +209,7 @@ export class Model {
    })
  }

-  static input(model: Model): Model.ConstructorInput {
+  static input<Options extends ProviderOptions>(model: Model<Options>): Model.ConstructorInput {
    return {
      id: model.id,
      provider: model.provider,
@@ -213,9 +219,9 @@ export class Model {
    }
  }

-  static update(model: Model, patch: Partial<Model.Input>) {
+  static update<Options extends ProviderOptions>(model: Model<Options>, patch: Partial<Model.Input>) {
    if (Object.keys(patch).length === 0) return model
-    return Model.make({
+    return Model.make<Options>({
      ...Model.input(model),
      ...patch,
    })
@@ -241,6 +247,8 @@ export namespace Model {

 export type ModelInput = Model.Input

+export type ModelProviderOptions<SelectedModel> = SelectedModel extends Model<infer Options> ? Options : never
+
 export const ModelSchema = Schema.declare((value): value is Model => value instanceof Model, { expected: "LLM.Model" })

 export class CacheHint extends Schema.Class<CacheHint>("LLM.CacheHint")({
@@ -248,18 +256,17 @@ export class CacheHint extends Schema.Class<CacheHint>("LLM.CacheHint")({
  ttlSeconds: Schema.optional(Schema.Number),
 }) {}

-// Auto-placement policy for prompt caching. The protocol-neutral lowering step
-// reads this and injects `CacheHint`s at the configured boundaries; the
-// per-protocol body builders then translate those hints into wire markers as
-// usual. `"auto"` is the recommended default for agent loops — it places
-// breakpoints at the last tool definition, the first and last distinct system
-// parts, and the conversation tail. The rolling message breakpoint keeps a
-// prior cache entry within Anthropic/Bedrock's 20-block lookback during long
-// tool loops.
-//
-// Pass `"none"` to opt out entirely (the legacy behavior). Pass the granular
-// object form to override individual choices.
-export const CachePolicyObject = Schema.Struct({
+const CacheKey = { key: Schema.optional(Schema.String) }
+
+export const CacheAuto = Schema.Struct({
+  mode: Schema.Literal("auto"),
+  ...CacheKey,
+})
+export type CacheAuto = Schema.Schema.Type<typeof CacheAuto>
+
+export const CacheExplicit = Schema.Struct({
+  mode: Schema.Literal("explicit"),
+  ...CacheKey,
  tools: Schema.optional(Schema.Boolean),
  system: Schema.optional(Schema.Boolean),
  messages: Schema.optional(
@@ -271,7 +278,12 @@ export const CachePolicyObject = Schema.Struct({
  ),
  ttlSeconds: Schema.optional(Schema.Number),
 })
-export type CachePolicyObject = Schema.Schema.Type<typeof CachePolicyObject>
+export type CacheExplicit = Schema.Schema.Type<typeof CacheExplicit>

-export const CachePolicy = Schema.Union([Schema.Literal("auto"), Schema.Literal("none"), CachePolicyObject])
+// Omitted configuration uses automatic provider behavior and OpenCode's
+// automatic breakpoint placement where required. `"none"` sends no cache key
+// or explicit controls; providers may still cache implicitly.
+export const CachePolicy = Schema.Union([Schema.Literal("none"), CacheAuto, CacheExplicit])
 export type CachePolicy = Schema.Schema.Type<typeof CachePolicy>
+
+export const cacheKey = (cache: CachePolicy | undefined) => (cache && cache !== "none" ? cache.key : undefined)
@@ -99,7 +99,6 @@ export const layer = (options: LayerOptions = {}) =>
        )
      }) as LLMClientShape["stream"]
      const client = LLMClient.Service.of({
-        prepare: () => Effect.die("TestLLM does not prepare provider-native requests"),
        stream,
        generate: (request) =>
          stream(request).pipe(
@@ -2,6 +2,7 @@ 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 { compileRequest } from "../src/route/client"
 import { Model } from "../src/schema"
 import { testEffect } from "./lib/effect"
 import { dynamicResponse } from "./lib/http"
@@ -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 }) }),
      )

@@ -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: [
@@ -63,7 +64,7 @@ describe("applyCachePolicy", () => {
            Message.assistant("assistant reply"),
            Message.user("latest user message"),
          ],
-          cache: "auto",
+          cache: { mode: "auto" },
        }),
      )

@@ -87,12 +88,12 @@ 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",
          prompt: "hi",
-          cache: "auto",
+          cache: { mode: "auto" },
        }),
      )

@@ -106,12 +107,12 @@ 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",
          prompt: "hi",
-          cache: "auto",
+          cache: { mode: "auto" },
        }),
      )

@@ -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: [
@@ -132,7 +133,7 @@ describe("applyCachePolicy", () => {
          ],
          tools: [{ name: "t1", description: "t1", inputSchema: { type: "object", properties: {} } }],
          messages: [Message.user("first user"), Message.assistant("reply"), Message.user("latest user")],
-          cache: "auto",
+          cache: { mode: "auto" },
        }),
      )

@@ -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",
@@ -176,13 +177,13 @@ 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",
          tools: [{ name: "t1", description: "t1", inputSchema: { type: "object", properties: {} } }],
          prompt: "hi",
-          cache: { tools: true },
+          cache: { mode: "explicit", tools: true },
        }),
      )

@@ -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: [
@@ -203,7 +204,7 @@ describe("applyCachePolicy", () => {
            { type: "text", text: "last system" },
          ],
          prompt: "hi",
-          cache: "auto",
+          cache: { mode: "auto" },
        }),
      )

@@ -232,7 +233,7 @@ describe("applyCachePolicy", () => {
        ],
        tools: [{ name: "t1", description: "t1", inputSchema: { type: "object", properties: {} } }],
        prompt: "hi",
-        cache: "auto",
+          cache: { mode: "auto" },
      })
      const applied = applyCachePolicy(request)
      expect(applied.tools[0]?.cache).toBeDefined()
@@ -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 }>
@@ -261,12 +262,12 @@ 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",
          prompt: "hi",
-          cache: { system: true, ttlSeconds: 3600 },
+          cache: { mode: "explicit", system: true, ttlSeconds: 3600 },
        }),
      )

@@ -278,11 +279,11 @@ 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")],
-          cache: { messages: { tail: 2 } },
+          cache: { mode: "explicit", messages: { tail: 2 } },
        }),
      )

@@ -296,11 +297,11 @@ 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")],
-          cache: { messages: "latest-assistant" },
+          cache: { mode: "explicit", messages: "latest-assistant" },
        }),
      )

@@ -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 } }),
      )

@@ -53,9 +53,7 @@ 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")
@@ -0,0 +1,47 @@
+import { Schema } from "effect"
+import { LLM, type Model, type ModelProviderOptions, 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: Model
+LLM.request({ model: generic, prompt: "Hello", providerOptions: { arbitrary: { option: true } } })
+
+const options: ModelProviderOptions<typeof model> = { example: { mode: "fast" } }
+void options
@@ -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 } },
+})
@@ -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" } } },
+})
@@ -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" } },
+})
@@ -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", cache: { mode: "auto", key: "cache" } })
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  // @ts-expect-error Prompt cache keys must be strings.
+  cache: { mode: "auto", key: 1 },
+})
@@ -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" } },
+})
@@ -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" } },
+})
@@ -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 } },
+})
@@ -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" } },
+})
@@ -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" } } },
+})
@@ -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" } } },
+})
@@ -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" } },
+})
@@ -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" } },
+})
@@ -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 } },
+})
@@ -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" } },
+})
@@ -0,0 +1,13 @@
+import { LLM } from "../../src"
+import { XAI } from "../../src/providers"
+
+const model = XAI.provider.model("grok-4")
+
+LLM.request({ model, prompt: "Hello", providerOptions: { xai: { reasoningEffort: "high" } } })
+
+LLM.request({
+  model,
+  prompt: "Hello",
+  // @ts-expect-error xAI's OpenAI-compatible reasoning effort must be a string.
+  providerOptions: { xai: { reasoningEffort: true } },
+})
@@ -21,6 +21,8 @@ describe("provider package entrypoints", () => {
      import("@opencode-ai/ai/providers/google-vertex/chat"),
      import("@opencode-ai/ai/providers/google-vertex/responses"),
      import("@opencode-ai/ai/providers/google-vertex/messages"),
+      import("@opencode-ai/ai/providers/openrouter"),
+      import("@opencode-ai/ai/providers/xai"),
    ])

    for (const module of modules) expect(module.model).toBeFunction()
@@ -29,6 +31,35 @@ describe("provider package entrypoints", () => {
    expect(modules[12].model).toBe(modules[13].model)
  })

+  test("maps OpenRouter and xAI package settings onto executable models", async () => {
+    const OpenRouter = await import("@opencode-ai/ai/providers/openrouter")
+    const XAI = await import("@opencode-ai/ai/providers/xai")
+    const settings = {
+      apiKey: "fixture",
+      baseURL: "https://provider.example.test/v1",
+      headers: { "x-application": "opencode" },
+      body: { service_tier: "priority" },
+      limits: { context: 200_000, output: 64_000 },
+    }
+    const openrouter = OpenRouter.model("anthropic/claude-sonnet-4", {
+      ...settings,
+      providerOptions: { openrouter: { usage: true } },
+    })
+    const xai = XAI.model("grok-4", {
+      ...settings,
+      providerOptions: { xai: { reasoningEffort: "high" } },
+    })
+
+    for (const selected of [openrouter, xai]) {
+      expect(selected.route.endpoint.baseURL).toBe(settings.baseURL)
+      expect(selected.route.defaults.headers).toEqual(settings.headers)
+      expect(selected.route.defaults.http?.body).toEqual(settings.body)
+      expect(selected.route.defaults.limits).toEqual(settings.limits)
+    }
+    expect(openrouter.route.defaults.providerOptions).toEqual({ openrouter: { usage: true } })
+    expect(xai.route.defaults.providerOptions).toMatchObject({ xai: { reasoningEffort: "high", store: false } })
+  })
+
  test("maps package settings onto the executable model", () => {
    const selected = model("gpt-5", {
      apiKey: "fixture",
@@ -3,6 +3,7 @@ import { Effect } from "effect"
 import { HttpClientRequest } from "effect/unstable/http"
 import { CacheHint, LLM, LLMError, LLMRequest, Message, ToolCallPart, ToolDefinition, Usage } from "../../src"
 import { Auth, LLMClient } from "../../src/route"
+import { compileRequest } from "../../src/route/client"
 import * as AnthropicMessages from "../../src/protocols/anthropic-messages"
 import { continuationRequest, nativeAnthropicMessagesContinuation } from "../continuation-scenarios"
 import { it } from "../lib/effect"
@@ -44,7 +45,7 @@ const expectToolResult = (body: AnthropicMessages.AnthropicMessagesBody): Anthro
 describe("Anthropic Messages route", () => {
  it.effect("prepares Anthropic Messages target", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(request)
+      const prepared = yield* compileRequest(request)

      expect(prepared.body).toEqual({
        model: "claude-sonnet-4-5",
@@ -59,7 +60,7 @@ describe("Anthropic Messages route", () => {

  it.effect("lowers adaptive thinking settings with effort", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
+      const prepared = yield* compileRequest(
        LLMRequest.update(request, {
          providerOptions: {
            anthropic: { thinking: { type: "adaptive", display: "summarized" }, effort: "low" },
@@ -76,17 +77,17 @@ describe("Anthropic Messages route", () => {

  it.effect("normalizes enabled and disabled thinking settings", () =>
    Effect.gen(function* () {
-      const enabled = yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
+      const enabled = yield* compileRequest(
        LLMRequest.update(request, {
          providerOptions: { anthropic: { thinking: { type: "enabled", budgetTokens: 1_024 } } },
        }),
      )
-      const legacy = yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
+      const legacy = yield* compileRequest(
        LLMRequest.update(request, {
          providerOptions: { anthropic: { thinking: { type: "enabled", budget_tokens: 2_048 } } },
        }),
      )
-      const disabled = yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
+      const disabled = yield* compileRequest(
        LLMRequest.update(request, {
          providerOptions: { anthropic: { thinking: { type: "disabled" } } },
        }),
@@ -100,7 +101,7 @@ describe("Anthropic Messages route", () => {

  it.effect("rejects enabled thinking without a budget", () =>
    Effect.gen(function* () {
-      const error = yield* LLMClient.prepare(
+      const error = yield* compileRequest(
        LLMRequest.update(request, {
          providerOptions: { anthropic: { thinking: { type: "enabled" } } },
        }),
@@ -112,7 +113,7 @@ describe("Anthropic Messages route", () => {

  it.effect("lowers chronological system updates natively for Claude Opus 4.8 with cache hints", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model: opus48,
          messages: [
@@ -137,7 +138,7 @@ describe("Anthropic Messages route", () => {

  it.effect("lowers chronological system updates to wrapped user text for unsupported Anthropic models", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -164,7 +165,7 @@ describe("Anthropic Messages route", () => {

  it.effect("rejects non-text chronological system update content before send", () =>
    Effect.gen(function* () {
-      const error = yield* LLMClient.prepare(
+      const error = yield* compileRequest(
        LLM.request({
          model: opus48,
          messages: [
@@ -181,7 +182,7 @@ describe("Anthropic Messages route", () => {
  it.effect("falls back for unsupported native chronological system update placement", () =>
    Effect.gen(function* () {
      expect(
-        (yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
+        (yield* compileRequest(
          LLM.request({
            model: opus48,
            messages: [Message.assistant("Plain."), Message.system("After plain assistant.")],
@@ -196,12 +197,11 @@ describe("Anthropic Messages route", () => {
        },
      ])
      expect(
-        (yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
-          LLM.request({ model: opus48, messages: [Message.system("First.")], cache: "none" }),
-        )).body.messages,
+        (yield* compileRequest(LLM.request({ model: opus48, messages: [Message.system("First.")], cache: "none" })))
+          .body.messages,
      ).toEqual([{ role: "user", content: [{ type: "text", text: "<system-update>\nFirst.\n</system-update>" }] }])
      expect(
-        (yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
+        (yield* compileRequest(
          LLM.request({
            model: opus48,
            messages: [Message.user("Before."), Message.system("One."), Message.system("Two.")],
@@ -223,7 +223,7 @@ describe("Anthropic Messages route", () => {

  it.effect("rejects a system update between a local tool call and its result", () =>
    Effect.gen(function* () {
-      const error = yield* LLMClient.prepare(
+      const error = yield* compileRequest(
        LLM.request({
          model: opus48,
          messages: [
@@ -242,7 +242,7 @@ describe("Anthropic Messages route", () => {

  it.effect("prepares tool call and tool result messages", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_tool_result",
          model,
@@ -273,7 +273,7 @@ describe("Anthropic Messages route", () => {

  it.effect("keeps tools and sends tool_choice none", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_tool_choice_none",
          model,
@@ -303,7 +303,7 @@ describe("Anthropic Messages route", () => {
  // not JSON-stringified into `tool_result.content`.
  it.effect("lowers media tool-result content as structured blocks", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_tool_result_image",
          model,
@@ -335,7 +335,7 @@ describe("Anthropic Messages route", () => {

  it.effect("lowers single-image tool-result content as a structured image block", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_tool_result_image_only",
          model,
@@ -360,7 +360,7 @@ describe("Anthropic Messages route", () => {

  it.effect("rejects unsupported media in tool-result content with a clear error", () =>
    Effect.gen(function* () {
-      const error = yield* LLMClient.prepare(
+      const error = yield* compileRequest(
        LLM.request({
          id: "req_tool_result_unsupported_media",
          model,
@@ -384,7 +384,7 @@ describe("Anthropic Messages route", () => {

  it.effect("prepares the composed native continuation request", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
+      const prepared = yield* compileRequest(
        continuationRequest({
          id: "req_native_continuation_anthropic",
          model,
@@ -428,7 +428,7 @@ describe("Anthropic Messages route", () => {

  it.effect("lowers preserved Anthropic reasoning signature metadata", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -447,7 +447,7 @@ describe("Anthropic Messages route", () => {

  it.effect("round-trips redacted thinking as redacted_thinking blocks", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -628,9 +628,7 @@ describe("Anthropic Messages route", () => {
        { type: "reasoning", text: "", providerMetadata: { anthropic: { signature: "sig_1" } } },
      ])

-      const prepared = yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
-        LLM.request({ model, messages: [response.message], cache: "none" }),
-      )
+      const prepared = yield* compileRequest(LLM.request({ model, messages: [response.message], cache: "none" }))
      expect(prepared.body.messages).toEqual([
        { role: "assistant", content: [{ type: "thinking", thinking: "", signature: "sig_1" }] },
      ])
@@ -773,7 +771,7 @@ describe("Anthropic Messages route", () => {
          ),
        ),
      )
-      const prepared = yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -1133,7 +1131,7 @@ describe("Anthropic Messages route", () => {

  it.effect("round-trips provider-executed assistant content into server tool blocks", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_round_trip",
          model,
@@ -1184,7 +1182,7 @@ describe("Anthropic Messages route", () => {

  it.effect("rejects round-trip for unknown server tool names", () =>
    Effect.gen(function* () {
-      const error = yield* LLMClient.prepare(
+      const error = yield* compileRequest(
        LLM.request({
          id: "req_unknown_server_tool",
          model,
@@ -1261,7 +1259,7 @@ describe("Anthropic Messages route", () => {

  it.effect("maps ttlSeconds >= 3600 to cache_control ttl: '1h'", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          system: { type: "text", text: "system", cache: new CacheHint({ type: "ephemeral", ttlSeconds: 3600 }) },
@@ -1277,7 +1275,7 @@ describe("Anthropic Messages route", () => {

  it.effect("emits cache_control on tool definitions and tool-result blocks", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          tools: [
@@ -1318,7 +1316,7 @@ describe("Anthropic Messages route", () => {
  it.effect("drops cache_control breakpoints past the 4-per-request cap", () =>
    Effect.gen(function* () {
      const hint = new CacheHint({ type: "ephemeral" })
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          system: [
@@ -1344,7 +1342,7 @@ describe("Anthropic Messages route", () => {
  it.effect("spends breakpoint budget on tools before system before messages", () =>
    Effect.gen(function* () {
      const hint = new CacheHint({ type: "ephemeral" })
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          tools: [
@@ -13,6 +13,7 @@ import {
  ToolDefinition,
 } from "../../src"
 import { LLMClient } from "../../src/route"
+import { compileRequest } from "../../src/route/client"
 import { AmazonBedrock } from "../../src/providers"
 import * as BedrockConverse from "../../src/protocols/bedrock-converse"
 import { it } from "../lib/effect"
@@ -101,7 +102,7 @@ const baseRequest = LLM.request({
 describe("Bedrock Converse route", () => {
  it.effect("prepares Converse target with system, inference config, and messages", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(baseRequest)
+      const prepared = yield* compileRequest(baseRequest)

      expect(prepared.body).toEqual({
        modelId: "anthropic.claude-3-5-sonnet-20240620-v1:0",
@@ -114,7 +115,7 @@ describe("Bedrock Converse route", () => {

  it.effect("passes topK through additionalModelRequestFields as top_k", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<BedrockConverse.BedrockConverseBody>(
+      const prepared = yield* compileRequest(
        LLMRequest.update(baseRequest, {
          generation: GenerationOptions.make({ maxTokens: 64, temperature: 0, topK: 40 }),
        }),
@@ -129,14 +130,14 @@ describe("Bedrock Converse route", () => {

  it.effect("omits additionalModelRequestFields when topK is unset", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<BedrockConverse.BedrockConverseBody>(baseRequest)
+      const prepared = yield* compileRequest(baseRequest)
      expect(prepared.body.additionalModelRequestFields).toBeUndefined()
    }),
  )

  it.effect("lowers chronological system updates to wrapped user text in order", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<BedrockConverse.BedrockConverseBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [Message.user("Before."), Message.system("Update."), Message.assistant("After.")],
@@ -153,7 +154,7 @@ describe("Bedrock Converse route", () => {

  it.effect("prepares tool config with toolSpec and toolChoice", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLMRequest.update(baseRequest, {
          tools: [
            ToolDefinition.make({
@@ -187,7 +188,7 @@ describe("Bedrock Converse route", () => {

  it.effect("keeps tools and omits the unsupported choice when tool choice is none", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLMRequest.update(baseRequest, {
          tools: [
            ToolDefinition.make({
@@ -217,7 +218,7 @@ describe("Bedrock Converse route", () => {

  it.effect("lowers assistant tool-call + tool-result message history", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_history",
          model,
@@ -256,7 +257,7 @@ describe("Bedrock Converse route", () => {

  it.effect("lowers image content in tool-result messages", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_tool_image",
          model,
@@ -491,7 +492,7 @@ describe("Bedrock Converse route", () => {
        providerMetadata: { bedrock: { signature: "sig_1" } },
      })

-      const prepared = yield* LLMClient.prepare<BedrockConverse.BedrockConverseBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -546,9 +547,7 @@ describe("Bedrock Converse route", () => {
        },
      ])

