test(core): limit process group survival check to POSIX

Node assigns non-detached Windows children to a kill-on-parent-exit job, so the held-stdio mcp fixture cannot assert POSIX process-group survival there. Keep detached descendant, capture deadline, and success-policy coverage enabled on Windows.
fix(util): separate process exit from capture completion
2026-09-01 06:26:24 +00:00 · 2026-08-31 15:35:32 -04:00 · 2026-08-31 15:12:01 -04:00 · 2026-08-31 14:55:04 -04:00 · 2026-08-31 14:31:03 -04:00 · 2026-08-31 14:28:19 -04:00
220 changed files with 4137 additions and 4508 deletions
@@ -241,121 +241,14 @@ Constructing `stream()` or `generate()` does not record a request, invoke a resp
 Each execution does. An exhausted queue without a fallback defects immediately rather than waiting for a
 future reply.

-Generation responses remain canonical event arrays or arbitrary `Stream<LLMEvent, AIError>` values. The client consumes
+Responses remain canonical event arrays or arbitrary `Stream<LLMEvent, AIError>` values. The client consumes
 supplied streams directly, preserving failure identity, finalizers, incomplete output, and post-finish tails;
 it does not repair or truncate them.

-For explicit compaction, script a `CompactionResponse` through `push`, `always`, or `serve`. The client returns that replacement window directly, including retained user messages and usage, with the same lazy request recording and gates. Generation and compaction reject fixtures for the wrong operation instead of converting between response shapes.
-
 The published legacy `Service`, `layer`, `clientLayer`, and module-level controls remain available as adapters
 over the same implementation, including the legacy live `requests` array. New tests should use `Test` and
 `testLayer`.

-## Provider compaction
-
-Compaction is opt-in. The package supports automatic compaction in OpenAI/Azure Responses and Anthropic Messages (including Claude on Vertex and Bedrock Messages), and explicit compaction calls in OpenAI/Azure/xAI Responses. Model and deployment support still depends on the provider.
-
-This is different from prompt caching, server-side history storage, or truncation. Compaction returns provider-owned context that must be replayed to continue the conversation.
-
-### Automatic compaction
-
-Inside an `Effect.gen`, enable OpenAI compaction with typed provider options:
-
-```ts
-import { LLM, LLMClient, LLMRequest, Message } from "@opencode-ai/ai"
-import { OpenAI } from "@opencode-ai/ai/providers"
-
-const request = LLM.request({
-  model: OpenAI.configure({ apiKey }).responses("gpt-5.3-codex"),
-  messages,
-  providerOptions: {
-    contextManagement: [{ type: "compaction", compactThreshold: 200_000 }],
-  },
-})
-const response = yield * LLMClient.generate(request)
-const next = LLMRequest.update(request, {
-  messages: [...request.messages, response.message, Message.user("Continue")],
-})
-```
-
-`store: false` remains the default. Keep the entire `response.message`, not just `response.text`. Compaction events become ordered `CompactionPart`s alongside text and reasoning. The conversation contains everything needed to continue; there is no separate replay object or hidden provider transcript.
-
-A compaction part has `provider` and exactly one representation: `encrypted` for Responses, or `text` for Anthropic. Responses also preserves the optional checkpoint `id`. These fields survive message serialization without becoming visible assistant text. Sending a checkpoint to another provider or an incompatible API fails rather than silently losing context.
-
-```ts
-import { CompactionPart, ProviderID } from "@opencode-ai/ai"
-
-CompactionPart.make({ provider: ProviderID.make("openai"), id: "cmp_123", encrypted: "..." })
-CompactionPart.make({ provider: ProviderID.make("anthropic"), text: "Summary of the conversation..." })
-```
-
-For Anthropic, use:
-
-```ts
-providerOptions: {
-  contextManagement: {
-    edits: [{
-      type: "compact_20260112",
-      trigger: { type: "input_tokens", value: 150_000 },
-      pauseAfterCompaction: true,
-      instructions: "Summarize the task and decisions. Do not call tools while summarizing.",
-    }],
-  },
-}
-```
-
- The trigger is optional (provider default: 150,000 tokens), with a minimum of 50,000.
- Custom instructions replace Anthropic's default summarization instructions.
- The route adds `compact-2026-01-12` to existing beta headers, including when replaying a checkpoint without enabling new compactions.
- A pause is exposed as `response.finishReason.raw === "compaction"`. The caller explicitly issues the next request; the package never automatically resumes.
- Anthropic can return a compaction block with `content: null` when summarization fails. This becomes a compaction part with `text: null`, which is **not** a successful replacement for prior history. The package never prunes history automatically.
- `Usage` totals include all reported Anthropic `usage.iterations`, including compaction. `contextTokens` separately reports the final message iteration's inclusive input size, when available. A compaction-only pause does not report a post-compaction context size. Raw iteration usage remains in `providerMetadata`.
-
-Bedrock's Converse API does not support this feature. Select the native Claude Messages route explicitly; the default `.model(...)` remains Converse:
-
-```ts
-import { AmazonBedrock } from "@opencode-ai/ai/providers"
-
-const model = AmazonBedrock.configure({ region: "us-east-1", credentials }).messages("us.anthropic.claude-opus-4-6-v1")
-```
-
-The corresponding package entrypoint is `@opencode-ai/ai/providers/amazon-bedrock/messages`. It uses InvokeModelWithResponseStream, AWS event-stream framing, bearer or SigV4 auth, and `anthropic_beta` in the request body.
-
-### Explicit compaction
-
-`LLMClient.compact(request)` performs exactly one HTTP call to `/responses/compact`, using the selected route's endpoint, credentials, query, and HTTP middleware. It returns a `CompactionResponse` containing replacement `messages` and usage, not a normal generation response.
-
-The selected model carries explicit-compaction capability through request construction and updates. Calls using unsupported routes fail type checking. When the model is selected dynamically, narrow the request with `LLMClient.canCompact(request)` before calling `LLMClient.compact`; a model or route switch does not inherit the old capability. Runtime validation still rejects unsupported calls from untyped consumers. Capability describes the route's API, not whether every model or custom deployment supports the operation.
-
-```ts
-const compacted = yield * LLMClient.compact(request)
-const next = LLMRequest.update(request, {
-  messages: [...compacted.messages, Message.user("Continue")],
-})
-const response = yield * LLMClient.generate(next)
-```
-
-Replace the prior window with `compacted.messages`. Do not append it to the original transcript or extract only the encrypted item: the provider may retain additional messages in its output. Retained user and assistant messages remain ordinary messages with typed text, media, or reasoning parts, in their original order. Provider-specific message IDs, status, and phase use `providerMetadata`, not a raw output array hidden in an assistant message. Unsupported returned item types fail explicitly. Generation-only body overlays such as `stream` and `store` are not sent to the compact endpoint.
-
-Supported compact controls such as service tier and prompt-cache settings preserve request defaults and HTTP-overlay precedence. Retained image and file detail settings survive serialization and replay.
-
-The input must still fit the model's context window. Explicit compaction is not an overflow-recovery operation. xAI supports this explicit path, not the automatic OpenAI option. Unsupported routes, including Bedrock Mantle, do not inherit an explicit compact endpoint simply because they use a Responses protocol.
-
-### Ownership and verification
-
-The AI package transports options and typed conversation parts. It does not schedule compaction, persist Session checkpoints, select history, switch providers, or replace Core's existing local compaction policy. Native compaction is not enabled for OpenCode Sessions by this feature; Session integration must persist these parts before enabling it. The AI SDK bridge rejects native compaction parts rather than dropping them. Provider-executed tool APIs and persistence changes are a separate follow-up.
-
-Tests cover serialized round trips, real local HTTP plus a tool loop, AWS binary frames and signing, provider errors, malformed blocks, and usage accounting. Live provider tests are gated by `RECORD=true` and the relevant API keys:
-
-```sh
-# Run from packages/ai. Only records the selected new cassette group.
-RECORD=true RECORDED_PREFIX=openai-compaction bun test test/provider/compaction.recorded.test.ts
-RECORD=true RECORDED_PREFIX=xai-compaction bun test test/provider/compaction.recorded.test.ts
-RECORD=true RECORDED_PREFIX=anthropic-compaction bun test test/provider/compaction.recorded.test.ts
-```
-
-Provider references: [OpenAI](https://developers.openai.com/api/docs/guides/compaction), [Azure](https://learn.microsoft.com/en-us/azure/foundry/openai/how-to/responses#server-side-compaction), [Anthropic](https://platform.claude.com/docs/en/build-with-claude/compaction), [Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/claude-messages-compaction.html), [xAI](https://docs.x.ai/developers/advanced-api-usage/context-compaction).
-
 ## Caching

 Prompt caching is **on by default**. Every `LLMRequest` resolves to `cache: "auto"` unless the caller opts out with `cache: "none"`. Each protocol translates `CacheHint`s to its wire format (`cache_control` on Anthropic, `cachePoint` on Bedrock; OpenAI and Gemini do implicit caching server-side and don't need inline markers — auto is a no-op there).
@@ -40,7 +40,6 @@ const RESPECTS_INLINE_HINTS = new Set([
  "anthropic-messages",
  "google-vertex-messages",
  "bedrock-converse",
-  "bedrock-messages",
  "openrouter",
 ])

@@ -6,12 +6,8 @@ import { Auth } from "../route/auth.js"
 import { Endpoint } from "../route/endpoint.js"
 import { Framing } from "../route/framing.js"
 import { Protocol } from "../route/protocol.js"
-import { Headers } from "effect/unstable/http"
-import { HttpTransport } from "../route/transport/index.js"
 import {
  AIError,
-  HttpOptions,
-  LLMRequest,
  LLMEvent,
  mergeJsonRecords,
  Usage,
@@ -19,6 +15,7 @@ import {
  type FinishReasonDetails,
  type FinishReason,
  type JsonSchema,
+  type LLMRequest,
  type MediaPart,
  type ProviderMetadata,
  type ToolCallPart,
@@ -64,7 +61,6 @@ export type ThinkingInput =
    ))

 export interface OptionsInput {
-  readonly contextManagement?: ContextManagement
  readonly [key: string]: unknown
  readonly thinking?: ThinkingInput
  readonly effort?: string
@@ -93,23 +89,6 @@ export interface OptionsInput {

 export type ProviderOptionsInput = OptionsInput

-export const ContextManagement = Schema.Struct({
-  edits: Schema.Array(
-    Schema.Struct({
-      type: Schema.Literal("compact_20260112"),
-      trigger: Schema.optional(
-        Schema.Struct({
-          type: Schema.Literal("input_tokens"),
-          value: Schema.Int.check(Schema.isGreaterThanOrEqualTo(50000)),
-        }),
-      ),
-      pauseAfterCompaction: Schema.optional(Schema.Boolean),
-      instructions: Schema.optional(Schema.String),
-    }),
-  ),
-})
-export type ContextManagement = typeof ContextManagement.Type
-
 // =============================================================================
 // Request Body Schema
 // =============================================================================
@@ -257,12 +236,7 @@ const AnthropicUserBlock = Schema.Union([
  AnthropicToolResultBlock,
 ])
 type AnthropicUserBlock = Schema.Schema.Type<typeof AnthropicUserBlock>
-const AnthropicCompactionBlock = Schema.Struct({
-  type: Schema.Literal("compaction"),
-  content: Schema.NullOr(Schema.String),
-})
 const AnthropicAssistantBlock = Schema.Union([
-  AnthropicCompactionBlock,
  AnthropicTextBlock,
  AnthropicThinkingBlock,
  AnthropicRedactedThinkingBlock,
@@ -338,18 +312,6 @@ const AnthropicContainer = Schema.Union([
 ])

 const AnthropicBodyFields = {
-  context_management: Schema.optional(
-    Schema.Struct({
-      edits: Schema.Array(
-        Schema.Struct({
-          type: Schema.Literal("compact_20260112"),
-          trigger: ContextManagement.fields.edits.value.fields.trigger,
-          pause_after_compaction: Schema.optional(Schema.Boolean),
-          instructions: Schema.optional(Schema.String),
-        }),
-      ),
-    }),
-  ),
  model: Schema.String,
  system: optionalArray(AnthropicTextBlock),
  messages: Schema.Array(AnthropicMessage),
@@ -373,7 +335,7 @@ const AnthropicBodyFields = {
 export const AnthropicMessagesBody = Schema.Struct(AnthropicBodyFields)
 export type AnthropicMessagesBody = Schema.Schema.Type<typeof AnthropicMessagesBody>

-const AnthropicIterationUsage = Schema.StructWithRest(
+const AnthropicUsage = Schema.StructWithRest(
  Schema.Struct({
    input_tokens: optionalNull(Schema.Number),
    output_tokens: Schema.optional(Schema.Number),
@@ -392,13 +354,6 @@ const AnthropicIterationUsage = Schema.StructWithRest(
  }),
  [Schema.Record(Schema.String, Schema.Unknown)],
 )
-const AnthropicUsage = Schema.StructWithRest(
-  Schema.Struct({
-    ...AnthropicIterationUsage.schema.fields,
-    iterations: Schema.optional(Schema.Array(AnthropicIterationUsage)),
-  }),
-  [JsonObject],
-)
 type AnthropicUsage = Schema.Schema.Type<typeof AnthropicUsage>

 const AnthropicStreamBlock = Schema.Struct({
@@ -422,7 +377,6 @@ type AnthropicStreamBlock = Schema.Schema.Type<typeof AnthropicStreamBlock>
 const decodeAnthropicStreamBlock = Schema.decodeUnknownOption(AnthropicStreamBlock)

 const AnthropicStreamDelta = Schema.Struct({
-  content: optionalNull(Schema.String),
  type: Schema.optional(Schema.String),
  text: Schema.optional(Schema.String),
  thinking: Schema.optional(Schema.String),
@@ -452,8 +406,6 @@ const AnthropicEvent = Schema.Struct({
 type AnthropicEvent = Schema.Schema.Type<typeof AnthropicEvent>

 interface ParserState {
-  readonly provider: LLMRequest["model"]["provider"]
-  readonly compactions: Readonly<Record<number, string | null>>
  readonly providerMetadataKey: string
  readonly tools: ToolStream.State<number>
  readonly reasoningSignatures: Readonly<Record<number, string>>
@@ -896,12 +848,6 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
    if (message.role === "assistant") {
      const content: AnthropicAssistantBlock[] = []
      for (const part of message.content) {
-        if (part.type === "compaction") {
-          if (part.provider !== request.model.provider || part.text === undefined)
-            return yield* invalid("Compaction state must be replayed to its originating provider and API")
-          content.push({ type: "compaction", content: part.text })
-          continue
-        }
        if (part.type === "text") {
          if (part.text.trim().length === 0) continue
          content.push({ type: "text", text: part.text, cache_control: cacheControl(breakpoints, part.cache) })
@@ -1057,9 +1003,6 @@ const resolveThinking = Effect.fn("AnthropicMessages.resolveThinking")(function*
 })

 const fromRequest = Effect.fn("AnthropicMessages.fromRequest")(function* (request: LLMRequest) {
-  const management = yield* ProviderShared.validateWith(
-    Schema.decodeUnknownEffect(Schema.UndefinedOr(ContextManagement)),
-  )(request.providerOptions?.contextManagement)
  const generation = request.generation
  const toolSchemaCompatibility = request.model.compatibility?.toolSchema
  // Allocate the 4-breakpoint budget in invalidation order: tools → system →
@@ -1094,7 +1037,7 @@ const fromRequest = Effect.fn("AnthropicMessages.fromRequest")(function* (reques
    )
  }
  const options = yield* resolveOptions(request)
-  const body = {
+  return {
    model: request.model.id,
    system,
    messages,
@@ -1115,18 +1058,6 @@ const fromRequest = Effect.fn("AnthropicMessages.fromRequest")(function* (reques
    metadata: options.metadata,
    service_tier: options.service_tier,
  }
-  if (!management) return body
-  return {
-    ...body,
-    context_management: {
-      edits: management.edits.map((edit) => ({
-        type: edit.type,
-        trigger: edit.trigger,
-        pause_after_compaction: edit.pauseAfterCompaction,
-        instructions: edit.instructions,
-      })),
-    },
-  }
 })

 // =============================================================================
@@ -1148,31 +1079,18 @@ const mapFinishReason = (reason: string | null | undefined): FinishReason => {
 // expose that subset through `output_tokens_details.thinking_tokens`.
 const mapUsage = (usage: AnthropicUsage | undefined, providerMetadataKey: string): Usage | undefined => {
  if (!usage) return undefined
-  const iterations = usage.iterations?.length ? usage.iterations : [usage]
-  const last = usage.iterations?.at(-1)
-  const nonCached = ProviderShared.sumTokens(...iterations.map((item) => item.input_tokens ?? undefined))
-  const cacheRead = ProviderShared.sumTokens(...iterations.map((item) => item.cache_read_input_tokens ?? undefined))
-  const cacheWrite = ProviderShared.sumTokens(
-    ...iterations.map((item) => item.cache_creation_input_tokens ?? undefined),
-  )
+  const nonCached = usage.input_tokens ?? undefined
+  const cacheRead = usage.cache_read_input_tokens ?? undefined
+  const cacheWrite = usage.cache_creation_input_tokens ?? undefined
  const inputTokens = ProviderShared.sumTokens(nonCached, cacheRead, cacheWrite)
-  const outputTokens = ProviderShared.sumTokens(...iterations.map((item) => item.output_tokens))
  return new Usage({
    inputTokens,
-    outputTokens,
-    contextTokens:
-      last?.type === "message"
-        ? ProviderShared.sumTokens(
-            last.input_tokens ?? undefined,
-            last.cache_read_input_tokens ?? undefined,
-            last.cache_creation_input_tokens ?? undefined,
-          )
-        : undefined,
+    outputTokens: usage.output_tokens,
    nonCachedInputTokens: nonCached,
    cacheReadInputTokens: cacheRead,
    cacheWriteInputTokens: cacheWrite,
-    reasoningTokens: ProviderShared.sumTokens(...iterations.map((item) => item.output_tokens_details?.thinking_tokens)),
-    totalTokens: ProviderShared.totalTokens(inputTokens, outputTokens, undefined),
+    reasoningTokens: usage.output_tokens_details?.thinking_tokens,
+    totalTokens: ProviderShared.totalTokens(inputTokens, usage.output_tokens, undefined),
    providerMetadata: { [providerMetadataKey]: usage },
  })
 }
@@ -1194,7 +1112,6 @@ const mergeUsage = (left: Usage | undefined, right: Usage | undefined, providerM
  return new Usage({
    inputTokens,
    outputTokens,
-    contextTokens: right.contextTokens ?? left.contextTokens,
    nonCachedInputTokens,
    cacheReadInputTokens,
    cacheWriteInputTokens,
@@ -1253,6 +1170,7 @@ const onContentBlockStart = (
  event: AnthropicEvent & { readonly content_block: AnthropicStreamBlock },
 ): StepResult => {
  const block = event.content_block
+  if (!block) return [state, NO_EVENTS]

  if (block.type === "tool_use" || block.type === "server_tool_use") {
    if (event.index === undefined || !block.id) return [state, NO_EVENTS]
@@ -1347,16 +1265,7 @@ const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(f
 ) {
  const delta = event.delta

-  if (delta.type === "compaction_delta") {
-    if (event.index === undefined || !(event.index in state.compactions) || delta.content === undefined)
-      return yield* ProviderShared.eventError(ADAPTER, "Compaction delta is missing its block or content")
-    return [
-      { ...state, compactions: { ...state.compactions, [event.index]: delta.content } },
-      NO_EVENTS,
-    ] satisfies StepResult
-  }
-
-  if (delta.type === "text_delta" && delta.text) {
+  if (delta?.type === "text_delta" && delta.text) {
    if (!state.lifecycle.text.has(`text-${event.index ?? 0}`)) return [state, NO_EVENTS] satisfies StepResult
    const events: LLMEvent[] = []
    return [
@@ -1365,7 +1274,7 @@ const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(f
    ] satisfies StepResult
  }

-  if (delta.type === "thinking_delta" && delta.thinking) {
+  if (delta?.type === "thinking_delta" && delta.thinking) {
    if (!state.lifecycle.reasoning.has(`reasoning-${event.index ?? 0}`)) return [state, NO_EVENTS] satisfies StepResult
    const events: LLMEvent[] = []
    return [
@@ -1377,7 +1286,7 @@ const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(f
    ] satisfies StepResult
  }

-  if (delta.type === "signature_delta" && delta.signature) {
+  if (delta?.type === "signature_delta" && delta.signature) {
    const index = event.index ?? 0
    if (!state.lifecycle.reasoning.has(`reasoning-${index}`)) return [state, NO_EVENTS] satisfies StepResult
    return [
@@ -1389,7 +1298,7 @@ const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(f
    ] satisfies StepResult
  }

-  if (delta.type === "input_json_delta" && event.index !== undefined) {
+  if (delta?.type === "input_json_delta" && event.index !== undefined) {
    if (!delta.partial_json) return [state, NO_EVENTS] satisfies StepResult
    if (!state.tools[event.index]) return [state, NO_EVENTS] satisfies StepResult
    const result = ToolStream.appendExisting(
@@ -1414,18 +1323,6 @@ const onContentBlockStop = Effect.fn("AnthropicMessages.onContentBlockStop")(fun
  event: AnthropicEvent,
 ) {
  if (event.index === undefined) return [state, NO_EVENTS] satisfies StepResult
-  if (event.index in state.compactions) {
-    const { [event.index]: content, ...compactions } = state.compactions
-    const events: LLMEvent[] = []
-    const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
-    events.push(
-      LLMEvent.compaction({
-        provider: state.provider,
-        text: content,
-      }),
-    )
-    return [{ ...state, compactions, lifecycle }, events] satisfies StepResult
-  }
  const result = yield* ToolStream.finish(ADAPTER, state.tools, event.index)
  const events: LLMEvent[] = []
  const resultEvents = result.events ?? []
@@ -1477,8 +1374,6 @@ const onMessageDelta = (
 }

 const onMessageStop = Effect.fn("AnthropicMessages.onMessageStop")(function* (state: ParserState) {
-  if (Object.keys(state.compactions).length)
-    return yield* ProviderShared.eventError(ADAPTER, "Response ended with an incomplete compaction block")
  const result = yield* ToolStream.finishAll(ADAPTER, state.tools)
  const events: LLMEvent[] = []
  const lifecycle = result.events.length ? Lifecycle.stepStart(state.lifecycle, events) : state.lifecycle
@@ -1523,21 +1418,16 @@ const onError = (event: AnthropicEvent) => {
  )
 }

-const STREAM_BLOCK_TYPES = new Set([
-  "compaction",
-  "text",
-  "thinking",
-  "redacted_thinking",
-  "tool_use",
-  "server_tool_use",
-])
-const STREAM_DELTA_TYPES = new Set([
-  "compaction_delta",
-  "text_delta",
-  "thinking_delta",
-  "signature_delta",
-  "input_json_delta",
-])
+const isKnownStreamBlockType = (type: string) =>
+  type === "text" ||
+  type === "thinking" ||
+  type === "redacted_thinking" ||
+  type === "tool_use" ||
+  type === "server_tool_use" ||
+  isServerToolResultType(type)
+
+const isKnownStreamDeltaType = (type: string) =>
+  type === "text_delta" || type === "thinking_delta" || type === "signature_delta" || type === "input_json_delta"

 const invalidStreamEvent = (event: AnthropicEvent) =>
  Effect.fail(
@@ -1566,16 +1456,7 @@ const step = (state: ParserState, event: AnthropicEvent) => {
  if (event.type === "content_block_start") {
    if (!ProviderShared.isRecord(event.content_block) || typeof event.content_block.type !== "string")
      return invalidStreamEvent(event)
-    if (event.content_block.type === "compaction") {
-      const decoded = Schema.decodeUnknownOption(AnthropicCompactionBlock)(event.content_block)
-      if (event.index === undefined || Option.isNone(decoded)) return invalidStreamEvent(event)
-      return Effect.succeed<StepResult>([
-        { ...state, compactions: { ...state.compactions, [event.index]: decoded.value.content } },
-        NO_EVENTS,
-      ])
-    }
-    if (!STREAM_BLOCK_TYPES.has(event.content_block.type) && !isServerToolResultType(event.content_block.type))
-      return Effect.succeed<StepResult>([state, NO_EVENTS])
+    if (!isKnownStreamBlockType(event.content_block.type)) return Effect.succeed<StepResult>([state, NO_EVENTS])
    const decoded = decodeAnthropicStreamBlock(event.content_block)
    if (Option.isNone(decoded)) return invalidStreamEvent(event)
    const block = decoded.value
@@ -1589,7 +1470,7 @@ const step = (state: ParserState, event: AnthropicEvent) => {
  }
  if (event.type === "content_block_delta") {
    if (!ProviderShared.isRecord(event.delta)) return invalidStreamEvent(event)
-    if (typeof event.delta.type === "string" && !STREAM_DELTA_TYPES.has(event.delta.type))
+    if (typeof event.delta.type === "string" && !isKnownStreamDeltaType(event.delta.type))
      return Effect.succeed<StepResult>([state, NO_EVENTS])
    const decoded = decodeAnthropicStreamDelta(event.delta)
    if (Option.isNone(decoded)) return invalidStreamEvent(event)
@@ -1623,8 +1504,6 @@ export const protocol = Protocol.make({
  stream: {
    event: Protocol.jsonEvent(AnthropicEvent),
    initial: (request) => ({
-      provider: request.model.provider,
-      compactions: {},
      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
      tools: ToolStream.empty<number>(),
      reasoningSignatures: {},
@@ -1634,37 +1513,6 @@ export const protocol = Protocol.make({
  },
 })

-export const transport = <Body extends Pick<AnthropicMessagesBody, "messages" | "context_management">>() => {
-  const http = HttpTransport.httpJson<Body, string>({ framing })
-  return {
-    ...http,
-    prepare: (input: Parameters<typeof http.prepare>[0]) => {
-      if (
-        !input.body.context_management?.edits.length &&
-        !input.body.messages.some((message) => message.content.some((block) => block.type === "compaction"))
-      )
-        return http.prepare(input)
-      const headers = Headers.fromInput(input.request.http?.headers)
-      const betas = new Set(
-        (headers["anthropic-beta"] ?? "")
-          .split(",")
-          .map((item) => item.trim())
-          .filter(Boolean),
-      )
-      betas.add("compact-2026-01-12")
-      return http.prepare({
-        ...input,
-        request: LLMRequest.update(input.request, {
-          http: new HttpOptions({
-            ...input.request.http,
-            headers: { ...headers, "anthropic-beta": [...betas].join(",") },
-          }),
-        }),
-      })
-    },
-  }
-}
-
 export const route = Route.make({
  id: ADAPTER,
  provider: "anthropic",
@@ -1674,7 +1522,7 @@ export const route = Route.make({
    baseURL: DEFAULT_BASE_URL,
  }),
  auth: Auth.none,
-  transport: transport<AnthropicMessagesBody>(),
+  framing,
  headers: () => ({ "anthropic-version": "2023-06-01" }),
 })

@@ -510,9 +510,10 @@ interface ParserState {
  readonly tools: ToolStream.State<number>
  readonly finishedTools: ReadonlySet<number>
  // Bedrock splits the finish into `messageStop` (carries `stopReason`) and
-  // `metadata` (carries usage). Hold the terminal event in state so `onHalt`
-  // can emit exactly one finish after both chunks have had a chance to arrive.
-  readonly pendingFinish: { readonly reason: FinishReasonDetails; readonly usage?: Usage } | undefined
+  // `metadata` (carries usage). Hold both in state so `onHalt` can emit exactly
+  // one finish after both chunks have had a chance to arrive.
+  readonly finishReason: FinishReasonDetails | undefined
+  readonly usage: Usage | undefined
  readonly hasToolCalls: boolean
  readonly lifecycle: Lifecycle.State
  readonly reasoningSignatures: Readonly<Record<number, string>>
@@ -692,12 +693,9 @@ const step = (state: ParserState, event: BedrockEvent) =>
      return [
        {
          ...state,
-          pendingFinish: {
-            reason: {
-              normalized: mapFinishReason(event.messageStop.stopReason),
-              raw: event.messageStop.stopReason,
-            },
-            usage: state.pendingFinish?.usage,
+          finishReason: {
+            normalized: mapFinishReason(event.messageStop.stopReason),
+            raw: event.messageStop.stopReason,
          },
        },
        [],
@@ -705,14 +703,11 @@ const step = (state: ParserState, event: BedrockEvent) =>
    }

    if (event.metadata) {
-      const usage = mapUsage(event.metadata.usage, state.providerMetadataKey) ?? state.pendingFinish?.usage
+      const usage = mapUsage(event.metadata.usage, state.providerMetadataKey) ?? state.usage
      return [
        {
          ...state,
-          pendingFinish: {
-            reason: state.pendingFinish?.reason ?? { normalized: "stop" },
-            usage,
-          },
+          usage,
        },
        [],
      ] as const
@@ -736,18 +731,18 @@ const step = (state: ParserState, event: BedrockEvent) =>
 const framing = BedrockEventStream.framing(ADAPTER)

 const onHalt = (state: ParserState): ReadonlyArray<LLMEvent> => {
-  if (!state.pendingFinish) return []
+  if (!state.finishReason) return []
  const normalized = (() => {
-    if (state.pendingFinish.reason.normalized === "stop" && state.hasToolCalls) return "tool-calls"
-    return state.pendingFinish.reason.normalized
+    if (state.finishReason.normalized === "stop" && state.hasToolCalls) return "tool-calls"
+    return state.finishReason.normalized
  })()
  const events: LLMEvent[] = []
  Lifecycle.finish(state.lifecycle, events, {
    reason: {
-      ...state.pendingFinish.reason,
+      ...state.finishReason,
      normalized,
    },
-    usage: state.pendingFinish.usage,
+    usage: state.usage,
  })
  return events
 }
@@ -771,7 +766,8 @@ export const protocol = Protocol.make({
      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
      tools: ToolStream.empty<number>(),
      finishedTools: new Set<number>(),
-      pendingFinish: undefined,
+      finishReason: undefined,
+      usage: undefined,
      hasToolCalls: false,
      lifecycle: Lifecycle.initial(),
      reasoningSignatures: {},
@@ -1,121 +0,0 @@
-import { Effect, Encoding, Schema, Struct } from "effect"
-import { Headers } from "effect/unstable/http"
-import { AIError } from "../schema/index.js"
-import { Route } from "../route/client.js"
-import { Endpoint } from "../route/endpoint.js"
-import { Protocol } from "../route/protocol.js"
-import { classifyProviderFailure } from "../provider-error.js"
-import { AnthropicMessages } from "./anthropic-messages.js"
-import { BedrockEventStream } from "./bedrock-event-stream.js"
-import { BedrockAuth } from "./utils/bedrock-auth.js"
-import { JsonObject, ProviderShared } from "./shared.js"
-
-const ID = "bedrock-messages"
-const VERSION = "bedrock-2023-05-31"
-const Body = Schema.Struct({
-  ...Struct.omit(AnthropicMessages.AnthropicMessagesBody.fields, ["model", "stream"]),
-  anthropic_version: Schema.Literal(VERSION),
-  anthropic_beta: Schema.optional(Schema.Array(Schema.String)),
-}).check(
-  Schema.makeFilter(
-    (body) =>
-      body.messages.flatMap((message) => message.content.map(mediaIssue)).find((issue) => issue !== undefined) ?? true,
-  ),
-)
-const Event = Schema.Struct({
-  chunk: Schema.optional(Schema.Struct({ bytes: Schema.String })),
-  exception: Schema.optional(
-    Schema.Struct({
-      type: Schema.String,
-      details: Schema.StructWithRest(
-        Schema.Struct({ message: Schema.optional(Schema.String), originalMessage: Schema.optional(Schema.String) }),
-        [JsonObject],
-      ),
-    }),
-  ),
-})
-
-export const protocol = Protocol.make({
-  id: ID,
-  body: {
-    schema: Body,
-    from: Effect.fn("BedrockMessages.fromRequest")(function* (request) {
-      const body = yield* AnthropicMessages.protocol.body.from(request)
-      const headers = Headers.fromInput(request.http?.headers)
-      const betas = new Set(
-        (headers["anthropic-beta"] ?? "")
-          .split(",")
-          .map((value) => value.trim())
-          .filter(Boolean),
-      )
-      if (
-        body.context_management?.edits.length ||
-        body.messages.some((message) => message.content.some((block) => block.type === "compaction"))
-      )
-        betas.add("compact-2026-01-12")
-      return {
-        ...Struct.omit(body, ["model", "stream"]),
-        anthropic_version: VERSION,
-        anthropic_beta: betas.size ? [...betas] : undefined,
-      } satisfies typeof Body.Type
-    }),
-  },
-  stream: {
-    event: Event,
-    initial: AnthropicMessages.protocol.stream.initial,
-    step: Effect.fn("BedrockMessages.step")(function* (state, event) {
-      if (event.exception)
-        return yield* new AIError({
-          reason: classifyProviderFailure({
-            message: event.exception.details.message ?? event.exception.details.originalMessage ?? event.exception.type,
-            rawBody: ProviderShared.encodeJson(event),
-          }),
-        })
-      if (!event.chunk) return yield* ProviderShared.eventError(ID, "Bedrock Messages event is missing its chunk")
-      const text = yield* Effect.fromResult(Encoding.decodeBase64String(event.chunk.bytes)).pipe(
-        Effect.mapError((cause) =>
-          ProviderShared.eventError(ID, "Invalid Bedrock Messages chunk encoding", undefined, cause),
-        ),
-      )
-      const decoded = yield* Schema.decodeUnknownEffect(AnthropicMessages.protocol.stream.event)(text).pipe(
-        Effect.mapError((cause) => ProviderShared.eventError(ID, "Invalid Bedrock Messages event", undefined, cause)),
-      )
-      return yield* AnthropicMessages.protocol.stream.step(state, decoded)
-    }),
-  },
-})
-
-function mediaIssue(
-  block: AnthropicMessages.AnthropicMessagesBody["messages"][number]["content"][number],
-): string | undefined {
-  if (block.type === "tool_result")
-    return typeof block.content === "string"
-      ? undefined
-      : block.content.map(mediaIssue).find((issue) => issue !== undefined)
-  if (block.type !== "image" && block.type !== "document") return undefined
-  if (block.source.type === "url" || block.source.type === "file")
-    return "Bedrock Messages does not support URL or file-ID media sources"
-  if (
-    block.type === "image" &&
-    !["image/jpeg", "image/png", "image/webp", "image/gif"].includes(block.source.media_type)
-  )
-    return "Bedrock Messages requires a JPEG, PNG, WebP, or GIF image"
-  if (block.source.type === "base64" && Encoding.decodeBase64(block.source.data)._tag === "Failure")
-    return "Bedrock Messages media data must be valid base64"
-  return undefined
-}
-
-export const route = Route.make({
-  id: ID,
-  provider: "amazon-bedrock",
-  providerMetadataKey: "anthropic",
-  protocol,
-  endpoint: Endpoint.path(
-    ({ request }) => `/model/${encodeURIComponent(request.model.id)}/invoke-with-response-stream`,
-    { baseURL: "https://bedrock-runtime.us-east-1.amazonaws.com" },
-  ),
-  auth: BedrockAuth.auth,
-  framing: BedrockEventStream.framing(ID),
-})
-
-export * as BedrockMessages from "./bedrock-messages.js"
@@ -609,18 +609,27 @@ const finish = (state: ParserState): ReadonlyArray<LLMEvent> => {
 }

 const step = (state: ParserState, event: GeminiEvent) => {
-  if (ProviderShared.isRecord(event.error) && typeof event.error.message === "string") {
+  if (ProviderShared.isRecord(event.error)) {
    const body = ProviderShared.encodeJson(event)
    return Effect.fail(
      new AIError({
        reason: classifyProviderFailure({
-          message: event.error.message,
+          message:
+            typeof event.error.message === "string" && event.error.message.length > 0
+              ? event.error.message
+              : typeof event.error.status === "string" && event.error.status.length > 0
+                ? event.error.status
+                : "Gemini provider error",
          status: typeof event.error.code === "number" ? event.error.code : undefined,
          rawBody: body,
        }),
      }),
    )
  }
+  if ("error" in event)
+    return Effect.fail(
+      ProviderShared.eventError(state.route, `Invalid ${state.route} stream event`, ProviderShared.encodeJson(event)),
+    )
  const nextState = {
    ...state,
    promptFeedback: event.promptFeedback ?? state.promptFeedback,
@@ -1,6 +1,5 @@
 export * as AnthropicMessages from "./anthropic-messages.js"
 export * as BedrockConverse from "./bedrock-converse.js"
-export { BedrockMessages } from "./bedrock-messages.js"
 export * as Gemini from "./gemini.js"
 export * as MistralChat from "./mistral-chat.js"
 export * as OpenAIChat from "./openai-chat.js"
@@ -32,19 +32,17 @@ export const PATH = "/responses"
 // =============================================================================
 // Request Body Schema
 // =============================================================================
-export const OpenResponsesInputText = Schema.Struct({
+const OpenResponsesInputText = Schema.Struct({
  type: Schema.tag("input_text"),
  text: Schema.String,
 })
-export const OpenResponsesInputImage = Schema.Struct({
+const OpenResponsesInputImage = Schema.Struct({
  type: Schema.tag("input_image"),
  image_url: Schema.String,
-  detail: Schema.optional(Schema.String),
 })
-export const OpenResponsesInputFile = Schema.Struct({
+const OpenResponsesInputFile = Schema.Struct({
  type: Schema.tag("input_file"),
  filename: Schema.String,
-  detail: Schema.optional(Schema.String),
  file_data: Schema.optional(Schema.String),
  file_url: Schema.optional(Schema.String),
 })
@@ -56,7 +54,7 @@ const MediaInput = Schema.Union([OpenResponsesInputImage, OpenResponsesInputFile
 export type MediaInput = Schema.Schema.Type<typeof MediaInput>
 const OpenResponsesInputContent = Schema.Union([OpenResponsesInputText, MediaInput])

-export const OpenResponsesOutputText = Schema.Struct({
+const OpenResponsesOutputText = Schema.Struct({
  type: Schema.tag("output_text"),
  text: Schema.String,
 })
@@ -64,13 +62,6 @@ export const OpenResponsesOutputText = Schema.Struct({
 export const MessagePhase = Schema.NullOr(Schema.Literals(["commentary", "final_answer"]))
 type MessagePhase = Schema.Schema.Type<typeof MessagePhase>

-export const MessageMetadata = Schema.Struct({
-  itemId: Schema.optional(Schema.String),
-  type: Schema.optional(Schema.Literal("message")),
-  status: Schema.optional(Schema.String),
-  phase: Schema.optional(MessagePhase),
-})
-
 const messagePhase = (value: unknown): MessagePhase | undefined => {
  if (value === null || value === "commentary" || value === "final_answer") return value
  return undefined
@@ -81,7 +72,7 @@ const OpenResponsesReasoningSummaryText = Schema.Struct({
  text: Schema.String,
 })

-export const OpenResponsesReasoningItem = Schema.Struct({
+const OpenResponsesReasoningItem = Schema.Struct({
  type: Schema.tag("reasoning"),
  id: Schema.optionalKey(Schema.String),
  summary: Schema.Array(OpenResponsesReasoningSummaryText),
@@ -158,30 +149,16 @@ const OpenResponsesFunctionCallOutput = Schema.Union([
  Schema.Array(OpenResponsesFunctionCallOutputContent),
 ])

-export const CompactionItem = Schema.Struct({
-  type: Schema.Literal("compaction"),
-  id: optionalNull(Schema.String),
-  encrypted_content: Schema.String,
-})
-
 export const InputItem = Schema.Union([
-  CompactionItem,
  Schema.Struct({ role: Schema.tag("system"), content: Schema.String }),
  Schema.Struct({ role: Schema.tag("developer"), content: Schema.String }),
-  Schema.Struct({
-    role: Schema.tag("user"),
-    content: Schema.Array(OpenResponsesInputContent),
-    type: Schema.optional(Schema.Literal("message")),
-    id: Schema.optional(Schema.String),
-    status: Schema.optional(Schema.String),
-  }),
+  Schema.Struct({ role: Schema.tag("user"), content: Schema.Array(OpenResponsesInputContent) }),
  Schema.Struct({
    type: Schema.tag("message"),
    id: Schema.optionalKey(Schema.String),
    role: Schema.tag("assistant"),
    content: Schema.Array(OpenResponsesOutputText),
    phase: Schema.optionalKey(MessagePhase),
-    status: Schema.optional(Schema.String),
  }),
  OpenResponsesReasoningItem,
  Schema.Struct({
@@ -290,7 +267,7 @@ const OpenResponsesBody = Schema.Struct({
 })
 export type OpenResponsesBody = Schema.Schema.Type<typeof OpenResponsesBody>

-export const OpenResponsesUsage = Schema.Struct({
+const OpenResponsesUsage = Schema.Struct({
  input_tokens: Schema.optional(Schema.Number),
  input_tokens_details: optionalNull(
    Schema.Struct({
@@ -410,8 +387,6 @@ export interface Extension {
 const BASE: Extension = { id: ADAPTER, name: NAME }

 export interface ParserState {
-  readonly provider: LLMRequest["model"]["provider"]
-  readonly completedCompactions: ReadonlySet<string>
  readonly id: string
  readonly name: string
  readonly providerMetadataKey: string
@@ -513,9 +488,6 @@ const lowerMedia = Effect.fn("OpenResponses.lowerMedia")(function* (
  const media = ProviderShared.normalizeMedia(part)
  const extended = extension.lowerMedia?.({ part, media, request })
  if (extended) return extended
-  const detail = yield* ProviderShared.validateWith(Schema.decodeUnknownEffect(OpenResponsesInputImage.fields.detail))(
-    part.providerMetadata?.[metadataKey(request.model)]?.detail,
-  )
  const url =
    typeof part.data === "string" && (part.data.startsWith("https://") || part.data.startsWith("http://"))
      ? part.data
@@ -526,15 +498,10 @@ const lowerMedia = Effect.fn("OpenResponses.lowerMedia")(function* (
    return {
      type: "input_file" as const,
      filename: part.filename ?? (media.mime === "application/pdf" ? "document.pdf" : "file"),
-      detail,
      ...(url ? { file_url: url } : { file_data: media.dataUrl }),
    }
  }
-  return {
-    type: "input_image" as const,
-    image_url: url ?? media.dataUrl,
-    detail,
-  }
+  return { type: "input_image" as const, image_url: url ?? media.dataUrl }
 })

 const lowerUserContent = Effect.fnUntraced(function* (
@@ -598,12 +565,9 @@ const lowerToolResultOutput = Effect.fnUntraced(function* (

 const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (request: LLMRequest, extension: Extension) {
  const input: LoweredInputItem[] = []
-  const providerMetadataKey = metadataKey(request.model)
+  const providerMetadataKey = request.model.route.providerMetadataKey ?? "openresponses"

  for (const message of request.messages) {
-    const metadata = yield* ProviderShared.validateWith(
-      Schema.decodeUnknownEffect(Schema.UndefinedOr(MessageMetadata)),
-    )(message.providerMetadata?.[providerMetadataKey])
    if (message.role === "system") {
      input.push({
        role: "developer",
@@ -614,8 +578,7 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques

    if (message.role === "user") {
      const content = yield* Effect.forEach(message.content, (part) => lowerUserContent(part, request, extension))
-      if (content.length > 0)
-        input.push({ role: "user", content, type: metadata?.type, id: metadata?.itemId, status: metadata?.status })
+      if (content.length > 0) input.push({ role: "user", content })
      continue
    }

@@ -628,10 +591,9 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
        const groups = content.reduce<
          Array<{ id: string | undefined; phase: MessagePhase | null | undefined; parts: TextPart[] }>
        >((groups, part) => {
-          const partMetadata = part.providerMetadata?.[providerMetadataKey]
-          const id = itemID(part.providerMetadata, providerMetadataKey) ?? metadata?.itemId
-          const partPhase = messagePhase(partMetadata?.phase)
-          const phase = partPhase === undefined ? metadata?.phase : partPhase
+          const metadata = part.providerMetadata?.[providerMetadataKey]
+          const id = itemID(part.providerMetadata, providerMetadataKey)
+          const phase = ProviderShared.isRecord(metadata) ? messagePhase(metadata.phase) : undefined
          const group = groups.at(-1)
          if (group && group.id === id && group.phase === phase) group.parts.push(part)
          else groups.push({ id, phase, parts: [part] })
@@ -642,7 +604,6 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
            type: "message" as const,
            ...(group.id === undefined ? {} : { id: group.id }),
            role: "assistant" as const,
-            status: metadata?.status,
            content: group.parts.map((part) => ({ type: "output_text" as const, text: part.text })),
            ...(group.phase === undefined ? {} : { phase: group.phase }),
          })),
@@ -650,15 +611,6 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
        content.splice(0, content.length)
      }
      for (const part of message.content) {
-        if (part.type === "compaction") {
-          flushText()
-          if (part.provider !== request.model.provider || part.encrypted === undefined)
-            return yield* ProviderShared.invalidRequest(
-              "Compaction state must be replayed to its originating provider and API",
-            )
-          input.push({ type: "compaction", id: part.id, encrypted_content: part.encrypted })
-          continue
-        }
        if (part.type === "text") {
          content.push(part)
          continue
@@ -737,30 +689,13 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
  return input
 })

-export const lowerConversation = Effect.fn("OpenResponses.lowerConversation")(function* (
-  request: LLMRequest,
-  extension: Extension,
-) {
-  const instructions = ProviderShared.joinText(request.system)
-  return {
-    model: request.model.id,
-    input: yield* lowerMessages(request, extension),
-    ...(instructions ? { instructions } : {}),
-  }
-})
-
-export const lowerGeneration = (request: LLMRequest) => {
+const lowerOptions = (request: LLMRequest) => {
  const options = OpenResponsesOptions.resolve(request)
-  const generation = request.generation
+  const instructions = ProviderShared.joinText(request.system)
  const cacheKey = ProviderShared.promptCacheKey(request)
  const parallelToolCalls = resolveParallelToolCalls(request)
  return {
-    stream: true as const,
-    max_output_tokens: generation?.maxTokens,
-    temperature: generation?.temperature,
-    top_p: generation?.topP,
-    presence_penalty: generation?.presencePenalty,
-    frequency_penalty: generation?.frequencyPenalty,
+    ...(instructions ? { instructions } : {}),
    ...(options.store !== undefined ? { store: options.store } : {}),
    ...(options.metadata ? { metadata: options.metadata } : {}),
    ...(options.safetyIdentifier ? { safety_identifier: options.safetyIdentifier } : {}),
@@ -788,7 +723,7 @@ export const resolveParallelToolCalls = (request: LLMRequest) => {
  return disabled === undefined ? undefined : !disabled
 }

-export const allowedToolChoice = (request: LLMRequest) => {
+const allowedToolChoice = (request: LLMRequest) => {
  const allowed = OpenResponsesOptions.resolve(request).allowedTools
  if (!allowed) return undefined
  return {
@@ -802,10 +737,11 @@ export const fromRequestWithExtension = Effect.fn("OpenResponses.fromRequestWith
  request: LLMRequest,
  extension: Extension,
 ) {
+  const generation = request.generation
  const toolSchemaCompatibility = request.model.compatibility?.toolSchema
  return {
-    ...(yield* lowerConversation(request, extension)),
-    ...lowerGeneration(request),
+    model: request.model.id,
+    input: yield* lowerMessages(request, extension),
    tools:
      request.tools.length === 0
        ? undefined
@@ -819,6 +755,13 @@ export const fromRequestWithExtension = Effect.fn("OpenResponses.fromRequestWith
    tool_choice:
      allowedToolChoice(request) ??
      (request.toolChoice ? yield* lowerToolChoice(extension.name, request.toolChoice) : undefined),
+    stream: true as const,
+    max_output_tokens: generation?.maxTokens,
+    temperature: generation?.temperature,
+    top_p: generation?.topP,
+    presence_penalty: generation?.presencePenalty,
+    frequency_penalty: generation?.frequencyPenalty,
+    ...lowerOptions(request),
  }
 })

@@ -835,7 +778,7 @@ export const fromRequest = Effect.fn("OpenResponses.fromRequest")(function* (req
 // cached-read and cache-write subsets, and `output_tokens` (inclusive total)
 // with a `reasoning_tokens` subset. Pass the totals through and derive the
 // non-cached breakdown.
-export const mapUsage = (usage: OpenResponsesUsage | null | undefined, providerMetadataKey: string) => {
+const mapUsage = (usage: OpenResponsesUsage | null | undefined, providerMetadataKey: string) => {
  if (!usage) return undefined
  const cached = usage.input_tokens_details?.cached_tokens
  const cacheWrite = usage.input_tokens_details?.cache_write_tokens
@@ -865,8 +808,6 @@ const mapFinishReason = (event: Event, hasFunctionCall: boolean): FinishReason =
  return hasFunctionCall ? "tool-calls" : "unknown"
 }