-      const prepared = yield* LLMClient.prepare<BedrockConverse.BedrockConverseBody>(
-        LLM.request({ model, messages: [response.message], cache: "none" }),
-      )
+      const prepared = yield* compileRequest(LLM.request({ model, messages: [response.message], cache: "none" }))
      expect(prepared.body.messages).toEqual([
        {
          role: "assistant",
@@ -639,7 +638,7 @@ describe("Bedrock Converse route", () => {
        text: "",
        providerMetadata: { bedrock: { redactedData } },
      })
-      const prepared = yield* LLMClient.prepare<BedrockConverse.BedrockConverseBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -754,7 +753,7 @@ describe("Bedrock Converse route", () => {
          secretAccessKey: "wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY",
        },
      }).model("anthropic.claude-3-5-sonnet-20240620-v1:0")
-      const prepared = yield* LLMClient.prepare(LLMRequest.update(baseRequest, { model: signed }))
+      const prepared = yield* compileRequest(LLMRequest.update(baseRequest, { model: signed }))

      expect(prepared.route).toBe("bedrock-converse")
      expect(prepared.model).toBe(signed)
@@ -764,7 +763,7 @@ describe("Bedrock Converse route", () => {
  it.effect("emits cachePoint markers after system, user-text, and assistant-text with cache hints", () =>
    Effect.gen(function* () {
      const cache = new CacheHint({ type: "ephemeral" })
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_cache",
          model,
@@ -796,7 +795,7 @@ describe("Bedrock Converse route", () => {

  it.effect("does not emit cachePoint when no cache hint is set", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(baseRequest)
+      const prepared = yield* compileRequest(baseRequest)
      expect(prepared.body).toMatchObject({
        system: [{ text: "You are concise." }],
        messages: [{ role: "user", content: [{ text: "Say hello." }] }],
@@ -806,7 +805,7 @@ describe("Bedrock Converse route", () => {

  it.effect("lowers image media into Bedrock image blocks", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_image",
          model,
@@ -843,7 +842,7 @@ describe("Bedrock Converse route", () => {

  it.effect("base64-encodes Uint8Array image bytes", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_image_bytes",
          model,
@@ -865,7 +864,7 @@ describe("Bedrock Converse route", () => {

  it.effect("lowers document media into Bedrock document blocks with format and name", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_doc",
          model,
@@ -897,7 +896,7 @@ describe("Bedrock Converse route", () => {

  it.effect("requires names for document media", () =>
    Effect.gen(function* () {
-      const error = yield* LLMClient.prepare(
+      const error = yield* compileRequest(
        LLM.request({
          model,
          messages: [Message.user({ type: "media", mediaType: "application/pdf", data: "UERGREFUQQ==" })],
@@ -910,7 +909,7 @@ describe("Bedrock Converse route", () => {

  it.effect("passes named document-only messages through for provider validation", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<BedrockConverse.BedrockConverseBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          cache: "none",
@@ -936,7 +935,7 @@ describe("Bedrock Converse route", () => {

  it.effect("lowers document media in tool results", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<BedrockConverse.BedrockConverseBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          cache: "none",
@@ -988,7 +987,7 @@ describe("Bedrock Converse route", () => {

  it.effect("rejects unsupported image media types", () =>
    Effect.gen(function* () {
-      const error = yield* LLMClient.prepare(
+      const error = yield* compileRequest(
        LLM.request({
          id: "req_bad_image",
          model,
@@ -1002,7 +1001,7 @@ describe("Bedrock Converse route", () => {

  it.effect("rejects unsupported document media types", () =>
    Effect.gen(function* () {
-      const error = yield* LLMClient.prepare(
+      const error = yield* compileRequest(
        LLM.request({
          id: "req_bad_doc",
          model,
@@ -1017,7 +1016,7 @@ describe("Bedrock Converse route", () => {
  it.effect("maps ttlSeconds >= 3600 to cachePoint ttl: '1h'", () =>
    Effect.gen(function* () {
      const cache = new CacheHint({ type: "ephemeral", ttlSeconds: 3600 })
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          system: [{ type: "text", text: "system", cache }],
@@ -1034,7 +1033,7 @@ describe("Bedrock Converse route", () => {
  it.effect("appends cachePoint after marked tool definitions and tool-result blocks", () =>
    Effect.gen(function* () {
      const cache = new CacheHint({ type: "ephemeral" })
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          tools: [{ name: "lookup", description: "lookup", inputSchema: { type: "object", properties: {} }, cache }],
@@ -1066,7 +1065,7 @@ describe("Bedrock Converse route", () => {
  it.effect("drops cachePoint markers past the 4-per-request cap", () =>
    Effect.gen(function* () {
      const cache = new CacheHint({ type: "ephemeral" })
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          system: [
@@ -3,7 +3,7 @@ import { ConfigProvider, Effect, Schema } from "effect"
 import { HttpClientRequest } from "effect/unstable/http"
 import { LLM, LLMEvent } from "../../src"
 import { CloudflareAIGateway, CloudflareWorkersAI } from "../../src/providers/cloudflare"
-import { LLMClient } from "../../src/route"
+import { compileRequest } from "../../src/route/client"
 import { it } from "../lib/effect"
 import { dynamicResponse } from "../lib/http"
 import { sseEvents } from "../lib/sse"
@@ -34,7 +34,7 @@ describe("Cloudflare", () => {
      })
      expect(model.route.endpoint.baseURL).toBe("https://gateway.ai.cloudflare.com/v1/test-account/test-gateway/compat")

-      const prepared = yield* LLMClient.prepare(LLM.request({ model, prompt: "Say hello." }))
+      const prepared = yield* compileRequest(LLM.request({ model, prompt: "Say hello." }))

      expect(prepared.route).toBe("cloudflare-ai-gateway")
      expect(prepared.body).toMatchObject({
@@ -129,7 +129,7 @@ describe("Cloudflare", () => {
        openai: { reasoningField: "reasoning", reasoningDetails: merged },
      })

-      const replay = yield* LLMClient.prepare(LLM.request({ model, messages: [response.message] }))
+      const replay = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
      expect(replay.body.messages).toEqual([
        { role: "assistant", content: "Hello", reasoning: "Thinking", reasoning_details: merged },
      ])
@@ -180,7 +180,7 @@ describe("Cloudflare", () => {

  it.effect("allows a fully configured baseURL override", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          model: CloudflareAIGateway.configure({
            baseURL: "https://gateway.proxy.test/v1/custom/compat",
@@ -208,7 +208,7 @@ describe("Cloudflare", () => {
      })
      expect(model.route.endpoint.baseURL).toBe("https://api.cloudflare.com/client/v4/accounts/test-account/ai/v1")

-      const prepared = yield* LLMClient.prepare(LLM.request({ model, prompt: "Say hello." }))
+      const prepared = yield* compileRequest(LLM.request({ model, prompt: "Say hello." }))

      expect(prepared.route).toBe("cloudflare-workers-ai")
      expect(prepared.body).toMatchObject({
@@ -2,6 +2,7 @@ import { describe, expect } from "bun:test"
 import { Effect } from "effect"
 import { LLM, LLMError, LLMRequest, Message, ToolCallPart, ToolDefinition, Usage } from "../../src"
 import { Auth, LLMClient } from "../../src/route"
+import { compileRequest } from "../../src/route/client"
 import * as Gemini from "../../src/protocols/gemini"
 import { ProviderShared } from "../../src/protocols/shared"
 import { it } from "../lib/effect"
@@ -26,7 +27,7 @@ const request = LLM.request({
 describe("Gemini route", () => {
  it.effect("prepares Gemini target", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(request)
+      const prepared = yield* compileRequest(request)

      expect(prepared.body).toEqual({
        contents: [{ role: "user", parts: [{ text: "Say hello." }] }],
@@ -38,28 +39,56 @@ describe("Gemini route", () => {

  it.effect("normalizes Gemini thinking options", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<Gemini.GeminiBody>(
+      const prepared = yield* compileRequest(
        LLMRequest.update(request, {
-          providerOptions: { gemini: { thinkingConfig: { thinkingBudget: 0, includeThoughts: false } } },
+          providerOptions: {
+            gemini: {
+              cachedContent: "cachedContents/example",
+              safetySettings: [{ category: "HARM_CATEGORY_HATE_SPEECH", threshold: "BLOCK_ONLY_HIGH" }],
+              serviceTier: "priority",
+              thinkingConfig: { thinkingBudget: 0, includeThoughts: false, thinkingLevel: "high" },
+            },
+          },
        }),
      )
-      const filtered = yield* LLMClient.prepare<Gemini.GeminiBody>(
+      const filtered = yield* compileRequest(
        LLMRequest.update(request, {
          providerOptions: { gemini: { thinkingConfig: { thinkingBudget: "invalid", includeThoughts: false } } },
        }),
      )
+      const defaulted = yield* compileRequest(
+        LLMRequest.update(request, {
+          providerOptions: { gemini: { thinkingConfig: { thinkingLevel: "high" } } },
+        }),
+      )
+      const emptySafetySettings = yield* compileRequest(
+        LLMRequest.update(request, {
+          providerOptions: { gemini: { safetySettings: [] } },
+        }),
+      )

      expect(prepared.body.generationConfig?.thinkingConfig).toEqual({
        thinkingBudget: 0,
        includeThoughts: false,
+        thinkingLevel: "high",
      })
+      expect(prepared.body.cachedContent).toBe("cachedContents/example")
+      expect(prepared.body.safetySettings).toEqual([
+        { category: "HARM_CATEGORY_HATE_SPEECH", threshold: "BLOCK_ONLY_HIGH" },
+      ])
+      expect(prepared.body.serviceTier).toBe("priority")
      expect(filtered.body.generationConfig?.thinkingConfig).toEqual({ includeThoughts: false })
+      expect(defaulted.body.generationConfig?.thinkingConfig).toEqual({
+        includeThoughts: true,
+        thinkingLevel: "high",
+      })
+      expect(emptySafetySettings.body.safetySettings).toEqual([])
    }),
  )

  it.effect("lowers chronological system updates to wrapped user text in order", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<Gemini.GeminiBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [Message.user("Before."), Message.system("Update."), Message.assistant("After.")],
@@ -75,7 +104,7 @@ describe("Gemini route", () => {

  it.effect("prepares multimodal user input and tool history", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_tool_result",
          model,
@@ -143,7 +172,7 @@ describe("Gemini route", () => {

  it.effect("continues media tool results as inline model input without base64 text", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<Gemini.GeminiBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -188,7 +217,7 @@ describe("Gemini route", () => {

  it.effect("strips matching data URLs to raw base64 inlineData", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<Gemini.GeminiBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -229,7 +258,7 @@ describe("Gemini route", () => {
  ] as const)
    it.effect(`rejects ${name}`, () =>
      Effect.gen(function* () {
-        const error = yield* LLMClient.prepare(
+        const error = yield* compileRequest(
          LLM.request({ model, messages: [Message.user({ type: "media", ...media })] }),
        ).pipe(Effect.flip)
        expect(error.message).toMatch(/does not support|does not match|valid base64/)
@@ -238,7 +267,7 @@ describe("Gemini route", () => {

  it.effect("rejects oversized image input", () =>
    Effect.gen(function* () {
-      const error = yield* LLMClient.prepare(
+      const error = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -256,7 +285,7 @@ describe("Gemini route", () => {

  it.effect("keeps tools and sends function calling mode NONE", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_tool_choice_none",
          model,
@@ -276,7 +305,7 @@ describe("Gemini route", () => {

  it.effect("sanitizes integer enums, dangling required, untyped arrays, and scalar object keys", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_schema_patch",
          model,
@@ -457,7 +486,7 @@ describe("Gemini route", () => {
        response.events.findIndex((event) => event.type === "tool-call"),
      )

-      const prepared = yield* LLMClient.prepare<Gemini.GeminiBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -691,7 +720,7 @@ describe("Gemini route", () => {

  it.effect("rejects unsupported assistant media content", () =>
    Effect.gen(function* () {
-      const error = yield* LLMClient.prepare(
+      const error = yield* compileRequest(
        LLM.request({
          id: "req_media",
          model,
@@ -4,6 +4,7 @@ import { HttpClientRequest } from "effect/unstable/http"
 import { LLM } from "../../src"
 import { GoogleVertex, GoogleVertexChat, GoogleVertexMessages, GoogleVertexResponses } from "../../src/providers"
 import { LLMClient } from "../../src/route"
+import { compileRequest } from "../../src/route/client"
 import { it } from "../lib/effect"
 import { dynamicResponse } from "../lib/http"
 import { deltaChunk, finishChunk } from "../lib/openai-chunks"
@@ -180,23 +181,6 @@ describe("Google Vertex providers", () => {
    }),
  )

-  it.effect("protects the Vertex Messages API version from body overlays", () =>
-    Effect.gen(function* () {
-      const error = yield* LLMClient.prepare(
-        LLM.request({
-          model: GoogleVertexMessages.configure({
-            accessToken: "vertex-token",
-            http: { body: { anthropic_version: "wrong" } },
-            project: "vertex-project",
-          }).model("claude-sonnet-4-6"),
-          prompt: "Say hello.",
-        }),
-      ).pipe(Effect.flip)
-
-      expect(error.message).toContain("http.body cannot overlay protocol-owned field(s): anthropic_version")
-    }),
-  )
-
  it.effect("routes tuned Gemini models through their deployed endpoint", () =>
    Effect.gen(function* () {
      const response = yield* LLMClient.generate(
@@ -5,6 +5,7 @@ import { OpenAIChat } from "../../src/protocols/openai-chat"
 import * as OpenAICompatible from "../../src/providers/openai-compatible"
 import * as OpenRouter from "../../src/providers/openrouter"
 import { LLMClient } from "../../src/route"
+import { compileRequest } from "../../src/route/client"
 import { recordedTests } from "../recorded-test"
 import { expectWeatherToolLoop, goldenWeatherToolLoopRequest, runWeatherToolLoop } from "../recorded-scenarios"

@@ -84,9 +85,7 @@ for (const item of cases) {
            ),
          ).toBe(true)

-          const replay = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
-            LLM.request({ model: item.model, messages: [response.message] }),
-          )
+          const replay = yield* compileRequest(LLM.request({ model: item.model, messages: [response.message] }))
          expect(replay.body.messages).toMatchObject([
            { role: "assistant", content: response.text, reasoning: response.reasoning },
          ])
@@ -18,6 +18,7 @@ import * as OpenAI from "../../src/providers/openai"
 import * as OpenAIChat from "../../src/protocols/openai-chat"
 import { ProviderShared } from "../../src/protocols/shared"
 import { Auth, LLMClient } from "../../src/route"
+import { compileRequest } from "../../src/route/client"
 import { it } from "../lib/effect"
 import { dynamicResponse, fixedResponse, truncatedStream } from "../lib/http"
 import { deltaChunk, usageChunk } from "../lib/openai-chunks"
@@ -42,11 +43,7 @@ const request = LLM.request({
 describe("OpenAI Chat route", () => {
  it.effect("prepares OpenAI Chat payload", () =>
    Effect.gen(function* () {
-      // Pass the OpenAIChat payload type so `prepared.body` is statically
-      // typed to the route's native shape — the assertions below read field
-      // names without `unknown` casts.
-      const prepared = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(request)
-      const _typed: { readonly model: string; readonly stream: true } = prepared.body
+      const prepared = yield* compileRequest(request)

      expect(prepared.body).toEqual({
        model: "gpt-4o-mini",
@@ -64,7 +61,7 @@ describe("OpenAI Chat route", () => {

  it.effect("lowers chronological system updates to escaped user wrappers in order", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -87,7 +84,7 @@ describe("OpenAI Chat route", () => {

  it.effect("replays canonical reasoning as OpenAI-compatible reasoning_content", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -105,7 +102,7 @@ describe("OpenAI Chat route", () => {

  it.effect("writes reasoning to a configured custom field on every assistant message", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model: Model.update(model, { compatibility: { reasoningField: "vendor_reasoning" } }),
          messages: [
@@ -131,7 +128,7 @@ describe("OpenAI Chat route", () => {

  it.effect("rejects reasoning fields that conflict with assistant message fields", () =>
    Effect.gen(function* () {
-      const error = yield* LLMClient.prepare(
+      const error = yield* compileRequest(
        LLM.request({
          model: Model.update(model, { compatibility: { reasoningField: "content" } }),
          messages: [Message.assistant([{ type: "reasoning", text: "thinking" }])],
@@ -144,7 +141,7 @@ describe("OpenAI Chat route", () => {

  it.effect("maps OpenAI provider options to Chat options", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model: OpenAI.configure({ baseURL: "https://api.openai.test/v1/", apiKey: "test" }).chat("gpt-4o-mini"),
          prompt: "think",
@@ -159,7 +156,7 @@ describe("OpenAI Chat route", () => {

  it.effect("passes through custom OpenAI-compatible reasoning effort strings", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          prompt: "think",
@@ -253,7 +250,7 @@ describe("OpenAI Chat route", () => {

  it.effect("prepares assistant tool-call and tool-result messages", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_tool_result",
          model,
@@ -291,7 +288,7 @@ describe("OpenAI Chat route", () => {
  it.effect("preserves structured tool errors for the model", () =>
    Effect.gen(function* () {
      const error = { error: { type: "unknown", message: "Tool execution interrupted" } }
-      const prepared = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -311,7 +308,7 @@ describe("OpenAI Chat route", () => {

  it.effect("continues image tool results as vision input without base64 text", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -355,7 +352,7 @@ describe("OpenAI Chat route", () => {

  it.effect("orders parallel tool responses before one aggregated vision message", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -405,7 +402,7 @@ describe("OpenAI Chat route", () => {

  it.effect("aggregates consecutive tool images with a following system update", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -446,7 +443,7 @@ describe("OpenAI Chat route", () => {

  it.effect("appends system updates without replacing multipart user content", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -474,7 +471,7 @@ describe("OpenAI Chat route", () => {
  ] as const)
    it.effect(`rejects ${name}`, () =>
      Effect.gen(function* () {
-        const error = yield* LLMClient.prepare(
+        const error = yield* compileRequest(
          LLM.request({ model, messages: [Message.user({ type: "media", ...media })] }),
        ).pipe(Effect.flip)
        expect(error.message).toMatch(/does not support|does not match|valid base64/)
@@ -483,7 +480,7 @@ describe("OpenAI Chat route", () => {

  it.effect("rejects oversized image input", () =>
    Effect.gen(function* () {
-      const error = yield* LLMClient.prepare(
+      const error = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -501,7 +498,7 @@ describe("OpenAI Chat route", () => {

  it.effect("prepares raw and data URL image media as vision input", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_media",
          model,
@@ -528,7 +525,7 @@ describe("OpenAI Chat route", () => {

  it.effect("lowers reasoning-only assistant history", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_reasoning",
          model,
@@ -619,9 +616,7 @@ describe("OpenAI Chat route", () => {
          openai: { reasoningField: field },
        })

-        const replay = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
-          LLM.request({ model, messages: [response.message] }),
-        )
+        const replay = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
        expect(replay.body.messages).toEqual([{ role: "assistant", content: "Hello", [field]: "thinking" }])
      }
    }),
@@ -647,9 +642,7 @@ describe("OpenAI Chat route", () => {
        openai: { reasoningField: "vendor_reasoning" },
      })

-      const replay = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
-        LLM.request({ model: custom, messages: [response.message] }),
-      )
+      const replay = yield* compileRequest(LLM.request({ model: custom, messages: [response.message] }))
      expect(replay.body.messages).toEqual([{ role: "assistant", content: "Hello", vendor_reasoning: "thinking" }])
    }),
  )
@@ -692,9 +685,7 @@ describe("OpenAI Chat route", () => {
        openai: { reasoningField: "reasoning", reasoningDetails: details },
      })

-      const replay = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
-        LLM.request({ model, messages: [response.message] }),
-      )
+      const replay = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
      expect(replay.body.messages).toEqual([
        {
          role: "assistant",
@@ -737,9 +728,7 @@ describe("OpenAI Chat route", () => {
        openai: { reasoningDetails: details },
      })

-      const replay = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
-        LLM.request({ model, messages: [response.message] }),
-      )
+      const replay = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
      expect(replay.body.messages).toEqual([{ role: "assistant", content: "Hello", reasoning_details: details }])
    }),
  )
@@ -764,9 +753,7 @@ describe("OpenAI Chat route", () => {
        openai: { reasoningField: "reasoning", reasoningDetails: details },
      })

-      const replay = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
-        LLM.request({ model, messages: [response.message] }),
-      )
+      const replay = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
      expect(replay.body.messages).toEqual([
        { role: "assistant", content: "Hello", reasoning: "thinking", reasoning_details: details },
      ])
@@ -839,9 +826,7 @@ describe("OpenAI Chat route", () => {
        openai: { reasoningDetails: [] },
      })

-      const replay = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
-        LLM.request({ model, messages: [response.message] }),
-      )
+      const replay = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
      expect(replay.body.messages).toEqual([{ role: "assistant", content: "Hello", reasoning_details: [] }])
    }),
  )
@@ -889,9 +874,7 @@ describe("OpenAI Chat route", () => {
        response.events.findIndex(LLMEvent.is.textStart),
      )

-      const replay = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
-        LLM.request({ model, messages: [response.message] }),
-      )
+      const replay = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
      expect(replay.body.messages).toEqual([
        { role: "assistant", content: "Hello", reasoning: "thinking", reasoning_details: merged },
      ])
@@ -918,9 +901,7 @@ describe("OpenAI Chat route", () => {
      expect(response.events.filter(LLMEvent.is.reasoningStart)).toHaveLength(1)
      expect(response.events.filter(LLMEvent.is.reasoningEnd)).toHaveLength(1)

-      const replay = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
-        LLM.request({ model, messages: [response.message] }),
-      )
+      const replay = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
      expect(replay.body.messages).toEqual([{ role: "assistant", content: null, reasoning_details: details }])
    }),
  )
@@ -950,7 +931,7 @@ describe("OpenAI Chat route", () => {
    Effect.gen(function* () {
      const first = { type: "reasoning.text", text: "first", signature: "signed-0", index: 0 }
      const second = { type: "reasoning.text", text: "second", signature: "signed-1", index: 1 }
-      const replay = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
+      const replay = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -979,7 +960,7 @@ describe("OpenAI Chat route", () => {
  it.effect("retains scalar replay for mixed structured reasoning parts", () =>
    Effect.gen(function* () {
      const detail = { type: "reasoning.encrypted", data: "opaque", index: 0 }
-      const replay = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
+      const replay = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -1004,7 +985,7 @@ describe("OpenAI Chat route", () => {
  it.effect("replays native scalar reasoning alongside native details", () =>
    Effect.gen(function* () {
      const details = [{ type: "reasoning.encrypted", data: "opaque", index: 0 }]
-      const replay = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
+      const replay = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -3,6 +3,7 @@ import { Effect, Schema } from "effect"
 import { HttpClientRequest } from "effect/unstable/http"
 import { LLM, LLMRequest, Message, ToolCallPart, ToolChoice, ToolDefinition } from "../../src"
 import { Auth, LLMClient } from "../../src/route"
+import { compileRequest } from "../../src/route/client"
 import * as OpenAICompatible from "../../src/providers/openai-compatible"
 import * as OpenAICompatibleChat from "../../src/protocols/openai-compatible-chat"
 import { it } from "../lib/effect"
@@ -52,7 +53,7 @@ const providerFamilies = [
 describe("OpenAI-compatible Chat route", () => {
  it.effect("prepares generic Chat target", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLMRequest.update(request, {
          tools: [ToolDefinition.make({ name: "lookup", description: "Lookup data", inputSchema: { type: "object" } })],
          toolChoice: ToolChoice.make({ type: "required" }),
@@ -127,7 +128,7 @@ describe("OpenAI-compatible Chat route", () => {

  it.effect("matches AI SDK compatible basic request body fixture", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(request)
+      const prepared = yield* compileRequest(request)

      expect(prepared.body).toEqual({
        model: "deepseek-chat",
@@ -143,9 +144,23 @@ describe("OpenAI-compatible Chat route", () => {
    }),
  )