-export const metadataKey = (model: LLMRequest["model"]) => model.route.providerMetadataKey ?? "openresponses"
-
 export const providerMetadata = (state: ParserState, metadata: Record<string, unknown>): ProviderMetadata => ({
  [state.providerMetadataKey]: metadata,
 })
@@ -1143,25 +1084,6 @@ const onOutputItemDone = Effect.fn("OpenResponses.onOutputItemDone")(function* (
 ) {
  if (!item) return [state, NO_EVENTS] satisfies StepResult

-  if (item.type === "compaction") {
-    if (!item.id || typeof item.encrypted_content !== "string")
-      return yield* ProviderShared.eventError(state.id, "Compaction output is missing its id or encrypted content")
-    if (state.completedCompactions.has(item.id)) return [state, NO_EVENTS] satisfies StepResult
-    const events: LLMEvent[] = []
-    const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
-    events.push(
-      LLMEvent.compaction({
-        provider: state.provider,
-        id: item.id,
-        encrypted: item.encrypted_content,
-      }),
-    )
-    return [
-      { ...state, lifecycle, completedCompactions: new Set([...state.completedCompactions, item.id]) },
-      events,
-    ] satisfies StepResult
-  }
-
  if (item.type === "message" && item.id !== undefined) {
    const message = state.message?.id === item.id ? state.message : undefined
    const itemPhase = messagePhase(item.phase)
@@ -1321,34 +1243,26 @@ const onResponseFinish = Effect.fn("OpenResponses.onResponseFinish")(function* (
  const events: LLMEvent[] = []
  if (event.type === "response.completed") {
    for (const item of event.response?.output ?? []) {
-      if (item.type !== "compaction" && item.type !== "function_call") continue
-      if (item.type === "compaction") {
-        // Terminal recovery cannot insert a checkpoint before already-emitted content.
-        if (state.lifecycle.stepStarted && !state.completedCompactions.has(item.id ?? ""))
-          return yield* ProviderShared.eventError(
-            state.id,
-            "Cannot recover a compaction checkpoint after output has been emitted",
-          )
-      }
-      if (item.type === "function_call" && !current.tools[item.id ?? item.call_id ?? ""]) continue
+      const id = item.id ?? (item.type === "function_call" ? item.call_id : undefined)
+      if (id === undefined) continue
+      if (item.type !== "function_call" || !current.tools[id]) continue
      const [next, emitted] = yield* onOutputItemDone(current, item)
      current = next
      events.push(...emitted)
    }
-    // Some compatible providers omit output_item.done even after completing the response.
-    const pending = yield* ToolStream.finishAll(current.id, current.tools)
-    current = {
-      ...current,
-      tools: pending.tools,
-      hasFunctionCall:
-        current.hasFunctionCall ||
-        pending.events.some((event) => LLMEvent.is.toolCall(event) || LLMEvent.is.toolInputError(event)),
-    }
-    events.push(...pending.events)
  }
+  // Some compatible providers omit output_item.done even after completing the response.
+  const pending =
+    event.type === "response.completed"
+      ? yield* ToolStream.finishAll(current.id, current.tools)
+      : { tools: current.tools, events: NO_EVENTS }
+  events.push(...pending.events)
+  const hasFunctionCall =
+    pending.events.some((event) => LLMEvent.is.toolCall(event) || LLMEvent.is.toolInputError(event)) ||
+    current.hasFunctionCall
  const lifecycle = Lifecycle.finish(current.lifecycle, events, {
    reason: {
-      normalized: mapFinishReason(event, current.hasFunctionCall),
+      normalized: mapFinishReason(event, hasFunctionCall),
      raw: event.response?.incomplete_details?.reason,
    },
    usage: mapUsage(event.response?.usage, current.providerMetadataKey),
@@ -1360,7 +1274,7 @@ const onResponseFinish = Effect.fn("OpenResponses.onResponseFinish")(function* (
          })
        : undefined,
  })
-  return [{ ...current, lifecycle }, events] satisfies StepResult
+  return [{ ...current, lifecycle, hasFunctionCall, tools: pending.tools }, events] satisfies StepResult
 })

 // Build the prettiest summary available from whatever the provider supplied.
@@ -1495,11 +1409,9 @@ export const step = (state: ParserState, input: Event) => {
 * implementations compose this baseline with their own tools and event variants.
 */
 export const initial = (request: LLMRequest, extension: Extension = BASE): ParserState => ({
-  provider: request.model.provider,
-  completedCompactions: new Set<string>(),
  id: extension.id,
  name: extension.name,
-  providerMetadataKey: metadataKey(request.model),
+  providerMetadataKey: request.model.route.providerMetadataKey ?? "openresponses",
  hasFunctionCall: false,
  tools: ToolStream.empty<string>(),
  completedTools: new Set<string>(),
@@ -5,14 +5,13 @@ import { Auth } from "../route/auth.js"
 import { Endpoint } from "../route/endpoint.js"
 import { Protocol } from "../route/protocol.js"
 import { HttpTransport } from "../route/transport/index.js"
-import type { LLMRequest, JsonSchema, ToolDefinition } from "../schema/index.js"
+import { LLMRequest, type JsonSchema, type ToolDefinition } from "../schema/index.js"
 import { OpenResponses } from "./open-responses.js"
 import { JsonObject, optionalArray, optionalNull, ProviderShared } from "./shared.js"
 import { OpenAIImage } from "./utils/openai-image.js"
 import { ResponsesHostedTools } from "./utils/responses-hosted-tools.js"
 import { ToolSchemaProjection } from "./utils/tool-schema.js"
 import { OpenResponsesChannel } from "./open-responses-channel.js"
-import { ResponsesCompaction } from "./utils/responses-compaction.js"

 const ADAPTER = "openai-responses"
 const NAME = "OpenAI Responses"
@@ -21,14 +20,6 @@ const WEBSOCKET_ROTATE_AFTER_MS = 55 * 60 * 1000
 export const DEFAULT_BASE_URL = "https://api.openai.com/v1"
 export const PATH = OpenResponses.PATH

-export const ContextManagement = Schema.Array(
-  Schema.Struct({
-    type: Schema.Literal("compaction"),
-    compactThreshold: Schema.optional(Schema.Int.check(Schema.isGreaterThan(0))),
-  }),
-)
-export type ContextManagement = typeof ContextManagement.Type
-
 const OpenAIResponsesImageGenerationTool = Schema.Struct({
  type: Schema.tag("image_generation"),
  action: Schema.optional(Schema.Literals(["auto", "generate", "edit"])),
@@ -87,14 +78,6 @@ const OpenAIResponsesCoreFields = {
  input: Schema.Array(Schema.Union([OpenResponses.InputItem, OpenAIResponsesHostedToolItem])),
  tools: optionalArray(OpenAIResponsesTools),
  tool_choice: Schema.optional(OpenAIResponsesToolChoice),
-  context_management: Schema.optional(
-    Schema.Array(
-      Schema.Struct({
-        type: Schema.Literal("compaction"),
-        compact_threshold: Schema.optional(Schema.Int.check(Schema.isGreaterThan(0))),
-      }),
-    ),
-  ),
 }

 const OpenAIResponsesBody = Schema.Struct({
@@ -142,14 +125,15 @@ const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>, tool
 const decodeBody = ProviderShared.validateWith(Schema.decodeUnknownEffect(OpenAIResponsesBody))

 const fromRequest = Effect.fn("OpenAIResponses.fromRequest")(function* (request: LLMRequest) {
-  const management = yield* ProviderShared.validateWith(
-    Schema.decodeUnknownEffect(Schema.UndefinedOr(ContextManagement)),
-  )(request.providerOptions?.contextManagement)
+  const body = yield* OpenResponses.fromRequestWithExtension(
+    LLMRequest.update(request, { tools: [], toolChoice: undefined }),
+    extension,
+  )
  const toolSchemaCompatibility = request.model.compatibility?.toolSchema
+  const parallelToolCalls = OpenResponses.resolveParallelToolCalls(request)
  return yield* decodeBody({
-    ...(yield* OpenResponses.lowerConversation(request, extension)),
-    ...OpenResponses.lowerGeneration(request),
-    context_management: management?.map((edit) => ({ type: edit.type, compact_threshold: edit.compactThreshold })),
+    ...body,
+    ...(parallelToolCalls === undefined ? {} : { parallel_tool_calls: parallelToolCalls }),
    tools:
      request.tools.length === 0
        ? undefined
@@ -157,8 +141,7 @@ const fromRequest = Effect.fn("OpenAIResponses.fromRequest")(function* (request:
            lowerTool(tool, ToolSchemaProjection.modelCompatibility(tool.inputSchema, toolSchemaCompatibility)),
          ),
    tool_choice:
-      OpenResponses.allowedToolChoice(request) ??
-      (request.toolChoice ? yield* lowerToolChoice(request.toolChoice, request.tools) : undefined),
+      body.tool_choice ?? (request.toolChoice ? yield* lowerToolChoice(request.toolChoice, request.tools) : undefined),
  })
 })

@@ -240,7 +223,6 @@ export const transport = channelTransport({
 })

 export const route = Route.make({
-  compact: ResponsesCompaction.make(extension),
  id: ADAPTER,
  provider: "openai",
  providerMetadataKey: "openai",
@@ -1,174 +0,0 @@
-import { Effect, Schema, Stream } from "effect"
-import {
-  AIError,
-  InvalidProviderOutputError,
-  CompactionPart,
-  CompactionResponse,
-  HttpOptions,
-  LLMRequest,
-  Message,
-  type ContentPart,
-  mergeJsonRecords,
-} from "../../schema/index.js"
-import type { CompactOperation } from "../../route/client.js"
-import { Endpoint } from "../../route/endpoint.js"
-import { RequestExecutor } from "../../route/executor.js"
-import { HttpTransport } from "../../route/transport/index.js"
-import { OpenResponses } from "../open-responses.js"
-import { JsonObject, optionalNull, ProviderShared } from "../shared.js"
-
-const Body = Schema.Struct({
-  model: Schema.String,
-  input: Schema.Array(Schema.Unknown),
-  instructions: optionalNull(Schema.String),
-  previous_response_id: optionalNull(Schema.String),
-  service_tier: optionalNull(Schema.String),
-  prompt_cache_key: optionalNull(Schema.String),
-  prompt_cache_retention: optionalNull(Schema.String),
-  prompt_cache_options: optionalNull(
-    Schema.Struct({ mode: Schema.optional(Schema.String), ttl: Schema.optional(Schema.String) }),
-  ),
-})
-
-const Text = Schema.Union([OpenResponses.OpenResponsesInputText, OpenResponses.OpenResponsesOutputText])
-const File = Schema.Union([
-  Schema.Struct({
-    ...OpenResponses.OpenResponsesInputFile.fields,
-    file_url: Schema.String,
-    file_data: Schema.optional(Schema.Never),
-  }),
-  Schema.Struct({
-    ...OpenResponses.OpenResponsesInputFile.fields,
-    file_data: Schema.String,
-    file_url: Schema.optional(Schema.Never),
-  }),
-])
-const MessageFields = {
-  type: Schema.Literal("message"),
-  id: Schema.optional(Schema.String),
-  status: Schema.optional(Schema.String),
-  phase: Schema.optional(OpenResponses.MessagePhase),
-}
-const Response = Schema.Struct({
-  object: Schema.Literal("response.compaction"),
-  output: Schema.Array(
-    Schema.Union([
-      OpenResponses.CompactionItem,
-      OpenResponses.OpenResponsesReasoningItem,
-      Schema.Struct({
-        ...MessageFields,
-        role: Schema.Literal("user"),
-        content: Schema.Array(Schema.Union([Text, OpenResponses.OpenResponsesInputImage, File])).check(
-          Schema.isMinLength(1),
-        ),
-      }),
-      Schema.Struct({
-        ...MessageFields,
-        role: Schema.Literal("assistant"),
-        content: Schema.Array(Text).check(Schema.isMinLength(1)),
-      }),
-    ]),
-  ),
-  usage: Schema.optional(Schema.StructWithRest(OpenResponses.OpenResponsesUsage, [JsonObject])),
-})
-
-export const make = (extension: OpenResponses.Extension): CompactOperation =>
-  Effect.fn("ResponsesCompaction.execute")(function* (request, executor, options) {
-    const route = request.model.route
-    const native = yield* OpenResponses.lowerConversation(request, extension)
-    const body = yield* ProviderShared.validateWith(Schema.decodeUnknownEffect(Body))(
-      mergeJsonRecords(
-        {
-          ...native,
-          service_tier: request.providerOptions?.serviceTier,
-          prompt_cache_key: ProviderShared.promptCacheKey(request),
-        },
-        request.http?.body,
-      ),
-    )
-    const url = Endpoint.render(route.endpoint, { request, body: native })
-    url.pathname = `${url.pathname.replace(/\/$/, "")}/compact`
-    const parts = yield* HttpTransport.jsonRequestParts({
-      request: LLMRequest.update(request, {
-        http: request.http === undefined ? undefined : new HttpOptions({ ...request.http, body: undefined }),
-      }),
-      body,
-      endpoint: Endpoint.path(url.toString()),
-      auth: route.auth,
-      encodeBody: Schema.encodeSync(Schema.fromJsonString(Body)),
-    })
-    const response = yield* executor.execute(
-      ProviderShared.jsonPost({ url: parts.url, body: parts.bodyText, headers: parts.headers }),
-      options?.http,
-    )
-    const text = yield* RequestExecutor.responseStream(response).pipe(
-      Stream.decodeText(),
-      Stream.runFold(
-        () => "",
-        (text, chunk) => text + chunk,
-      ),
-    )
-    const invalid = (message: string, cause?: unknown) =>
-      new AIError({
-        reason: new InvalidProviderOutputError({
-          route: route.id,
-          message,
-          body: text,
-          cause,
-          http: RequestExecutor.responseHttp(response),
-        }),
-      })
-    const result = yield* Schema.decodeUnknownEffect(Schema.fromJsonString(Response))(text).pipe(
-      Effect.mapError((cause) => invalid("Invalid compaction response", cause)),
-    )
-    if (!result.output.some((item) => item.type === "compaction"))
-      return yield* invalid("Compaction response did not contain a checkpoint")
-    return new CompactionResponse({
-      messages: result.output.map((item) => toMessage(item, request.model)),
-      usage: OpenResponses.mapUsage(result.usage, OpenResponses.metadataKey(request.model)),
-    })
-  })
-
-function toMessage(item: (typeof Response.Type.output)[number], model: LLMRequest["model"]): Message {
-  if (item.type === "compaction")
-    return Message.assistant(
-      CompactionPart.make({ provider: model.provider, id: item.id ?? undefined, encrypted: item.encrypted_content }),
-    )
-
-  const key = OpenResponses.metadataKey(model)
-  if (item.type === "reasoning") {
-    const summary = item.summary.length ? item.summary : [{ text: "" }]
-    return Message.assistant(
-      summary.map((part) => ({
-        type: "reasoning" as const,
-        text: part.text,
-        providerMetadata: { [key]: { itemId: item.id, reasoningEncryptedContent: item.encrypted_content } },
-      })),
-    )
-  }
-
-  return Message.make({
-    role: item.role,
-    providerMetadata: { [key]: { itemId: item.id, type: item.type, status: item.status, phase: item.phase } },
-    content: item.content.map((part): ContentPart => {
-      if (part.type === "input_text" || part.type === "output_text") return { type: "text", text: part.text }
-      if (part.type === "input_image")
-        return {
-          type: "media",
-          data: part.image_url,
-          mediaType: /^data:([^;,]+)/.exec(part.image_url)?.[1] ?? "image/*",
-          providerMetadata: part.detail === undefined ? undefined : { [key]: { detail: part.detail } },
-        }
-      const data = part.file_url === undefined ? part.file_data : part.file_url
-      return {
-        type: "media",
-        data,
-        filename: part.filename,
-        mediaType: /^data:([^;,]+)/.exec(data)?.[1] ?? "application/octet-stream",
-        providerMetadata: part.detail === undefined ? undefined : { [key]: { detail: part.detail } },
-      }
-    }),
-  })
-}
-
-export * as ResponsesCompaction from "./responses-compaction.js"
@@ -4,7 +4,6 @@ import type { LLMRequest } from "../schema/index.js"
 import { OpenResponses } from "./open-responses.js"
 import { JsonObject, optionalNull, ProviderShared } from "./shared.js"
 import { ResponsesHostedTools } from "./utils/responses-hosted-tools.js"
-import { ResponsesCompaction } from "./utils/responses-compaction.js"

 const ADAPTER = "xai-responses"
 const NAME = "xAI Responses"
@@ -45,10 +44,6 @@ const extension = {

 const decodeBody = ProviderShared.validateWith(Schema.decodeUnknownEffect(XAIResponsesBody))
 const fromRequest = Effect.fn("XAIResponses.fromRequest")(function* (request: LLMRequest) {
-  if (request.providerOptions?.contextManagement !== undefined)
-    return yield* ProviderShared.invalidRequest(
-      "xAI requires explicit compaction through LLMClient.compact; automatic context management is not supported",
-    )
  return yield* decodeBody(yield* OpenResponses.fromRequestWithExtension(request, extension))
 })

@@ -89,6 +84,4 @@ export const protocol = Protocol.make({
  },
 })

-export const compact = ResponsesCompaction.make(extension)
-
 export * as XAIResponses from "./xai-responses.js"
@@ -1,5 +1,4 @@
 import type { LanguageModel, ProviderOptions } from "./schema/index.js"
-import type { CompactOperation } from "./route/client.js"

 export interface Settings extends Readonly<Record<string, unknown>> {
  readonly baseURL?: string
@@ -10,9 +9,8 @@ export interface Settings extends Readonly<Record<string, unknown>> {
 export interface Definition<
  ProviderSettings extends Settings = Settings,
  Options extends ProviderOptions = ProviderOptions,
-  Compact extends CompactOperation | undefined = CompactOperation | undefined,
 > {
-  readonly model: (modelID: string, settings: ProviderSettings) => LanguageModel<Options, Compact>
+  readonly model: (modelID: string, settings: ProviderSettings) => LanguageModel<Options>
 }

 export * as ProviderPackage from "./provider-package.js"
@@ -1,11 +1,9 @@
-import type { Route, RouteDefaultsInput } from "../route/client.js"
+import type { RouteDefaultsInput } from "../route/client.js"
 import { Auth } from "../route/auth.js"
 import type { ProviderPackage } from "../provider-package.js"
 import { ProviderID, type ModelID } from "../schema/index.js"
 import * as BedrockConverse from "../protocols/bedrock-converse.js"
 import type { BedrockCredentials } from "../protocols/bedrock-converse.js"
-import { BedrockMessages } from "../protocols/bedrock-messages.js"
-import type { AnthropicMessages } from "../protocols/anthropic-messages.js"

 export const id = ProviderID.make("amazon-bedrock")

@@ -27,40 +25,38 @@ export interface Settings extends ProviderPackage.Settings {
  readonly region?: string
  readonly topP?: number
 }
-export const routes = [BedrockConverse.route, BedrockMessages.route]
+export const routes = [BedrockConverse.route]

 const bedrockBaseURL = (region: string) => `https://bedrock-runtime.${region}.amazonaws.com`

-const configuredRoute = <Body, Prepared>(route: Route<Body, Prepared>, input: Config) => {
+const configuredRoute = (input: Config) => {
  const { apiKey, credentials, region, baseURL, ...rest } = input
  const resolvedRegion = region ?? credentials?.region ?? "us-east-1"
-  return route.with({
+  return BedrockConverse.route.with({
    ...rest,
    provider: id,
-    providerMetadataKey: route.providerMetadataKey,
+    providerMetadataKey: "bedrock",
    endpoint: { baseURL: baseURL ?? bedrockBaseURL(resolvedRegion) },
    auth: apiKey === undefined ? BedrockConverse.sigV4Auth(credentials) : Auth.bearer(apiKey),
  })
 }

 export const configure = (input: Config = {}) => {
-  const route = configuredRoute(BedrockConverse.route, input)
-  const messages = configuredRoute(BedrockMessages.route, input)
+  const route = configuredRoute(input)
  return {
    id,
    model: (modelID: string | ModelID) => route.model({ id: modelID }),
-    messages: (modelID: string | ModelID) => messages.model<AnthropicMessages.ProviderOptionsInput>({ id: modelID }),
    configure,
  }
 }

 export const provider = configure()
-const config = (settings: Settings): Config => {
+export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) => {
  if (settings.auth === "bearer" && settings.apiKey === undefined)
    throw new Error("Amazon Bedrock bearer auth requires apiKey")
  if (settings.auth === "sigv4" && settings.apiKey !== undefined)
    throw new Error("Amazon Bedrock SigV4 auth does not accept apiKey")
-  return {
+  return configure({
    apiKey: settings.auth === "sigv4" ? undefined : settings.apiKey,
    baseURL: settings.baseURL,
    credentials: settings.credentials,
@@ -68,13 +64,5 @@ const config = (settings: Settings): Config => {
    headers: settings.headers === undefined ? undefined : { ...settings.headers },
    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
    region: settings.region,
-  }
+  }).model(modelID)
 }
-
-export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) =>
-  configure(config(settings)).model(modelID)
-export const messagesModel: ProviderPackage.Definition<
-  Settings & { readonly providerOptions?: AnthropicMessages.ProviderOptionsInput },
-  AnthropicMessages.ProviderOptionsInput
->["model"] = (modelID, settings) =>
-  configure({ ...config(settings), providerOptions: settings.providerOptions }).messages(modelID)
@@ -1 +0,0 @@
-export { messagesModel as model } from "../amazon-bedrock.js"
@@ -1,7 +1,7 @@
 import { Headers } from "effect/unstable/http"
 import { Auth } from "../route/auth.js"
 import { type AtLeastOne, type ProviderAuthOption } from "../route/auth-options.js"
-import type { Route, RouteDefaultsInput, CompactOperation } from "../route/client.js"
+import type { Route as RouteDef, RouteDefaultsInput } from "../route/client.js"
 import type { ProviderPackage } from "../provider-package.js"
 import { ProviderID, type ModelID } from "../schema/index.js"
 import * as OpenAIChat from "../protocols/openai-chat.js"
@@ -102,11 +102,7 @@ const auth = (input: Config) => {
  )
 }

-const configuredRoute = <Body, Prepared, Compact extends CompactOperation | undefined>(
-  route: Route<Body, Prepared, Compact>,
-  input: Config,
-  modelID: string | ModelID,
-) =>
+const configuredRoute = <Body, Prepared>(route: RouteDef<Body, Prepared>, input: Config, modelID: string | ModelID) =>
  route.with({
    auth: auth(input),
    endpoint: endpoint(input, modelID),
@@ -165,11 +161,10 @@ const config = (settings: Settings): Config => {
  throw new Error("Azure requires resourceName or baseURL")
 }

-export const responsesModel: ProviderPackage.Definition<
-  Settings,
-  OpenAIProviderOptionsInput,
-  CompactOperation
->["model"] = (modelID, settings) => configure(config(settings)).responses(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,
@@ -57,9 +57,7 @@ const route = Route.make({
  }),
  endpoint: Endpoint.path(({ request }) => `/${request.model.id}:streamRawPredict`),
  auth: Auth.none,
-  transport: AnthropicMessages.transport<
-    Omit<AnthropicMessages.AnthropicMessagesBody, "model"> & { readonly anthropic_version: typeof VERSION }
-  >(),
+  framing: AnthropicMessages.framing,
  headers: () => ({ "anthropic-version": HEADER_VERSION }),
 })

@@ -1,12 +1,10 @@
 import { mergeProviderOptions, type ProviderOptions } from "../schema/index.js"
 import type { OpenAIServiceTier } from "../protocols/utils/openai-options.js"
 import type { Options } from "../protocols/utils/open-responses-options.js"
-import type { ContextManagement } from "../protocols/openai-responses.js"

 export type { OpenAIResponseIncludable, OpenAIServiceTier } from "../protocols/utils/openai-options.js"

 export type OpenAIOptionsInput = Omit<Options, "serviceTier"> & {
-  readonly contextManagement?: ContextManagement
  readonly serviceTier?: OpenAIServiceTier
  readonly [key: string]: unknown
 }
@@ -1,5 +1,5 @@
 import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
-import type { Route, RouteDefaultsInput, CompactOperation } from "../route/client.js"
+import type { Route, RouteDefaultsInput } from "../route/client.js"
 import type { ProviderPackage } from "../provider-package.js"
 import { HttpOptions, ProviderID, ToolDefinition, mergeHttpOptions, type ModelID } from "../schema/index.js"
 import * as OpenAIChat from "../protocols/openai-chat.js"
@@ -73,10 +73,7 @@ const defaults = (input: Config) => {
  return rest
 }

-const configuredRoute = <Body, Prepared, Compact extends CompactOperation | undefined>(
-  route: Route<Body, Prepared, Compact>,
-  input: Config,
-) =>
+const configuredRoute = <Body, Prepared>(route: Route<Body, Prepared>, input: Config) =>
  route.with({
    auth: auth(input),
    endpoint: { baseURL: input.baseURL, query: input.queryParams },
@@ -132,10 +129,7 @@ const config = (settings: Settings): Config => {
  }
 }