+  it.effect("configures the max tokens request field", () =>
+    Effect.gen(function* () {
+      const compatible = OpenAICompatibleChat.route
+        .with({ provider: "custom", endpoint: { baseURL: "https://api.custom.test/v1" } })
+        .model({ id: "custom-model", compatibility: { maxTokensField: "max_completion_tokens" } })
+      const prepared = yield* compileRequest(
+        LLM.request({ model: compatible, prompt: "Say hello.", generation: { maxTokens: 20 } }),
+      )
+
+      expect(prepared.body).toMatchObject({ max_completion_tokens: 20 })
+      expect(prepared.body).not.toHaveProperty("max_tokens")
+    }),
+  )
+
  it.effect("matches AI SDK compatible tool request body fixture", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_tool_parity",
          model,
@@ -7,6 +7,7 @@ import { OpenResponses } from "../../src/protocols/open-responses"
 import { OpenAICompatibleResponses } from "../../src/protocols/openai-compatible-responses"
 import { OpenAIResponses } from "../../src/protocols/openai-responses"
 import { LLMClient } from "../../src/route"
+import { compileRequest } from "../../src/route/client"
 import { it } from "../lib/effect"
 import { fixedResponse } from "../lib/http"
 import { sseEvents } from "../lib/sse"
@@ -23,7 +24,7 @@ describe("Open Responses-compatible route", () => {
        baseURL: "https://responses.example.test/v1",
        provider: "example",
      }).model("example-model")
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          system: "You are concise.",
@@ -61,7 +62,7 @@ describe("Open Responses-compatible route", () => {
        apiKey: "test-key",
        baseURL: "https://responses.example.test/v1",
      }).model("example-model")
-      const error = yield* LLMClient.prepare(
+      const error = yield* compileRequest(
        LLM.request({ model, prompt: "Draw.", tools: [OpenAI.imageGeneration()] }),
      ).pipe(Effect.flip)

@@ -76,7 +77,7 @@ describe("Open Responses-compatible route", () => {
        apiKey: "test-key",
        baseURL: "https://responses.example.test/v1",
      }).model("example-model")
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -102,7 +103,7 @@ describe("Open Responses-compatible route", () => {
        baseURL: "https://responses.example.test/v1",
        providerOptions: { openresponses: { reasoningEffort: "low", store: true } },
      }).model("example-model")
-      const prepared = yield* LLMClient.prepare(LLM.request({ model, prompt: "Think." }))
+      const prepared = yield* compileRequest(LLM.request({ model, prompt: "Think." }))

      expect(prepared.body).toMatchObject({
        reasoning: { effort: "low" },
@@ -20,7 +20,7 @@ const cacheRequest = LLM.request({
  system: LARGE_CACHEABLE_SYSTEM,
  prompt: "Say hi.",
  generation: { maxTokens: 16, temperature: 0 },
-  providerOptions: { openai: { promptCacheKey: "recorded-cache-test" } },
+  cache: { mode: "auto", key: "recorded-cache-test" },
 })

 const recorded = recordedTests({
@@ -14,6 +14,7 @@ import {
  Usage,
 } from "../../src"
 import { Auth, LLMClient, RequestExecutor, WebSocketExecutor } from "../../src/route"
+import { compileRequest } from "../../src/route/client"
 import * as Azure from "../../src/providers/azure"
 import * as OpenAI from "../../src/providers/openai"
 import * as XAI from "../../src/providers/xai"
@@ -56,7 +57,7 @@ const expectToolOutput = (body: OpenAIResponses.OpenAIResponsesBody): OpenAITool
 describe("OpenAI Responses route", () => {
  it.effect("prepares OpenAI Responses target", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(request)
+      const prepared = yield* compileRequest(request)

      expect(prepared.body).toEqual({
        model: "gpt-4.1-mini",
@@ -74,7 +75,7 @@ describe("OpenAI Responses route", () => {

  it.effect("lowers the hosted OpenAI image generation tool", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          prompt: "Show me a rooftop garden.",
@@ -92,7 +93,7 @@ describe("OpenAI Responses route", () => {

  it.effect("rejects invalid hosted image generation options locally", () =>
    Effect.gen(function* () {
-      const error = yield* LLMClient.prepare(
+      const error = yield* compileRequest(
        LLM.request({
          model,
          prompt: "Show me a rooftop garden.",
@@ -109,7 +110,7 @@ describe("OpenAI Responses route", () => {
    Effect.gen(function* () {
      const input = LLMRequest.update(request, { providerOptions: { openai: { serviceTier: "priority" } } })
      expect(input.providerOptions).toEqual({ openai: { serviceTier: "priority" } })
-      const prepared = yield* LLMClient.prepare(input)
+      const prepared = yield* compileRequest(input)

      expect(prepared.body).toMatchObject({ service_tier: "priority" })
      expect(prepared.body).not.toHaveProperty("serviceTier")
@@ -118,7 +119,7 @@ describe("OpenAI Responses route", () => {

  it.effect("passes through custom OpenAI reasoning effort strings", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLMRequest.update(request, { providerOptions: { openai: { reasoningEffort: "experimental" } } }),
      )

@@ -128,7 +129,7 @@ describe("OpenAI Responses route", () => {

  it.effect("omits unsupported semantic service tiers", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLMRequest.update(request, { providerOptions: { openai: { serviceTier: "unsupported" } } }),
      )

@@ -138,7 +139,7 @@ describe("OpenAI Responses route", () => {

  it.effect("flattens top-level object unions in function schemas", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLMRequest.update(request, {
          tools: [
            ToolDefinition.make({
@@ -191,7 +192,7 @@ describe("OpenAI Responses route", () => {

  it.effect("lowers chronological system updates to escaped user wrappers in order", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -217,7 +218,7 @@ describe("OpenAI Responses route", () => {

  it.effect("prepares OpenAI Responses WebSocket target", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLMRequest.update(request, {
          model: OpenAIResponses.webSocketRoute
            .with({ endpoint: { baseURL: "https://api.openai.test/v1/" }, auth: Auth.bearer("test") })
@@ -395,7 +396,7 @@ describe("OpenAI Responses route", () => {

  it.effect("prepares function call and function output input items", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_tool_result",
          model,
@@ -432,7 +433,7 @@ describe("OpenAI Responses route", () => {
        content: [],
        structured: {},
      }
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -453,7 +454,7 @@ describe("OpenAI Responses route", () => {

  it.effect("keeps primitive tool errors as plain text", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -469,7 +470,7 @@ describe("OpenAI Responses route", () => {

  it.effect("keeps non-JSON tool errors as plain text", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -487,7 +488,7 @@ describe("OpenAI Responses route", () => {
  // image data is not JSON-stringified into `function_call_output.output`.
  it.effect("lowers image tool-result content as structured input_image items", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_tool_result_image",
          model,
@@ -516,7 +517,7 @@ describe("OpenAI Responses route", () => {

  it.effect("lowers single-image tool-result content as structured input_image array", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_tool_result_image_only",
          model,
@@ -540,7 +541,7 @@ describe("OpenAI Responses route", () => {

  it.effect("lowers PDF tool-result content as structured input_file array", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_tool_result_pdf",
          model,
@@ -575,7 +576,7 @@ describe("OpenAI Responses route", () => {

  it.effect("uses xAI inline file encoding for PDF tool results", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model: xaiModel,
          messages: [
@@ -610,7 +611,7 @@ describe("OpenAI Responses route", () => {

  it.effect("rejects unsupported media in tool-result content with a clear error", () =>
    Effect.gen(function* () {
-      const error = yield* LLMClient.prepare(
+      const error = yield* compileRequest(
        LLM.request({
          id: "req_tool_result_unsupported_media",
          model,
@@ -633,7 +634,7 @@ describe("OpenAI Responses route", () => {

  it.effect("prepares the composed native continuation request", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        continuationRequest({
          id: "req_native_continuation_openai",
          model,
@@ -675,13 +676,13 @@ describe("OpenAI Responses route", () => {

  it.effect("maps OpenAI provider options to Responses options", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model: OpenAI.configure({ baseURL: "https://api.openai.test/v1/", apiKey: "test" }).model("gpt-5.2"),
          prompt: "think",
+          cache: { mode: "auto", key: "session_123" },
          providerOptions: {
            openai: {
-              promptCacheKey: "session_123",
              reasoningEffort: "high",
              reasoningSummary: "auto",
              include: ["reasoning.encrypted_content"],
@@ -700,7 +701,7 @@ describe("OpenAI Responses route", () => {

  it.effect("accepts the full ResponseIncludable union", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          prompt: "hi",
@@ -722,7 +723,7 @@ describe("OpenAI Responses route", () => {

  it.effect("filters unknown includable values out of the include array", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          prompt: "hi",
@@ -739,7 +740,7 @@ describe("OpenAI Responses route", () => {

  it.effect("treats an explicit empty include as no include at all", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({ model, prompt: "hi", providerOptions: { openai: { include: [] } } }),
      )

@@ -749,7 +750,7 @@ describe("OpenAI Responses route", () => {

  it.effect("treats an all-invalid include as no include at all", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({ model, prompt: "hi", providerOptions: { openai: { include: ["bogus.thing"] } } }),
      )

@@ -759,7 +760,7 @@ describe("OpenAI Responses route", () => {

  it.effect("omits include when no include is set", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({ model, prompt: "hi", providerOptions: { openai: { store: false } } }),
      )

@@ -773,7 +774,7 @@ describe("OpenAI Responses route", () => {
      // reasoningSummary: "auto" by default. Without `include`, a follow-up
      // turn cannot replay reasoning state, so the facade also opts into
      // `reasoning.encrypted_content` automatically.
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model: OpenAI.configure({ baseURL: "https://api.openai.test/v1/", apiKey: "test" }).responses("gpt-5.2"),
          prompt: "hi",
@@ -788,7 +789,7 @@ describe("OpenAI Responses route", () => {

  it.effect("lets callers opt out of the GPT-5 default include", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model: OpenAI.configure({ baseURL: "https://api.openai.test/v1/", apiKey: "test" }).responses("gpt-5.2"),
          prompt: "hi",
@@ -800,17 +801,16 @@ describe("OpenAI Responses route", () => {
    }),
  )

-  it.effect("request OpenAI provider options override route defaults", () =>
+  it.effect("maps the request prompt cache key", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model: OpenAI.configure({
            baseURL: "https://api.openai.test/v1/",
            apiKey: "test",
-            providerOptions: { openai: { promptCacheKey: "model_cache" } },
          }).model("gpt-4.1-mini"),
          prompt: "no cache",
-          providerOptions: { openai: { promptCacheKey: "request_cache" } },
+          cache: { mode: "auto", key: "request_cache" },
        }),
      )

@@ -934,9 +934,7 @@ describe("OpenAI Responses route", () => {
        },
      ])

-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
-        LLM.request({ model, messages: [response.message] }),
-      )
+      const prepared = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
      expect(prepared.body.input).toEqual([
        {
          role: "assistant",
@@ -1270,7 +1268,7 @@ describe("OpenAI Responses route", () => {

  it.effect("preserves assistant content order around reasoning items", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_reasoning_order",
          model,
@@ -1308,7 +1306,7 @@ describe("OpenAI Responses route", () => {

  it.effect("references stored reasoning items by id", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -1330,7 +1328,7 @@ describe("OpenAI Responses route", () => {

  it.effect("references stored provider-executed hosted tool results by id", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -1367,7 +1365,7 @@ describe("OpenAI Responses route", () => {
  it.effect("continues stateless hosted image generation with the generated image", () =>
    Effect.gen(function* () {
      const imageTool = OpenAI.imageGeneration({ action: "edit" })
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
@@ -1408,7 +1406,7 @@ describe("OpenAI Responses route", () => {

  it.effect("joins streamed summary blocks into one continuation reasoning item", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_multi_summary_continuation",
          model,
@@ -1445,7 +1443,7 @@ describe("OpenAI Responses route", () => {

  it.effect("skips non-persisted reasoning ids without encrypted state", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_reasoning_without_encrypted_state",
          model,
@@ -1762,7 +1760,7 @@ describe("OpenAI Responses route", () => {

  it.effect("lowers user image and PDF content", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          id: "req_media",
          model,
@@ -1793,7 +1791,7 @@ describe("OpenAI Responses route", () => {

  it.effect("uses xAI inline file encoding for user PDFs", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model: xaiModel,
          messages: [
@@ -1825,7 +1823,7 @@ describe("OpenAI Responses route", () => {

  it.effect("rejects unsupported user media content", () =>
    Effect.gen(function* () {
-      const error = yield* LLMClient.prepare(
+      const error = yield* compileRequest(
        LLM.request({
          id: "req_media",
          model,
@@ -1,7 +1,8 @@
 import { describe, expect } from "bun:test"
 import { Effect } from "effect"
-import { LLM, Message } from "../../src"
+import { CacheHint, LLM, Message } from "../../src"
 import { LLMClient } from "../../src/route"
+import { compileRequest } from "../../src/route/client"
 import * as OpenRouter from "../../src/providers/openrouter"
 import { it } from "../lib/effect"
 import { fixedResponse } from "../lib/http"
@@ -19,20 +20,141 @@ describe("OpenRouter", () => {
      })
      expect(model.route.endpoint.baseURL).toBe("https://openrouter.ai/api/v1")

-      const prepared = yield* LLMClient.prepare(LLM.request({ model, prompt: "Say hello." }))
+      const prepared = yield* compileRequest(LLM.request({ model, prompt: "Say hello." }))

      expect(prepared.route).toBe("openrouter")
      expect(prepared.body).toMatchObject({
        model: "openai/gpt-4o-mini",
        messages: [{ role: "user", content: "Say hello." }],
        stream: true,
+        usage: { include: true },
      })
    }),
  )

+  it.effect("lowers the native cache policy to OpenRouter cache controls", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: OpenRouter.configure({ apiKey: "test-key" }).model("anthropic/claude-sonnet-4.6"),
+          system: [
+            { type: "text", text: "Base agent", cache: new CacheHint({ type: "ephemeral", ttlSeconds: 3_600 }) },
+            { type: "text", text: "Project instructions" },
+          ],
+          tools: [{ name: "lookup", description: "Lookup", inputSchema: { type: "object", properties: {} } }],
+          prompt: "Hello",
+          cache: { mode: "explicit", tools: true, system: true, messages: { tail: 1 } },
+        }),
+      )
+
+      expect(prepared.body).toMatchObject({
+        tools: [{ cache_control: { type: "ephemeral" } }],
+        messages: [
+          {
+            role: "system",
+            content: [
+              { text: "Base agent", cache_control: { type: "ephemeral", ttl: "1h" } },
+              { text: "Project instructions", cache_control: { type: "ephemeral" } },
+            ],
+          },
+          {
+            role: "user",
+            content: [{ text: "Hello", cache_control: { type: "ephemeral" } }],
+          },
+        ],
+      })
+    }),
+  )
+
+  it.effect("lowers manual assistant and tool-result cache hints", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: OpenRouter.configure({ apiKey: "test-key" }).model("anthropic/claude-sonnet-4.6"),
+          cache: "none",
+          messages: [
+            Message.user("Call the tool"),
+            Message.assistant([
+              { type: "text", text: "Calling", cache: new CacheHint({ type: "ephemeral" }) },
+              { type: "tool-call", id: "call_1", name: "lookup", input: {} },
+            ]),
+            Message.tool({
+              id: "call_1",
+              name: "lookup",
+              result: "Done",
+              cache: new CacheHint({ type: "ephemeral", ttlSeconds: 3_600 }),
+            }),
+          ],
+        }),
+      )
+
+      expect(prepared.body.messages).toMatchObject([
+        { role: "user", content: "Call the tool" },
+        { role: "assistant", content: "Calling", cache_control: { type: "ephemeral" } },
+        { role: "tool", content: '"Done"', cache_control: { type: "ephemeral", ttl: "1h" } },
+      ])
+    }),
+  )
+
+  it.effect("caps manual cache controls at four breakpoints", () =>
+    Effect.gen(function* () {
+      const cache = new CacheHint({ type: "ephemeral" })
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: OpenRouter.configure({ apiKey: "test-key" }).model("anthropic/claude-sonnet-4.6"),
+          cache: "none",
+          system: [1, 2, 3, 4, 5].map((index) => ({ type: "text" as const, text: `System ${index}`, cache })),
+          prompt: "Hello",
+        }),
+      )
+
+      const system = prepared.body.messages[0]
+      expect(system?.role).toBe("system")
+      expect(
+        system && Array.isArray(system.content)
+          ? system.content.filter((part) => "cache_control" in part && part.cache_control !== undefined)
+          : [],
+      ).toHaveLength(4)
+    }),
+  )
+
+  it.effect("preserves cache policy hints on reasoning-only assistant messages", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: OpenRouter.configure({ apiKey: "test-key" }).model("anthropic/claude-sonnet-4.6"),
+          cache: { mode: "explicit", messages: "latest-assistant" },
+          messages: [Message.user("Think"), Message.assistant([{ type: "reasoning", text: "Reasoning" }])],
+        }),
+      )
+
+      expect(prepared.body.messages).toMatchObject([
+        { role: "user", content: "Think" },
+        { role: "assistant", cache_control: { type: "ephemeral" } },
+      ])
+    }),
+  )
+
+  it.effect("allows usage accounting to be disabled explicitly", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: OpenRouter.configure({
+            apiKey: "test-key",
+            providerOptions: { openrouter: { usage: false } },
+          }).model("openai/gpt-4o-mini"),
+          cache: "none",
+          prompt: "Hello",
+        }),
+      )
+
+      expect(prepared.body.usage).toEqual({ include: false })
+    }),
+  )
+
  it.effect("applies OpenRouter payload options from the model helper", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare(
+      const prepared = yield* compileRequest(
        LLM.request({
          model: OpenRouter.configure({
            apiKey: "test-key",
@@ -40,11 +162,18 @@ describe("OpenRouter", () => {
              openrouter: {
                usage: true,
                reasoning: { effort: "high" },
-                promptCacheKey: "session_123",
+                models: ["anthropic/claude-sonnet-4.6", "google/gemini-3.1-pro"],
+                provider: { order: ["anthropic", "google"], require_parameters: true },
+                plugins: [{ id: "response-healing" }],
+                web_search_options: { engine: "native", max_results: 3 },
+                debug: { echo_upstream_body: true },
+                user: "user_123",
+                future_option: { enabled: true },
              },
            },
          }).model("anthropic/claude-3.7-sonnet:thinking"),
          prompt: "Think briefly.",
+          cache: { mode: "auto", key: "session_123" },
        }),
      )

@@ -52,10 +181,54 @@ describe("OpenRouter", () => {
        usage: { include: true },
        reasoning: { effort: "high" },
        prompt_cache_key: "session_123",
+        models: ["anthropic/claude-sonnet-4.6", "google/gemini-3.1-pro"],
+        provider: { order: ["anthropic", "google"], require_parameters: true },
+        plugins: [{ id: "response-healing" }],
+        web_search_options: { engine: "native", max_results: 3 },
+        debug: { echo_upstream_body: true },
+        user: "user_123",
+        future_option: { enabled: true },
      })
    }),
  )