-export const model: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput, CompactOperation>["model"] = (
-  modelID,
-  settings,
-) => {
+export const model: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (modelID, settings) => {
  return configure(config(settings)).responses(modelID)
 }

@@ -1,5 +1,5 @@
 import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
-import { Route, type RouteDefaultsInput, type CompactOperation } from "../route/client.js"
+import { Route, type RouteDefaultsInput } from "../route/client.js"
 import { Endpoint } from "../route/endpoint.js"
 import { HttpOptions, ProviderID, type ModelID } from "../schema/index.js"
 import * as OpenAICompatibleProfiles from "./openai-compatible-profile.js"
@@ -13,7 +13,7 @@ import type { ProviderPackage } from "../provider-package.js"

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

-export type XAIProviderOptionsInput = OpenAIOptionsInput & { readonly contextManagement?: never }
+export type XAIProviderOptionsInput = OpenAIOptionsInput

 export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
  ProviderAuthOption<"optional"> & {
@@ -32,7 +32,6 @@ export type { XAIImageOptions } from "../protocols/xai-images.js"
 const RESPONSES_WEBSOCKET_ROTATE_AFTER_MS = 24 * 60 * 1000

 const responsesRoute = Route.make({
-  compact: XAIResponses.compact,
  id: "openai-responses",
  provider: id,
  providerMetadataKey: "xai",
@@ -103,10 +102,7 @@ export const configure = (input: LanguageModelOptions = {}) => {
 }

 export const provider = configure()
-export const model: ProviderPackage.Definition<Settings, XAIProviderOptionsInput, CompactOperation>["model"] = (
-  modelID,
-  settings,
-) =>
+export const model: ProviderPackage.Definition<Settings, XAIProviderOptionsInput>["model"] = (modelID, settings) =>
  configure({
    apiKey: settings.apiKey,
    baseURL: settings.baseURL,
@@ -13,7 +13,6 @@ import * as ProviderShared from "../protocols/shared.js"
 import type { ProtocolID, ProviderOptions } from "../schema/index.js"
 import {
  AIError,
-  CompactionResponse,
  AIErrorReason,
  GenerationOptions,
  HttpOptions,
@@ -35,12 +34,7 @@ export interface RouteBody<Body> {
  readonly from: (request: LLMRequest) => Effect.Effect<Body, AIError>
 }

-export interface Route<
-  Body,
-  Prepared = unknown,
-  Compact extends CompactOperation | undefined = CompactOperation | undefined,
-> {
-  readonly compact: Compact
+export interface Route<Body, Prepared = unknown> {
  readonly id: string
  readonly provider?: ProviderID
  /** ProviderMetadata namespace emitted and consumed by this route. */
@@ -48,15 +42,13 @@ export interface Route<
  readonly protocol: ProtocolID
  readonly endpoint: Endpoint.Definition<Body>
  readonly auth: Auth.Definition
-  /** Deployment headers resolved once for every operation, before transport authentication. */
-  readonly headers?: (input: { readonly request: LLMRequest }) => Record<string, string>
  readonly transport: Transport<Body, Prepared, unknown>
  readonly defaults: RouteDefaults
  readonly body: RouteBody<Body>
-  readonly with: (patch: RoutePatch<Body, Prepared>) => Route<Body, Prepared, Compact>
+  readonly with: (patch: RoutePatch<Body, Prepared>) => Route<Body, Prepared>
  readonly model: <Options extends ProviderOptions = ProviderOptions>(
    input: RouteMappedLanguageModelInput,
-  ) => LanguageModel<Options, Compact>
+  ) => LanguageModel<Options>
  readonly prepareTransport: (
    body: Body,
    request: LLMRequest,
@@ -74,11 +66,7 @@ export interface Route<
 // Normal call sites use `OpenAIChat.route`; callers only need body types
 // when preparing a request with a protocol-specific type assertion.
 // oxlint-disable-next-line typescript-eslint/no-explicit-any
-export type AnyRoute<Compact extends CompactOperation | undefined = CompactOperation | undefined> = Route<
-  any,
-  any,
-  Compact
->
+export type AnyRoute = Route<any, any>

 export type HttpOptionsInput = HttpOptions.Input

@@ -111,15 +99,15 @@ export interface RoutePatch<Body, Prepared> extends RouteDefaultsInput {

 type RouteMappedLanguageModelInput = RouteLanguageModelInput | RouteRoutedLanguageModelInput

-const makeRouteLanguageModel = <Options extends ProviderOptions, Compact extends CompactOperation | undefined>(
-  route: AnyRoute<Compact>,
+const makeRouteLanguageModel = <Options extends ProviderOptions = ProviderOptions>(
+  route: AnyRoute,
  mapped: RouteMappedLanguageModelInput,
 ) => {
  const provider = route.provider ?? ("provider" in mapped ? mapped.provider : undefined)
  if (!provider) throw new Error(`Route.model(${route.id}) requires a provider`)
  if (!endpointBaseURL(route.endpoint))
    throw new Error(`Route.model(${route.id}) requires an endpoint baseURL — configure it on the route first`)
-  return LanguageModel.make<Options, Compact>({
+  return LanguageModel.make<Options>({
    ...mapped,
    provider,
    route,
@@ -162,10 +150,6 @@ export const httpOptions = (input: HttpOptionsInput | undefined) => {
 }

 export interface Interface {
-  readonly compact: (
-    request: CompactionRequest,
-    options?: Pick<StreamOptions, "http">,
-  ) => Effect.Effect<CompactionResponse, AIError>
  readonly stream: StreamMethod
  readonly generate: GenerateMethod
 }
@@ -183,17 +167,6 @@ export interface GenerateMethod {
  (request: LLMRequest, options?: StreamOptions): Effect.Effect<LLMResponse, AIError>
 }

-export type CompactOperation = (
-  request: LLMRequest,
-  executor: RequestExecutor.Interface,
-  options?: Pick<StreamOptions, "http">,
-) => Effect.Effect<CompactionResponse, AIError>
-
-export type CompactionRequest = LLMRequest<LanguageModel<ProviderOptions, CompactOperation>>
-
-export const canCompact = (request: LLMRequest): request is CompactionRequest =>
-  request.model.route.compact !== undefined
-
 export class Service extends Context.Service<Service, Interface>()("@opencode/LLMClient") {}

 const resolveRequestOptions = (request: LLMRequest) => {
@@ -214,7 +187,6 @@ const resolveRequestOptions = (request: LLMRequest) => {
 }

 export interface MakeInput<Body, Frame, Event, State> {
-  readonly compact?: CompactOperation
  /** Route id used in diagnostics and prepared request metadata. */
  readonly id: string
  /** Provider identity for route-owned model construction. */
@@ -236,7 +208,6 @@ export interface MakeInput<Body, Frame, Event, State> {
 }

 export interface MakeTransportInput<Body, Prepared, Frame, Event, State> {
-  readonly compact?: CompactOperation
  /** Route id used in diagnostics and prepared request metadata. */
  readonly id: string
  /** Provider identity for route-owned model construction. */
@@ -312,14 +283,12 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(

  const build = (routeInput: BuiltRouteInput): Route<Body, Prepared> => {
    const route: Route<Body, Prepared> = {
-      compact: routeInput.compact,
      id: routeInput.id,
      provider: routeInput.provider === undefined ? undefined : ProviderID.make(routeInput.provider),
      providerMetadataKey: routeInput.providerMetadataKey,
      protocol: protocol.id,
      endpoint: routeInput.endpoint,
      auth: routeInput.auth ?? Auth.none,
-      headers: routeInput.headers,
      transport: routeInput.transport,
      defaults: routeInput.defaults ?? {},
      body: protocol.body,
@@ -341,7 +310,7 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
        })
      },
      model: <Options extends ProviderOptions = ProviderOptions>(input: RouteMappedLanguageModelInput) =>
-        makeRouteLanguageModel<Options, CompactOperation | undefined>(route, input),
+        makeRouteLanguageModel<Options>(route, input),
      prepareTransport: (body, request, options) =>
        routeInput.transport.prepare({
          body,
@@ -349,6 +318,7 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
          endpoint: routeInput.endpoint,
          auth: routeInput.auth ?? Auth.none,
          encodeBody,
+          headers: routeInput.headers,
          middleware: options?.http,
          webSocket: options?.webSocket,
        }),
@@ -438,12 +408,6 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
  return build({ ...input, defaults: mergeRouteDefaults(undefined, input.defaults ?? {}) })
 }

-export function make<Body, Prepared, Frame, Event, State>(
-  input: MakeTransportInput<Body, Prepared, Frame, Event, State> & { readonly compact: CompactOperation },
-): Route<Body, Prepared, CompactOperation>
-export function make<Body, Frame, Event, State>(
-  input: MakeInput<Body, Frame, Event, State> & { readonly compact: CompactOperation },
-): Route<Body, HttpTransport.HttpPrepared<Frame>, CompactOperation>
 export function make<Body, Prepared, Frame, Event, State>(
  input: MakeTransportInput<Body, Prepared, Frame, Event, State>,
 ): Route<Body, Prepared>
@@ -471,7 +435,6 @@ export function make<Body, Prepared, Frame, Event, State>(
  if ("transport" in input) return makeFromTransport(input)
  const protocol = input.protocol
  return makeFromTransport({
-    compact: input.compact,
    id: input.id,
    provider: input.provider,
    providerMetadataKey: input.providerMetadataKey,
@@ -484,19 +447,11 @@ export function make<Body, Prepared, Frame, Event, State>(
  })
 }

-const prepareRequest = (request: LLMRequest) => {
+const compile = Effect.fn("LLM.compile")(function* (request: LLMRequest, options?: StreamOptions) {
  const original = applyCachePolicy(resolveRequestOptions(request))
  const sanitized = LLMRequest.update(original, sanitizeSurrogates({ ...LLMRequest.input(original), model: undefined }))
  const tools = [...new Map(sanitized.tools.map((tool) => [tool.name, tool])).values()]
  const resolved = tools.length === sanitized.tools.length ? sanitized : LLMRequest.update(sanitized, { tools })
-  const headers = resolved.model.route.headers?.({ request: resolved })
-  return headers === undefined
-    ? resolved
-    : LLMRequest.update(resolved, { http: mergeHttpOptions(new HttpOptions({ headers }), resolved.http) })
-}
-
-const compile = Effect.fn("LLM.compile")(function* (request: LLMRequest, options?: StreamOptions) {
-  const resolved = prepareRequest(request)
  const route = resolved.model.route

  const body = yield* route.body
@@ -555,15 +510,6 @@ export function generate(request: LLMRequest, options?: StreamOptions): Effect.E
  })
 }

-export const compact = (
-  request: CompactionRequest,
-  options?: Pick<StreamOptions, "http">,
-): Effect.Effect<CompactionResponse, AIError, Service> =>
-  Effect.gen(function* () {
-    const client = yield* Service
-    return yield* client.compact(request, options)
-  })
-
 export const streamRequest = (request: LLMRequest, options?: StreamOptions) =>
  Stream.unwrap(
    Effect.gen(function* () {
@@ -574,29 +520,16 @@ export const streamRequest = (request: LLMRequest, options?: StreamOptions) =>
 export const layer: Layer.Layer<Service, never, RequestExecutor.Service> = Layer.effect(
  Service,
  Effect.gen(function* () {
-    const executor = yield* RequestExecutor.Service
-    const stream = streamRequestWith({ http: executor })
-    return Service.of({
-      stream,
-      generate: generateWith(stream),
-      compact: (request, options) =>
-        Effect.suspend(() => {
-          const operation = request.model.route.compact
-          if (!operation)
-            return ProviderShared.invalidRequest(
-              `${request.model.provider}/${request.model.route.id} does not support explicit compaction`,
-            )
-          return operation(prepareRequest(request), executor, options)
-        }),
+    const stream = streamRequestWith({
+      http: yield* RequestExecutor.Service,
    })
+    return Service.of({ stream, generate: generateWith(stream) })
  }),
 )

 export const Route = { make } as const

 export const LLMClient = {
-  canCompact,
-  compact,
  Service,
  layer,
  stream,
@@ -3,7 +3,6 @@ import { LLM } from "@opencode-ai/schema/llm"
 import { ContentBlockID, ToolCallID } from "./ids.js"
 import {
  Message,
-  CompactionPart,
  ProviderMetadata,
  ToolCallPart,
  ToolOutput,
@@ -63,8 +62,6 @@ export { ProviderMetadata } from "./messages.js"
 * Matches the same escape-hatch field on `LLMEvent`.
 */
 export class Usage extends Schema.Class<Usage>("AI.Usage")({
-  /** Effective input size of the final message iteration, when reported; not billed totals. */
-  contextTokens: Schema.optional(Schema.Number),
  inputTokens: Schema.optional(Schema.Number),
  outputTokens: Schema.optional(Schema.Number),
  nonCachedInputTokens: Schema.optional(Schema.Number),
@@ -75,7 +72,7 @@ export class Usage extends Schema.Class<Usage>("AI.Usage")({
  providerMetadata: Schema.optional(ProviderMetadata),
 }) {
  /**
-   * Non-reasoning output tokens (including compaction summaries) — `outputTokens` minus `reasoningTokens`, clamped
+   * Visible output tokens — `outputTokens` minus `reasoningTokens`, clamped
   * to zero. The one place subtraction happens in this contract; the clamp
   * means a provider reporting `reasoningTokens > outputTokens` produces a
   * harmless zero rather than a negative that crashes downstream schemas.
@@ -91,12 +88,6 @@ export class Usage extends Schema.Class<Usage>("AI.Usage")({

 export type UsageInput = Usage | ConstructorParameters<typeof Usage>[0]

-/** A replacement context window. Replace prior history with these messages. */
-export class CompactionResponse extends Schema.Class<CompactionResponse>("LLM.CompactionResponse")({
-  messages: Schema.Array(Message),
-  usage: Schema.optional(Usage),
-}) {}
-
 export const StepStart = Schema.Struct({
  type: Schema.tag("step-start"),
  index: Schema.Number,
@@ -250,7 +241,6 @@ export const ProviderErrorEvent = Schema.Struct({
 export type ProviderErrorEvent = Schema.Schema.Type<typeof ProviderErrorEvent>

 const llmEventTagged = Schema.Union([
-  CompactionPart,
  StepStart,
  TextStart,
  TextDelta,
@@ -284,7 +274,6 @@ const toolCallID = (value: ToolCallID | string) => ToolCallID.make(value)
 * `events.filter(LLMEvent.guards["tool-call"])`.
 */
 export const LLMEvent = Object.assign(llmEventTagged, {
-  compaction: CompactionPart.make,
  stepStart: StepStart.make,
  textStart: (input: WithID<TextStart, ContentBlockID>) => TextStart.make({ ...input, id: contentBlockID(input.id) }),
  textDelta: (input: WithID<TextDelta, ContentBlockID>) => TextDelta.make({ ...input, id: contentBlockID(input.id) }),
@@ -322,7 +311,6 @@ export const LLMEvent = Object.assign(llmEventTagged, {
    }),
  providerError: ProviderErrorEvent.make,
  is: {
-    compaction: llmEventTagged.guards.compaction,
    stepStart: llmEventTagged.guards["step-start"],
    textStart: llmEventTagged.guards["text-start"],
    textDelta: llmEventTagged.guards["text-delta"],
@@ -345,10 +333,10 @@ export const LLMEvent = Object.assign(llmEventTagged, {
 export type LLMEvent = Schema.Schema.Type<typeof llmEventTagged>

 /** Joins deltas per fragment, letting an authoritative end value replace that fragment's accumulated deltas. */
-const joinFragments = (
+const joinFragments = <Delta extends { id: string; text: string }, End extends { id: string; text?: string }>(
  events: ReadonlyArray<LLMEvent>,
-  isDelta: (event: LLMEvent) => event is LLMEvent & { id: string; text: string },
-  isEnd: (event: LLMEvent) => event is LLMEvent & { id: string; text?: string },
+  isDelta: (event: LLMEvent) => event is Extract<LLMEvent, Delta>,
+  isEnd: (event: LLMEvent) => event is Extract<LLMEvent, End>,
 ) => {
  const order: string[] = []
  const parts = new Map<string, string>()
@@ -575,8 +563,6 @@ const reduceToolCall = (state: ResponseState, event: ToolCall): ResponseState =>
 const reduceResponseState = (state: ResponseState, event: LLMEvent): ResponseState => {
  const next = appendEvent(state, event)
  switch (event.type) {
-    case "compaction":
-      return appendContent(next, event)
    case "text-start":
      return ensureText(next, event.id, event.providerMetadata)
    case "text-delta":
@@ -7,10 +7,8 @@ import {
  HttpOptions,
  JsonSchema,
  LanguageModelSchema,
-  type LanguageModel,
  ProviderOptions,
 } from "./options.js"
-import { ProviderID } from "./ids.js"

 export const MessageRole = Schema.Literals(["system", "user", "assistant", "tool"])
 export type MessageRole = Schema.Schema.Type<typeof MessageRole>
@@ -54,7 +52,6 @@ export const MediaPart = Schema.Struct({
  filename: 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.Media" })
 export type MediaPart = Schema.Schema.Type<typeof MediaPart>

@@ -188,40 +185,9 @@ export const ReasoningPart = Schema.Struct({
 }).annotate({ identifier: "LLM.Content.Reasoning" })
 export type ReasoningPart = Schema.Schema.Type<typeof ReasoningPart>

-/** A provider-generated context checkpoint, distinct from visible assistant text. */
-type CompactionContent =
-  | { readonly encrypted: string; readonly text?: never }
-  | { readonly text: string | null; readonly encrypted?: never }
-
-const compactionPartSchema = Schema.Struct({
-  type: Schema.Literal("compaction"),
-  provider: ProviderID,
-  id: Schema.optional(Schema.String),
-  encrypted: Schema.optional(Schema.String),
-  /** Null means the provider failed to produce a summary; prior history must be retained. */
-  text: Schema.optional(Schema.NullOr(Schema.String)),
-})
-  .pipe(
-    Schema.refine(
-      (part): part is typeof part & CompactionContent => (part.encrypted !== undefined) !== (part.text !== undefined),
-      { message: "Compaction requires either encrypted content or a summary" },
-    ),
-  )
-  .annotate({ identifier: "LLM.Content.Compaction" })
-export type CompactionPart = typeof compactionPartSchema.Type
-export const CompactionPart = Object.assign(compactionPartSchema, {
-  make: (input: Omit<CompactionPart, "type" | "encrypted" | "text"> & CompactionContent): CompactionPart =>
-    Schema.decodeUnknownSync(compactionPartSchema)({ type: "compaction", ...input }),
-})
-
-export const ContentPart = Schema.Union([
-  TextPart,
-  MediaPart,
-  ToolCallPart,
-  ToolResultPart,
-  ReasoningPart,
-  CompactionPart,
-]).pipe(Schema.toTaggedUnion("type"))
+export const ContentPart = Schema.Union([TextPart, MediaPart, ToolCallPart, ToolResultPart, ReasoningPart]).pipe(
+  Schema.toTaggedUnion("type"),
+)
 export type ContentPart = Schema.Schema.Type<typeof ContentPart>

 export class Message extends Schema.Class<Message>("LLM.Message")({
@@ -229,7 +195,6 @@ export class Message extends Schema.Class<Message>("LLM.Message")({
  role: MessageRole,
  content: Schema.Array(ContentPart),
  metadata: Schema.optional(Schema.Record(Schema.String, Schema.Unknown)),
-  providerMetadata: Schema.optional(ProviderMetadata),
  native: Schema.optional(Schema.Record(Schema.String, Schema.Unknown)),
 }) {}

@@ -307,7 +272,7 @@ export namespace ToolChoice {
  }
 }

-const requestSchema = Schema.Struct({
+export class LLMRequest extends Schema.Class<LLMRequest>("LLM.Request")({
  id: Schema.optional(Schema.String),
  model: LanguageModelSchema,
  system: Schema.Array(SystemPart),
@@ -321,26 +286,12 @@ const requestSchema = Schema.Struct({
  // Stable cache affinity for protocols that support provider-managed prompt caching.
  promptCacheKey: Schema.optional(Schema.String),
  metadata: Schema.optional(Schema.Record(Schema.String, Schema.Unknown)),
-})
-
-export class LLMRequest<Model extends LanguageModel = LanguageModel> extends Schema.Class<LLMRequest>("LLM.Request")(
-  requestSchema.fields,
-) {
-  declare readonly model: Model
-
-  // Preserve model inference instead of inheriting the schema's erased constructor signature.
-  // oxlint-disable-next-line no-useless-constructor
-  constructor(input: LLMRequest.Input<Model>) {
-    super(input)
-  }
-}
+}) {}

 export namespace LLMRequest {
-  export type Input<Model extends LanguageModel = LanguageModel> = Omit<typeof requestSchema.Type, "model"> & {
-    readonly model: Model
-  }
+  export type Input = ConstructorParameters<typeof LLMRequest>[0]

-  export const input = <Model extends LanguageModel>(request: LLMRequest<Model>): Input<Model> => ({
+  export const input = (request: LLMRequest): Input => ({
    id: request.id,
    model: request.model,
    system: request.system,
@@ -355,16 +306,7 @@ export namespace LLMRequest {
    metadata: request.metadata,
  })

-  export function update<Model extends LanguageModel>(
-    request: LLMRequest,
-    patch: Partial<Input<Model>> & { readonly model: Model },
-  ): LLMRequest<Model>
-  export function update<Model extends LanguageModel>(
-    request: LLMRequest<Model>,
-    patch: Partial<Omit<Input, "model">> & { readonly model?: undefined },
-  ): LLMRequest<Model>
-  export function update(request: LLMRequest, patch: Partial<Input>): LLMRequest
-  export function update(request: LLMRequest, patch: Partial<Input>) {
+  export const update = (request: LLMRequest, patch: Partial<Input>) => {
    if (Object.keys(patch).length === 0) return request
    return new LLMRequest({
      ...input(request),
@@ -1,6 +1,6 @@
 import { Schema } from "effect"
 import { ModelID, ProviderID } from "./ids.js"
-import type { AnyRoute, CompactOperation } from "../route/client.js"
+import type { AnyRoute } from "../route/client.js"
 import { isRecord } from "../utils/record.js"

 export const JsonSchema = Schema.Record(Schema.String, Schema.Unknown)
@@ -173,18 +173,15 @@ export namespace LanguageModelCompatibility {
    input instanceof LanguageModelCompatibility ? input : new LanguageModelCompatibility(input)
 }

-export class LanguageModel<
-  Options extends ProviderOptions = ProviderOptions,
-  Compact extends CompactOperation | undefined = CompactOperation | undefined,
-> {
+export class LanguageModel<Options extends ProviderOptions = ProviderOptions> {
  declare protected readonly _ProviderOptions: Options
  readonly id: ModelID
  readonly provider: ProviderID
-  readonly route: AnyRoute<Compact>
+  readonly route: AnyRoute
  readonly defaults?: LanguageModelDefaults
  readonly compatibility?: LanguageModelCompatibility

-  constructor(input: LanguageModel.ConstructorInput<Compact>) {
+  constructor(input: LanguageModel.ConstructorInput) {
    this.id = input.id
    this.provider = input.provider
    this.route = input.route
@@ -192,11 +189,8 @@ export class LanguageModel<
    this.compatibility = input.compatibility
  }

-  static make<
-    Options extends ProviderOptions = ProviderOptions,
-    Compact extends CompactOperation | undefined = CompactOperation | undefined,
-  >(input: LanguageModel.Input<Compact>) {
-    return new LanguageModel<Options, Compact>({
+  static make<Options extends ProviderOptions = ProviderOptions>(input: LanguageModel.Input) {
+    return new LanguageModel<Options>({
      id: ModelID.make(input.id),
      provider: ProviderID.make(input.provider),
      route: input.route,
@@ -206,9 +200,7 @@ export class LanguageModel<
    })
  }

-  static input<Options extends ProviderOptions, Compact extends CompactOperation | undefined>(
-    model: LanguageModel<Options, Compact>,
-  ): LanguageModel.ConstructorInput<Compact> {
+  static input<Options extends ProviderOptions>(model: LanguageModel<Options>): LanguageModel.ConstructorInput {
    return {
      id: model.id,
      provider: model.provider,
@@ -218,41 +210,25 @@ export class LanguageModel<
    }
  }

-  static update<Options extends ProviderOptions, Compact extends CompactOperation | undefined>(
-    model: LanguageModel<Options>,
-    patch: Partial<LanguageModel.Input<Compact>> & { readonly route: AnyRoute<Compact> },
-  ): LanguageModel<Options, Compact>
-  static update<Options extends ProviderOptions, Compact extends CompactOperation | undefined>(
-    model: LanguageModel<Options, Compact>,
-    patch: Partial<Omit<LanguageModel.Input, "route">> & { readonly route?: undefined },
-  ): LanguageModel<Options, Compact>
-  static update<Options extends ProviderOptions>(
-    model: LanguageModel<Options>,
-    patch: Partial<LanguageModel.Input>,
-  ): LanguageModel<Options>
  static update<Options extends ProviderOptions>(model: LanguageModel<Options>, patch: Partial<LanguageModel.Input>) {
    if (Object.keys(patch).length === 0) return model
    return LanguageModel.make<Options>({
      ...LanguageModel.input(model),
      ...patch,
-      route: patch.route ?? model.route,
    })
  }
 }

 export namespace LanguageModel {
-  export type ConstructorInput<Compact extends CompactOperation | undefined = CompactOperation | undefined> = {
+  export type ConstructorInput = {
    readonly id: ModelID
    readonly provider: ProviderID
-    readonly route: AnyRoute<Compact>
+    readonly route: AnyRoute
    readonly defaults?: LanguageModelDefaults
    readonly compatibility?: LanguageModelCompatibility
  }

-  export type Input<Compact extends CompactOperation | undefined = CompactOperation | undefined> = Omit<
-    ConstructorInput<Compact>,
-    "id" | "provider" | "defaults" | "compatibility"
-  > & {
+  export type Input = Omit<ConstructorInput, "id" | "provider" | "defaults" | "compatibility"> & {
    readonly id: string | ModelID
    readonly provider: string | ProviderID
    readonly defaults?: LanguageModelDefaults.Input
@@ -4,7 +4,6 @@ import { LLMClient } from "./route/client.js"
 import {
  LLMEvent,
  LLMResponse,
-  CompactionResponse,
  type FinishReasonDetails,
  type AIError,
  type LLMRequest,
@@ -13,7 +12,7 @@ import {
 } from "./schema/index.js"
 import { Context, Deferred, Effect, Latch, Layer, Queue, Scope, Stream } from "effect"

-export type Response = readonly LLMEvent[] | Stream.Stream<LLMEvent, AIError> | CompactionResponse
+export type Response = readonly LLMEvent[] | Stream.Stream<LLMEvent, AIError>

 export type Gate = Readonly<{ started: Effect.Effect<void>; release: Effect.Effect<void> }>

@@ -100,6 +99,8 @@ export const failAfter = (error: AIError, ...events: readonly LLMEvent[]) =>

 export const hangAfter = (...events: readonly LLMEvent[]) => Stream.concat(Stream.fromIterable(events), Stream.never)

+const toStream = (response: Response) => (Stream.isStream(response) ? response : Stream.fromIterable(response))
+
 const make = (options: LayerOptions) =>
  Effect.sync(() => {
    const requests: LLMRequest[] = []
@@ -112,41 +113,26 @@ const make = (options: LayerOptions) =>
        requests.length >= count ? Effect.void : Deferred.await(started).pipe(Effect.andThen(wait(count))),
      )

-    const take = (request: LLMRequest) =>
-      Effect.suspend(() => {
+    const stream: ClientInterface["stream"] = (request) =>
+      Stream.suspend(() => {
        const count = requests.push(options.transformRequest?.(request) ?? request)
        const waiting = started
        started = Deferred.makeUnsafe()
        const gate = activeGate
        try {
          const response = responses.shift() ?? (typeof fallback === "function" ? fallback(request) : fallback)
-          if (!response) return Effect.die(new Error(`TestLLM has no response for request ${count}`))
-          if (!gate) return Effect.succeed(response)
-          return Queue.offer(gate.started, undefined).pipe(Effect.andThen(gate.release.await), Effect.as(response))
+          if (!response) return Stream.die(new Error(`TestLLM has no response for request ${count}`))
+          const streamed = toStream(response)
+          if (!gate) return streamed
+          return Stream.unwrap(
+            Queue.offer(gate.started, undefined).pipe(Effect.andThen(gate.release.await), Effect.as(streamed)),
+          )
        } finally {
          // Waiters can resume synchronously; assign the reply and gate before notifying them.
          Deferred.doneUnsafe(waiting, Effect.void)
        }
      })
-    const stream: ClientInterface["stream"] = (request) =>
-      Stream.unwrap(
-        take(request).pipe(
-          Effect.map((response) => {
-            if (response instanceof CompactionResponse)
-              return Stream.die("TestLLM generation requires an event response")
-            return Stream.isStream(response) ? response : Stream.fromIterable(response)
-          }),
-        ),
-      )
    const test = Test.of({
-      compact: (request) =>
-        take(request).pipe(
-          Effect.flatMap((response) =>
-            response instanceof CompactionResponse
-              ? Effect.succeed(response)
-              : Effect.die("TestLLM compaction requires a CompactionResponse"),
-          ),
-        ),
      stream,
      generate: (request) =>
        stream(request).pipe(
@@ -56,7 +56,6 @@ function normalizeToolMessage(message: Message, pending: Map<string, ToolCallPar
    role: message.role,
    content,
    metadata: message.metadata,
-    providerMetadata: message.providerMetadata,
    native: message.native,
  })
 }
@@ -126,38 +126,6 @@ describe("applyCachePolicy", () => {
    }),
  )

-  for (const fixture of [
-    { name: "default", cache: undefined, control: { type: "ephemeral" } },
-    { name: "auto", cache: "auto", control: { type: "ephemeral" } },
-    {
-      name: "explicit one-hour",
-      cache: { tools: true, system: true, messages: { tail: 1 }, ttlSeconds: 3600 },
-      control: { type: "ephemeral", ttl: "1h" },
-    },
-    { name: "disabled", cache: "none", control: undefined },
-  ] as const) {
-    it.effect(`Bedrock Messages respects ${fixture.name} caching`, () =>
-      Effect.gen(function* () {
-        const prepared = yield* compileRequest(
-          LLM.request({
-            model: AmazonBedrock.configure({ apiKey: "test" }).messages("anthropic.claude-opus-4-6-v1"),
-            system: "Stable instructions",
-            tools: [
-              { name: "lookup", description: "Look up a value", inputSchema: { type: "object", properties: {} } },
-            ],
-            prompt: "hello",
-            cache: fixture.cache,
-          }),
-        )
-        expect(prepared.body).toMatchObject({
-          tools: [{ name: "lookup", cache_control: fixture.control }],
-          system: [{ type: "text", text: "Stable instructions", cache_control: fixture.control }],
-          messages: [{ role: "user", content: [{ type: "text", text: "hello", cache_control: fixture.control }] }],
-        })
-      }),
-    )
-  }
-
  it.effect("'auto' is a no-op on Gemini (out-of-band caching protocol)", () =>
    Effect.gen(function* () {
      const prepared = yield* compileRequest(
@@ -1,97 +0,0 @@
-import { expect } from "bun:test"
-import { Effect } from "effect"
-import { FetchHttpClient } from "effect/unstable/http"
-import { LLM, LLMRequest, Message } from "../src/index.js"
-import { LLMClient } from "../src/route/client.js"
-import { OpenAI } from "../src/providers.js"
-import { testEffect } from "./lib/effect.js"
-import { runtimeLayer } from "./lib/http.js"
-import { sseEvents } from "./lib/sse.js"
-
-testEffect(runtimeLayer(FetchHttpClient.layer)).live("compaction and a tool loop work end to end over HTTP", () =>
-  Effect.gen(function* () {
-    const checkpoint = { type: "compaction", id: "cmp_local", encrypted_content: "opaque-local-state" }
-    const calls: string[] = []
-    const server = yield* Effect.acquireRelease(
-      Effect.sync(() =>
-        Bun.serve({
-          hostname: "127.0.0.1",
-          port: 0,
-          async fetch(request) {
-            const path = new URL(request.url).pathname
-            calls.push(path)
-            const body = await request.json()
-            expect(request.headers.get("authorization")).toBe("Bearer fixture")
-            if (path === "/v1/responses/compact") {
-              expect(body.stream).toBeUndefined()
-              return Response.json({
-                object: "response.compaction",
-                output: [checkpoint],
-                usage: { input_tokens: 100, output_tokens: 10, total_tokens: 110 },
-              })
-            }
-            expect(body.input[0]).toEqual(checkpoint)
-            expect(body.stream).toBe(true)
-            if (calls.length === 2)
-              return new Response(
-                sseEvents(
-                  {
-                    type: "response.output_item.done",
-                    item: { type: "function_call", id: "fc_1", call_id: "call_1", name: "lookup", arguments: "{}" },
-                  },
-                  { type: "response.completed", response: { id: "resp_1" } },
-                ),
-                { headers: { "content-type": "text/event-stream" } },
-              )
-            expect(body.input.at(-2)).toMatchObject({ type: "function_call", call_id: "call_1" })
-            expect(body.input.at(-1)).toEqual({ type: "function_call_output", call_id: "call_1", output: "42" })
-            const output = sseEvents(
-              { type: "response.output_item.added", item: { type: "message", id: "msg_1" } },
-              { type: "response.output_text.delta", item_id: "msg_1", delta: "The answer is 42." },
-              {
-                type: "response.output_item.done",
-                item: { type: "message", id: "msg_1", content: [{ type: "output_text", text: "The answer is 42." }] },
-              },
-              { type: "response.completed", response: { id: "resp_2" } },
-            )
-            return new Response(
-              new ReadableStream({
-                start(controller) {
-                  controller.enqueue(new TextEncoder().encode(output.slice(0, 37)))
-                  controller.enqueue(new TextEncoder().encode(output.slice(37)))
-                  controller.close()
-                },
-              }),
-              { headers: { "content-type": "text/event-stream" } },
-            )
-          },
-        }),
-      ),
-      (server) => Effect.sync(() => server.stop(true)),
-    )
-    const model = OpenAI.configure({ apiKey: "fixture", baseURL: `http://127.0.0.1:${server.port}/v1` }).responses(
-      "fixture",
-    )
-    const request = LLM.request({
-      model,
-      prompt: "original",
-      tools: [{ name: "lookup", description: "Lookup a number", inputSchema: { type: "object", properties: {} } }],
-    })
-    const compacted = yield* LLMClient.compact(request)
-    const messages = [...compacted.messages, Message.user("Look up the answer")]
-    const first = yield* LLMClient.generate(LLMRequest.update(request, { messages }))
-    expect(first.toolCalls).toHaveLength(1)
-    const call = first.toolCalls[0]!
-    const last = yield* LLMClient.generate(
-      LLMRequest.update(request, {
-        messages: [
-          ...messages,
-          first.message,
-          Message.tool({ id: call.id, name: call.name, result: "42", resultType: "text" }),
-        ],
-      }),
-    )
-    expect(last.text).toBe("The answer is 42.")
-    expect(calls).toEqual(["/v1/responses/compact", "/v1/responses", "/v1/responses"])
-  }),
-)
@@ -1,64 +0,0 @@
-import { expect, test } from "bun:test"
-import { Schema } from "effect"
-import { CompactionPart, LLMEvent, LLMResponse, Message, ProviderID } from "../src/schema/index.js"
-import { LLM, LLMClient, LLMRequest, LanguageModel } from "../src/index.js"
-import { OpenAI, Anthropic } from "../src/providers.js"
-
-test("runtime capability checks follow model and route updates", () => {
-  const supported = OpenAI.configure({ apiKey: "test" }).responses("fixture")
-  const unsupported = Anthropic.configure({ apiKey: "test" }).model("fixture")
-  const request = LLM.request({ model: supported, prompt: "hello" })
-  expect(LLMClient.canCompact(request)).toBe(true)
-  expect(LLMClient.canCompact(LLMRequest.update(request, { messages: [] }))).toBe(true)
-  expect(LLMClient.canCompact(LLMRequest.update(request, { model: unsupported }))).toBe(false)
-  expect(
-    LLMClient.canCompact(LLM.request({ model: LanguageModel.update(supported, { route: unsupported.route }) })),
-  ).toBe(false)
-  expect(LLMClient.canCompact(LLM.request({ model: LanguageModel.update(supported, { route: undefined }) }))).toBe(true)
-})
-
-test("compaction survives event assembly and message serialization without becoming text", () => {
-  const part = CompactionPart.make({
-    provider: ProviderID.make("openai"),
-    id: "cmp_1",
-    encrypted: "opaque",
-  })
-  const response = LLMResponse.fromEvents([
-    LLMEvent.textStart({ id: "before" }),
-    LLMEvent.textDelta({ id: "before", text: "Before" }),
-    LLMEvent.textEnd({ id: "before" }),
-    part,
-    LLMEvent.textStart({ id: "after" }),
-    LLMEvent.textDelta({ id: "after", text: "After" }),
-    LLMEvent.textEnd({ id: "after" }),
-    LLMEvent.finish({ reason: { normalized: "stop" } }),
-  ])!
-  expect(response.message.content.map((part) => part.type)).toEqual(["text", "compaction", "text"])
-  expect(response.text).toBe("BeforeAfter")
-  expect(response.reasoning).toBe("")
-  expect(response.events.filter(LLMEvent.is.compaction)).toEqual([part])
-  const codec = Schema.fromJsonString(Message)
-  expect(Schema.decodeSync(codec)(Schema.encodeSync(codec)(response.message))).toEqual(response.message)
-})
-
-test("compaction requires exactly one typed representation", () => {
-  const provider = ProviderID.make("anthropic")
-  expect(CompactionPart.make({ provider, text: null })).toEqual({ type: "compaction", provider, text: null })
-  const decode = Schema.decodeUnknownSync(CompactionPart)
-  expect(() => decode({ type: "compaction", provider })).toThrow()
-  expect(() => decode({ type: "compaction", provider, text: "summary", encrypted: "opaque" })).toThrow()
-})
-
-test("tagged content and event guards accept both checkpoint representations", () => {
-  for (const part of [
-    CompactionPart.make({ provider: ProviderID.make("openai"), encrypted: "opaque" }),
-    CompactionPart.make({ provider: ProviderID.make("anthropic"), text: "summary" }),
-    CompactionPart.make({ provider: ProviderID.make("anthropic"), text: null }),
-  ]) {
-    expect(LLMEvent.is.compaction(part)).toBe(true)
-    expect(LLMEvent.guards.compaction(part)).toBe(true)
-    const codec = Schema.fromJsonString(Message)
-    const message = Message.assistant(part)
-    expect(Schema.decodeSync(codec)(Schema.encodeSync(codec)(message))).toEqual(message)
-  }
-})
@@ -1,154 +0,0 @@
-import { Effect } from "effect"
-import {
-  CompactionPart,
-  LanguageModel,
-  LLM,
-  LLMClient,
-  LLMEvent,
-  LLMRequest,
-  Message,
-  ProviderID,
-} from "../../src/index.js"
-import {
-  OpenAI,
-  Azure,
-  XAI,
-  Anthropic,
-  AmazonBedrock,
-  AmazonBedrockMantle,
-  OpenAICompatibleResponses,
-} from "../../src/providers.js"
-
-const openai = OpenAI.configure({
-  apiKey: "test",
-  providerOptions: { contextManagement: [{ type: "compaction", compactThreshold: 100000 }] },
-}).responses("gpt-5.3-codex")
-LLMClient.compact(LLM.request({ model: openai, prompt: "hello" }))
-
-for (const model of [
-  OpenAI.configure().responses("fixture"),
-  Azure.configure({ resourceName: "test" }).responses("fixture"),
-  XAI.configure().responses("fixture"),
-  OpenAI.model("fixture", {}),
-  Azure.responsesModel("fixture", { resourceName: "test" }),
-  XAI.model("fixture", {}),
-  openai.route.with({ headers: { "x-test": "test" } }).model({ id: "fixture" }),
-  LanguageModel.update(openai, { defaults: { generation: { maxTokens: 100 } } }),
-  LanguageModel.make(LanguageModel.input(openai)),
-]) {
-  LLMClient.compact(LLM.request({ model, prompt: "hello" }))
-}
-
-const unsupported = {
-  anthropic: LLM.request({ model: Anthropic.configure().model("fixture") }),
-  openaiChat: LLM.request({ model: OpenAI.configure().chat("fixture") }),
-  azureChat: LLM.request({ model: Azure.configure({ resourceName: "test" }).chat("fixture") }),
-  xaiChat: LLM.request({ model: XAI.configure().chat("fixture") }),
-  converse: LLM.request({ model: AmazonBedrock.configure().model("fixture") }),
-  bedrockMessages: LLM.request({ model: AmazonBedrock.configure().messages("fixture") }),
-  mantle: LLM.request({ model: AmazonBedrockMantle.configure().responses("fixture") }),
-  compatible: LLM.request({
-    model: OpenAICompatibleResponses.configure({ baseURL: "https://example.com" }).model("fixture"),
-  }),
-}
-// @ts-expect-error Anthropic has no standalone compact endpoint.
-LLMClient.compact(unsupported.anthropic)
-// @ts-expect-error Chat does not expose Responses compaction.
-LLMClient.compact(unsupported.openaiChat)
-// @ts-expect-error Azure Chat does not expose Responses compaction.
-LLMClient.compact(unsupported.azureChat)
-// @ts-expect-error xAI Chat does not expose Responses compaction.
-LLMClient.compact(unsupported.xaiChat)
-// @ts-expect-error Converse has no standalone compact endpoint.
-LLMClient.compact(unsupported.converse)
-// @ts-expect-error Bedrock Messages has no standalone compact endpoint.
-LLMClient.compact(unsupported.bedrockMessages)
-// @ts-expect-error Mantle does not inherit the OpenAI compact endpoint.
-LLMClient.compact(unsupported.mantle)
-// @ts-expect-error Protocol compatibility does not guarantee endpoint support.
-LLMClient.compact(unsupported.compatible)
-LLMClient.Service.use((client) => {
-  // @ts-expect-error The service enforces the same capability as the convenience function.
-  return client.compact(unsupported.anthropic)
-})
-
-const request = LLM.request({ model: openai, prompt: "hello" })
-LLMClient.compact(LLMRequest.update(request, { messages: [Message.user("continue")] }))
-LLMClient.compact(new LLMRequest(LLMRequest.input(request)))
-const switched = LLMRequest.update(request, { model: Anthropic.configure().model("fixture") })
-// @ts-expect-error Switching models replaces, rather than inherits, the capability.
-LLMClient.compact(switched)
-LLMClient.compact(LLMRequest.update(switched, { model: openai }))
-LLMClient.compact(
-  // @ts-expect-error Replacing the route also replaces compaction capability.
-  LLM.request({ model: LanguageModel.update(openai, { route: Anthropic.configure().model("fixture").route }) }),
-)
-
-declare const dynamicModel: LanguageModel
-declare const dynamicPatch: Partial<LLMRequest.Input>
-const dynamicRequest = LLM.request({ model: dynamicModel, prompt: "hello" })
-// @ts-expect-error A dynamically selected model must be narrowed first.
-LLMClient.compact(dynamicRequest)
-if (LLMClient.canCompact(dynamicRequest)) LLMClient.compact(dynamicRequest)
-// @ts-expect-error An optional model override cannot retain the old capability statically.
-LLMClient.compact(LLMRequest.update(request, dynamicPatch))
-
-const checkpoint = CompactionPart.make({ provider: ProviderID.make("openai"), id: "cmp_1", encrypted: "opaque" })
-const provider = ProviderID.make("anthropic")
-CompactionPart.make({ provider, text: "summary" })
-CompactionPart.make({ provider, text: null })
-// @ts-expect-error A checkpoint must have a representation.
-CompactionPart.make({ provider })
-// @ts-expect-error Encrypted and summary representations are mutually exclusive.
-CompactionPart.make({ provider, encrypted: "opaque", text: "summary" })
-// @ts-expect-error A failed summary cannot also carry encrypted content.
-LLMEvent.compaction({ provider, encrypted: "opaque", text: null })
-// @ts-expect-error The canonical message type also enforces the invariant.
-Message.assistant({ type: "compaction", provider })
-if (checkpoint.encrypted !== undefined) {
-  checkpoint.encrypted satisfies string
-  checkpoint.text satisfies undefined
-}
-if (checkpoint.text !== undefined) {
-  checkpoint.text satisfies string | null
-  checkpoint.encrypted satisfies undefined
-}
-checkpoint.encrypted
-// @ts-expect-error Compaction parts do not contain a generic provider payload.
-checkpoint.value
-LLMClient.compact(LLM.request({ model: openai, prompt: "hello" })).pipe(
-  Effect.map((result) => {
-    result.messages
-    // @ts-expect-error Compaction returns replacement history, not a synthetic assistant message.
-    result.message
-  }),
-)
-LLM.request({
-  model: openai,
-  providerOptions: {
-    // @ts-expect-error A token threshold is numeric.
-    contextManagement: [{ type: "compaction", compactThreshold: "100000" }],
-  },
-})
-for (const model of [
-  Anthropic.configure().model("claude-opus-4-6"),
-  AmazonBedrock.configure().messages("anthropic.claude-opus-4-6-v1"),
-]) {
-  LLM.request({
-    model,
-    providerOptions: {
-      contextManagement: {
-        edits: [
-          { type: "compact_20260112", pauseAfterCompaction: true, instructions: "Summarize without using tools" },
-        ],
-      },
-    },
-  })
-  LLM.request({
-    model,
-    providerOptions: {
-      // @ts-expect-error A pause setting is boolean.
-      contextManagement: { edits: [{ type: "compact_20260112", pauseAfterCompaction: "yes" }] },
-    },
-  })
-}
@@ -1,170 +0,0 @@
-import { expect } from "bun:test"
-import { Effect, Schema } from "effect"
-import { LLM, LLMRequest, Message } from "../../src/index.js"
-import { LLMClient } from "../../src/route/client.js"
-import { Anthropic, GoogleVertexMessages } from "../../src/providers/index.js"
-import { testEffect } from "../lib/effect.js"
-import { dynamicResponse, fixedResponse } from "../lib/http.js"
-import { sseEvents } from "../lib/sse.js"
-
-for (const fixture of [
-  {
-    name: "empty iterations fall back to top-level usage",
-    usage: { input_tokens: 2, output_tokens: 3, cache_read_input_tokens: null, iterations: [] },
-    expected: { inputTokens: 2, outputTokens: 3, totalTokens: 5, contextTokens: undefined },
-  },
-  {
-    name: "compaction-only usage has no post-compaction context size",
-    usage: {
-      input_tokens: 0,
-      output_tokens: 0,
-      iterations: [{ type: "compaction", input_tokens: 7, cache_read_input_tokens: 3, output_tokens: 2 }],
-    },
-    expected: { inputTokens: 10, outputTokens: 2, totalTokens: 12, contextTokens: undefined },
-  },
-  {
-    name: "partially reported iterations preserve known totals",
-    usage: {
-      iterations: [
-        { type: "compaction", input_tokens: 7, cache_creation_input_tokens: 2 },
-        { type: "message", output_tokens: 3 },
-      ],
-    },
-    expected: { inputTokens: 9, outputTokens: 3, totalTokens: 12, contextTokens: undefined },
-  },
-  {
-    name: "missing counters remain unknown rather than zero",
-    usage: { iterations: [{ type: "message" }] },
-    expected: { inputTokens: undefined, outputTokens: undefined, totalTokens: undefined, contextTokens: undefined },
-  },
-]) {
-  testEffect(
-    fixedResponse(
-      sseEvents(
-        { type: "message_start", message: { usage: fixture.usage } },
-        { type: "message_delta", delta: { stop_reason: "end_turn" } },
-        { type: "message_stop" },
-      ),
-    ),
-  ).effect(fixture.name, () =>
-    Effect.gen(function* () {
-      const result = yield* LLMClient.generate(
-        LLM.request({
-          model: Anthropic.configure({ apiKey: "test" }).model("claude-opus-4-6"),
-          prompt: "hello",
-        }),
-      )
-      expect(result.usage).toMatchObject(fixture.expected)
-    }),
-  )
-}
-
-for (const model of [
-  Anthropic.configure({ apiKey: "test" }).model("claude-opus-4-6"),
-  GoogleVertexMessages.configure({ accessToken: "test", project: "test" }).model("claude-opus-4-6"),
-]) {
-  for (const summary of ["Summary of the conversation", null]) {
-    const block = { type: "compaction", content: summary }
-    testEffect(
-      dynamicResponse(({ request, text, respond }) =>
-        Effect.sync(() => {
-          const body = JSON.parse(text)
-          expect(request.headers["anthropic-beta"]).toBe("existing-beta,compact-2026-01-12")
-          if (body.messages.length === 1) {
-            expect(body.context_management.edits).toEqual([
-              {
-                type: "compact_20260112",
-                trigger: { type: "input_tokens", value: 50000 },
-                pause_after_compaction: true,
-              },
-            ])
-          }
-          if (body.messages.length > 1) {
-            expect(body.messages[1].content).toEqual([block])
-            expect(body.context_management).toBeUndefined()
-          }
-          return respond(
-            sseEvents(
-              { type: "message_start", message: { usage: { input_tokens: 50000, output_tokens: 0 } } },
-              { type: "content_block_start", index: 0, content_block: { type: "compaction", content: null } },
-              { type: "content_block_delta", index: 0, delta: { type: "compaction_delta", content: summary } },
-              { type: "content_block_stop", index: 0 },
-              {
-                type: "message_delta",
-                delta: { stop_reason: "compaction" },
-                usage: {
-                  input_tokens: 1000,
-                  output_tokens: 5,
-                  iterations: [
-                    { type: "compaction", input_tokens: 50000, output_tokens: 1000, cache_read_input_tokens: 10 },
-                    { type: "message", input_tokens: 1000, output_tokens: 5 },
-                  ],
-                },
-              },
-              { type: "message_stop" },
-            ),
-            { headers: { "content-type": "text/event-stream" } },
-          )
-        }),
-      ),
-    ).effect(
-      `${model.provider} replays ${summary === null ? "failed" : "successful"} compaction with billing and context usage`,
-      () =>
-        Effect.gen(function* () {
-          const request = LLM.request({
-            model,
-            prompt: "hello",
-            http: { headers: { "anthropic-beta": "existing-beta" } },
-            providerOptions: {
-              contextManagement: {
-                edits: [
-                  {
-                    type: "compact_20260112",
-                    trigger: { type: "input_tokens", value: 50000 },
-                    pauseAfterCompaction: true,
-                  },
-                ],
-              },
-            },
-          })
-          const first = yield* LLMClient.generate(request)
-          expect(first.finishReason.raw).toBe("compaction")
-          expect(first.message.content).toEqual([{ type: "compaction", provider: model.provider, text: summary }])
-          expect(first.text).toBe("")
-          expect(first.usage?.inputTokens).toBe(51010)
-          expect(first.usage?.outputTokens).toBe(1005)
-          expect(first.usage?.totalTokens).toBe(52015)
-          expect(first.usage?.contextTokens).toBe(1000)
-          const codec = Schema.fromJsonString(Message)
-          const message = Schema.decodeSync(codec)(Schema.encodeSync(codec)(first.message))
-          yield* LLMClient.generate(
-            LLMRequest.update(request, {
-              providerOptions: {},
-              messages: [...request.messages, message, Message.user("continue")],
-            }),
-          )
-        }),
-    )
-  }
-}
-
-for (const events of [
-  [{ type: "content_block_start", index: 0, content_block: { type: "compaction", content: 42 } }],
-  [{ type: "content_block_delta", index: 0, delta: { type: "compaction_delta", content: "no start" } }],
-  [
-    { type: "content_block_start", index: 0, content_block: { type: "compaction", content: null } },
-    { type: "message_stop" },
-  ],
-]) {
-  testEffect(fixedResponse(sseEvents(...events))).effect(
-    `rejects malformed compaction lifecycle: ${JSON.stringify(events)}`,
-    () =>
-      Effect.gen(function* () {
-        const error = yield* LLMClient.generate(
-          LLM.request({ model: Anthropic.configure({ apiKey: "test" }).model("claude-opus-4-6"), prompt: "hello" }),
-        ).pipe(Effect.flip)
-        expect(error.reason._tag).toBe("InvalidProviderOutput")
-        expect(error.reason.http?.status).toBe(200)
-      }),
-  )
-}
@@ -1,69 +0,0 @@
-import { EventStreamCodec } from "@smithy/eventstream-codec"
-import { fromUtf8, toUtf8 } from "@smithy/util-utf8"
-import { expect } from "bun:test"
-import { Effect } from "effect"
-import { LLM, LLMRequest, Message } from "../../src/index.js"
-import { LLMClient } from "../../src/route/client.js"
-import { AmazonBedrock } from "../../src/providers/index.js"
-import { testEffect } from "../lib/effect.js"
-import { dynamicResponse } from "../lib/http.js"
-
-const codec = new EventStreamCodec(toUtf8, fromUtf8)
-const frame = (event: object) =>
-  codec.encode({
-    headers: { ":message-type": { type: "string", value: "event" }, ":event-type": { type: "string", value: "chunk" } },
-    body: new TextEncoder().encode(JSON.stringify({ bytes: Buffer.from(JSON.stringify(event)).toString("base64") })),
-  })
-const response = Buffer.concat(
-  [
-    { type: "message_start", message: { usage: { input_tokens: 60000 } } },
-    { type: "content_block_start", index: 0, content_block: { type: "compaction", content: null } },
-    { type: "content_block_delta", index: 0, delta: { type: "compaction_delta", content: "Summary" } },
-    { type: "content_block_stop", index: 0 },
-    { type: "content_block_start", index: 1, content_block: { type: "text", text: "" } },
-    { type: "content_block_delta", index: 1, delta: { type: "text_delta", text: "Hello" } },
-    { type: "content_block_stop", index: 1 },
-    { type: "message_delta", delta: { stop_reason: "end_turn" }, usage: { output_tokens: 10 } },
-    { type: "message_stop" },
-  ].map(frame),
-)
-
-for (const auth of [
-  { apiKey: "test" },
-  { credentials: { accessKeyId: "test", secretAccessKey: "test", region: "us-west-2" } },
-]) {
-  testEffect(
-    dynamicResponse(({ request, text, respond }) =>
-      Effect.sync(() => {
-        expect(request.url).toBe(
-          "https://bedrock-runtime.us-west-2.amazonaws.com/model/anthropic.claude-opus-4-6-v1%3A0/invoke-with-response-stream",
-        )
-        expect(request.headers.authorization).toStartWith(auth.apiKey ? "Bearer test" : "AWS4-HMAC-SHA256")
-        const body = JSON.parse(text)
-        expect(body.model).toBeUndefined()
-        expect(body.stream).toBeUndefined()
-        expect(body.anthropic_version).toBe("bedrock-2023-05-31")
-        expect(body.anthropic_beta).toEqual(["compact-2026-01-12"])
-        expect(body.context_management.edits).toEqual([{ type: "compact_20260112" }])
-        if (body.messages.length > 1)
-          expect(body.messages[1].content[0]).toEqual({ type: "compaction", content: "Summary" })
-        return respond(response, { headers: { "content-type": "application/vnd.amazon.eventstream" } })
-      }),
-    ),
-  ).effect(`Bedrock Messages compaction round trip with ${auth.apiKey ? "bearer" : "SigV4"} authentication`, () =>
-    Effect.gen(function* () {
-      const model = AmazonBedrock.configure({ ...auth, region: "us-west-2" }).messages("anthropic.claude-opus-4-6-v1:0")
-      const request = LLM.request({
-        model,
-        prompt: "hello",
-        providerOptions: { contextManagement: { edits: [{ type: "compact_20260112" }] } },
-      })
-      const first = yield* LLMClient.generate(request)
-      expect(first.text).toBe("Hello")
-      expect(first.message.content.map((part) => part.type)).toEqual(["compaction", "text"])
-      yield* LLMClient.generate(
-        LLMRequest.update(request, { messages: [...request.messages, first.message, Message.user("continue")] }),
-      )
-    }),
-  )
-}
@@ -631,10 +631,45 @@ describe("Bedrock Converse route", () => {
      )
      const response = yield* LLMClient.generate(baseRequest).pipe(Effect.provide(fixedBytes(body)))