+  it.effect("filters invalid known OpenRouter options while preserving extensions", () =>
+    Effect.gen(function* () {
+      const invalid: Record<string, unknown> = {
+        usage: "yes",
+        models: "anthropic/claude-sonnet-4.6",
+        provider: [],
+        plugins: {},
+        web_search_options: [],
+        debug: [],
+        user: 123,
+        reasoning: [],
+        promptCacheKey: 123,
+        future_option: { enabled: true },
+      }
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: OpenRouter.configure({
+            apiKey: "test-key",
+            providerOptions: { openrouter: invalid },
+          }).model("openai/gpt-4o-mini"),
+          prompt: "Hello",
+        }),
+      )
+
+      expect(prepared.body).toMatchObject({ future_option: { enabled: true } })
+      expect(prepared.body).not.toHaveProperty("usage")
+      expect(prepared.body).not.toHaveProperty("models")
+      expect(prepared.body).not.toHaveProperty("provider")
+      expect(prepared.body).not.toHaveProperty("plugins")
+      expect(prepared.body).not.toHaveProperty("web_search_options")
+      expect(prepared.body).not.toHaveProperty("debug")
+      expect(prepared.body).not.toHaveProperty("user")
+      expect(prepared.body).not.toHaveProperty("reasoning")
+      expect(prepared.body).not.toHaveProperty("prompt_cache_key")
+    }),
+  )
+
  it.effect("preserves the upstream provider finish reason", () =>
    Effect.gen(function* () {
      const model = OpenRouter.configure({ apiKey: "test-key" }).model("anthropic/claude-sonnet-4.6")
@@ -100,9 +273,10 @@ describe("OpenRouter", () => {
        { type: "reasoning.text", signature: "signed", format: "anthropic-claude-v1", index: 0 },
        { type: "reasoning.encrypted", data: "opaque", format: "openai-responses-v1", index: 1 },
      ]
-      const prepared = yield* LLMClient.prepare<OpenRouter.OpenRouterBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model: OpenRouter.configure({ apiKey: "test-key" }).model("anthropic/claude-sonnet-4.6"),
+          cache: "none",
          messages: [
            Message.assistant([
              {
@@ -133,9 +307,10 @@ describe("OpenRouter", () => {
        { type: "reasoning.encrypted", id: "state", data: "opaque" },
        { type: "reasoning.encrypted", id: "state", data: "opaque" },
      ]
-      const prepared = yield* LLMClient.prepare<OpenRouter.OpenRouterBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model: OpenRouter.configure({ apiKey: "test-key" }).model("anthropic/claude-sonnet-4.6"),
+          cache: "none",
          messages: [
            Message.assistant({
              type: "reasoning",
@@ -158,9 +333,10 @@ describe("OpenRouter", () => {
        { type: "reasoning.text", id: "first", index: 0, text: "A", opaque: "first" },
        { type: "reasoning.text", id: "second", index: 1, text: "B", opaque: "second" },
      ]
-      const prepared = yield* LLMClient.prepare<OpenRouter.OpenRouterBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model: OpenRouter.configure({ apiKey: "test-key" }).model("anthropic/claude-sonnet-4.6"),
+          cache: "none",
          messages: [
            Message.assistant({
              type: "reasoning",
@@ -179,9 +355,10 @@ describe("OpenRouter", () => {

  it.effect("omits scalar reasoning without continuation details", () =>
    Effect.gen(function* () {
-      const prepared = yield* LLMClient.prepare<OpenRouter.OpenRouterBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model: OpenRouter.configure({ apiKey: "test-key" }).model("anthropic/claude-sonnet-4.6"),
+          cache: "none",
          messages: [Message.assistant({ type: "reasoning", text: "Thinking" })],
        }),
      )
@@ -3,7 +3,8 @@ import { Effect } from "effect"
 import { LLM } from "../src"
 import { OpenAIChat } from "../src/protocols"
 import { ToolSchemaProjection } from "../src/protocols/utils/tool-schema"
-import { Auth, LLMClient } from "../src/route"
+import { Auth } from "../src/route"
+import { compileRequest } from "../src/route/client"
 import { it } from "./lib/effect"

 describe("tool schema projections", () => {
@@ -79,7 +80,7 @@ describe("tool schema projections", () => {
      const model = OpenAIChat.route
        .with({ endpoint: { baseURL: "https://api.openai.test/v1/" }, auth: Auth.bearer("test") })
        .model({ id: "kimi-k2", compatibility: { toolSchema: "moonshot" } })
-      const prepared = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
+      const prepared = yield* compileRequest(
        LLM.request({
          model,
          prompt: "Use the tool.",
@@ -59,6 +59,7 @@
    "@opencode-ai/sdk": "file:vendor/opencode-ai-sdk-1.18.8-dev.tgz",
    "@opencode-ai/session-ui": "workspace:*",
    "@opencode-ai/ui": "workspace:*",
+    "@opencode-ai/util": "workspace:*",
    "@pierre/trees": "1.0.0-beta.4",
    "@sentry/solid": "catalog:",
    "@shikijs/transformers": "3.9.2",
@@ -69,7 +69,7 @@ export const DialogFork: Component = () => {
    const dir = base64Encode(sdk().directory)

    sdk()
-      .api.session.fork({ sessionID, messageID: item.id })
+      .api.session.fork({ sessionID, boundary: { type: "before", messageID: item.id } })
      .then((forked) => {
        dialog.close()
        prompt.set(restored, undefined, { dir, id: forked.id })
@@ -4,6 +4,7 @@ import { createResizeObserver } from "@solid-primitives/resize-observer"
 import { createMutation } from "@tanstack/solid-query"
 import { IconButtonV2 } from "@opencode-ai/ui/v2/icon-button-v2"
 import { Icon as IconV2 } from "@opencode-ai/ui/v2/icon"
+import { displayLabel } from "@opencode-ai/util/session-title-fallback"
 import { useGlobal } from "@/context/global"
 import { ServerConnection, serverName } from "@/context/server"
 import { displayName, projectForSession } from "@/pages/layout/helpers"
@@ -54,7 +55,10 @@ export function TabNavItem(props: {
    if (!session) return
    return projectForSession(session, serverCtx()?.projects.list() ?? [])
  })
-  const title = createMemo(() => props.session()?.title ?? props.fallbackTitle)
+  const title = createMemo(() => {
+    const session = props.session()
+    return session ? displayLabel(session) : props.fallbackTitle
+  })

  const projectName = createMemo(() => {
    const session = props.session()
@@ -143,7 +147,7 @@ export function TabNavItem(props: {
    if (!canOpenTabRename(props.dragging, editing(), rename.isPending)) return
    const session = props.session()
    if (!session) return
-    titleEl.textContent = session.title
+    titleEl.textContent = session.title ?? ""
    setEditing(true)

    requestAnimationFrame(() => {
@@ -302,7 +306,7 @@ export function TabNavItem(props: {
      }}
      data={{
        projectName: projectName(),
-        title: props.session()?.title,
+        title: title(),
        path: previewPath(),
        serverName: serverLabel(),
      }}
@@ -129,8 +129,8 @@ export const loadProjectsQuery = (scope: ServerScope, api: ProjectApi) =>
        api.list().then((projects) => {
          return projects
            .filter((p) => !!p?.id)
-            .filter((p) => !!p.worktree && !p.worktree.includes("opencode-test"))
            .map(normalizeProjectInfo)
+            .filter((p) => !!p.worktree && !p.worktree.includes("opencode-test"))
            .slice()
            .sort((a, b) => cmp(a.id, b.id))
        }),
@@ -168,6 +168,7 @@ export function sanitizeProject(project: Project) {
 export function normalizeProjectInfo(project: Project | CurrentProject): Project {
  return {
    ...project,
+    worktree: "canonical" in project ? project.canonical : project.worktree,
    vcs: project.vcs === "git" ? "git" : undefined,
  }
 }
@@ -3,6 +3,7 @@ import { useLanguage } from "@/context/language"
 import { serverName } from "@/context/server"
 import { displayName } from "@/pages/layout/helpers"
 import { makeEventListener } from "@solid-primitives/event-listener"
+import { displayLabel } from "@opencode-ai/util/session-title-fallback"
 import { createMemo, onCleanup } from "solid-js"
 import { createStore } from "solid-js/store"
 import type { HomeController } from "./home-controller"
@@ -23,7 +24,7 @@ export function createHomeSessionSearchController(home: HomeController, sessions
    if (!value) return []
    return sessions.data
      .searchRecords()
-      .filter((record) => `${record.session.title} ${record.projectName}`.toLowerCase().includes(value))
+      .filter((record) => `${displayLabel(record.session)} ${record.projectName}`.toLowerCase().includes(value))
  })
  const active = createMemo(() => {
    const records = results()
@@ -6,10 +6,10 @@ import { ButtonV2 } from "@opencode-ai/ui/v2/button-v2"
 import { Icon as IconV2 } from "@opencode-ai/ui/v2/icon"
 import { IconButtonV2 } from "@opencode-ai/ui/v2/icon-button-v2"
 import { TooltipV2 } from "@opencode-ai/ui/v2/tooltip-v2"
+import { displayLabel } from "@opencode-ai/util/session-title-fallback"
 import { useLanguage } from "@/context/language"
 import { ServerConnection } from "@/context/server"
 import { SessionTabAvatarView } from "@/pages/layout/session-tab-avatar"
-import { sessionTitle } from "@/utils/session-title"
 import { shouldOpenSessionInBackground } from "../home-session-open"
 import {
  HomeSessionStatusController,
@@ -344,7 +344,7 @@ function HomeSessionSearchResultRow(
    selected: boolean
  },
 ) {
-  const title = createMemo(() => sessionTitle(props.record.session.title) || props.record.session.id)
+  const title = createMemo(() => displayLabel(props.record.session))
  const showProjectName = () => props.showProjectName() && props.record.projectName
  const key = () => homeSessionSearchKey(props.record)

@@ -415,7 +415,7 @@ function HomeSessionGroupHeader(props: {
 }

 function HomeSessionRow(props: HomeSessionsViewProps & { record: HomeSessionRecord }) {
-  const title = createMemo(() => sessionTitle(props.record.session.title) || props.record.session.id)
+  const title = createMemo(() => displayLabel(props.record.session))
  const showProjectName = () => props.showProjectName() && props.record.projectName

  return (
@@ -5,6 +5,7 @@ import { Icon as IconV2 } from "@opencode-ai/ui/v2/icon"
 import { IconButton } from "@opencode-ai/ui/icon-button"
 import { Spinner } from "@opencode-ai/ui/spinner"
 import { Tooltip } from "@opencode-ai/ui/tooltip"
+import { displayLabel } from "@opencode-ai/util/session-title-fallback"
 import { getFilename } from "@opencode-ai/core/util/path"
 import { A, useParams } from "@solidjs/router"
 import { type Accessor, createMemo, For, type JSX, Match, Show, Switch } from "solid-js"
@@ -14,7 +15,6 @@ import { getAvatarColors, type LocalProject, useLayout } from "@/context/layout"
 import { useNotification } from "@/context/notification"
 import { usePermission } from "@/context/permission"
 import { messageAgentColor } from "@/utils/agent"
-import { sessionTitle } from "@/utils/session-title"
 import { sessionPermissionRequest } from "../session/composer/session-request-tree"
 import { childSessionOnPath, getProjectAvatarSource, hasProjectPermissions } from "./helpers"

@@ -104,7 +104,7 @@ const SessionRow = (props: {
  warmPress: () => void
  warmFocus: () => void
 }): JSX.Element => {
-  const title = () => sessionTitle(props.session.title)
+  const title = () => displayLabel(props.session)

  return (
    <A
@@ -229,7 +229,7 @@ export const SessionItem = (props: SessionItemProps): JSX.Element => {
              fallback={
                <Tooltip
                  placement={props.mobile ? "bottom" : "right"}
-                  value={sessionTitle(props.session.title)}
+                  value={displayLabel(props.session)}
                  gutter={10}
                  class="min-w-0 w-full"
                >
@@ -57,7 +57,7 @@ import { ServerConnection, serverName, useServer } from "@/context/server"
 import { useSettings } from "@/context/settings"
 import { useSync } from "@/context/sync"
 import { useTabs } from "@/context/tabs"
-import { TerminalProvider } from "@/context/terminal"
+import { TerminalProvider, useTerminal } from "@/context/terminal"
 import { PromptInput } from "@/components/prompt-input"
 import { PromptInputV2Composer, usePromptInputV2Controller } from "@/components/prompt-input-v2"
 import { useSettingsCommand } from "@/components/settings-dialog"
@@ -70,11 +70,11 @@ import {
  createSessionComposerRegionController,
  SessionComposerRegion,
 } from "@/pages/session/composer"
-import { createOpenReviewFile, createSizing, shouldShowFileTree } from "@/pages/session/helpers"
+import { createOpenReviewFile, createSessionTabs, createSizing, shouldShowFileTree } from "@/pages/session/helpers"
 import { MessageTimeline } from "@/pages/session/timeline/message-timeline"
 import { createTimelineModel } from "@/pages/session/timeline/model"
 import { type DiffStyle, SessionReviewTab, type SessionReviewTabProps } from "@/pages/session/review-tab"
-import { createSessionController } from "@/pages/session/session-controller"
+import { useSessionLayout } from "@/pages/session/session-layout"
 import { restorePromptModel, syncPromptModel, syncSessionModel } from "@/pages/session/session-model-helpers"
 import {
  clampSessionPanelWidth,
@@ -101,6 +101,7 @@ import { extractPromptFromParts } from "@/utils/prompt"
 import { formatServerError, isLocalSessionNotFoundError, isSessionNotFoundError } from "@/utils/server-errors"
 import { legacySessionHref, requireServerKey, sessionHref } from "@/utils/session-route"
 import { useUsageExceededDialogs } from "./session/usage-exceeded-dialogs"
+import { createSessionOwnership } from "./session/session-ownership"
 import { createSessionLineage } from "./session/session-lineage"

 type FollowupItem = FollowupDraft & { id: string }
@@ -365,28 +366,15 @@ export default function Page() {
  const prompt = usePrompt()
  const comments = useComments()
  const command = useCommand()
+  const terminal = useTerminal()
  const [searchParams, setSearchParams] = useSearchParams<{ prompt?: string }>()
  const location = useLocation()
  const navigate = useNavigate()
-  const isDesktop = createMediaQuery("(min-width: 768px)")
-  const newSessionDesign = createMemo(() => settings.general.newLayoutDesigns())
-  const canReview = createMemo(() => !!sync().project)
-  const session = createSessionController({
-    review: isDesktop,
-    hasReview: canReview,
-    fileBrowser: (sessionID) => newSessionDesign() && isDesktop() && !!sessionID,
-  })
-  const params = session.identity.params
-  const sessionKey = session.identity.sessionKey
-  const workspaceKey = session.identity.workspaceKey
-  const tabs = session.layout.tabs
-  const view = session.layout.view
+  const { params, sessionKey, workspaceKey, tabs, view } = useSessionLayout()
  const reviewMode = () => view().review.mode() ?? "git"
  const reviewFile = () => view().review.file()
-  const sessionOwnership = session.ownership
-  const info = session.data.info
-  const isChildSession = session.data.isChild
-  const revertMessageID = session.data.revertMessageID
+  const sessionOwnership = createSessionOwnership(sessionKey)
+  const newSessionDesign = createMemo(() => settings.general.newLayoutDesigns())

  createEffect(() => {
    if (!prompt.ready()) return
@@ -445,8 +433,8 @@ export default function Page() {
        const current = tabs().tabs()
        if (current.all.length > 0 || current.active) return

-        const all = session.tabs.normalizeAll(from.all)
-        const active = from.active ? session.tabs.normalize(from.active) : undefined
+        const all = normalizeTabs(from.all)
+        const active = from.active ? normalizeTab(from.active) : undefined
        tabs().setAll(all)
        tabs().setActive(active && all.includes(active) ? active : all[0])

@@ -457,6 +445,7 @@ export default function Page() {
    ),
  )

+  const isDesktop = createMediaQuery("(min-width: 768px)")
  const size = createSizing()
  const desktopReviewOpen = createMemo(() => isDesktop() && view().reviewPanel.opened())
  const desktopV2ReviewOpen = createMemo(() => newSessionDesign() && desktopReviewOpen() && !!params.id)
@@ -521,16 +510,46 @@ export default function Page() {
    }),
  )

+  function normalizeTab(tab: string) {
+    if (!tab.startsWith("file://")) return tab
+    return file.tab(tab)
+  }
+
+  function normalizeTabs(list: string[]) {
+    const seen = new Set<string>()
+    const next: string[] = []
+    for (const item of list) {
+      const value = normalizeTab(item)
+      if (seen.has(value)) continue
+      seen.add(value)
+      next.push(value)
+    }
+    return next
+  }
+
  const openReviewPanel = () => {
    if (!view().reviewPanel.opened()) view().reviewPanel.open()
  }

-  const activeTab = session.tabs.activeTab
-  const activeFileTab = session.tabs.activeFileTab
-  const timeline = createTimelineModel({ session })
+  const info = createMemo(() => (params.id ? sync().session.get(params.id) : undefined))
+  const isChildSession = createMemo(() => !!info()?.parentID)
+  const canReview = createMemo(() => !!sync().project)
+  const reviewTab = createMemo(() => isDesktop())
+  const tabState = createSessionTabs({
+    tabs,
+    pathFromTab: file.pathFromTab,
+    normalizeTab,
+    review: reviewTab,
+    hasReview: canReview,
+  })
+  const activeTab = tabState.activeTab
+  const activeFileTab = tabState.activeFileTab
+  const revertMessageID = createMemo(() => info()?.revert?.messageID)
+  const timeline = createTimelineModel({ sessionID: () => params.id, revertMessageID })
  const historyLoading = timeline.history.loading
  const historyMore = timeline.history.more
  const lastUserMessage = timeline.lastUserMessage
+  const messages = timeline.messages
  const messagesReady = timeline.ready
  const sessionSync = timeline.resource
  const userMessages = timeline.userMessages
@@ -1119,10 +1138,11 @@ export default function Page() {

  useComposerCommands()
  useSessionCommands({
-    session,
    navigateMessageByOffset,
    setActiveMessage,
    focusInput,
+    review: reviewTab,
+    fileBrowser: () => newSessionDesign() && isDesktop() && !!params.id,
  })
  command.register("session-palette", () => [
    {
@@ -1672,6 +1692,8 @@ export default function Page() {
    })
  }

+  const merge = (next: NonNullable<ReturnType<typeof info>>, target = sync()) => target.session.remember(next)
+
  const roll = (sessionID: string, next: NonNullable<ReturnType<typeof info>>["revert"], target = sync()) => {
    const session = target.session.get(sessionID)
    if (!session) return
@@ -1732,7 +1754,7 @@ export default function Page() {
  const queueEnabled = createMemo(() => {
    const id = params.id
    if (!id) return false
-    return settings.general.followup() === "queue" && session.data.working() && !composer.blocked() && !isChildSession()
+    return settings.general.followup() === "queue" && busy(id) && !composer.blocked() && !isChildSession()
  })

  const followupText = (item: FollowupDraft) => {
@@ -1910,7 +1932,7 @@ export default function Page() {
    if (followup.paused[sessionID]) return
    if (isChildSession()) return
    if (composer.blocked()) return
-    if (session.data.working()) return
+    if (busy(sessionID)) return

    void sendFollowup(sessionID, item.id)
  })
@@ -2058,7 +2080,6 @@ export default function Page() {
            <Show when={messagesReady() ? params.id : undefined} keyed>
              {(_id) => (
                <MessageTimeline
-                  session={session}
                  actions={actions}
                  scroll={ui.scroll}
                  onResumeScroll={resumeScroll}
@@ -2136,7 +2157,7 @@ export default function Page() {
                : undefined,
            onResponseSubmit: resumeScroll,
            openParent: () => {
-              const id = session.data.parentID()
+              const id = info()?.parentID
              if (!id) return
              navigate(
                params.serverKey
@@ -1,71 +0,0 @@
-import { describe, expect, test } from "bun:test"
-import type { AssistantMessage, Message, UserMessage } from "@opencode-ai/sdk/v2"
-import { createRoot, createSignal } from "solid-js"
-import {
-  normalizeSessionTab,
-  normalizeSessionTabs,
-  selectSessionUserMessages,
-  selectVisibleSessionUserMessages,
-} from "./session-domain"
-import { createSessionOwnership } from "./session-ownership"
-
-const user = (id: string): UserMessage => ({
-  id,
-  sessionID: "session",
-  role: "user",
-  time: { created: 0 },
-  agent: "build",
-  model: { providerID: "provider", modelID: "model" },
-})
-
-const assistant: AssistantMessage = {
-  id: "msg_2",
-  sessionID: "session",
-  role: "assistant",
-  time: { created: 0 },
-  parentID: "msg_1",
-  modelID: "model",
-  providerID: "provider",
-  mode: "build",
-  agent: "build",
-  path: { cwd: "/workspace", root: "/workspace" },
-  cost: 0,
-  tokens: { input: 0, output: 0, reasoning: 0, cache: { read: 0, write: 0 } },
-}
-
-describe("session controller invariants", () => {
-  test("normalizes file tabs once while preserving non-file tabs and order", () => {
-    const normalize = (tab: string) => normalizeSessionTab(tab, (value) => value.toLowerCase())
-
-    expect(normalizeSessionTabs(["review", "file://SRC/A.TS", "file://src/a.ts", "context"], normalize)).toEqual([
-      "review",
-      "file://src/a.ts",
-      "context",
-    ])
-  })
-
-  test("selects user history strictly before the revert boundary", () => {
-    const messages: Message[] = [user("msg_1"), assistant, user("msg_3"), user("msg_5")]
-    const users = selectSessionUserMessages(messages)
-
-    expect(users.map((message) => message.id)).toEqual(["msg_1", "msg_3", "msg_5"])
-    expect(selectVisibleSessionUserMessages(users, "msg_3").map((message) => message.id)).toEqual(["msg_1"])
-    expect(selectVisibleSessionUserMessages(users)).toBe(users)
-  })
-
-  test("rejects work captured by a previous session", () => {
-    createRoot((dispose) => {
-      const [key, setKey] = createSignal("session-a")
-      const ownership = createSessionOwnership(key)
-      const captured = ownership.capture()
-      let ran = false
-
-      setKey("session-b")
-
-      expect(captured.current()).toBe(false)
-      expect(captured.run(() => (ran = true))).toBeUndefined()
-      expect(ran).toBe(false)
-      dispose()
-    })
-  })
-})
@@ -1,101 +0,0 @@
-import type { Message, UserMessage } from "@opencode-ai/sdk/v2"
-import { createMemo, type Accessor } from "solid-js"
-import { useFile } from "@/context/file"
-import { useSync } from "@/context/sync"
-import { same } from "@/utils/same"
-import { createSessionTabs } from "./helpers"
-import {
-  normalizeSessionTab,
-  normalizeSessionTabs,
-  selectSessionUserMessages,
-  selectVisibleSessionUserMessages,
-} from "./session-domain"
-import { useSessionLayout } from "./session-layout"
-import { createSessionOwnership } from "./session-ownership"
-
-const emptyMessages: Message[] = []
-const emptyUserMessages: UserMessage[] = []
-const idle = { type: "idle" as const }
-
-export function createSessionController(input: {
-  review?: Accessor<boolean>
-  hasReview?: Accessor<boolean>
-  fileBrowser?: (sessionID: string | undefined) => boolean
-}) {
-  const file = useFile()
-  const sync = useSync()
-  const layout = useSessionLayout()
-  const sessionID = createMemo(() => layout.params.id)
-  const info = createMemo(() => {
-    const id = sessionID()
-    return id ? sync().session.get(id) : undefined
-  })
-  const parentID = createMemo(() => info()?.parentID)
-  const parent = createMemo(() => {
-    const id = parentID()
-    return id ? sync().session.get(id) : undefined
-  })
-  const status = createMemo(() => {
-    const id = sessionID()
-    return id ? (sync().data.session_status[id] ?? idle) : idle
-  })
-  const messages = createMemo(() => {
-    const id = sessionID()
-    return id ? (sync().data.message[id] ?? emptyMessages) : emptyMessages
-  })
-  const userMessages = createMemo(() => selectSessionUserMessages(messages()), emptyUserMessages, { equals: same })
-  const revertMessageID = createMemo(() => info()?.revert?.messageID)
-  const visibleUserMessages = createMemo(
-    () => selectVisibleSessionUserMessages(userMessages(), revertMessageID()),
-    emptyUserMessages,
-    { equals: same },
-  )
-  const normalizeTab = (tab: string) => normalizeSessionTab(tab, file.tab)
-  const tabs = createSessionTabs({
-    tabs: layout.tabs,
-    pathFromTab: file.pathFromTab,
-    normalizeTab,
-    review: input.review,
-    hasReview: input.hasReview,
-    fileBrowser: input.fileBrowser ? () => input.fileBrowser?.(sessionID()) ?? false : undefined,
-  })
-
-  return {
-    identity: {
-      params: layout.params,
-      sessionID,
-      sessionKey: layout.sessionKey,
-      workspaceKey: layout.workspaceKey,
-    },
-    data: {
-      info,
-      parent,
-      parentID,
-      isChild: createMemo(() => !!parentID()),
-      status,
-      working: createMemo(() => {
-        const id = sessionID()
-        return id ? sync().data.session_working(id) : false
-      }),
-      revertMessageID,
-    },
-    history: {
-      messages,
-      userMessages,
-      visibleUserMessages,
-      lastUserMessage: createMemo(() => visibleUserMessages().at(-1)),
-    },
-    layout: {
-      tabs: layout.tabs,
-      view: layout.view,
-    },
-    ownership: createSessionOwnership(layout.sessionKey),
-    tabs: {
-      ...tabs,
-      normalize: normalizeTab,
-      normalizeAll: (values: string[]) => normalizeSessionTabs(values, normalizeTab),
-    },
-  }
-}
-
-export type SessionController = ReturnType<typeof createSessionController>
@@ -1,19 +0,0 @@
-import type { Message, UserMessage } from "@opencode-ai/sdk/v2"
-
-export function normalizeSessionTab(tab: string, normalizeFileTab: (tab: string) => string) {
-  if (!tab.startsWith("file://")) return tab
-  return normalizeFileTab(tab)
-}
-
-export function normalizeSessionTabs(tabs: string[], normalize: (tab: string) => string) {
-  return [...new Set(tabs.map(normalize))]
-}
-
-export function selectSessionUserMessages(messages: Message[]) {
-  return messages.filter((message): message is UserMessage => message.role === "user")
-}
-
-export function selectVisibleSessionUserMessages(messages: UserMessage[], revertMessageID?: string) {
-  if (!revertMessageID) return messages
-  return messages.filter((message) => message.id < revertMessageID)
-}
@@ -1,26 +0,0 @@
-export function timelineChildTitle(input: {
-  parentID?: string
-  taskDescription?: string
-  title?: string
-  fallback: string
-}) {
-  if (!input.parentID) return input.title ?? ""
-  if (input.taskDescription) return input.taskDescription
-  return input.title?.replace(/\s+\(@[^)]+ subagent\)$/, "") || input.fallback
-}
-
-export function timelineRemovedSessionIDs(sessions: readonly { id: string; parentID?: string }[], sessionID: string) {
-  const removed = new Set([sessionID])
-  const byParent = Map.groupBy(
-    sessions.filter((session) => session.parentID),
-    (session) => session.parentID!,
-  )
-  const visit = (id: string) =>
-    byParent.get(id)?.forEach((child) => {
-      if (removed.has(child.id)) return
-      removed.add(child.id)
-      visit(child.id)
-    })
-  visit(sessionID)
-  return removed
-}
@@ -1,30 +0,0 @@
-import { describe, expect, test } from "bun:test"
-import { timelineChildTitle, timelineRemovedSessionIDs } from "./controller-projection"
-
-describe("timeline controller", () => {
-  test("projects child titles from task descriptions and session fallbacks", () => {
-    expect(timelineChildTitle({ title: "Root", fallback: "New session" })).toBe("Root")
-    expect(
-      timelineChildTitle({ parentID: "parent", taskDescription: "Investigate timeline", fallback: "New session" }),
-    ).toBe("Investigate timeline")
-    expect(
-      timelineChildTitle({ parentID: "parent", title: "Fallback (@build subagent)", fallback: "New session" }),
-    ).toBe("Fallback")
-    expect(timelineChildTitle({ parentID: "parent", fallback: "New session" })).toBe("New session")
-  })
-
-  test("collects the removed session and all descendants", () => {
-    const removed = timelineRemovedSessionIDs(
-      [
-        { id: "root" },
-        { id: "child", parentID: "root" },
-        { id: "grandchild", parentID: "child" },
-        { id: "sibling", parentID: "root" },
-        { id: "unrelated" },
-      ],
-      "root",
-    )
-
-    expect([...removed]).toEqual(["root", "child", "grandchild", "sibling"])
-  })
-})
@@ -1,328 +0,0 @@
-import type { Message, Part, UserMessage } from "@opencode-ai/sdk/v2"
-import { Button } from "@opencode-ai/ui/button"
-import { Dialog } from "@opencode-ai/ui/dialog"
-import { DialogFooter, DialogHeader, DialogTitleGroup, DialogV2 } from "@opencode-ai/ui/v2/dialog-v2"
-import { ButtonV2 } from "@opencode-ai/ui/v2/button-v2"
-import { useNavigate } from "@solidjs/router"
-import { createEffect, createMemo, on, type Accessor } from "solid-js"
-import { createStore, produce } from "solid-js/store"
-import { notifySessionTabsRemoved } from "@/components/titlebar-session-events"
-import { useDialog } from "@opencode-ai/ui/context/dialog"
-import { useLanguage } from "@/context/language"
-import { usePlatform } from "@/context/platform"
-import { useServerSDK } from "@/context/server-sdk"
-import { useSettings } from "@/context/settings"
-import { useSDK } from "@/context/sdk"
-import { useSync } from "@/context/sync"
-import { useTabs } from "@/context/tabs"
-import type { SessionController } from "@/pages/session/session-controller"
-import { legacySessionHref, requireServerKey, sessionHref } from "@/utils/session-route"
-import { sessionTitle } from "@/utils/session-title"
-import { showToast } from "@/utils/toast"
-import { timelineChildTitle, timelineRemovedSessionIDs } from "./controller-projection"
-import { createTimelineProjection } from "./projection"
-
-const emptyMessages: Message[] = []
-const emptyParts: Part[] = []
-const taskDescription = (part: Part, sessionID: string): string | undefined => {
-  if (part.type !== "tool" || part.tool !== "task") return undefined
-  const metadata = "metadata" in part.state ? part.state.metadata : undefined
-  if (metadata?.sessionId !== sessionID) return undefined
-  const value = part.state.input?.description
-  if (typeof value === "string" && value) return value
-  return undefined
-}
-
-export type TimelineSessionSource = {
-  identity: Pick<SessionController["identity"], "params" | "sessionID" | "sessionKey">
-  data: Pick<SessionController["data"], "info" | "parent" | "parentID" | "status">
-  history: Pick<SessionController["history"], "messages">
-}
-
-export function createTimelineController(input: {
-  session: TimelineSessionSource
-  userMessages: Accessor<UserMessage[]>
-}) {
-  const navigate = useNavigate()
-  const serverSDK = useServerSDK()
-  const sdk = useSDK()
-  const sync = useSync()
-  const settings = useSettings()
-  const tabs = useTabs()
-  const dialog = useDialog()
-  const language = useLanguage()
-  const platform = usePlatform()
-  const params = input.session.identity.params
-  const sessionKey = input.session.identity.sessionKey
-  const sessionID = input.session.identity.sessionID
-  const status = input.session.data.status
-  const messages = input.session.history.messages
-  const projectedMessages = createMemo(() => {
-    const id = sessionID()
-    if (!id) return []
-    const visible = new Set(input.userMessages().map((message) => message.id))
-    const boundary = messages().find((message) => message.role === "user" && !visible.has(message.id))?.id
-    const projected = sync().data.session_message[id] ?? []
-    return boundary ? projected.filter((message) => message.id < boundary) : projected
-  })
-  const info = input.session.data.info
-  const titleValue = createMemo(() => info()?.title)
-  const titleLabel = createMemo(() => sessionTitle(titleValue()))
-  const shareUrl = createMemo(() => info()?.share?.url)
-  const shareEnabled = createMemo(() => sync().data.config.share !== "disabled")
-  const parentID = input.session.data.parentID
-  const parent = input.session.data.parent
-  const parentMessages = createMemo(() => {
-    const id = parentID()
-    return id ? (sync().data.message[id] ?? emptyMessages) : emptyMessages
-  })
-  const parentTitle = createMemo(() => sessionTitle(parent()?.title) ?? language.t("command.session.new"))
-  const parts = (messageID: string) => sync().data.part[messageID] ?? emptyParts
-  const part = (messageID: string, partID: string) => parts(messageID).find((item) => item.id === partID)
-  const childTaskDescription = createMemo(() => {
-    const id = sessionID()
-    if (!id) return undefined
-    return parentMessages()
-      .flatMap((message) => parts(message.id))
-      .map((item) => taskDescription(item, id))
-      .findLast((value): value is string => !!value)
-  })
-  const childTitle = createMemo(() => {
-    return timelineChildTitle({
-      parentID: parentID(),
-      taskDescription: childTaskDescription(),
-      title: titleLabel(),
-      fallback: language.t("command.session.new"),
-    })
-  })
-  const showHeader = createMemo(() => !!(titleValue() || parentID()))
-  const projection = createTimelineProjection({
-    messages,
-    userMessages: input.userMessages,
-    sessionMessages: projectedMessages,
-    parts,
-    status,
-    showReasoningSummaries: settings.general.showReasoningSummaries,
-    inlineComments: settings.general.newLayoutDesigns,
-  })
-  const [pending, setPending] = createStore({ rename: false, share: false, unshare: false })
-
-  const errorMessage = (error: unknown) => {
-    if (error && typeof error === "object" && "data" in error) {
-      const data = error.data
-      if (data && typeof data === "object" && "message" in data && typeof data.message === "string") return data.message
-    }
-    if (error instanceof Error) return error.message
-    return language.t("common.requestFailed")
-  }
-  const rename = async (title: string) => {
-    const id = sessionID()
-    if (!id || pending.rename) return false
-    const next = title.trim()
-    if (!next || next === (titleLabel() ?? "")) return true
-    setPending("rename", true)
-    const success = await sdk()
-      .api.session.rename({ sessionID: id, title: next })
-      .then(() => true)
-      .catch((error) => {
-        showToast({ title: language.t("common.requestFailed"), description: errorMessage(error) })
-        return false
-      })
-    setPending("rename", false)
-    if (!success) return false
-    sync().set(
-      produce((draft) => {
-        const index = draft.session.findIndex((session) => session.id === id)
-        if (index !== -1) draft.session[index].title = next
-      }),
-    )
-    return true
-  }
-  const share = async () => {
-    const id = sessionID()
-    if (!id || pending.share || !shareEnabled()) return
-    setPending("share", true)
-    await serverSDK()
-      .client.session.share({ sessionID: id })
-      .catch((error) => console.error("Failed to share session", error))
-    setPending("share", false)
-  }
-  const unshare = async () => {
-    const id = sessionID()
-    if (!id || pending.unshare || !shareEnabled()) return
-    setPending("unshare", true)
-    await serverSDK()
-      .client.session.unshare({ sessionID: id })
-      .catch((error) => console.error("Failed to unshare session", error))
-    setPending("unshare", false)
-  }
-  const href = (id: string) =>
-    params.serverKey ? sessionHref(requireServerKey(params.serverKey), id) : legacySessionHref(sdk().directory, id)
-  const navigateAfterRemoval = (id: string, parent?: string, next?: string) => {
-    if (params.id !== id) return
-    if (parent) return navigate(href(parent))
-    if (next) return navigate(href(next))
-    if (params.serverKey)
-      return tabs.newDraft({ server: requireServerKey(params.serverKey), directory: sdk().directory })
-    navigate(`/${params.dir}/session`)
-  }
-  const archive = async (id: string) => {
-    const session = sync().session.get(id)
-    if (!session || (await sdk().protocol) !== "v1") return
-    const index = sync().data.session.findIndex((item) => item.id === id)
-    const next = index === -1 ? undefined : (sync().data.session[index + 1] ?? sync().data.session[index - 1])
-    await sdk()
-      .client.session.update({ sessionID: id, directory: sdk().directory, time: { archived: Date.now() } })
-      .then(() => {
-        sync().set(
-          produce((draft) => {
-            const index = draft.session.findIndex((item) => item.id === id)
-            if (index !== -1) draft.session.splice(index, 1)
-          }),
-        )
-        sync().session.evict(id)
-        void navigateAfterRemoval(id, session.parentID, next?.id)
-        notifySessionTabsRemoved({ directory: sdk().directory, sessionIDs: [id] })
-      })
-      .catch((error) => showToast({ title: language.t("common.requestFailed"), description: errorMessage(error) }))
-  }
-  const remove = async (id: string) => {
-    const session = sync().session.get(id)
-    if (!session) return false
-    const sessions = sync().data.session.filter((item) => !item.parentID && !item.time?.archived)
-    const index = sessions.findIndex((item) => item.id === id)
-    const next = index === -1 ? undefined : (sessions[index + 1] ?? sessions[index - 1])
-    const success = await sdk()
-      .api.session.remove({ sessionID: id })
-      .then(() => true)
-      .catch((error) => {
-        showToast({ title: language.t("session.delete.failed.title"), description: errorMessage(error) })
-        return false
-      })
-    if (!success) return false
-    const removed = timelineRemovedSessionIDs(sync().data.session, id)
-    void navigateAfterRemoval(id, session.parentID, next?.id)
-    sync().set(produce((draft) => void (draft.session = draft.session.filter((item) => !removed.has(item.id)))))
-    removed.forEach((sessionID) => sync().session.evict(sessionID))
-    notifySessionTabsRemoved({ directory: sdk().directory, sessionIDs: [...removed] })
-    return true
-  }
-
-  function DeleteDialog(props: { sessionID: string }) {
-    const name = createMemo(
-      () => sessionTitle(sync().session.get(props.sessionID)?.title) ?? language.t("command.session.new"),
-    )
-    const confirm = async () => {
-      await remove(props.sessionID)
-      dialog.close()
-    }
-    if (settings.general.newLayoutDesigns())
-      return (
-        <DialogV2 fit>
-          <DialogHeader hideClose>
-            <DialogTitleGroup
-              title={language.t("session.delete.title")}
-              description={language.t("session.delete.confirm", { name: name() })}
-            />
-          </DialogHeader>
-          <DialogFooter>
-            <ButtonV2 variant="ghost" onClick={() => dialog.close()}>
-              {language.t("common.cancel")}
-            </ButtonV2>
-            <ButtonV2 variant="danger" onClick={confirm}>
-              {language.t("session.delete.button")}
-            </ButtonV2>
-          </DialogFooter>
-        </DialogV2>
-      )
-    return (
-      <Dialog title={language.t("session.delete.title")} fit>
-        <div class="flex flex-col gap-4 pl-6 pr-2.5 pb-3">
-          <div class="flex flex-col gap-1">
-            <span class="text-14-regular text-text-strong">
-              {language.t("session.delete.confirm", { name: name() })}
-            </span>
-          </div>
-          <div class="flex justify-end gap-2">
-            <Button variant="ghost" size="large" onClick={() => dialog.close()}>
-              {language.t("common.cancel")}
-            </Button>
-            <Button variant="primary" size="large" onClick={confirm}>
-              {language.t("session.delete.button")}
-            </Button>
-          </div>
-        </div>
-      </Dialog>
-    )
-  }
-
-  createEffect(
-    on(
-      () => [parentID(), childTaskDescription()] as const,
-      ([id, description]) => {
-        if (!id || description || sync().data.message[id] !== undefined) return
-        void sync().session.sync(id)
-      },
-      { defer: true },
-    ),
-  )
-
-  return {
-    data: {
-      sessionKey,
-      sessionID,
-      status,
-      titleValue,
-      titleLabel,
-      shareUrl,
-      shareEnabled,
-      parentID,
-      parentTitle,
-      childTitle,
-      showHeader,
-      parts,
-      part,
-      projection,
-      newLayoutDesigns: settings.general.newLayoutDesigns,
-      showReasoningSummaries: settings.general.showReasoningSummaries,
-      shellToolPartsExpanded: settings.general.shellToolPartsExpanded,
-      editToolPartsExpanded: settings.general.editToolPartsExpanded,
-    },
-    pending: {
-      rename: () => pending.rename,
-      share: () => pending.share,
-      unshare: () => pending.unshare,
-    },
-    action: {
-      rename,
-      share,
-      unshare,
-      archive,
-      showDelete: (id: string) => dialog.show(() => <DeleteDialog sessionID={id} />),
-      navigateParent: () => {
-        const id = parentID()
-        if (id) navigate(href(id))
-      },
-      viewShare: () => {
-        const url = shareUrl()
-        if (url) platform.openLink(url)
-      },
-      copyShareUrl: async () => {
-        const url = shareUrl()
-        if (!url) return
-        await navigator.clipboard.writeText(url).then(
-          () =>
-            showToast({
-              variant: "success",
-              icon: "circle-check",
-              title: language.t("session.share.copy.copied"),
-              description: url,
-            }),
-          (error) => showToast({ title: language.t("common.requestFailed"), description: errorMessage(error) }),
-        )
-      },
-    },
-  }
-}
-
-export type TimelineController = ReturnType<typeof createTimelineController>
@@ -11,8 +11,10 @@ import {
  type Accessor,
  type JSX,
 } from "solid-js"
-import { createStore } from "solid-js/store"
+import { createStore, produce } from "solid-js/store"
 import { Dynamic } from "solid-js/web"
+import { useNavigate } from "@solidjs/router"
+import { useMutation } from "@tanstack/solid-query"
 import { createVirtualizer, defaultRangeExtractor, elementScroll, type VirtualItem } from "@tanstack/solid-virtual"
 import { Accordion } from "@opencode-ai/ui/accordion"
 import { Button } from "@opencode-ai/ui/button"
@@ -33,6 +35,8 @@ import { Icon as IconV2 } from "@opencode-ai/ui/v2/icon"
 import { IconButtonV2 } from "@opencode-ai/ui/v2/icon-button-v2"
 import { DropdownMenu } from "@opencode-ai/ui/dropdown-menu"
 import { MenuV2 } from "@opencode-ai/ui/v2/menu-v2"
+import { Dialog } from "@opencode-ai/ui/dialog"
+import { DialogFooter, DialogHeader, DialogTitleGroup, DialogV2 } from "@opencode-ai/ui/v2/dialog-v2"
 import { InlineInput } from "@opencode-ai/ui/inline-input"
 import { ButtonV2 } from "@opencode-ai/ui/v2/button-v2"
 import { SessionRetry } from "@opencode-ai/session-ui/session-retry"
@@ -41,22 +45,43 @@ import { StickyAccordionHeader } from "@opencode-ai/ui/sticky-accordion-header"
 import { TextField } from "@opencode-ai/ui/text-field"
 import { TextReveal } from "@opencode-ai/ui/text-reveal"
 import { TextShimmer } from "@opencode-ai/ui/text-shimmer"
-import type { AssistantMessage, ToolPart, UserMessage } from "@opencode-ai/sdk/v2"
+import type {
+  AssistantMessage,
+  Message as MessageType,
+  Part as PartType,
+  ToolPart,
+  UserMessage,
+} from "@opencode-ai/sdk/v2"
+import { showToast } from "@/utils/toast"
 import { getDirectory, getFilename } from "@opencode-ai/core/util/path"
 import { Popover as KobaltePopover } from "@kobalte/core/popover"
 import { normalize } from "@opencode-ai/session-ui/session-diff"
 import { useFileComponent } from "@opencode-ai/ui/context/file"
 import { shouldMarkBoundaryGesture, normalizeWheelDelta } from "@/pages/session/message-gesture"
 import { SessionContextUsage } from "@/components/session-context-usage"
+import { useDialog } from "@opencode-ai/ui/context/dialog"
 import { useLanguage } from "@/context/language"
+import { useSessionKey } from "@/pages/session/session-layout"
+import { useServerSDK } from "@/context/server-sdk"
+import { usePlatform } from "@/context/platform"
+import { useSettings } from "@/context/settings"
+import { useTabs } from "@/context/tabs"
+import { legacySessionHref, requireServerKey, sessionHref } from "@/utils/session-route"
+import { useSDK } from "@/context/sdk"
+import { useSync } from "@/context/sync"
+import { notifySessionTabsRemoved } from "@/components/titlebar-session-events"
+import { displayLabel } from "@opencode-ai/util/session-title-fallback"
 import { scheduleConnectedMeasure } from "./measure"
 import { observeElementOffsetReconnectAware } from "./observe-element-offset"
+import { createTimelineProjection } from "./projection"
 import { MessageComment, SummaryDiff, TimelineRow, TimelineRowMap } from "./rows"
 import { filterVirtualIndexes } from "./virtual-items"
-import { createTimelineController, type TimelineController, type TimelineSessionSource } from "./controller"