+      expect(response.events.filter((event) => event.type === "finish")).toHaveLength(1)
      expect(response.usage).toMatchObject({ inputTokens: 5, outputTokens: 2, totalTokens: 7 })
    }),
  )

+  it.effect("retains metadata usage that arrives before messageStop", () =>
+    Effect.gen(function* () {
+      const body = eventStreamBody(
+        ["metadata", { usage: { inputTokens: 5, outputTokens: 2, totalTokens: 7 } }],
+        ["messageStop", { stopReason: "end_turn" }],
+      )
+      const response = yield* LLMClient.generate(baseRequest).pipe(Effect.provide(fixedBytes(body)))
+
+      expect(response.events.filter((event) => event.type === "finish")).toHaveLength(1)
+      expect(response.finishReason).toEqual({ normalized: "stop", raw: "end_turn" })
+      expect(response.usage).toMatchObject({ inputTokens: 5, outputTokens: 2, totalTokens: 7 })
+    }),
+  )
+
+  it.effect("rejects metadata-only streams as incomplete with HTTP context", () =>
+    Effect.gen(function* () {
+      const error = yield* LLMClient.generate(baseRequest).pipe(
+        Effect.provide(
+          fixedBytes(eventStreamBody(["metadata", { usage: { inputTokens: 5, outputTokens: 2, totalTokens: 7 } }])),
+        ),
+        Effect.flip,
+      )
+
+      expect(error.reason).toMatchObject({
+        _tag: "InvalidProviderOutput",
+        classification: "incomplete-stream",
+        http: {
+          status: 200,
+          headers: { "content-type": "application/vnd.amazon.eventstream" },
+        },
+      })
+    }),
+  )
+
  it.effect("assembles streamed tool call input", () =>
    Effect.gen(function* () {
      const body = eventStreamBody(
@@ -1,166 +0,0 @@
-import { EventStreamCodec } from "@smithy/eventstream-codec"
-import { fromUtf8, toUtf8 } from "@smithy/util-utf8"
-import { expect } from "bun:test"
-import { Effect } from "effect"
-import { LLM, LLMRequest, Message } from "../../src/index.js"
-import { LLMClient, compileRequest } from "../../src/route/client.js"
-import { AmazonBedrock } from "../../src/providers/index.js"
-import { it, testEffect } from "../lib/effect.js"
-import { dynamicResponse, fixedResponse } from "../lib/http.js"
-
-const codec = new EventStreamCodec(toUtf8, fromUtf8)
-const response = Buffer.concat(
-  [
-    { type: "message_start", message: { usage: { input_tokens: 10 } } },
-    { type: "content_block_start", index: 0, content_block: { type: "text", text: "" } },
-    { type: "content_block_delta", index: 0, delta: { type: "text_delta", text: "Hello" } },
-    { type: "content_block_stop", index: 0 },
-    { type: "message_delta", delta: { stop_reason: "end_turn" }, usage: { output_tokens: 2 } },
-    { type: "message_stop" },
-  ].map((event) =>
-    codec.encode({
-      headers: {
-        ":message-type": { type: "string", value: "event" },
-        ":event-type": { type: "string", value: "chunk" },
-      },
-      body: new TextEncoder().encode(JSON.stringify({ bytes: Buffer.from(JSON.stringify(event)).toString("base64") })),
-    }),
-  ),
-)
-
-for (const auth of [
-  { apiKey: "test" },
-  { credentials: { accessKeyId: "test", secretAccessKey: "test", region: "us-west-2" } },
-]) {
-  testEffect(
-    dynamicResponse(({ request, text, respond }) =>
-      Effect.sync(() => {
-        expect(request.url).toBe(
-          "https://bedrock-runtime.us-west-2.amazonaws.com/model/anthropic.claude-opus-4-6-v1%3A0/invoke-with-response-stream",
-        )
-        expect(request.headers.authorization).toStartWith(auth.apiKey ? "Bearer test" : "AWS4-HMAC-SHA256")
-        const body = JSON.parse(text)
-        expect(body.model).toBeUndefined()
-        expect(body.stream).toBeUndefined()
-        expect(body.anthropic_version).toBe("bedrock-2023-05-31")
-        expect(body.anthropic_beta).toEqual(["existing-beta"])
-        if (body.messages.length > 1) expect(body.messages[1].content).toEqual([{ type: "text", text: "Hello" }])
-        return respond(response, { headers: { "content-type": "application/vnd.amazon.eventstream" } })
-      }),
-    ),
-  ).effect(`Bedrock Messages text round trip with ${auth.apiKey ? "bearer" : "SigV4"} authentication`, () =>
-    Effect.gen(function* () {
-      const provider = AmazonBedrock.configure({ ...auth, region: "us-west-2" })
-      expect(provider.model("fixture").route.id).toBe("bedrock-converse")
-      const request = LLM.request({
-        model: provider.messages("anthropic.claude-opus-4-6-v1:0"),
-        prompt: "hello",
-        http: { headers: { "anthropic-beta": "existing-beta, existing-beta" } },
-      })
-      const first = yield* LLMClient.generate(request)
-      expect(first.text).toBe("Hello")
-      expect(first.usage?.totalTokens).toBe(12)
-      yield* LLMClient.generate(
-        LLMRequest.update(request, {
-          messages: [...request.messages, first.message, Message.user("continue")],
-        }),
-      )
-    }),
-  )
-}
-
-testEffect(
-  fixedResponse(
-    codec.encode({
-      headers: {
-        ":message-type": { type: "string", value: "exception" },
-        ":exception-type": { type: "string", value: "throttlingException" },
-      },
-      body: new TextEncoder().encode(JSON.stringify({ message: "Too many requests", trace: "keep-original" })),
-    }),
-  ),
-).effect("Bedrock Messages retains the original exception frame", () =>
-  Effect.gen(function* () {
-    const error = yield* LLMClient.generate(
-      LLM.request({
-        model: AmazonBedrock.configure({ apiKey: "test" }).messages("claude"),
-        prompt: "hello",
-      }),
-    ).pipe(Effect.flip)
-    expect(error.reason.body).toContain("keep-original")
-    expect(error.reason.http?.status).toBe(200)
-  }),
-)
-
-for (const mediaType of ["image/png", "application/pdf"]) {
-  for (const data of ["https://example.com/media", "invalid base64!"]) {
-    for (const role of ["user", "tool"] as const) {
-      it.effect(`rejects ${role} ${mediaType} with ${data}`, () =>
-        Effect.gen(function* () {
-          const error = yield* compileRequest(
-            LLM.request({
-              model: AmazonBedrock.configure({ apiKey: "test" }).messages("claude"),
-              messages:
-                role === "user"
-                  ? [Message.user({ type: "media", mediaType, data })]
-                  : [
-                      Message.tool({
-                        id: "call_1",
-                        name: "read",
-                        result: { type: "content", value: [{ type: "file", mime: mediaType, uri: data }] },
-                      }),
-                    ],
-            }),
-          ).pipe(Effect.flip)
-          expect(error.reason._tag).toBe("InvalidRequest")
-          expect(error.message).toContain("Bedrock Messages")
-        }),
-      )
-    }
-  }
-}
-
-it.effect("accepts inline image and document sources", () =>
-  Effect.gen(function* () {
-    const prepared = yield* compileRequest(
-      LLM.request({
-        model: AmazonBedrock.configure({ apiKey: "test" }).messages("claude"),
-        messages: [
-          Message.user([
-            { type: "media", mediaType: "image/png", data: "data:image/png;base64,AQID" },
-            { type: "media", mediaType: "application/pdf", data: "data:application/pdf;base64,AQID" },
-          ]),
-        ],
-      }),
-    )
-    expect(prepared.body.messages[0].content.map((block: { source: { type: string } }) => block.source.type)).toEqual([
-      "base64",
-      "base64",
-    ])
-  }),
-)
-
-it.effect("rejects Anthropic file IDs", () =>
-  Effect.gen(function* () {
-    const error = yield* compileRequest(
-      LLM.request({
-        model: AmazonBedrock.configure({ apiKey: "test" }).messages("claude"),
-        messages: [Message.user({ type: "media", mediaType: "image/png", data: "", metadata: { file_id: "file_1" } })],
-      }),
-    ).pipe(Effect.flip)
-    expect(error.message).toContain("file-ID")
-  }),
-)
-
-it.effect("rejects unsupported image formats", () =>
-  Effect.gen(function* () {
-    const error = yield* compileRequest(
-      LLM.request({
-        model: AmazonBedrock.configure({ apiKey: "test" }).messages("claude"),
-        messages: [Message.user({ type: "media", mediaType: "image/svg+xml", data: "AQID" })],
-      }),
-    ).pipe(Effect.flip)
-    expect(error.reason._tag).toBe("InvalidRequest")
-    expect(error.message).toContain("JPEG, PNG, WebP, or GIF")
-  }),
-)
@@ -1,50 +0,0 @@
-import { expect } from "bun:test"
-import { Effect, Stream } from "effect"
-import { LLM, LLMRequest, Message } from "../../src/index.js"
-import { LLMClient, WebSocketTransport } from "../../src/route.js"
-import { OpenAI } from "../../src/providers.js"
-import { testEffect } from "../lib/effect.js"
-import { fixedResponse } from "../lib/http.js"
-
-testEffect(fixedResponse("unexpected HTTP fallback")).effect(
-  "WebSocket responses preserve compaction options and replay state",
-  () =>
-    Effect.gen(function* () {
-      const checkpoint = { type: "compaction", id: "cmp_ws", encrypted_content: "opaque" }
-      const sent: unknown[] = []
-      const webSocket = WebSocketTransport.makeDirect({
-        open: () =>
-          Effect.succeed({
-            sendText: (message) =>
-              Effect.sync(() => {
-                const body = JSON.parse(message)
-                expect(body.context_management).toEqual([{ type: "compaction", compact_threshold: 100000 }])
-                expect(body.stream).toBeUndefined()
-                if (sent.length) expect(body.input[1]).toEqual(checkpoint)
-                sent.push(body)
-              }),
-            messages: Stream.fromIterable(
-              [
-                { type: "response.created", response: { id: "resp_ws" } },
-                { type: "response.output_item.done", item: checkpoint },
-                { type: "response.completed", response: { id: "resp_ws", output: [checkpoint] } },
-              ].map((event) => JSON.stringify(event)),
-            ),
-            close: Effect.void,
-          }),
-      })
-      const request = LLM.request({
-        model: OpenAI.configure({ apiKey: "test" }).responses("gpt-5.3-codex"),
-        prompt: "hello",
-        providerOptions: { contextManagement: [{ type: "compaction", compactThreshold: 100000 }] },
-      })
-      const first = yield* LLMClient.generate(request, { webSocket })
-      expect(first.message.content).toHaveLength(1)
-      expect(first.message.content[0]?.type).toBe("compaction")
-      yield* LLMClient.generate(
-        LLMRequest.update(request, { messages: [...request.messages, first.message, Message.user("continue")] }),
-        { webSocket },
-      )
-      expect(sent).toHaveLength(2)
-    }),
-)
@@ -1,91 +0,0 @@
-import { expect } from "bun:test"
-import { Effect } from "effect"
-import { LLM, LLMRequest, Message } from "../../src/index.js"
-import { LLMClient } from "../../src/route/client.js"
-import { OpenAI, XAI, Anthropic } from "../../src/providers.js"
-import { recordedTests } from "../recorded-test.js"
-
-const history = [
-  Message.user("Remember the project codename COPPER-ORBIT-42."),
-  Message.assistant(
-    "The project codename is COPPER-ORBIT-42. " + "We reviewed the implementation and tests. ".repeat(1000),
-  ),
-]
-
-for (const provider of [
-  {
-    id: "openai",
-    key: "OPENAI_API_KEY",
-    model: OpenAI.configure({ apiKey: process.env.OPENAI_API_KEY ?? "fixture" }).responses("gpt-5.3-codex"),
-  },
-  {
-    id: "xai",
-    key: "XAI_API_KEY",
-    model: XAI.configure({ apiKey: process.env.XAI_API_KEY ?? "fixture" }).responses("grok-4.6"),
-  },
-]) {
-  recordedTests({ prefix: `${provider.id}-compaction`, provider: provider.id, requires: [provider.key] }).effect(
-    "compacts and continues with the provider checkpoint",
-    () =>
-      Effect.gen(function* () {
-        const request = LLM.request({ model: provider.model, messages: history, generation: { maxTokens: 1024 } })
-        const compacted = yield* LLMClient.compact(request)
-        const result = yield* LLMClient.generate(
-          LLMRequest.update(request, {
-            messages: [
-              ...compacted.messages,
-              Message.user("What is the project codename? Reply only with the codename."),
-            ],
-          }),
-        )
-        expect(result.text).toContain("COPPER-ORBIT-42")
-      }),
-    120000,
-  )
-}
-
-recordedTests({
-  prefix: "anthropic-compaction",
-  provider: "anthropic",
-  requires: ["ANTHROPIC_API_KEY"],
-  options: { redact: { allowRequestHeaders: ["anthropic-version", "anthropic-beta"] } },
-}).effect(
-  "automatically compacts and continues after a pause",
-  () =>
-    Effect.gen(function* () {
-      const model = Anthropic.configure({ apiKey: process.env.ANTHROPIC_API_KEY ?? "fixture" }).model(
-        "claude-sonnet-4-6",
-      )
-      const request = LLM.request({
-        model,
-        messages: [
-          Message.user(
-            "Remember the project codename COPPER-ORBIT-42. " +
-              "The implementation and tests were reviewed. ".repeat(10000),
-          ),
-        ],
-        generation: { maxTokens: 4096 },
-        providerOptions: {
-          contextManagement: {
-            edits: [
-              { type: "compact_20260112", trigger: { type: "input_tokens", value: 50000 }, pauseAfterCompaction: true },
-            ],
-          },
-        },
-      })
-      const first = yield* LLMClient.generate(request)
-      expect(first.finishReason.raw).toBe("compaction")
-      expect(first.message.content.some((part) => part.type === "compaction")).toBe(true)
-      const result = yield* LLMClient.generate(
-        LLMRequest.update(request, {
-          messages: [
-            ...request.messages,
-            first.message,
-            Message.user("What is the project codename? Reply only with the codename."),
-          ],
-        }),
-      )
-      expect(result.text).toContain("COPPER-ORBIT-42")
-    }),
-  120000,
-)
@@ -1,136 +0,0 @@
-import { expect } from "bun:test"
-import { Effect, Schema } from "effect"
-import { LLM, LLMRequest, Message } from "../../src/index.js"
-import { LLMClient } from "../../src/route/client.js"
-import { OpenAI, Azure, XAI } from "../../src/providers/index.js"
-import { testEffect } from "../lib/effect.js"
-import { dynamicResponse, fixedResponse } from "../lib/http.js"
-import { sseEvents } from "../lib/sse.js"
-
-const checkpoint = { type: "compaction", id: "cmp_1", encrypted_content: "opaque" }
-const response = sseEvents(
-  { type: "response.output_item.done", item: checkpoint },
-  {
-    type: "response.completed",
-    response: { id: "resp_1", output: [checkpoint], usage: { input_tokens: 10, output_tokens: 2, total_tokens: 12 } },
-  },
-)
-
-for (const model of [
-  OpenAI.configure({ apiKey: "test" }).responses("gpt-5.3-codex"),
-  Azure.configure({ apiKey: "test", resourceName: "test" }).responses("deployment"),
-]) {
-  testEffect(
-    dynamicResponse(({ text, respond }) =>
-      Effect.sync(() => {
-        const body = JSON.parse(text)
-        expect(body.context_management).toEqual([{ type: "compaction", compact_threshold: 100000 }])
-        expect(body.store).toBe(false)
-        if (body.input.length > 1) expect(body.input[1]).toEqual(checkpoint)
-        return respond(response, { headers: { "content-type": "text/event-stream" } })
-      }),
-    ),
-  ).effect(`${model.provider} compaction survives generation, serialization, and a second request`, () =>
-    Effect.gen(function* () {
-      const request = LLM.request({
-        model,
-        prompt: "hello",
-        providerOptions: { contextManagement: [{ type: "compaction", compactThreshold: 100000 }] },
-      })
-      const first = yield* LLMClient.generate(request)
-      expect(first.message.content).toHaveLength(1)
-      expect(first.message.content[0]?.type).toBe("compaction")
-      expect(first.text).toBe("")
-      const codec = Schema.fromJsonString(Message)
-      const message = Schema.decodeSync(codec)(Schema.encodeSync(codec)(first.message))
-      yield* LLMClient.generate(
-        LLMRequest.update(request, { messages: [...request.messages, message, Message.user("continue")] }),
-      )
-      const rejected = yield* LLMClient.generate(
-        LLMRequest.update(request, {
-          model: XAI.configure({ apiKey: "test" }).responses("grok-4.6"),
-          providerOptions: {},
-          messages: [message],
-        }),
-      ).pipe(Effect.flip)
-      expect(rejected.reason._tag).toBe("InvalidRequest")
-    }),
-  )
-}
-
-testEffect(fixedResponse(sseEvents({ type: "response.completed", response: { output: [checkpoint] } }))).effect(
-  "recovers compaction from the terminal output when item completion is absent",
-  () =>
-    Effect.gen(function* () {
-      const result = yield* LLMClient.generate(
-        LLM.request({ model: OpenAI.configure({ apiKey: "test" }).responses("gpt-5.3-codex"), prompt: "hello" }),
-      )
-      expect(result.message.content).toHaveLength(1)
-      expect(result.message.content[0]?.type).toBe("compaction")
-    }),
-)
-
-testEffect(
-  fixedResponse(sseEvents({ type: "response.output_item.done", item: { type: "compaction", id: "cmp_bad" } })),
-).effect("rejects incomplete compaction payloads without publishing a checkpoint", () =>
-  Effect.gen(function* () {
-    const error = yield* LLMClient.generate(
-      LLM.request({ model: OpenAI.configure({ apiKey: "test" }).responses("gpt-5.3-codex"), prompt: "hello" }),
-    ).pipe(Effect.flip)
-    expect(error.reason._tag).toBe("InvalidProviderOutput")
-    expect(error.reason.body).toContain("cmp_bad")
-  }),
-)
-
-const textItem = {
-  type: "message",
-  id: "msg_after",
-  role: "assistant",
-  content: [{ type: "output_text", text: "After checkpoint" }],
-}
-
-for (const completed of [false, true]) {
-  testEffect(
-    fixedResponse(
-      sseEvents(
-        { type: "response.output_item.added", output_index: 0, item: { type: "compaction", id: checkpoint.id } },
-        ...(completed ? [{ type: "response.output_item.done", output_index: 0, item: checkpoint }] : []),
-        { type: "response.output_item.added", output_index: 1, item: textItem },
-        { type: "response.output_text.delta", output_index: 1, item_id: textItem.id, delta: "After checkpoint" },
-        { type: "response.output_item.done", output_index: 1, item: textItem },
-        { type: "response.completed", response: { id: "resp_1", output: [checkpoint, textItem] } },
-      ),
-    ),
-  ).effect(completed ? "keeps streamed checkpoints before later text" : "rejects order-unsafe terminal recovery", () =>
-    Effect.gen(function* () {
-      const request = LLM.request({ model: OpenAI.configure({ apiKey: "test" }).responses("fixture"), prompt: "hello" })
-      if (completed) {
-        const response = yield* LLMClient.generate(request)
-        expect(response.message.content.map((part) => part.type)).toEqual(["compaction", "text"])
-        return
-      }
-      const error = yield* LLMClient.generate(request).pipe(Effect.flip)
-      expect(error.reason._tag).toBe("InvalidProviderOutput")
-      expect(error.message).toContain("Cannot recover a compaction checkpoint")
-      expect(error.reason.body).toContain("response.completed")
-      expect(error.reason.http?.status).toBe(200)
-    }),
-  )
-}
-
-testEffect(
-  fixedResponse(
-    sseEvents({
-      type: "response.completed",
-      response: { output: [{ type: "compaction", encrypted_content: "opaque" }] },
-    }),
-  ),
-).effect("rejects terminal checkpoints missing an id", () =>
-  Effect.gen(function* () {
-    const error = yield* LLMClient.generate(
-      LLM.request({ model: OpenAI.configure({ apiKey: "test" }).responses("fixture"), prompt: "hello" }),
-    ).pipe(Effect.flip)
-    expect(error.reason._tag).toBe("InvalidProviderOutput")
-    expect(error.message).toContain("missing its id")
-  }),
-)
@@ -1,30 +0,0 @@
-import { expect } from "bun:test"
-import { Effect } from "effect"
-import { LLM, Message } from "../../src/index.js"
-import { OpenAI } from "../../src/providers.js"
-import { OpenResponses } from "../../src/protocols/open-responses.js"
-import { it } from "../lib/effect.js"
-
-it.effect("conversation lowering excludes generation settings and tool definitions", () =>
-  Effect.gen(function* () {
-    const body = yield* OpenResponses.lowerConversation(
-      LLM.request({
-        model: OpenAI.configure({ apiKey: "test" }).responses("fixture"),
-        system: "Keep the context",
-        messages: [Message.user("hello"), Message.assistant("hi")],
-        generation: { maxTokens: 100, temperature: 0.5 },
-        providerOptions: { store: false },
-        tools: [{ name: "unsupported", description: "Generation only", inputSchema: {}, native: { unsupported: {} } }],
-      }),
-      { id: "open-responses", name: "Open Responses" },
-    )
-    expect(body).toEqual({
-      model: "fixture",
-      instructions: "Keep the context",
-      input: [
-        { role: "user", content: [{ type: "input_text", text: "hello" }] },
-        { type: "message", role: "assistant", content: [{ type: "output_text", text: "hi" }] },
-      ],
-    })
-  }),
-)
@@ -62,6 +62,62 @@ describe("provider error retention", () => {
    )
  }