+const emptyMessages: MessageType[] = []
+const emptyParts: PartType[] = []
 const emptyTools: ToolPart[] = []
 const emptyAssistantMessages: AssistantMessage[] = []
+const idle = { type: "idle" as const }

 type FramedTimelineRow = Exclude<TimelineRow.TimelineRow, { _tag: "TurnGap" }>
 type TimelineRowByTag<T extends TimelineRow.TimelineRow["_tag"]> = Extract<TimelineRow.TimelineRow, { _tag: T }>
@@ -64,6 +89,14 @@ type TimelineRowByTag<T extends TimelineRow.TimelineRow["_tag"]> = Extract<Timel
 const timelineFallbackItemSize = 60
 const timelineCache = new Map<string, { measurements: VirtualItem[]; toolOpen: Record<string, boolean | undefined> }>()

+const taskDescription = (part: PartType, sessionID: string) => {
+  if (part.type !== "tool" || part.tool !== "task") return
+  const metadata = "metadata" in part.state ? part.state.metadata : undefined
+  if (metadata?.sessionId !== sessionID) return
+  const value = part.state.input?.description
+  if (typeof value === "string" && value) return value
+}
+
 const boundaryTarget = (root: HTMLElement, target: EventTarget | null) => {
  const current = target instanceof Element ? target : undefined
  const nested = current?.closest("[data-scrollable]")
@@ -204,8 +237,7 @@ function TimelineDiffView(props: { diff: SummaryDiff }) {
  )
 }

-type MessageTimelineProps = {
-  session: TimelineSessionSource
+export function MessageTimeline(props: {
  actions?: UserActions
  scroll: { overflow: boolean; bottom: boolean; jump: boolean }
  onResumeScroll: () => void
@@ -225,42 +257,94 @@ type MessageTimelineProps = {
  setRevealMessage?: (fn: (id: string) => void) => void
  setScrollToEnd?: (fn: () => void) => void
  setHistoryAnchor?: (handlers: { capture: () => void; restore: (done: boolean) => void }) => void
-}
-
-export function MessageTimeline(props: MessageTimelineProps) {
-  const controller = createTimelineController({ session: props.session, userMessages: () => props.userMessages })
-  return (
-    <MessageTimelineView {...props} data={controller.data} action={controller.action} pending={controller.pending} />
-  )
-}
-
-function MessageTimelineView(
-  props: MessageTimelineProps & {
-    data: TimelineController["data"]
-    action: TimelineController["action"]
-    pending: TimelineController["pending"]
-  },
-) {
+}) {
  let touchGesture: number | undefined
+
+  const navigate = useNavigate()
+  const serverSDK = useServerSDK()
+  const sdk = useSDK()
+  const sync = useSync()
+  const settings = useSettings()
+  const tabs = useTabs()
+  const dialog = useDialog()
  const language = useLanguage()
-  const ownerSessionKey = props.data.sessionKey()
+  const { params, sessionKey } = useSessionKey()
+  const ownerSessionKey = sessionKey()
  const cached = timelineCache.get(ownerSessionKey)
  const initialMeasurements = cached?.measurements
  const coldBottomMount = !initialMeasurements?.length && props.shouldAnchorBottom()
+  const platform = usePlatform()

  const [listRoot, setListRoot] = createSignal<HTMLDivElement>()
-  const sessionID = props.data.sessionID
-  const sessionStatus = props.data.status
-  const titleLabel = props.data.titleLabel
-  const shareUrl = props.data.shareUrl
-  const shareEnabled = props.data.shareEnabled
-  const parentID = props.data.parentID
-  const parentTitle = props.data.parentTitle
-  const childTitle = props.data.childTitle
-  const showHeader = props.data.showHeader
-  const getMsgParts = props.data.parts
-  const getMsgPart = props.data.part
-  const projection = props.data.projection
+  const sessionID = createMemo(() => params.id)
+  const sessionStatus = createMemo(() => {
+    const id = sessionID()
+    if (!id) return idle
+    return sync().data.session_status[id] ?? idle
+  })
+  const sessionMessages = createMemo(() => (sessionID() ? (sync().data.message[sessionID()!] ?? []) : []))
+  const projectedMessages = createMemo(() => {
+    const id = sessionID()
+    if (!id) return []
+    const visible = new Set(props.userMessages.map((message) => message.id))
+    const boundary = sessionMessages().find((message) => message.role === "user" && !visible.has(message.id))?.id
+    const messages = sync().data.session_message[id] ?? []
+    return boundary ? messages.filter((message) => message.id < boundary) : messages
+  })
+  const info = createMemo(() => {
+    const id = sessionID()
+    if (!id) return
+    return sync().session.get(id)
+  })
+  const titleLabel = createMemo(() => {
+    const session = info()
+    if (!session) return
+    return displayLabel(session)
+  })
+  const shareUrl = createMemo(() => info()?.share?.url)
+  const shareEnabled = createMemo(() => sync().data.config.share !== "disabled")
+  const parentID = createMemo(() => info()?.parentID)
+  const parent = createMemo(() => {
+    const id = parentID()
+    if (!id) return
+    return sync().session.get(id)
+  })
+  const parentMessages = createMemo(() => {
+    const id = parentID()
+    if (!id) return emptyMessages
+    return sync().data.message[id] ?? emptyMessages
+  })
+  const parentTitle = createMemo(() => {
+    const session = parent()
+    return session ? displayLabel(session) : language.t("command.session.new")
+  })
+  const getMsgParts = (msgId: string) => sync().data.part[msgId] ?? emptyParts
+  const getMsgPart = (messageID: string, partID: string) => getMsgParts(messageID).find((part) => part.id === partID)
+  const childTaskDescription = createMemo(() => {
+    const id = sessionID()
+    if (!id) return
+    return parentMessages()
+      .flatMap((message) => getMsgParts(message.id))
+      .map((part) => taskDescription(part, id))
+      .findLast((value): value is string => !!value)
+  })
+  const childTitle = createMemo(() => {
+    if (!parentID()) return titleLabel() ?? ""
+    if (childTaskDescription()) return childTaskDescription()
+    const value = titleLabel()?.replace(/\s+\(@[^)]+ subagent\)$/, "")
+    if (value) return value
+    return language.t("command.session.new")
+  })
+  const showHeader = createMemo(() => !!(titleLabel() || parentID()))
+  const projection = createTimelineProjection({
+    messages: sessionMessages,
+    userMessages: () => props.userMessages,
+    sessionMessages: projectedMessages,
+    parts: getMsgParts,
+    status: sessionStatus,
+    showReasoningSummaries: settings.general.showReasoningSummaries,
+    inlineComments: settings.general.newLayoutDesigns,
+  })
  const activeMessageID = projection.activeMessageID
  const assistantMessagesByParent = projection.assistantMessagesByParent
  const lastAssistantGroupKey = projection.lastAssistantGroupKey
@@ -450,9 +534,9 @@ function MessageTimelineView(
    virtualizer.scrollToEnd()
  }

-  let measuredSessionKey = props.data.sessionKey()
+  let measuredSessionKey = sessionKey()
  createEffect(() => {
-    const key = props.data.sessionKey()
+    const key = sessionKey()
    timelineRows().length
    if (measuredSessionKey !== key) {
      measuredSessionKey = key
@@ -565,6 +649,88 @@ function MessageTimelineView(
    props.setScrollRef(undefined)
  })

+  const viewShare = () => {
+    const url = shareUrl()
+    if (!url) return
+    platform.openLink(url)
+  }
+
+  const errorMessage = (err: unknown) => {
+    if (err && typeof err === "object" && "data" in err) {
+      const data = (err as { data?: { message?: string } }).data
+      if (data?.message) return data.message
+    }
+    if (err instanceof Error) return err.message
+    return language.t("common.requestFailed")
+  }
+
+  const shareMutation = useMutation(() => ({
+    mutationFn: (id: string) => serverSDK().client.session.share({ sessionID: id }),
+    onError: (err) => {
+      console.error("Failed to share session", err)
+    },
+  }))
+
+  const unshareMutation = useMutation(() => ({
+    mutationFn: (id: string) => serverSDK().client.session.unshare({ sessionID: id }),
+    onError: (err) => {
+      console.error("Failed to unshare session", err)
+    },
+  }))
+
+  const titleMutation = useMutation(() => ({
+    mutationFn: (input: { id: string; title: string }) =>
+      sdk().api.session.rename({ sessionID: input.id, title: input.title }),
+    onSuccess: (_, input) => {
+      sync().set(
+        produce((draft) => {
+          const index = draft.session.findIndex((s) => s.id === input.id)
+          if (index !== -1) draft.session[index].title = input.title
+        }),
+      )
+      setTitle("editing", false)
+    },
+    onError: (err) => {
+      showToast({
+        title: language.t("common.requestFailed"),
+        description: errorMessage(err),
+      })
+    },
+  }))
+
+  const shareSession = () => {
+    const id = sessionID()
+    if (!id || shareMutation.isPending) return
+    if (!shareEnabled()) return
+    shareMutation.mutate(id)
+  }
+
+  const unshareSession = () => {
+    const id = sessionID()
+    if (!id || unshareMutation.isPending) return
+    if (!shareEnabled()) return
+    unshareMutation.mutate(id)
+  }
+  const copyShareUrl = () => {
+    const url = shareUrl()
+    if (!url) return
+    void navigator.clipboard
+      .writeText(url)
+      .then(() =>
+        showToast({
+          variant: "success",
+          icon: "circle-check",
+          title: language.t("session.share.copy.copied"),
+          description: url,
+        }),
+      )
+      .catch((err: unknown) =>
+        showToast({
+          title: language.t("common.requestFailed"),
+          description: errorMessage(err),
+        }),
+      )
+  }
  const selectShareUrlText: JSX.EventHandler<HTMLDivElement, MouseEvent> = (event) => {
    const selection = window.getSelection()
    if (!selection) return
@@ -576,7 +742,7 @@ function MessageTimelineView(

  createEffect(
    on(
-      props.data.sessionKey,
+      sessionKey,
      () =>
        setTitle({
          draft: "",
@@ -589,6 +755,18 @@ function MessageTimelineView(
    ),
  )

+  createEffect(
+    on(
+      () => [parentID(), childTaskDescription()] as const,
+      ([id, description]) => {
+        if (!id || description) return
+        if (sync().data.message[id] !== undefined) return
+        void sync().session.sync(id)
+      },
+      { defer: true },
+    ),
+  )
+
  const openTitleEditor = () => {
    if (!sessionID() || parentID()) return
    setTitle({ editing: true, draft: titleLabel() ?? "" })
@@ -600,12 +778,194 @@ function MessageTimelineView(
  }

  const closeTitleEditor = () => {
-    if (props.pending.rename()) return
+    if (titleMutation.isPending) return
    setTitle("editing", false)
  }

-  const saveTitleEditor = async () => {
-    if (await props.action.rename(title.draft)) setTitle("editing", false)
+  const saveTitleEditor = () => {
+    const id = sessionID()
+    if (!id) return
+    if (titleMutation.isPending) return
+
+    const next = title.draft.trim()
+    if (!next || next === (titleLabel() ?? "")) {
+      setTitle("editing", false)
+      return
+    }
+
+    titleMutation.mutate({ id, title: next })
+  }
+
+  const navigateAfterSessionRemoval = (sessionID: string, parentID?: string, nextSessionID?: string) => {
+    if (params.id !== sessionID) return
+    const href = (id: string) =>
+      params.serverKey ? sessionHref(requireServerKey(params.serverKey), id) : legacySessionHref(sdk().directory, id)
+    if (parentID) {
+      navigate(href(parentID))
+      return
+    }
+    if (nextSessionID) {
+      navigate(href(nextSessionID))
+      return
+    }
+    if (params.serverKey) {
+      tabs.newDraft({ server: requireServerKey(params.serverKey), directory: sdk().directory })
+      return
+    }
+    navigate(`/${params.dir}/session`)
+  }
+
+  const archiveSession = async (sessionID: string) => {
+    const session = sync().session.get(sessionID)
+    if (!session) return
+    if ((await sdk().protocol) !== "v1") return
+
+    const sessions = sync().data.session ?? []
+    const index = sessions.findIndex((s) => s.id === sessionID)
+    const nextSession = index === -1 ? undefined : (sessions[index + 1] ?? sessions[index - 1])
+
+    await sdk()
+      .client.session.update({ sessionID, directory: sdk().directory, time: { archived: Date.now() } })
+      .then(() => {
+        sync().set(
+          produce((draft) => {
+            const index = draft.session.findIndex((s) => s.id === sessionID)
+            if (index !== -1) draft.session.splice(index, 1)
+          }),
+        )
+        sync().session.evict(sessionID)
+        navigateAfterSessionRemoval(sessionID, session.parentID, nextSession?.id)
+        notifySessionTabsRemoved({ directory: sdk().directory, sessionIDs: [sessionID] })
+      })
+      .catch((err) => {
+        showToast({
+          title: language.t("common.requestFailed"),
+          description: errorMessage(err),
+        })
+      })
+  }
+
+  const deleteSession = async (sessionID: string) => {
+    const session = sync().session.get(sessionID)
+    if (!session) return false
+
+    const sessions = (sync().data.session ?? []).filter((s) => !s.parentID && !s.time?.archived)
+    const index = sessions.findIndex((s) => s.id === sessionID)
+    const nextSession = index === -1 ? undefined : (sessions[index + 1] ?? sessions[index - 1])
+
+    const result = await sdk()
+      .api.session.remove({ sessionID })
+      .then(() => true)
+      .catch((err) => {
+        showToast({
+          title: language.t("session.delete.failed.title"),
+          description: errorMessage(err),
+        })
+        return false
+      })
+
+    if (!result) return false
+
+    const removed = new Set<string>([sessionID])
+    const byParent = new Map<string, string[]>()
+    for (const item of sync().data.session) {
+      const parentID = item.parentID
+      if (!parentID) continue
+      const existing = byParent.get(parentID)
+      if (existing) {
+        existing.push(item.id)
+        continue
+      }
+      byParent.set(parentID, [item.id])
+    }
+
+    const stack = [sessionID]
+    while (stack.length) {
+      const parentID = stack.pop()
+      if (!parentID) continue
+
+      const children = byParent.get(parentID)
+      if (!children) continue
+
+      for (const child of children) {
+        if (removed.has(child)) continue
+        removed.add(child)
+        stack.push(child)
+      }
+    }
+
+    navigateAfterSessionRemoval(sessionID, session.parentID, nextSession?.id)
+
+    sync().set(
+      produce((draft) => {
+        draft.session = draft.session.filter((s) => !removed.has(s.id))
+      }),
+    )
+
+    for (const id of removed) {
+      sync().session.evict(id)
+    }
+    notifySessionTabsRemoved({ directory: sdk().directory, sessionIDs: [...removed] })
+    return true
+  }
+
+  const navigateParent = () => {
+    const id = parentID()
+    if (!id) return
+    navigate(
+      params.serverKey ? sessionHref(requireServerKey(params.serverKey), id) : legacySessionHref(sdk().directory, id),
+    )
+  }
+
+  function DialogDeleteSession(props: { sessionID: string }) {
+    const name = createMemo(() => {
+      const session = sync().session.get(props.sessionID)
+      return session ? displayLabel(session) : language.t("command.session.new")
+    })
+    const handleDelete = async () => {
+      await deleteSession(props.sessionID)
+      dialog.close()
+    }
+
+    if (settings.general.newLayoutDesigns())
+      return (
+        <DialogV2 fit>
+          <DialogHeader hideClose>
+            <DialogTitleGroup
+              title={language.t("session.delete.title")}
+              description={language.t("session.delete.confirm", { name: name() })}
+            />
+          </DialogHeader>
+          <DialogFooter>
+            <ButtonV2 variant="ghost" onClick={() => dialog.close()}>
+              {language.t("common.cancel")}
+            </ButtonV2>
+            <ButtonV2 variant="danger" onClick={handleDelete}>
+              {language.t("session.delete.button")}
+            </ButtonV2>
+          </DialogFooter>
+        </DialogV2>
+      )
+
+    return (
+      <Dialog title={language.t("session.delete.title")} fit>
+        <div class="flex flex-col gap-4 pl-6 pr-2.5 pb-3">
+          <div class="flex flex-col gap-1">
+            <span class="text-14-regular text-text-strong">
+              {language.t("session.delete.confirm", { name: name() })}
+            </span>
+          </div>
+          <div class="flex justify-end gap-2">
+            <Button variant="ghost" size="large" onClick={() => dialog.close()}>
+              {language.t("common.cancel")}
+            </Button>
+            <Button variant="primary" size="large" onClick={handleDelete}>
+              {language.t("session.delete.button")}
+            </Button>
+          </div>
+        </div>
+      </Dialog>
+    )
  }

  const workingTurn = (userMessageID: string) => sessionStatus().type !== "idle" && activeMessageID() === userMessageID
@@ -684,7 +1044,7 @@ function MessageTimelineView(
    const defaultOpen = createMemo(() => {
      const item = part()
      if (!item) return
-      return partDefaultOpen(item, props.data.shellToolPartsExpanded(), props.data.editToolPartsExpanded())
+      return partDefaultOpen(item, settings.general.shellToolPartsExpanded(), settings.general.editToolPartsExpanded())
    })

    return (
@@ -697,7 +1057,7 @@ function MessageTimelineView(
                message={message()}
                showAssistantCopyPartID={assistantCopyPartID(row().userMessageID)}
                turnDurationMs={turnDurationMs(row().userMessageID)}
-                useV2Actions={props.data.newLayoutDesigns()}
+                useV2Actions={settings.general.newLayoutDesigns()}
                defaultOpen={defaultOpen()}
                toolOpen={toolOpen[part().id] ?? defaultOpen()}
                onToolOpenChange={(open) => setToolOpen(part().id, open)}
@@ -760,8 +1120,8 @@ function MessageTimelineView(
                      <div
                        classList={{
                          "shrink-0 max-w-[260px] rounded-[6px] border-border-weak-base bg-background-stronger px-2.5 py-2": true,
-                          "border-[0.5px]": props.data.newLayoutDesigns(),
-                          border: !props.data.newLayoutDesigns(),
+                          "border-[0.5px]": settings.general.newLayoutDesigns(),
+                          border: !settings.general.newLayoutDesigns(),
                        }}
                      >
                        <div class="flex items-center gap-1.5 min-w-0 text-11-medium text-text-strong">
@@ -796,7 +1156,7 @@ function MessageTimelineView(
          if (m?.role === "user") return m
        })
        const messageComments = createMemo(() => {
-          if (!props.data.newLayoutDesigns()) return []
+          if (!settings.general.newLayoutDesigns()) return []
          return getMsgParts(userMessageRow().userMessageID).flatMap((part) => MessageComment.fromPart(part) ?? [])
        })
        return (
@@ -809,7 +1169,7 @@ function MessageTimelineView(
                      message={message()}
                      parts={getMsgParts(userMessageRow().userMessageID)}
                      actions={props.actions}
-                      useV2Actions={props.data.newLayoutDesigns()}
+                      useV2Actions={settings.general.newLayoutDesigns()}
                      comments={messageComments()}
                    />
                  </div>
@@ -857,7 +1217,7 @@ function MessageTimelineView(
            <div data-slot="session-turn-message-container" class="w-full px-4 md:px-5">
              <TimelineThinkingRow
                reasoningHeading={thinkingRow().reasoningHeading}
-                showReasoningSummaries={props.data.showReasoningSummaries()}
+                showReasoningSummaries={settings.general.showReasoningSummaries()}
              />
            </div>
          </TimelineRowFrame>
@@ -973,16 +1333,16 @@ function MessageTimelineView(
      <div
        class="absolute left-1/2 -translate-x-1/2 z-[60] pointer-events-none transition-all duration-200 ease-out"
        classList={{
-          "bottom-8": props.data.newLayoutDesigns(),
-          "bottom-6": !props.data.newLayoutDesigns(),
+          "bottom-8": settings.general.newLayoutDesigns(),
+          "bottom-6": !settings.general.newLayoutDesigns(),
          "opacity-100 translate-y-0 scale-100": props.scroll.overflow && props.scroll.jump,
          "opacity-0 translate-y-2 pointer-events-none": !props.scroll.overflow || !props.scroll.jump,
-          "scale-[0.8]": (!props.scroll.overflow || !props.scroll.jump) && props.data.newLayoutDesigns(),
-          "scale-95": (!props.scroll.overflow || !props.scroll.jump) && !props.data.newLayoutDesigns(),
+          "scale-[0.8]": (!props.scroll.overflow || !props.scroll.jump) && settings.general.newLayoutDesigns(),
+          "scale-95": (!props.scroll.overflow || !props.scroll.jump) && !settings.general.newLayoutDesigns(),
        }}
      >
        <Show
-          when={props.data.newLayoutDesigns()}
+          when={settings.general.newLayoutDesigns()}
          fallback={
            <button
              type="button"
@@ -1045,22 +1405,22 @@ function MessageTimelineView(
            classList={{
              "sticky top-0 z-30": true,
              "bg-[linear-gradient(to_bottom,var(--v2-background-bg-base)_48px,transparent)]":
-                props.data.newLayoutDesigns(),
+                settings.general.newLayoutDesigns(),
              "bg-[linear-gradient(to_bottom,var(--background-stronger)_48px,transparent)]":
-                !props.data.newLayoutDesigns(),
+                !settings.general.newLayoutDesigns(),
              "w-full": true,
              "pb-4": true,
              "pr-3": true,
-              "pl-2.5": props.data.newLayoutDesigns(),
-              "pl-2 md:pl-4": !props.data.newLayoutDesigns(),
-              "md:max-w-200 md:mx-auto 2xl:max-w-[1000px]": props.centered && !props.data.newLayoutDesigns(),
+              "pl-2.5": settings.general.newLayoutDesigns(),
+              "pl-2 md:pl-4": !settings.general.newLayoutDesigns(),
+              "md:max-w-200 md:mx-auto 2xl:max-w-[1000px]": props.centered && !settings.general.newLayoutDesigns(),
            }}
          >
            <div class="h-12 w-full flex items-center justify-between gap-2">
              <div
                classList={{
                  "flex items-center gap-1 min-w-0 flex-1": true,
-                  "pr-3": !props.data.newLayoutDesigns(),
+                  "pr-3": !settings.general.newLayoutDesigns(),
                }}
              >
                <div class="flex items-center min-w-0 flex-1 w-full">
@@ -1069,7 +1429,7 @@ function MessageTimelineView(
                      type="button"
                      data-slot="session-title-parent"
                      class="min-w-0 max-w-[40%] truncate pl-2 text-[13px] font-[530] leading-4 tracking-[-0.04px] text-v2-text-text-faint transition-colors hover:text-v2-text-text-muted"
-                      onClick={props.action.navigateParent}
+                      onClick={navigateParent}
                    >
                      {parentTitle()}
                    </button>
@@ -1090,8 +1450,8 @@ function MessageTimelineView(
                          classList={{
                            "truncate text-[13px] font-[530] leading-4 tracking-[-0.04px] text-v2-text-text-base": true,
                            "w-fit rounded-[6px] px-2 py-1 hover:bg-v2-overlay-simple-overlay-hover":
-                              props.data.newLayoutDesigns(),
-                            "grow-1 min-w-0": !props.data.newLayoutDesigns(),
+                              settings.general.newLayoutDesigns(),
+                            "grow-1 min-w-0": !settings.general.newLayoutDesigns(),
                          }}
                          onClick={openTitleEditor}
                        >
@@ -1105,14 +1465,15 @@ function MessageTimelineView(
                        }}
                        data-slot="session-title-child"
                        value={title.draft}
-                        disabled={props.pending.rename()}
+                        disabled={titleMutation.isPending}
                        classList={{
                          "block text-[13px] font-[530] leading-4 tracking-[-0.04px] text-v2-text-text-base": true,
-                          "w-full flex-1 grow-1 min-w-0 pl-1 -ml-1 rounded-[6px]": !props.data.newLayoutDesigns(),
-                          "field-sizing-content self-start rounded-[6px] px-2 py-1 ": props.data.newLayoutDesigns(),
+                          "w-full flex-1 grow-1 min-w-0 pl-1 -ml-1 rounded-[6px]": !settings.general.newLayoutDesigns(),
+                          "field-sizing-content self-start rounded-[6px] px-2 py-1 ":
+                            settings.general.newLayoutDesigns(),
                        }}
                        style={{
-                          "--inline-input-shadow": props.data.newLayoutDesigns()
+                          "--inline-input-shadow": settings.general.newLayoutDesigns()
                            ? "none"
                            : "var(--shadow-xs-border-select)",
                        }}
@@ -1140,17 +1501,17 @@ function MessageTimelineView(
                  <div
                    classList={{
                      "shrink-0 flex items-center": true,
-                      "gap-2": props.data.newLayoutDesigns(),
-                      "gap-3": !props.data.newLayoutDesigns(),
+                      "gap-2": settings.general.newLayoutDesigns(),
+                      "gap-3": !settings.general.newLayoutDesigns(),
                    }}
                  >
                    <SessionContextUsage
                      placement="bottom"
-                      buttonAppearance={props.data.newLayoutDesigns() ? "v2" : "default"}
+                      buttonAppearance={settings.general.newLayoutDesigns() ? "v2" : "default"}
                    />
                    <Show when={!parentID()}>
                      <Show
-                        when={props.data.newLayoutDesigns()}
+                        when={settings.general.newLayoutDesigns()}
                        fallback={
                          <DropdownMenu
                            gutter={4}
@@ -1213,11 +1574,13 @@ function MessageTimelineView(
                                    </DropdownMenu.ItemLabel>
                                  </DropdownMenu.Item>
                                </Show>
-                                <DropdownMenu.Item onSelect={() => void props.action.archive(id)}>
+                                <DropdownMenu.Item onSelect={() => void archiveSession(id)}>
                                  <DropdownMenu.ItemLabel>{language.t("common.archive")}</DropdownMenu.ItemLabel>
                                </DropdownMenu.Item>
                                <DropdownMenu.Separator />
-                                <DropdownMenu.Item onSelect={() => props.action.showDelete(id)}>
+                                <DropdownMenu.Item
+                                  onSelect={() => dialog.show(() => <DialogDeleteSession sessionID={id} />)}
+                                >
                                  <DropdownMenu.ItemLabel>{language.t("common.delete")}</DropdownMenu.ItemLabel>
                                </DropdownMenu.Item>
                              </DropdownMenu.Content>
@@ -1282,11 +1645,11 @@ function MessageTimelineView(
                                  {language.t("session.share.action.share")}...
                                </MenuV2.Item>
                              </Show>
-                              <MenuV2.Item onSelect={() => void props.action.archive(id)}>
+                              <MenuV2.Item onSelect={() => void archiveSession(id)}>
                                {language.t("common.archive")}
                              </MenuV2.Item>
                              <MenuV2.Separator />
-                              <MenuV2.Item onSelect={() => props.action.showDelete(id)}>
+                              <MenuV2.Item onSelect={() => dialog.show(() => <DialogDeleteSession sessionID={id} />)}>
                                {language.t("common.delete")}...
                              </MenuV2.Item>
                            </MenuV2.Content>
@@ -1298,7 +1661,7 @@ function MessageTimelineView(
                        open={share.open}
                        anchorRef={() => more}
                        placement="bottom-end"
-                        gutter={props.data.newLayoutDesigns() ? 6 : 4}
+                        gutter={settings.general.newLayoutDesigns() ? 6 : 4}
                        modal={false}
                        onOpenChange={(open) => {
                          if (open) setShare("dismiss", null)
@@ -1310,7 +1673,7 @@ function MessageTimelineView(
                            data-component="popover-content"
                            classList={{
                              "flex w-80 max-w-none flex-col items-start gap-3 rounded-[10px] border-0 bg-v2-background-bg-layer-01 p-3 shadow-[var(--v2-elevation-floating)]":
-                                props.data.newLayoutDesigns(),
+                                settings.general.newLayoutDesigns(),
                            }}
                            style={{ "min-width": "320px" }}
                            onEscapeKeyDown={(event) => {
@@ -1330,7 +1693,7 @@ function MessageTimelineView(
                            }}
                          >
                            <Show
-                              when={props.data.newLayoutDesigns()}
+                              when={settings.general.newLayoutDesigns()}
                              fallback={
                                <div class="flex flex-col p-3">
                                  <div class="flex flex-col gap-1">
@@ -1351,10 +1714,10 @@ function MessageTimelineView(
                                          size="large"
                                          variant="primary"
                                          class="w-full"
-                                          onClick={() => void props.action.share()}
-                                          disabled={props.pending.share()}
+                                          onClick={shareSession}
+                                          disabled={shareMutation.isPending}
                                        >
-                                          {props.pending.share()
+                                          {shareMutation.isPending
                                            ? language.t("session.share.action.publishing")
                                            : language.t("session.share.action.publish")}
                                        </Button>
@@ -1374,10 +1737,10 @@ function MessageTimelineView(
                                            size="large"
                                            variant="secondary"
                                            class="w-full shadow-none border border-border-weak-base"
-                                            onClick={() => void props.action.unshare()}
-                                            disabled={props.pending.unshare()}
+                                            onClick={unshareSession}
+                                            disabled={unshareMutation.isPending}
                                          >
-                                            {props.pending.unshare()
+                                            {unshareMutation.isPending
                                              ? language.t("session.share.action.unpublishing")
                                              : language.t("session.share.action.unpublish")}
                                          </Button>
@@ -1385,8 +1748,8 @@ function MessageTimelineView(
                                            size="large"
                                            variant="primary"
                                            class="w-full"
-                                            onClick={props.action.viewShare}
-                                            disabled={props.pending.unshare()}
+                                            onClick={viewShare}
+                                            disabled={unshareMutation.isPending}
                                          >
                                            {language.t("session.share.action.view")}
                                          </Button>
@@ -1414,10 +1777,10 @@ function MessageTimelineView(
                                    <ButtonV2
                                      variant="contrast"
                                      class="w-full"
-                                      onClick={() => void props.action.share()}
-                                      disabled={props.pending.share()}
+                                      onClick={shareSession}
+                                      disabled={shareMutation.isPending}
                                    >
-                                      {props.pending.share()
+                                      {shareMutation.isPending
                                        ? language.t("session.share.action.publishing")
                                        : language.t("session.share.action.publish")}
                                    </ButtonV2>
@@ -1443,7 +1806,7 @@ function MessageTimelineView(
                                        variant="ghost-muted"
                                        icon={<IconV2 name="outline-copy" />}
                                        aria-label={language.t("session.share.copy.copyLink")}
-                                        onClick={() => void props.action.copyShareUrl()}
+                                        onClick={copyShareUrl}
                                      />
                                      <IconButtonV2
                                        type="button"
@@ -1451,18 +1814,18 @@ function MessageTimelineView(
                                        variant="ghost-muted"
                                        icon={<IconV2 name="outline-square-arrow" />}
                                        aria-label={language.t("session.share.action.view")}
-                                        onClick={props.action.viewShare}
-                                        disabled={props.pending.unshare()}
+                                        onClick={viewShare}
+                                        disabled={unshareMutation.isPending}
                                      />
                                    </div>
                                    <div class="flex w-full">
                                      <ButtonV2
                                        variant="outline"
                                        class="w-full"
-                                        onClick={() => void props.action.unshare()}
-                                        disabled={props.pending.unshare()}
+                                        onClick={unshareSession}
+                                        disabled={unshareMutation.isPending}
                                      >
-                                        {props.pending.unshare()
+                                        {unshareMutation.isPending
                                          ? language.t("session.share.action.unpublishing")
                                          : language.t("session.share.action.unpublish")}
                                      </ButtonV2>
@@ -1,24 +1,23 @@
-import type { Message } from "@opencode-ai/sdk/v2"
+import type { Message, UserMessage } from "@opencode-ai/sdk/v2"
 import { createMemo, createResource, onCleanup, untrack, type Accessor } from "solid-js"
 import { useServerSync } from "@/context/server-sync"
 import { useSync } from "@/context/sync"
-import type { SessionController } from "../session-controller"
-
-export {
-  selectSessionUserMessages as selectUserMessages,
-  selectVisibleSessionUserMessages as selectVisibleUserMessages,
-} from "../session-domain"
+import { same } from "@/utils/same"

+const emptyUserMessages: UserMessage[] = []
 const sessionFreshness = 15_000

-export function createTimelineModel(input: { session: Pick<SessionController, "identity" | "history"> }) {
+export function createTimelineModel(input: {
+  sessionID: Accessor<string | undefined>
+  revertMessageID: Accessor<string | undefined>
+}) {
  const serverSync = useServerSync()
  const sync = useSync()
  let refreshFrame: number | undefined
  let refreshTimer: number | undefined

  const [resource] = createResource(
-    () => input.session.identity.sessionID(),
+    () => input.sessionID(),
    (id) => {
      clearRefresh()
      if (!id) return
@@ -30,7 +29,7 @@ export function createTimelineModel(input: { session: Pick<SessionController, "i
        refreshFrame = undefined
        refreshTimer = window.setTimeout(() => {
          refreshTimer = undefined
-          if (input.session.identity.sessionID() !== id) return
+          if (input.sessionID() !== id) return
          untrack(() => {
            if (stale) void sync().session.sync(id, { force: true })
          })
@@ -40,21 +39,33 @@ export function createTimelineModel(input: { session: Pick<SessionController, "i
      return sync().session.sync(id)
    },
  )
+  const messages = createMemo(() => {
+    const id = input.sessionID()
+    return id ? (sync().data.message[id] ?? []) : []
+  })
  const ready = createMemo(() => {
-    const id = input.session.identity.sessionID()
+    const id = input.sessionID()
    return !id || isTimelineReady(sync().data.message[id], serverSync().session.history.loading(id))
  })
+  const userMessages = createMemo(() => selectUserMessages(messages()), emptyUserMessages, { equals: same })
+  const visibleUserMessages = createMemo(
+    () => {
+      return selectVisibleUserMessages(userMessages(), input.revertMessageID())
+    },
+    emptyUserMessages,
+    { equals: same },
+  )
  const more = createMemo(() => {
-    const id = input.session.identity.sessionID()
+    const id = input.sessionID()
    return id ? sync().session.history.more(id) : false
  })
  const loading = createMemo(() => {
-    const id = input.session.identity.sessionID()
+    const id = input.sessionID()
    return id ? sync().session.history.loading(id) : false
  })
  const loadOlder = async (options?: { before?: () => void; after?: (done: boolean) => void }) => {
    return loadOlderTimeline({
-      sessionID: input.session.identity.sessionID,
+      sessionID: input.sessionID,
      more,
      loading,
      loadMore: (sessionID) => sync().session.history.loadMore(sessionID),
@@ -67,12 +78,12 @@ export function createTimelineModel(input: { session: Pick<SessionController, "i

  return {
    history: { loadOlder, loading, more },
-    lastUserMessage: input.session.history.lastUserMessage,
-    messages: input.session.history.messages,
+    lastUserMessage: createMemo(() => visibleUserMessages().at(-1)),
+    messages,
    ready,
    resource,
-    userMessages: input.session.history.userMessages,
-    visibleUserMessages: input.session.history.visibleUserMessages,
+    userMessages,
+    visibleUserMessages,
  }

  function clearRefresh() {
@@ -83,10 +94,19 @@ export function createTimelineModel(input: { session: Pick<SessionController, "i
  }
 }

+export function selectUserMessages(messages: Message[]) {
+  return messages.filter((message): message is UserMessage => message.role === "user")
+}
+
 export function isTimelineReady(messages: Message[] | undefined, loading: boolean) {
  return messages !== undefined && (messages.some((message) => message.role === "user") || !loading)
 }

+export function selectVisibleUserMessages(messages: UserMessage[], revertMessageID?: string) {
+  if (!revertMessageID) return messages
+  return messages.filter((message) => message.id < revertMessageID)
+}
+
 export async function loadOlderTimeline(input: {
  sessionID: Accessor<string | undefined>
  more: Accessor<boolean>
@@ -38,17 +38,19 @@ test("reports a divergent native offset once and ignores equal offsets and unrel
    instance.scrollOffset = offset
  })

-  document.body.append(unrelated)
-  unrelated.remove()
-  await frames(2)
-  expect(calls).toEqual([])
-
  route.remove()
  document.body.append(route)
  await new Promise((resolve) => setTimeout(resolve, 0))
  await frames(3)
  expect(calls).toEqual([[0, false]])

+  instance.scrollOffset = 79_400
+  document.body.append(unrelated)
+  unrelated.remove()
+  await frames(2)
+  expect(calls).toEqual([[0, false]])
+
+  instance.scrollOffset = 0
  route.remove()
  document.body.append(route)
  await new Promise((resolve) => setTimeout(resolve, 0))
@@ -13,25 +13,19 @@ import { useSync } from "@/context/sync"
 import { useTerminal } from "@/context/terminal"
 import { showToast } from "@/utils/toast"
 import { findLast } from "@opencode-ai/core/util/array"
+import { createSessionTabs } from "@/pages/session/helpers"
 import { extractPromptFromParts } from "@/utils/prompt"
-import type { UserMessage } from "@opencode-ai/sdk/v2"
+import { UserMessage } from "@opencode-ai/sdk/v2"
+import { useSessionLayout } from "@/pages/session/session-layout"
+import { createSessionOwnership } from "./session-ownership"
 import { useLocal } from "@/context/local"
-import type { SessionController } from "./session-controller"
-
-type SessionCommandSource = {
-  identity: SessionController["identity"]
-  data: Pick<SessionController["data"], "info" | "revertMessageID">
-  history: Pick<SessionController["history"], "userMessages" | "visibleUserMessages">
-  layout: SessionController["layout"]
-  ownership: SessionController["ownership"]
-  tabs: Pick<SessionController["tabs"], "activeFileTab" | "closableTab">
-}

 export type SessionCommandContext = {
-  session: SessionCommandSource
  navigateMessageByOffset: (offset: number) => void
  setActiveMessage: (message: UserMessage | undefined) => void
  focusInput: () => void
+  review?: () => boolean
+  fileBrowser?: () => boolean
 }

 const withCategory = (category: string) => {
@@ -55,17 +49,15 @@ export const useSessionCommands = (actions: SessionCommandContext) => {
  const layout = useLayout()
  const local = useLocal()
  const navigate = useNavigate()
-  const params = actions.session.identity.params
-  const tabs = actions.session.layout.tabs
-  const view = actions.session.layout.view
-  const sessionOwnership = actions.session.ownership
+  const { params, sessionKey, tabs, view } = useSessionLayout()
+  const sessionOwnership = createSessionOwnership(sessionKey)
  const openDialog = async <T,>(load: () => Promise<T>, show: (value: T) => void) => {
    const owner = sessionOwnership.capture()
    const value = await load()
    owner.run(() => show(value))
  }
  const runCommand = async <T,>(input: {
-    owner: ReturnType<SessionController["ownership"]["capture"]>
+    owner: ReturnType<ReturnType<typeof createSessionOwnership>["capture"]>
    prompt: T
    request: () => Promise<unknown>
    updatePrompt: (prompt: T) => void
@@ -76,13 +68,39 @@ export const useSessionCommands = (actions: SessionCommandContext) => {
    input.owner.run(input.updateViewport)
  }

-  const info = actions.session.data.info
-  const activeFileTab = actions.session.tabs.activeFileTab
-  const closableTab = actions.session.tabs.closableTab
+  const info = () => {
+    const id = params.id
+    if (!id) return
+    return sync().session.get(id)
+  }
+  const hasReview = () => !!params.id
+  const normalizeTab = (tab: string) => {
+    if (!tab.startsWith("file://")) return tab
+    return file.tab(tab)
+  }
+  const tabState = createSessionTabs({
+    tabs,
+    pathFromTab: file.pathFromTab,
+    normalizeTab,
+    review: actions.review,
+    hasReview,
+    fileBrowser: actions.fileBrowser,
+  })
+  const activeFileTab = tabState.activeFileTab
+  const closableTab = tabState.closableTab
  const shown = settings.visibility.fileTree

-  const userMessages = actions.session.history.userMessages
-  const visibleUserMessages = actions.session.history.visibleUserMessages
+  const messages = () => {
+    const id = params.id
+    if (!id) return []
+    return sync().data.message[id] ?? []
+  }
+  const userMessages = () => messages().filter((m) => m.role === "user") as UserMessage[]
+  const visibleUserMessages = () => {
+    const revert = info()?.revert?.messageID
+    if (!revert) return userMessages()
+    return userMessages().filter((m) => m.id < revert)
+  }

  const showAllFiles = () => {
    if (layout.fileTree.tab() !== "changes") return
@@ -291,7 +309,7 @@ export const useSessionCommands = (actions: SessionCommandContext) => {
    const session = sdk().api.session
    const directory = sdk().directory
    const promptSession = prompt.capture()
-    const revert = actions.session.data.revertMessageID()
+    const revert = info()?.revert?.messageID
    const messages = userMessages()
    const message = findLast(messages, (x) => !revert || x.id < revert)
    if (!message) return
@@ -320,7 +338,7 @@ export const useSessionCommands = (actions: SessionCommandContext) => {
    const messages = userMessages()
    const promptSession = prompt.capture()

-    const revertMessageID = actions.session.data.revertMessageID()
+    const revertMessageID = info()?.revert?.messageID
    if (!revertMessageID) return

    const next = messages.find((x) => x.id > revertMessageID)
@@ -26,7 +26,6 @@ function setup(
        return new Response(undefined, { status: 204 })
      if (request.method === "POST" && request.url.endsWith("/prompt")) {
        return Response.json({
-          admittedSeq: 1,
          id: "msg_1",
          sessionID: "ses_1",
          timeCreated: 1,
@@ -128,7 +128,7 @@ function createV1Api(input: CompatibleInput): CompatibleApi {
  const located = <T>(data: T, value?: { directory?: string }) => ({
    location: {
      directory: directory(value) ?? "",
-      project: { id: "", directory: directory(value) ?? "" },
+      project: { id: "", directory: directory(value) ?? "", canonical: directory(value) ?? "" },
    },
    data,
  })
@@ -229,7 +229,6 @@ function createV1Api(input: CompatibleInput): CompatibleApi {
          ],
        })
        return {
-          admittedSeq: 0,
          id: value.id ?? "",
          sessionID: value.sessionID,
          timeCreated: Date.now(),
@@ -255,7 +254,6 @@ function createV1Api(input: CompatibleInput): CompatibleApi {
          })),
        })
        return {
-          admittedSeq: 0,
          id: value.id ?? "",
          sessionID: value.sessionID,
          timeCreated: Date.now(),
@@ -280,7 +278,6 @@ function createV1Api(input: CompatibleInput): CompatibleApi {
          modelID: value.model.modelID,
        })
        return {
-          admittedSeq: 0,
          id: value.id ?? "",
          sessionID: value.sessionID,
          timeCreated: Date.now(),
@@ -301,12 +298,19 @@ function createV1Api(input: CompatibleInput): CompatibleApi {
    project: {
      ...input.current.project,
      async list() {
-        return ((await legacy().project.list()).data ?? []) as Project[]
+        return ((await legacy().project.list()).data ?? []).map((project) => ({
+          ...project,
+          canonical: project.worktree,
+        }))
      },
      async current(value?: Parameters<ServerApi["project"]["current"]>[0]) {
        const result = await legacy(value?.location).project.current()
        if (!result.data) throw new Error("Project not found")
-        return { id: result.data.id, directory: result.data.worktree } satisfies ProjectCurrent
+        return {
+          id: result.data.id,
+          directory: result.data.worktree,
+          canonical: result.data.worktree,
+        } satisfies ProjectCurrent
      },
      // async update(value: Parameters<ServerApi["project"]["update"]>[0]) {
      //   const project = (await legacy().project.list()).data?.find((item) => item.id === value.projectID)
@@ -1,7 +0,0 @@
-const pattern = /^(New session|Child session) - \d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}\.\d{3}Z$/
-
-export function sessionTitle(title?: string) {
-  if (!title) return title
-  const match = title.match(pattern)
-  return match?.[1] ?? title
-}
@@ -1,5 +1,6 @@
 import type { SessionApi, SessionInfo, SessionListInput } from "@opencode-ai/client/promise"
 import type { Session } from "@opencode-ai/sdk/v2/client"
+import { withTimestampedFallback } from "@opencode-ai/util/session-title-fallback"

 export function normalizeSessionInfo(input: SessionInfo | Session): Session {
  if (!("location" in input)) return input
@@ -13,7 +14,7 @@ export function normalizeSessionInfo(input: SessionInfo | Session): Session {
    parentID: input.parentID,
    cost: input.cost,
    tokens: input.tokens,
-    title: input.title,
+    title: withTimestampedFallback(input),
    agent: input.agent,
    model: input.model,
    version: "",
@@ -40,6 +40,9 @@
    "open": "10.1.2",
    "semver": "catalog:",
    "solid-js": "catalog:",
+    "tree-sitter-bash": "0.25.0",
+    "tree-sitter-powershell": "0.25.10",
+    "web-tree-sitter": "0.25.10",
    "uqr": "0.1.3",
    "ws": "8.21.0"
  },
@@ -62,8 +62,8 @@ for (const target of targets) {
  await rm("dist-node", { recursive: true, force: true })
  const assetHash = await hashNodeAssets(assets)
  const input = { version: Script.version, channel: Script.channel, models: modelsData, assetHash, target }
-  await build(mainConfig(input))
  await copyNodeAssets(assets)
+  await build(mainConfig(input))

  const host = target.platform === process.platform && target.arch === process.arch
  if (host) {
@@ -3,7 +3,7 @@ import { copyFile, mkdir, readdir, readFile, stat } from "node:fs/promises"
 import path from "node:path"
 import { fileURLToPath } from "node:url"
 import { getNodeAssets } from "@opentui/core/node-assets"
-import { attentionSoundAssets, type NodeTarget, photonWasmAsset } from "../src/node/target"
+import { attentionSoundAssets, type NodeTarget, photonWasmAsset, shellParserWasmAssets } from "../src/node/target"

 const dir = path.resolve(import.meta.dirname, "..")