+  it.effect("classifies a message-less Gemini 429 and retains its event and HTTP context", () =>
+    Effect.gen(function* () {
+      const body = JSON.stringify({
+        error: { code: 429, status: "RESOURCE_EXHAUSTED", details: { opaque: [1, 2] } },
+        trace: { opaque: "outer" },
+      })
+      const error = yield* LLMClient.generate(
+        LLM.request({ model: Google.configure(options).model("gemini"), prompt: "hello" }),
+      ).pipe(
+        Effect.provide(
+          fixedResponse(sseEvents(body), {
+            headers: { "content-type": "text/event-stream", "x-provider-trace": "trace-1" },
+          }),
+        ),
+        Effect.flip,
+      )
+
+      expect(error.message).toBe("RESOURCE_EXHAUSTED")
+      expect(error.reason._tag).toBe("RateLimit")
+      expect(error.reason.body).toBe(body)
+      expect(error.reason.http).toMatchObject({ status: 200, headers: { "x-provider-trace": "trace-1" } })
+      expect(error.reason.http?.url).toStartWith("https://provider.test/")
+    }),
+  )
+
+  it.effect("rejects a malformed non-record Gemini error", () =>
+    Effect.gen(function* () {
+      const body = JSON.stringify({ error: "RESOURCE_EXHAUSTED", trace: { opaque: "outer" } })
+      const error = yield* LLMClient.generate(
+        LLM.request({ model: Google.configure(options).model("gemini"), prompt: "hello" }),
+      ).pipe(Effect.provide(fixedResponse(sseEvents(body))), Effect.flip)
+
+      expect(error.reason._tag).toBe("InvalidProviderOutput")
+      expect(error.message).toContain("Invalid google/gemini stream event")
+      expect(error.reason.body).toBe(body)
+      expect(error.reason.http?.status).toBe(200)
+    }),
+  )
+
+  it.effect("rejects and retains an explicit null Gemini error", () =>
+    Effect.gen(function* () {
+      const body = JSON.stringify({ error: null, trace: { opaque: "outer" } })
+      const error = yield* LLMClient.generate(
+        LLM.request({ model: Google.configure(options).model("gemini"), prompt: "hello" }),
+      ).pipe(
+        Effect.provide(fixedResponse(sseEvents(body), { headers: { "x-provider-trace": "trace-null" } })),
+        Effect.flip,
+      )
+
+      expect(error.reason._tag).toBe("InvalidProviderOutput")
+      expect(error.reason.body).toBe(body)
+      expect(error.reason.http).toMatchObject({ status: 200, headers: { "x-provider-trace": "trace-null" } })
+      expect(error.reason.http?.url).toStartWith("https://provider.test/")
+    }),
+  )
+
  it.effect("retains malformed provider frames and the original decode cause", () =>
    Effect.gen(function* () {
      const body = '{"type":"error","error":{"message":42,"opaque":{"nested":true}},"trace":"outer"}'
@@ -1,447 +0,0 @@
-import { expect } from "bun:test"
-import { Effect, Schema } from "effect"
-import { LLM, LLMRequest, Message } from "../../src/index.js"
-import { LLMClient, Route } from "../../src/route/client.js"
-import { Auth } from "../../src/route/auth.js"
-import { Endpoint } from "../../src/route/endpoint.js"
-import { OpenAIResponses } from "../../src/protocols/openai-responses.js"
-import { OpenAI, Azure, XAI, Anthropic, AmazonBedrockMantle } from "../../src/providers/index.js"
-import { testEffect } from "../lib/effect.js"
-import { dynamicResponse, fixedResponse } from "../lib/http.js"
-import { sseEvents } from "../lib/sse.js"
-
-const checkpoint = { type: "compaction", id: "cmp_1", encrypted_content: "opaque" }
-const retained = {
-  type: "message",
-  role: "user",
-  id: "msg_1",
-  status: "completed",
-  content: [{ type: "input_text", text: "retained" }],
-}
-const output = [retained, checkpoint]
-
-testEffect(
-  dynamicResponse(({ request, text, respond }) =>
-    Effect.sync(() => {
-      expect(request.headers["x-deployment"]).toBe("fixture")
-      expect(request.headers["x-override"]).toBe("request")
-      expect(request.headers["x-default"]).toBe("configured")
-      expect(request.headers.authorization).toBe("Bearer test")
-      expect(new URL(request.url).searchParams.get("api-version")).toBe("fixture")
-      expect(new URL(request.url).searchParams.get("trace")).toBe("request")
-      if (new URL(request.url).pathname.endsWith("/compact")) {
-        expect(JSON.parse(text)).toEqual({
-          model: "overlaid",
-          input: [{ role: "user", content: [{ type: "input_text", text: "hello" }] }],
-          instructions: "request instructions",
-          previous_response_id: "resp_previous",
-        })
-        return respond(JSON.stringify({ object: "response.compaction", output }))
-      }
-      return respond(sseEvents({ type: "response.completed", response: { id: "resp_1" } }))
-    }),
-  ),
-).effect("generation and compaction share deployment headers, defaults, auth, query, and middleware", () =>
-  Effect.gen(function* () {
-    const headers: string[] = []
-    const middleware: string[] = []
-    const route = Route.make({
-      id: "compaction-headers",
-      provider: "openai",
-      protocol: OpenAIResponses.protocol,
-      compact: OpenAIResponses.route.compact,
-      transport: OpenAIResponses.httpTransport,
-      endpoint: Endpoint.path(({ body }) => `/${body.model}/responses`, {
-        baseURL: "https://example.com",
-        query: { "api-version": "fixture" },
-      }),
-      auth: Auth.bearer("test"),
-      headers: ({ request }) => {
-        expect(request.providerOptions?.store).toBe(false)
-        headers.push(String(request.model.id))
-        return { "x-deployment": "fixture", "x-override": "route" }
-      },
-      defaults: {
-        headers: { "x-default": "configured", "x-override": "configured" },
-        providerOptions: { store: false },
-        http: { body: { instructions: "default instructions" } },
-      },
-    })
-    const request = LLM.request({
-      model: route.model({ id: "fixture" }),
-      prompt: "hello",
-      system: "system instructions",
-      http: {
-        headers: { "x-override": "request" },
-        query: { trace: "request" },
-        body: {
-          model: "overlaid",
-          instructions: "request instructions",
-          previous_response_id: "resp_previous",
-          store: false,
-          stream: true,
-        },
-      },
-    })
-    const options: Parameters<typeof LLMClient.compact>[1] = {
-      http: (request, next) => {
-        middleware.push(new URL(request.url).pathname)
-        return next(request)
-      },
-    }
-    yield* LLMClient.generate(request, options)
-    yield* LLMClient.compact(request, options)
-    expect(headers).toEqual(["fixture", "fixture"])
-    expect(middleware).toEqual(["/fixture/responses", "/fixture/responses/compact"])
-  }),
-)
-
-for (const model of [
-  OpenAI.configure({ apiKey: "test" }).responses("fixture"),
-  Azure.configure({ apiKey: "test", resourceName: "test" }).responses("fixture"),
-  XAI.configure({ apiKey: "test" }).responses("fixture"),
-]) {
-  const item = {
-    type: model.provider === "xai" ? "x_search_call" : "computer_call",
-    id: "hosted_1",
-    status: "completed",
-  }
-  testEffect(
-    dynamicResponse(({ request, text, respond }) =>
-      Effect.sync(() => {
-        expect(new URL(request.url).pathname).toEndWith("/responses/compact")
-        expect(JSON.parse(text)).toEqual({ model: "fixture", input: [item], instructions: "Keep the context" })
-        return respond(JSON.stringify({ object: "response.compaction", output: [checkpoint] }))
-      }),
-    ),
-  ).effect(`${model.provider} compacts provider-specific history without lowering generation settings`, () =>
-    Effect.gen(function* () {
-      const request = LLM.request({
-        model,
-        system: "Keep the context",
-        messages: [
-          Message.assistant({
-            type: "tool-result",
-            id: item.id,
-            name: item.type,
-            result: { type: "json", value: item },
-            providerExecuted: true,
-            providerMetadata: { [model.route.providerMetadataKey ?? model.provider]: { itemId: item.id } },
-          }),
-        ],
-      })
-      for (const candidate of [
-        LLMRequest.update(request, {
-          tools: [
-            { name: "unsupported", description: "Generation only", inputSchema: {}, native: { unsupported: {} } },
-          ],
-        }),
-        LLMRequest.update(request, { providerOptions: { contextManagement: "invalid-generation-option" } }),
-      ]) {
-        const error = yield* LLMClient.generate(candidate).pipe(Effect.flip)
-        expect(error.reason._tag).toBe("InvalidRequest")
-        const response = yield* LLMClient.compact(candidate)
-        expect(response.messages[0]?.content[0]?.type).toBe("compaction")
-      }
-    }),
-  )
-}
-
-const retainedItems = [
-  retained,
-  {
-    type: "message",
-    id: "msg_assistant",
-    role: "assistant",
-    status: "completed",
-    phase: "commentary",
-    content: [
-      { type: "output_text", text: "First" },
-      { type: "output_text", text: "Second" },
-    ],
-  },
-  {
-    type: "reasoning",
-    id: "rs_1",
-    summary: [
-      { type: "summary_text", text: "Thinking" },
-      { type: "summary_text", text: "More thinking" },
-    ],
-    encrypted_content: "reasoning-state",
-  },
-  { type: "reasoning", id: "rs_2", summary: [], encrypted_content: "hidden-reasoning" },
-  {
-    type: "message",
-    id: "msg_media",
-    role: "user",
-    content: [
-      { type: "input_image", image_url: "https://example.com/image.png" },
-      { type: "input_file", filename: "report.pdf", file_data: "data:application/pdf;base64,cGRm", detail: "high" },
-      { type: "input_file", filename: "other.pdf", file_url: "https://example.com/report.pdf", detail: "low" },
-    ],
-  },
-  checkpoint,
-]
-
-for (const model of [
-  OpenAI.configure({ apiKey: "test" }).responses("gpt-5.3-codex"),
-  ...[undefined, "custom"].map((providerMetadataKey) =>
-    Route.make({
-      id: providerMetadataKey ?? "default-metadata",
-      provider: "openai",
-      providerMetadataKey,
-      protocol: OpenAIResponses.protocol,
-      compact: OpenAIResponses.route.compact,
-      endpoint: OpenAIResponses.route.endpoint,
-      transport: OpenAIResponses.httpTransport,
-    }).model({ id: "fixture" }),
-  ),
-]) {
-  testEffect(
-    dynamicResponse(({ request, text, respond }) =>
-      Effect.sync(() => {
-        if (new URL(request.url).pathname.endsWith("/compact"))
-          return respond(JSON.stringify({ object: "response.compaction", output: retainedItems }))
-        expect(JSON.parse(text).input).toEqual(retainedItems)
-        return respond(sseEvents({ type: "response.completed", response: { id: "resp_1" } }), {
-          headers: { "content-type": "text/event-stream" },
-        })
-      }),
-    ),
-  ).effect(`${model.route.id} retains messages, reasoning, and media through typed conversation parts`, () =>
-    Effect.gen(function* () {
-      const request = LLM.request({
-        model,
-        prompt: "hello",
-      })
-      const compacted = yield* LLMClient.compact(request)
-      expect(compacted.messages.map((message) => message.role)).toEqual([
-        "user",
-        "assistant",
-        "assistant",
-        "assistant",
-        "user",
-        "assistant",
-      ])
-      expect(compacted.messages[1]?.content).toEqual([
-        { type: "text", text: "First" },
-        { type: "text", text: "Second" },
-      ])
-      expect(compacted.messages[2]?.content.map((part) => part.type)).toEqual(["reasoning", "reasoning"])
-      expect(compacted.messages[4]?.content.map((part) => part.type)).toEqual(["media", "media", "media"])
-      const codec = Schema.fromJsonString(Schema.Array(Message))
-      const messages = Schema.decodeSync(codec)(Schema.encodeSync(codec)(compacted.messages))
-      yield* LLMClient.generate(LLMRequest.update(request, { messages }))
-    }),
-  )
-}
-
-for (const overlay of [undefined, { service_tier: "priority", prompt_cache_key: "overridden" }]) {
-  testEffect(
-    dynamicResponse(({ text, respond }) =>
-      Effect.sync(() => {
-        expect(JSON.parse(text)).toEqual({
-          model: "fixture",
-          input: [{ role: "user", content: [{ type: "input_text", text: "hello" }] }],
-          service_tier: overlay?.service_tier ?? "flex",
-          prompt_cache_key: overlay?.prompt_cache_key ?? "affinity",
-          prompt_cache_retention: "24h",
-          prompt_cache_options: { mode: "explicit", ttl: "30m" },
-        })
-        return respond(JSON.stringify({ object: "response.compaction", output: [checkpoint] }))
-      }),
-    ),
-  ).effect(`compact preserves supported request controls${overlay ? " with HTTP overrides" : ""}`, () =>
-    LLMClient.compact(
-      LLM.request({
-        model: OpenAI.configure({ apiKey: "test" }).responses("fixture"),
-        prompt: "hello",
-        promptCacheKey: "affinity",
-        providerOptions: { serviceTier: "flex" },
-        generation: { maxTokens: 100 },
-        http: {
-          body: {
-            stream: true,
-            store: false,
-            prompt_cache_retention: "24h",
-            prompt_cache_options: { mode: "explicit", ttl: "30m" },
-            ...overlay,
-          },
-        },
-      }),
-    ),
-  )
-}
-
-for (const item of [
-  { type: "unknown_provider_item", data: "do not hide in a compaction part" },
-  {
-    type: "message",
-    role: "user",
-    content: [{ type: "input_image", image_url: "https://example.com/image.png", detail: 42 }],
-  },
-  { type: "message", role: "user", content: [] },
-  {
-    type: "message",
-    role: "assistant",
-    content: [{ type: "input_image", image_url: "https://example.com/image.png" }],
-  },
-  { type: "message", role: "user", content: [{ type: "input_file", filename: "missing.pdf" }] },
-  {
-    type: "message",
-    role: "user",
-    content: [{ type: "input_file", filename: "bad.pdf", file_url: "https://example.com/report.pdf", detail: 42 }],
-  },
-  {
-    type: "message",
-    role: "user",
-    content: [
-      {
-        type: "input_file",
-        filename: "both.pdf",
-        file_url: "https://example.com/report.pdf",
-        file_data: "data:application/pdf;base64,cGRm",
-      },
-    ],
-  },
-]) {
-  testEffect(fixedResponse(JSON.stringify({ object: "response.compaction", output: [item, checkpoint] }))).effect(
-    `rejects unsupported compact output: ${JSON.stringify(item)}`,
-    () =>
-      Effect.gen(function* () {
-        const error = yield* LLMClient.compact(
-          LLM.request({ model: OpenAI.configure({ apiKey: "test" }).responses("gpt-5.3-codex"), prompt: "hello" }),
-        ).pipe(Effect.flip)
-        expect(error.reason._tag).toBe("InvalidProviderOutput")
-        expect(error.reason.body).toContain(JSON.stringify(item))
-        expect(error.reason.http?.status).toBe(200)
-      }),
-  )
-}
-
-for (const model of [
-  OpenAI.configure({ apiKey: "test" }).responses("fixture"),
-  Azure.configure({ apiKey: "test", resourceName: "test" }).responses("fixture"),
-  XAI.configure({ apiKey: "test" }).responses("fixture"),
-]) {
-  const images = [undefined, "low", "high", "auto"].map((detail) => ({
-    type: "input_image",
-    image_url: "https://example.com/image.png",
-    ...(detail === undefined ? {} : { detail }),
-  }))
-  testEffect(
-    dynamicResponse(({ request, text, respond }) =>
-      Effect.sync(() => {
-        if (new URL(request.url).pathname.endsWith("/compact"))
-          return respond(
-            JSON.stringify({
-              object: "response.compaction",
-              output: [{ type: "message", role: "user", content: images }, checkpoint],
-            }),
-          )
-        expect(JSON.parse(text).input[0].content).toEqual(images)
-        return respond(sseEvents({ type: "response.completed", response: { id: "resp_1" } }))
-      }),
-    ),
-  ).effect(`${model.provider} preserves retained image detail through serialization and replay`, () =>
-    Effect.gen(function* () {
-      const request = LLM.request({ model, prompt: "hello" })
-      const compacted = yield* LLMClient.compact(request)
-      const codec = Schema.fromJsonString(Schema.Array(Message))
-      const messages = Schema.decodeSync(codec)(Schema.encodeSync(codec)(compacted.messages))
-      yield* LLMClient.generate(LLMRequest.update(request, { messages }))
-    }),
-  )
-}
-
-testEffect(fixedResponse("must not execute")).effect("xAI rejects automatic compaction options", () =>
-  Effect.gen(function* () {
-    const request = LLMRequest.update(
-      LLM.request({ model: XAI.configure({ apiKey: "test" }).responses("grok-4.6"), prompt: "hello" }),
-      { providerOptions: { contextManagement: [{ type: "compaction" }] } },
-    )
-    const error = yield* LLMClient.generate(request).pipe(Effect.flip)
-    expect(error.reason._tag).toBe("InvalidRequest")
-    expect(error.message).toContain("LLMClient.compact")
-  }),
-)
-
-for (const model of [
-  OpenAI.configure({ apiKey: "test" }).responses("gpt-5.3-codex"),
-  Azure.configure({ apiKey: "test", resourceName: "test" }).responses("deployment"),
-  XAI.configure({ apiKey: "test" }).responses("grok-4.6"),
-]) {
-  testEffect(
-    dynamicResponse(({ request, text, respond }) =>
-      Effect.sync(() => {
-        const body = JSON.parse(text)
-        expect(request.method).toBe("POST")
-        expect(request.headers[model.provider === "azure" ? "api-key" : "authorization"]).toBe(
-          model.provider === "azure" ? "test" : "Bearer test",
-        )
-        if (new URL(request.url).pathname.endsWith("/responses/compact")) {
-          expect(body).toEqual({
-            model: model.id,
-            input: [{ role: "user", content: [{ type: "input_text", text: "original" }] }],
-            instructions: "system",
-          })
-          return respond(
-            JSON.stringify({
-              object: "response.compaction",
-              output,
-              usage: { input_tokens: 1000, output_tokens: 10, total_tokens: 1010 },
-            }),
-            { headers: { "content-type": "application/json" } },
-          )
-        }
-        expect(new URL(request.url).pathname.endsWith("/responses")).toBe(true)
-        expect(body.input).toEqual([...output, { role: "user", content: [{ type: "input_text", text: "continue" }] }])
-        return respond(sseEvents({ type: "response.completed", response: { id: "resp_1", output: [] } }), {
-          headers: { "content-type": "text/event-stream" },
-        })
-      }),
-    ),
-  ).effect(`${model.provider} explicitly compacts and replays the entire canonical window`, () =>
-    Effect.gen(function* () {
-      const request = LLM.request({ model, prompt: "original", system: "system", http: { body: { store: false } } })
-      const compacted = yield* LLMClient.compact(request)
-      expect(compacted.usage?.totalTokens).toBe(1010)
-      expect(compacted.messages.map((message) => message.role)).toEqual(["user", "assistant"])
-      expect(compacted.messages[0]?.content).toEqual([{ type: "text", text: "retained" }])
-      expect(compacted.messages[1]?.content).toEqual([
-        { type: "compaction", provider: model.provider, id: "cmp_1", encrypted: "opaque" },
-      ])
-      const codec = Schema.fromJsonString(Schema.Array(Message))
-      const messages = Schema.decodeSync(codec)(Schema.encodeSync(codec)(compacted.messages))
-      yield* LLMClient.generate(LLMRequest.update(request, { messages: [...messages, Message.user("continue")] }))
-    }),
-  )
-}
-
-for (const model of [
-  Anthropic.configure({ apiKey: "test" }).model("claude-opus-4-6"),
-  AmazonBedrockMantle.configure({ apiKey: "test" }).responses("model"),
-]) {
-  testEffect(fixedResponse("must not execute")).effect(
-    `${model.route.id} does not inherit an unsupported compact endpoint`,
-    () =>
-      Effect.gen(function* () {
-        // @ts-expect-error Untyped callers must still receive the runtime capability error.
-        const error = yield* LLMClient.compact(LLM.request({ model, prompt: "hello" })).pipe(Effect.flip)
-        expect(error.reason._tag).toBe("InvalidRequest")
-      }),
-  )
-}
-
-testEffect(
-  fixedResponse(JSON.stringify({ object: "response.compaction", output: [retained], debug: "original payload" })),
-).effect("invalid explicit compaction preserves the original response and HTTP context", () =>
-  Effect.gen(function* () {
-    const error = yield* LLMClient.compact(
-      LLM.request({ model: OpenAI.configure({ apiKey: "test" }).responses("gpt-5.3-codex"), prompt: "hello" }),
-    ).pipe(Effect.flip)
-    expect(error.reason._tag).toBe("InvalidProviderOutput")
-    expect(error.reason.body).toContain("original payload")
-    expect(error.reason.http?.status).toBe(200)
-  }),
-)
@@ -1,60 +0,0 @@
-import { expect } from "bun:test"
-import { Effect, Schema } from "effect"
-import { LLM, Message } from "../../src/index.js"
-import { OpenAI, Azure, XAI } from "../../src/providers.js"
-import { compileRequest } from "../../src/route/client.js"
-import { it } from "../lib/effect.js"
-
-for (const model of [
-  OpenAI.configure({ apiKey: "test" }).responses("fixture"),
-  Azure.configure({ apiKey: "test", resourceName: "test" }).responses("fixture"),
-  XAI.configure({ apiKey: "test" }).responses("fixture"),
-]) {
-  it.effect(`${model.provider} preserves image detail through message serialization and lowering`, () =>
-    Effect.gen(function* () {
-      const details = [undefined, "low", "high", "auto"]
-      const message = Message.user(
-        details.map((detail) => ({
-          type: "media",
-          mediaType: "image/png",
-          data: "https://example.com/image.png",
-          providerMetadata:
-            detail === undefined ? undefined : { [model.route.providerMetadataKey ?? model.provider]: { detail } },
-        })),
-      )
-      const codec = Schema.fromJsonString(Message)
-      const prepared = yield* compileRequest(
-        LLM.request({
-          model,
-          messages: [Schema.decodeSync(codec)(Schema.encodeSync(codec)(message))],
-        }),
-      )
-      expect(prepared.body.input[0].content).toEqual(
-        details.map((detail) => ({
-          type: "input_image",
-          image_url: "https://example.com/image.png",
-          detail,
-        })),
-      )
-    }),
-  )
-}
-
-it.effect("rejects malformed image detail instead of silently discarding it", () =>
-  Effect.gen(function* () {
-    const error = yield* compileRequest(
-      LLM.request({
-        model: OpenAI.configure({ apiKey: "test" }).responses("fixture"),
-        messages: [
-          Message.user({
-            type: "media",
-            mediaType: "image/png",
-            data: "https://example.com/image.png",
-            providerMetadata: { openai: { detail: 42 } },
-          }),
-        ],
-      }),
-    ).pipe(Effect.flip)
-    expect(error.reason._tag).toBe("InvalidRequest")
-  }),
-)
@@ -1,18 +1,5 @@
 import { describe, expect } from "bun:test"
-import {
-  AIError,
-  CompactionPart,
-  CompactionResponse,
-  LanguageModel,
-  LLM,
-  LLMClient,
-  LLMEvent,
-  LLMRequest,
-  Message,
-  ProviderID,
-  RateLimitError,
-} from "../src/index.js"
-import { OpenAI } from "../src/providers.js"
+import { AIError, LanguageModel, LLM, LLMClient, LLMEvent, LLMRequest, RateLimitError } from "../src/index.js"
 import { OpenAIChat } from "../src/protocols/openai-chat.js"
 import { TestLLM } from "../src/testing.js"
 import { Effect, Fiber, Latch, Stream } from "effect"
@@ -79,54 +66,6 @@ describe("TestLLM legacy client", () => {
 })