@@ -43,6 +43,10 @@ export async function collectNodeAssets(target: NodeTarget) {
      key: photonWasmAsset,
      source: fileURLToPath(import.meta.resolve(photonWasmAsset)),
    },
+    ...Object.values(shellParserWasmAssets).map((key) => ({
+      key,
+      source: fileURLToPath(import.meta.resolve(key)),
+    })),
    ...attentionSoundAssets.map((key) => ({
      key,
      source: path.resolve(dir, "../ui/src/assets/audio", path.basename(key)),
@@ -54,8 +58,9 @@ export async function collectNodeAssets(target: NodeTarget) {
        source: path.join(ptyRoot, relative),
      })),
  ]
-  await Promise.all(assets.map((asset) => stat(asset.source)))
-  return assets
+  const unique = [...new Map(assets.map((asset) => [asset.key, asset])).values()]
+  await Promise.all(unique.map((asset) => stat(asset.source)))
+  return unique
 }

 export async function hashNodeAssets(assets: readonly NodeAsset[]) {
@@ -8,6 +8,7 @@ import {
  type SessionMessageInfo,
  type SkillInfo,
 } from "@opencode-ai/client/promise"
+import { withTimestampedFallback } from "@opencode-ai/util/session-title-fallback"
 import type {
  AgentSideConnection,
  AuthenticateRequest,
@@ -213,7 +214,7 @@ export function make(input: { readonly client: OpenCodeClient; readonly connecti
        sessions: page.data.map((session) => ({
          sessionId: session.id,
          cwd: session.location.directory,
-          title: session.title,
+          title: withTimestampedFallback(session),
          updatedAt: new Date(session.time.updated).toISOString(),
        })),
        ...(page.cursor.next ? { nextCursor: page.cursor.next } : {}),
@@ -244,7 +245,10 @@ export function make(input: { readonly client: OpenCodeClient; readonly connecti
      return {}
    },
    forkSession: async (params) => {
-      const forked = await input.client.session.fork({ sessionID: params.sessionId })
+      const forked = await input.client.session.fork({
+        sessionID: params.sessionId,
+        boundary: { type: "through" },
+      })
      const state = await attach(forked, forked.location.directory, params.mcpServers ?? [])
      await replay(state)
      return { sessionId: state.id, configOptions: configOptions(state) }
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Aiden Cline	a214ac39de	refactor(ai): unify prompt cache configuration	2026-07-31 17:52:32 -05:00
Aiden Cline	f6ea6b1762	refactor(ai): promote prompt cache key	2026-07-31 16:59:24 -05:00
opencode-agent[bot]GitHub Aiden Cline	56c6add5c3	refactor(core): remove unused helpers (#39943 ) Co-authored-by: Aiden Cline <rekram1-node@users.noreply.github.com>	2026-07-31 16:17:14 -05:00
Kit LangtonandGitHub	2cfe8883ea	feat(plugin): expose session rename and wait to plugins (#39932 )	2026-07-31 21:08:33 +00:00
Kit LangtonandGitHub	102086c50f	feat(tui): hot-reload local TUI plugins (#39776 )	2026-07-31 16:35:05 -04:00
Kit LangtonandGitHub	06e26d89ad	feat(protocol): accept title on session create (#39935 )	2026-07-31 19:39:54 +00:00
Aiden ClineandGitHub	84dd56ed34	fix(core): bound outbound image history (#39929 )	2026-07-31 13:57:04 -05:00
opencode-agent[bot]GitHub James Long	f92e490bb6	fix(tui): preserve diff hunk boundaries (#39693 ) Co-authored-by: James Long <17031+jlongster@users.noreply.github.com>	2026-07-31 14:47:36 -04:00
Kit LangtonandGitHub	0481dba88e	fix(tui): improve narrow layouts (#39918 )	2026-07-31 14:44:13 -04:00
Kit LangtonandGitHub	dd0e41c633	fix(tui): render plugin slots and routes with component semantics (#39917 )	2026-07-31 14:30:57 -04:00
opencode-agent[bot]GitHub Aiden Cline	b69e51d835	refactor(ai): simplify provider options (#39924 ) Co-authored-by: Aiden Cline <63023139+rekram1-node@users.noreply.github.com>	2026-07-31 13:27:32 -05:00
opencode-agent[bot]GitHub Aiden Cline	e60e8d9387	refactor(core): rename attachments config to media (#39927 ) Co-authored-by: Aiden Cline <rekram1-node@users.noreply.github.com>	2026-07-31 13:26:53 -05:00
Kit LangtonandGitHub	3bfce3fd2d	feat(tui): expand pasted text (#39920 )	2026-07-31 18:17:35 +00:00
Aiden ClineandGitHub	b498a5c6c4	feat(ai): expand OpenRouter native support (#39907 )	2026-07-31 12:57:18 -05:00
Kit LangtonandGitHub	db9e942398	fix(core): resize large image attachments (#39919 )	2026-07-31 13:24:05 -04:00
Kit LangtonandGitHub	0e26116f68	fix(tui): stabilize open picker	2026-07-31 16:38:48 +00:00
Aiden ClineandGitHub	3468aa0140	feat(ai): configure chat max tokens field (#39909 )	2026-07-31 11:14:08 -05:00
Kit LangtonandGitHub	7757f712a2	test(app): restore stable offset observer ordering (#39901 )	2026-07-31 12:13:16 -04:00
Aiden ClineandGitHub	7129ed6e62	fix(core): respect model input limits (#39797 )	2026-07-31 10:41:36 -05:00
Kit LangtonandGitHub	6c37842520	fix(cli): deduplicate Node asset destinations (#39900 )	2026-07-31 11:40:38 -04:00
Kit LangtonandGitHub	dc3c996892	fix(tui): stabilize generated session titles (#39894 )	2026-07-31 11:13:25 -04:00
Kit LangtonandGitHub	7fd12c560c	fix(tui): fit project paths in open menu (#39887 )	2026-07-31 11:07:12 -04:00
Kit LangtonandGitHub	7aaf4e7750	refactor(session): centralize fallback title policy (#39890 )	2026-07-31 10:56:38 -04:00
Kit LangtonandGitHub	1311d909a7	fix(tui): preserve session location during handoff (#39886 )	2026-07-31 10:54:36 -04:00
Kit LangtonandGitHub	d07d9ae0da	fix(tui): remove shells from their location (#39885 )	2026-07-31 10:46:43 -04:00
Kit LangtonandGitHub	1d18459cdc	fix(tui): align session picker scope (#39784 )	2026-07-31 10:32:16 -04:00
Kit LangtonandGitHub	db45026c6c	fix(tui): default tabs to global scope (#39783 )	2026-07-31 10:15:53 -04:00
Kit Langton	77df98db51	Revert "feat(tui): delete current session (#39750 )" This reverts commit `7814568ba0`.	2026-07-31 10:01:35 -04:00
Aiden ClineandGitHub	ff2b184af1	feat(ai): support Gemini thinking levels (#39796 )	2026-07-30 23:38:00 -05:00
Aiden ClineandGitHub	671e164f8d	refactor(core): contain Codex in OpenAI plugin (#39734 )	2026-07-30 21:43:38 -05:00
Aiden ClineandGitHub	5b4ebf3e9d	fix(core): map xAI native options (#39787 )	2026-07-30 21:42:53 -05:00
Aiden ClineandGitHub	98229d466d	feat(plugin): add session request hook (#39764 )	2026-07-30 21:11:22 -05:00
Kit LangtonandGitHub	5c7e2b5042	fix(tui): clarify open menu project labels (#39780 )	2026-07-30 21:34:29 -04:00
Kit LangtonandGitHub	0a6a5d3e80	fix(session): retry failed title generation (#39748 )	2026-07-30 21:26:44 -04:00
Kit LangtonandGitHub	eb95bd27fe	feat(session): make generated titles optional (#39747 )	2026-07-30 21:24:34 -04:00
Kit LangtonandGitHub	865f512a44	fix(tui): preserve current selection across list updates (#39774 )	2026-07-31 00:48:51 +00:00
Kit LangtonandGitHub	a460f02f67	feat(tui): inherit session directory when creating a new session (#39753 )	2026-07-30 20:31:38 -04:00
Kit LangtonandGitHub	146fdb9de1	feat(tui): add open menu for sessions and projects (#39752 )	2026-07-30 20:08:12 -04:00
Kit LangtonandGitHub	b866900417	fix(tui): name deleted session in toast (#39768 )	2026-07-30 19:59:27 -04:00
Kit LangtonandGitHub	8e3e94aa26	fix(tui): focus palette settings after layout (#39585 )	2026-07-30 19:39:16 -04:00
Aiden ClineandGitHub	cc1289048e	refactor(core): isolate AI SDK native mappings (#39761 )	2026-07-30 18:38:29 -05:00
Kit LangtonandGitHub	7814568ba0	feat(tui): delete current session (#39750 )	2026-07-30 17:06:41 -04:00
Kit LangtonandGitHub	16b247f756	fix(tui): smooth new session tab handoff (#39745 )	2026-07-30 16:17:01 -04:00
Kit LangtonandGitHub	9abd9594de	fix(simulation): mirror kitty keyboard protocol (#39741 )	2026-07-30 20:03:21 +00:00
Kit LangtonandGitHub	22d2012f75	fix(tui): register storybook only for story runs (#39733 )	2026-07-30 15:55:05 -04:00
Kit LangtonandGitHub	cc883058df	feat(tui): reopen closed session tabs (#39731 )	2026-07-30 15:54:43 -04:00
Kit LangtonandGitHub	8ebc4c6f85	fix(tui): call session tabs just tabs (#39730 )	2026-07-30 15:52:58 -04:00
Kit LangtonandGitHub	b0ed9990b3	feat(tui): add option tab shortcuts (#39725 )	2026-07-30 15:51:10 -04:00
Kit LangtonandGitHub	e1c04dcce6	feat(tui): add temporary new session tab (#39735 )	2026-07-30 15:05:34 -04:00
Kit LangtonandGitHub	188c642d8c	refactor(tui): flatten project picker list (#39726 )	2026-07-30 14:18:16 -04:00
Kit LangtonandGitHub	9d55b223bb	fix(core): stop repository discovery at nearest marker (#39714 )	2026-07-30 13:28:29 -04:00
Kit LangtonandGitHub	ce7a7e4e23	fix(tui): truncate project picker paths (#39678 )	2026-07-30 13:13:31 -04:00
Kit LangtonandGitHub	cba5ba03e3	fix(tui): show shells from session location (#39691 )	2026-07-30 13:12:01 -04:00
Kit LangtonandGitHub	49081a4e24	fix(tui): flash tabs on prompt admission (#39702 )	2026-07-30 16:57:36 +00:00
opencode-agent[bot]GitHub James Long	d6371f2fcd	fix(session): update activity on prompt (#39539 ) Co-authored-by: James Long <17031+jlongster@users.noreply.github.com>	2026-07-30 16:28:10 +00:00
Aiden ClineandGitHub	7625cbdf47	feat(core): route providers through native AI (#39615 )	2026-07-30 11:26:18 -05:00
opencode-agent[bot]GitHub Aiden Cline	0ffec67ca3	refactor: remove unused V2 code (#39699 ) Co-authored-by: Aiden Cline <rekram1-node@users.noreply.github.com>	2026-07-30 11:25:03 -05:00
Kit LangtonandGitHub	bd132f2614	refactor(core): skip redundant worktree lookup (#39692 )	2026-07-30 16:15:07 +00:00
opencode-agent[bot]GitHub Aiden Cline	dcb0df4ac1	fix(core): stop MCP SSE error reconnect loops (#39671 ) Co-authored-by: Aiden Cline <rekram1-node@users.noreply.github.com>	2026-07-30 10:53:33 -05:00
Kit LangtonandGitHub	02f3504055	fix(tui): hide redundant session directory labels (#39686 )	2026-07-30 11:37:33 -04:00
James LongandGitHub	f23ee5e4a8	refactor(core): simplify formatter selection (#39575 )	2026-07-30 10:59:01 -04:00
Kit LangtonandGitHub	77aa85c589	feat(tui): project picker with footer crossfade (#39566 )	2026-07-30 10:39:22 -04:00
Dax Raad	a618946b7e	fix(tui): reload plugins with config	2026-07-30 10:31:09 -04:00
opencode-agent[bot]GitHub Aiden Cline	f9de608dea	chore: refresh bun lockfile (#39617 ) Co-authored-by: Aiden Cline <rekram1-node@users.noreply.github.com>	2026-07-30 04:36:21 +00:00
Aiden ClineandGitHub	906dc8f5b2	fix(ai): apply catalog settings to provider models (#39613 )	2026-07-29 22:42:47 -05:00
Aiden ClineandGitHub	12971935ab	feat(core): parse shell permission commands (#39567 )	2026-07-29 22:42:18 -05:00
Kit LangtonandGitHub	78c139b8b5	feat(plugin): add ui.tabs API for session tab control (#39591 )	2026-07-29 22:57:39 -04:00
Kit LangtonandGitHub	c161978852	feat(tui): prefetch open session tabs after connect (#39589 )	2026-07-29 22:23:13 -04:00
Kit LangtonandGitHub	cb37a7166a	feat(tui): make session tab switching fast for long transcripts (#39568 )	2026-07-29 22:23:10 -04:00
Dax Raad	488445a679	fix(tui): correct project-aware session lists	2026-07-29 21:59:33 -04:00
opencode-agent[bot]GitHub Aiden Cline	97786afdd8	refactor(core): share file diff construction (#39586 ) Co-authored-by: Aiden Cline <63023139+rekram1-node@users.noreply.github.com>	2026-07-30 00:28:33 +00:00
Aiden ClineandGitHub	b1a5e8a6ae	test(tui): restore compaction event lifecycle (#39581 )	2026-07-29 18:11:58 -05:00
Aiden ClineandGitHub	0ece10af43	fix(core): add mutation permission previews (#39578 )	2026-07-29 18:04:13 -05:00
Aiden ClineandGitHub	d1a02b149c	fix(core): clarify subagent tool guidance (#39572 )	2026-07-29 16:48:49 -05:00
James LongandGitHub	94ee274aeb	feat(core): add V2 formatter runtime (#39564 )	2026-07-29 17:23:04 -04:00
Kit LangtonandGitHub	3c259fc552	feat(tui): replace scrap screen with component storybook (#39548 )	2026-07-29 15:51:15 -04:00
Aiden ClineandGitHub	210be4b749	fix(core): preserve shell output on timeout (#39559 )	2026-07-29 14:31:26 -05:00
James LongandGitHub	464649e67e	feat(tui): batch event delivery (#39551 )	2026-07-29 15:20:39 -04:00
Aiden ClineandGitHub	c8dca936b1	feat(plugin): add shell.create.before hook (#39547 )	2026-07-29 14:11:52 -05:00
Kit LangtonandGitHub	4f871906bc	fix(tui): support cd before session creation (#39555 )	2026-07-29 18:45:42 +00:00
Aiden ClineandGitHub	f7ea2fc346	test(cli): update ACP fork expectation (#39554 )	2026-07-29 13:21:34 -05:00
Aiden ClineandGitHub	5cb633a48e	feat(core): support pinned Code Mode tools (#39550 )	2026-07-29 13:16:49 -05:00
Kit LangtonandGitHub	b985d2eb8e	feat(tui): polish session tab animations (#39542 )	2026-07-29 13:50:49 -04:00
Dax Raad	014908a8d7	feat(tui): reload config file changes	2026-07-29 13:05:48 -04:00
Dax Raad	f599f8f3d3	fix(tui): guard Bun runtime plugin support	2026-07-29 12:57:35 -04:00
Aiden ClineandGitHub	b2010220f9	fix(core): clarify Code Mode tool boundary (#39540 )	2026-07-29 11:56:53 -05:00
James LongandGitHub	c2e975c4e6	refactor(plugin): expose resolved TUI theme (#39536 )	2026-07-29 12:51:55 -04:00
Dax Raad	6aa250ee5d	refactor(tui): flatten state storage path	2026-07-29 12:39:21 -04:00
Dax Raad	8f1e3ff75c	docs: record v1 to v2 database migration decisions	2026-07-29 12:37:00 -04:00
Dax Raad	5438dfb751	fix(tui): remove invalid model toasts	2026-07-29 12:36:56 -04:00
Dax Raad	4bd16d6f47	feat(tui): default tabs to cwd scope	2026-07-29 12:35:32 -04:00
Dax Raad	9ee337469d	feat(tui): add persistent storage context	2026-07-29 12:33:18 -04:00
Aiden ClineandGitHub	813c41ff6c	fix(core): simplify shell execution boundary (#39530 )	2026-07-29 11:16:45 -05:00
Dax Raad	247f14f955	feat(tui): add replaceable prompt footer slot	2026-07-29 10:16:34 -04:00
Dax Raad	bd906d468d	refactor(core): make watcher subscription effectful	2026-07-29 09:50:24 -04:00
Dax Raad	fc11ed3838	feat(session): define explicit fork boundaries	2026-07-29 09:50:24 -04:00
Dax Raad	2a85c861e0	fix(session): hide pending admission sequence	2026-07-29 09:50:24 -04:00
Shoubhit DashandGitHub	9554f9a16e	feat(ai): type Vertex request options (#39499 )	2026-07-29 18:23:00 +05:30
Shoubhit DashandGitHub	d72b428061	feat(ai): type Vertex Chat request options (#39503 )	2026-07-29 18:22:41 +05:30
Shoubhit DashandGitHub	fea17b4a0e	feat(ai): type Cloudflare request options (#39507 )	2026-07-29 18:22:16 +05:30
Shoubhit DashandGitHub	9038e44a68	feat(ai): type Copilot request options (#39496 )	2026-07-29 18:22:01 +05:30
Shoubhit DashandGitHub	3b8299e3f2	feat(ai): type Azure request options (#39498 )	2026-07-29 18:21:43 +05:30
Shoubhit DashandGitHub	f5cdf0f056	feat(ai): type Anthropic request options (#39502 )	2026-07-29 18:21:19 +05:30
Shoubhit DashandGitHub	224feff7c4	feat(ai): type compatible Responses options (#39506 )	2026-07-29 18:20:55 +05:30
Shoubhit DashandGitHub	ce2c9e7e26	feat(ai): type Google request options (#39504 )	2026-07-29 18:20:30 +05:30
Shoubhit DashandGitHub	cb80f47112	feat(ai): type Vertex Messages request options (#39501 )	2026-07-29 18:20:04 +05:30
Shoubhit DashandGitHub	b4ac939537	feat(ai): type xAI request options (#39505 )	2026-07-29 18:19:44 +05:30
Shoubhit DashandGitHub	8a96b80aec	feat(ai): type OpenAI request options (#39495 )	2026-07-29 18:19:26 +05:30
Shoubhit DashandGitHub	333a090975	feat(ai): type Vertex Responses request options (#39500 )	2026-07-29 18:19:02 +05:30
Shoubhit DashandGitHub	5a78a17e49	feat(ai): type OpenRouter request options (#39508 )	2026-07-29 18:18:43 +05:30
Shoubhit DashandGitHub	9d6af6afa4	feat(ai): type compatible request options (#39509 )	2026-07-29 18:18:24 +05:30
Shoubhit DashandGitHub	8c3e06798c	refactor(ai): limit provider option inference (#39510 )	2026-07-29 18:18:00 +05:30
Shoubhit DashandGitHub	5882b64612	feat(ai): type Anthropic-compatible request options (#39497 )	2026-07-29 18:07:24 +05:30
Shoubhit DashandGitHub	309c4fe6f0	feat(ai): infer model provider options (#39493 )	2026-07-29 18:05:56 +05:30
Shoubhit DashandGitHub	d9555f138b	refactor(ai): internalize request compilation (#39132 )	2026-07-29 16:51:11 +05:30