 describe("TestLLM first-class client", () => {
-  it.effect("rejects response fixtures for the wrong operation", () =>
-    Effect.gen(function* () {
-      const client = yield* TestLLM.Test
-      const request = LLM.request({ model: OpenAI.configure({ apiKey: "test" }).responses("fixture"), prompt: "hello" })
-      yield* client.push(TestLLM.stop(), new CompactionResponse({ messages: [] }))
-      expect(yield* client.compact(request).pipe(Effect.catchDefect(Effect.succeed))).toBe(
-        "TestLLM compaction requires a CompactionResponse",
-      )
-      expect(yield* client.generate(request).pipe(Effect.catchDefect(Effect.succeed))).toBe(
-        "TestLLM generation requires an event response",
-      )
-    }),
-  )
-
-  it.effect("scripts replacement windows with the same lazy recording, gates, and fallback controls", () =>
-    Effect.gen(function* () {
-      const client = yield* TestLLM.Test
-      const request = LLM.request({ model: OpenAI.configure({ apiKey: "test" }).responses("fixture"), prompt: "hello" })
-      const compacted = new CompactionResponse({
-        messages: [
-          Message.user("retained input"),
-          Message.assistant(CompactionPart.make({ provider: ProviderID.make("openai"), encrypted: "checkpoint" })),
-          Message.user("retained tail"),
-        ],
-      })
-      yield* client.push(compacted, TestLLM.text("continued", "answer"))
-      const operation = LLMClient.compact(request)
-      expect(yield* client.requests()).toEqual([])
-      const gate = yield* client.gate()
-      const fiber = yield* operation.pipe(Effect.forkChild({ startImmediately: true }))
-      yield* gate.started
-      yield* client.wait(1)
-      expect(fiber.pollUnsafe()).toBeUndefined()
-      yield* gate.release
-      expect(yield* Fiber.join(fiber)).toBe(compacted)
-      const next = LLMRequest.update(request, { messages: compacted.messages })
-      expect((yield* LLMClient.generate(next)).text).toBe("continued")
-      yield* client.serve((observed) => {
-        expect(observed).toBe(next)
-        return compacted
-      })
-      expect(yield* LLMClient.compact(next)).toBe(compacted)
-      yield* client.always(compacted)
-      expect(yield* LLMClient.compact(next)).toBe(compacted)
-      expect(yield* client.requests()).toEqual([request, next, next, next])
-    }),
-  )
-
  it.effect("provides the same object under normal and test tags with snapshot observations", () =>
    Effect.gen(function* () {
      const llm = yield* TestLLM.Test
@@ -9,13 +9,13 @@
      "src": "/web-app-manifest-192x192.png",
      "sizes": "192x192",
      "type": "image/png",
-      "purpose": "any"
+      "purpose": "maskable"
    },
    {
      "src": "/web-app-manifest-512x512.png",
      "sizes": "512x512",
      "type": "image/png",
-      "purpose": "any"
+      "purpose": "maskable"
    }
  ],
  "theme_color": "#080808",
@@ -6,6 +6,7 @@ import { AppBaseProviders, AppInterface } from "@/app"
 import { loadInitialLocale } from "@/runtime/i18n/language"
 import { PlatformProvider } from "@/runtime/platform/platform"
 import { createWebPlatform } from "@/runtime/platform/web"
+import { isStandalone, PwaRoutePersistence, restorePwaRoute } from "@/runtime/platform/pwa"
 import en from "@/runtime/i18n/en"
 import zh from "@/runtime/i18n/zh"
 import { authFromToken } from "@/runtime/server/api"
@@ -71,6 +72,8 @@ if (root instanceof HTMLElement && root.dataset.opencodeMounted === undefined) {
  void loadInitialLocale().then((locale) => {
    const auth = authFromToken(new URLSearchParams(location.search).get("auth_token"))
    clearAuthToken()
+    const standalone = isStandalone()
+    if (standalone) restorePwaRoute()
    const server: ServerConnection.Http = {
      type: "http",
      authToken: !!auth,
@@ -87,7 +90,9 @@ if (root instanceof HTMLElement && root.dataset.opencodeMounted === undefined) {
              defaultServer={ServerConnection.Key.make(web.defaultServerUrl)}
              canonicalLocalServer={ServerConnection.key(server)}
              servers={[server]}
-            />
+            >
+              {standalone && <PwaRoutePersistence />}
+            </AppInterface>
          </AppBaseProviders>
        </PlatformProvider>
      ),
@@ -0,0 +1,43 @@
+import { useLocation } from "@solidjs/router"
+import { createEffect } from "solid-js"
+
+const LAST_ROUTE_KEY = "opencode.pwa.last-route"
+
+export function isStandalone() {
+  return (
+    window.matchMedia("(display-mode: standalone)").matches ||
+    ("standalone" in navigator && navigator.standalone === true)
+  )
+}
+
+export function restorePwaRoute() {
+  if (location.pathname !== "/" || location.search || location.hash) return
+  try {
+    const value = localStorage.getItem(LAST_ROUTE_KEY)
+    if (!value) return
+    const url = new URL(value, location.origin)
+    if (url.origin !== location.origin || url.searchParams.has("auth_token")) return
+    if (
+      url.pathname !== "/" &&
+      url.pathname !== "/new-session" &&
+      !/^\/server\/[^/]+\/session\/[^/]+$/.test(url.pathname)
+    )
+      return
+    history.replaceState(history.state, "", url.pathname + url.search + url.hash)
+  } catch {
+    // Storage may be unavailable; keep the launch URL in that case.
+  }
+}
+
+export function PwaRoutePersistence() {
+  const location = useLocation()
+  createEffect(() => {
+    const value = location.pathname + location.search + location.hash
+    try {
+      localStorage.setItem(LAST_ROUTE_KEY, value)
+    } catch {
+      // Navigation must still work when storage is unavailable or full.
+    }
+  })
+  return null
+}
@@ -0,0 +1,79 @@
+import { afterEach, beforeEach, expect, test } from "bun:test"
+import { MemoryRouter, createMemoryHistory } from "@solidjs/router"
+import { createComponent, render } from "solid-js/web"
+import { isStandalone, PwaRoutePersistence, restorePwaRoute } from "../src/runtime/platform/pwa"
+
+const key = "opencode.pwa.last-route"
+const originalUrl = window.location.href
+
+beforeEach(() => {
+  window.location.href = "http://localhost/"
+})
+
+afterEach(() => {
+  localStorage.removeItem(key)
+  window.location.href = originalUrl
+})
+
+test("normal browser windows are not standalone", () => {
+  expect(isStandalone()).toBe(false)
+})
+
+test("restores the last PWA route including query and hash without adding history", () => {
+  window.history.replaceState({ retained: true }, "", "http://localhost/")
+  const length = window.history.length
+  localStorage.setItem(key, "/server/local/session/session-1?view=files#file")
+
+  restorePwaRoute()
+
+  expect(window.location.pathname + window.location.search + window.location.hash).toBe(
+    "/server/local/session/session-1?view=files#file",
+  )
+  expect(window.history.length).toBe(length)
+  expect(window.history.state).toEqual({ retained: true })
+})
+
+test("preserves explicit launch routes, queries, and hashes", () => {
+  localStorage.setItem(key, "/server/local/session/saved")
+  for (const route of ["/server/local/session/linked", "/new-session?draftId=123", "/?launch=1", "/#launch"]) {
+    window.history.replaceState(null, "", `http://localhost${route}`)
+    restorePwaRoute()
+    expect(window.location.pathname + window.location.search + window.location.hash).toBe(route)
+  }
+})
+
+test("ignores missing, invalid, external, and auth-bearing saved routes", () => {
+  window.history.replaceState(null, "", "http://localhost/")
+  restorePwaRoute()
+  expect(window.location.pathname).toBe("/")
+
+  for (const value of [
+    "/removed-route",
+    "https://example.com/new-session",
+    "//example.com/new-session",
+    "http://[",
+    "/new-session?auth_token=secret",
+  ]) {
+    localStorage.setItem(key, value)
+    restorePwaRoute()
+    expect(window.location.href).toBe("http://localhost/")
+  }
+})
+
+test("persists router navigation including returning home", async () => {
+  const host = document.createElement("div")
+  const history = createMemoryHistory()
+  history.set({ value: "/new-session?draftId=123", replace: true, scroll: false })
+  const dispose = render(() => createComponent(MemoryRouter, { history, root: PwaRoutePersistence }), host)
+  try {
+    expect(localStorage.getItem(key)).toBe("/new-session?draftId=123")
+    history.set({ value: "/server/local/session/next#file", scroll: false })
+    await new Promise((resolve) => setTimeout(resolve, 0))
+    expect(localStorage.getItem(key)).toBe("/server/local/session/next#file")
+    history.set({ value: "/", scroll: false })
+    await new Promise((resolve) => setTimeout(resolve, 0))
+    expect(localStorage.getItem(key)).toBe("/")
+  } finally {
+    dispose()
+  }
+})
@@ -59,6 +59,7 @@ test.each(["dev", "beta", "prod"])("serves %s app icons", async (channel) => {
 async function check(channel: string, read: (path: string) => Promise<Uint8Array>) {
  const html = new TextDecoder().decode(await read("/index.html"))
  const actual: typeof manifest = JSON.parse(new TextDecoder().decode(await read("/site.webmanifest")))
+  expect(actual.icons.every((icon) => icon.purpose === "maskable")).toBe(true)
  expect(actual).toEqual({
    ...manifest,
    icons: manifest.icons.map((icon) => ({ ...icon, src: `/icons/${channel}${icon.src}` })),
@@ -98,12 +98,13 @@ Effect.gen(function* () {
  Effect.provide(Config.layer),
  Effect.provide(Updater.layer),
  Effect.provide(
-    LayerNode.compile(LayerNode.group([Global.node, AppProcess.node, Npm.node]), [
-      [
-        Global.node,
-        Global.layerWith(process.env.OPENCODE_CONFIG_DIR ? { config: process.env.OPENCODE_CONFIG_DIR } : {}),
+    LayerNode.compile(LayerNode.group([Global.node, AppProcess.node, Npm.node]), {
+      replacements: [
+        Global.node.replace(
+          Global.layerWith(process.env.OPENCODE_CONFIG_DIR ? { config: process.env.OPENCODE_CONFIG_DIR } : {}),
+        ),
      ],
-    ]),
+    }),
  ),
  Effect.provide(
    Observability.layer({
@@ -30,12 +30,13 @@ export const run = Effect.fnUntraced(function* (options: Options) {
  return yield* processEffect(options).pipe(
    Effect.provide(Updater.layer),
    Effect.provide(
-      LayerNode.compile(LayerNode.group([Global.node, AppProcess.node]), [
-        [
-          Global.node,
-          Global.layerWith(process.env.OPENCODE_CONFIG_DIR ? { config: process.env.OPENCODE_CONFIG_DIR } : {}),
+      LayerNode.compile(LayerNode.group([Global.node, AppProcess.node]), {
+        replacements: [
+          Global.node.replace(
+            Global.layerWith(process.env.OPENCODE_CONFIG_DIR ? { config: process.env.OPENCODE_CONFIG_DIR } : {}),
+          ),
        ],
-      ]),
+      }),
    ),
    Effect.provide(NodeServices.layer),
  )
@@ -180,7 +180,7 @@ export const toTypeScript = (schema: Schema.Top, decoded = false, pretty = false

 export const jsonSchemaToTypeScript = (schema: JsonSchema, pretty = false): string => {
  try {
-    return renderSchema(schema, { definitions: { ...(schema.definitions ?? {}), ...(schema.$defs ?? {}) }, pretty })
+    return renderSchema(schema, { definitions: {}, pretty })
  } catch {
    return "unknown"
  }
@@ -216,6 +216,35 @@ describe("pretty signature rendering", () => {
  })
 })

+describe("JSON Schema definition scope", () => {
+  test.each(["definitions", "$defs"])("resolves root %s and lets $defs take precedence", (key) => {
+    const schema = { $ref: `#/${key}/Value`, [key]: { Value: { type: "string" } } }
+    expect(jsonSchemaToTypeScript(schema)).toBe("string")
+    expect(jsonSchemaToTypeScript(schema, true)).toBe("string")
+
+    const overridden = { ...schema, $defs: { Value: { type: "number" } } }
+    expect(jsonSchemaToTypeScript(overridden)).toBe("number")
+    expect(jsonSchemaToTypeScript(overridden, true)).toBe("number")
+  })
+
+  test.each(["definitions", "$defs"])("nested %s shadow inherited definitions without affecting siblings", (key) => {
+    const schema = {
+      type: "object",
+      definitions: { Inherited: { type: "string" } },
+      $defs: { Value: { type: "number" } },
+      properties: {
+        nested: { $ref: `#/${key}/Value`, [key]: { Value: { type: "boolean" } } },
+        inherited: { $ref: "#/definitions/Inherited" },
+        sibling: { $ref: "#/$defs/Value" },
+      },
+    }
+    expect(jsonSchemaToTypeScript(schema)).toBe("{ nested?: boolean; inherited?: string; sibling?: number }")
+    expect(jsonSchemaToTypeScript(schema, true)).toBe(
+      ["{", "  nested?: boolean,", "  inherited?: string,", "  sibling?: number,", "}"].join("\n"),
+    )
+  })
+})
+
 describe("non-identifier property names render as quoted keys", () => {
  // MCP-style schemas routinely carry property names that are not bare TS identifiers
  // (`foo-bar`, `@type`, dotted names); the rendered signature must quote them so the
@@ -303,7 +303,6 @@ function modelFromLanguage(info: Info, language: LanguageModelV3) {
  const projected = mapBodyToProviderOptions(info, packageName)
  const optionKey = providerOptionKey(packageName, info.providerID)
  const route: AnyRoute = {
-    compact: undefined,
    id: `ai-sdk:${packageName}`,
    provider: ProviderID.make(info.providerID),
    providerMetadataKey: optionKey,
@@ -328,12 +327,7 @@ function modelFromLanguage(info: Info, language: LanguageModelV3) {
    },
    body: {
      schema: Schema.Unknown,
-      from: (request) =>
-        Effect.try({
-          try: () => callOptions(request, packageName, info.modelID ?? info.id, optionKey),
-          catch: (cause) =>
-            cause instanceof AIError ? cause : ProviderShared.invalidRequest("Invalid AI SDK request", cause),
-        }),
+      from: (request) => Effect.succeed(callOptions(request, packageName, info.modelID ?? info.id, optionKey)),
    },
    with: () => route,
    model: (input) =>
@@ -518,8 +512,6 @@ function userPart(part: ContentPart): UserContent {

 function assistantPart(part: ContentPart): AssistantContent {
  switch (part.type) {
-    case "compaction":
-      throw ProviderShared.invalidRequest("AI SDK routes cannot replay native provider compaction state")
    case "text":
      return [{ type: "text", text: part.text, providerOptions: metadataProviderOptions(part.providerMetadata) }]
    case "media":
@@ -1,20 +1,11 @@
 import { buildLocationServiceMap } from "../location-services.js"
 import { LocationServiceMap } from "../location-service-map.js"
 import { LayerNode } from "@opencode-ai/util/effect/layer-node"
-import { makeGlobalNode } from "@opencode-ai/util/effect/app-node"

-export function build<A, E>(root: LayerNode.Node<A, E, any>, replacements: LayerNode.Replacements = []) {
-  // Only build the location service map if it's actually needed
-  if (!LayerNode.hasUnbound(root, LocationServiceMap.node) || hasReplacement(replacements, LocationServiceMap.node))
-    return LayerNode.compile(root, replacements)
-
-  const locationMap = buildLocationServiceMap(replacements)
-  const locationMapNode = makeGlobalNode({ service: LocationServiceMap.Service, layer: locationMap, deps: [] })
-  return LayerNode.compile(root, replacements.concat([[LocationServiceMap.node, locationMapNode]]))
-}
-
-function hasReplacement(replacements: LayerNode.Replacements, node: LayerNode.Node<unknown, unknown, any>) {
-  return replacements.some(([source]) => source.name === node.name)
+export function build<A, E>(root: LayerNode.Graph<A, E>, replacements: LayerNode.Replacements = []) {
+  return LayerNode.compile(root, {
+    replacements: [LocationServiceMap.node.replace(buildLocationServiceMap(replacements)), ...replacements],
+  })
 }

 export * as AppNodeBuilder from "./app-node-builder.js"
@@ -54,9 +54,8 @@ const layer = Layer.effect(
    })

    const file = Effect.fn("Formatter.file")(function* (filepath: string) {
-      const matching = state
-        .get()
-        .formatters.filter((formatter) => formatter.extensions.includes(path.extname(filepath)))
+      const extension = path.extname(filepath)
+      const matching = state.get().formatters.filter((formatter) => formatter.extensions.includes(extension))

      for (const formatter of matching) {
        const enabled = yield* command(formatter)
@@ -55,6 +55,7 @@ import { ToolOutput } from "./tool-output.js"
 import { Vcs } from "./vcs.js"

 export * as Instance from "./instance.js"
+export { Service, byLocationNode, type Interface } from "./instance/service.js"

 const nodes = [
  Location.node,
@@ -110,9 +111,9 @@ const nodes = [
  Vcs.node,
  // Start repository watches only after boot-critical filesystem and Git work.
  LocationWatcher.node,
-] as const satisfies readonly Node.LocationNode<unknown, unknown>[]
+] as const satisfies readonly Node.LocationGraph<never, unknown>[]

-export const graph = LayerNode.group<typeof nodes>(nodes)
+export const graph = LayerNode.group(nodes)

 export type Services = LayerNode.Output<typeof graph>
 export type Error = LayerNode.Error<typeof graph>
@@ -141,29 +142,23 @@ export interface Options {
 // source still honors explicit plugin operations from wellknown and
 // host-injected config.
 const vanillaReplacements: LayerNode.Replacements = [
-  [Config.node, Config.configured({ project: false, global: false })],
-  [InstructionDiscovery.node, InstructionDiscovery.configured({ project: false, global: false })],
+  Config.node.replace(Config.configured({ project: false, global: false })),
+  InstructionDiscovery.node.replace(InstructionDiscovery.configured({ project: false, global: false })),
 ]

 // One instance is one compiled, fresh copy of the graph standing on a directory.
 export function layer(ref: Location.Ref, options: Options = {}) {
  const startedAt = performance.now()
  // Ordered: vanilla defaults, then caller replacements (which win over the
-  // defaults), then bound pairs (which win over everything).
-  const allReplacements: LayerNode.Replacements = [
+  // defaults), then instance bindings (which win over everything).
+  const replacements: LayerNode.Replacements = [
    ...(options.discovery === false ? vanillaReplacements : []),
    ...(options.replacements ?? []),
-    [Location.node, Location.boundNode(ref, { discovery: options.discovery })],
-    [InstancePlugins.node, InstancePlugins.bound(options.plugins ?? [])],
+    Location.node.replace(Location.boundNode(ref, { discovery: options.discovery })),
+    InstancePlugins.node.replace(InstancePlugins.bound(options.plugins ?? [])),
  ]
-  // Apply replacements during hoist, not afterward: replacements can
-  // introduce new tagged dependencies (Location.boundNode depends on
-  // Project), and the hoist walk is the only pass that can still slice
-  // those back out.
-  const location = LayerNode.hoist(graph, Node.tags.values.global, allReplacements)

-  return LayerNode.compile(location.node).pipe(
-    Layer.fresh,
+  return LayerNode.compile(graph, { replacements, shared: Node.tags.values.global }).pipe(
    Layer.tap(() =>
      Effect.logInfo("location services booted", {
        directory: ref.directory,
@@ -171,6 +166,5 @@ export function layer(ref: Location.Ref, options: Options = {}) {
        durationMs: Math.round(performance.now() - startedAt),
      }),
    ),
-    Layer.provide(LayerNode.compile(location.hoisted)),
  )
 }
@@ -0,0 +1,42 @@
+export * as Instance from "./service.js"
+export type { Services } from "../instance.js"
+
+import { Context, Effect, Layer, Option, Scope } from "effect"
+import type { Session } from "@opencode-ai/schema/session"
+import { makeGlobalNode } from "@opencode-ai/util/effect/app-node"
+import type { Services } from "../instance.js"
+import { LocationServiceMap } from "../location-service-map.js"
+
+/** Selects Session capabilities; implementations own caching and lifetime. */
+export interface Interface {
+  readonly provide: (
+    session: Session.Info,
+  ) => <A, E, R>(effect: Effect.Effect<A, E, R>) => Effect.Effect<A, E, Exclude<R, Services>>
+  /** Borrow a cached instance without initializing one when it is absent. */
+  readonly provideIfLoaded: (
+    session: Session.Info,
+  ) => <A, E, R>(effect: Effect.Effect<A, E, R>) => Effect.Effect<Option.Option<A>, E, Exclude<R, Services>>
+}
+
+export class Service extends Context.Service<Service, Interface>()("@opencode/Instance") {}
+
+const layer = Layer.effect(
+  Service,
+  Effect.gen(function* () {
+    const locations = yield* LocationServiceMap.Service
+    return Service.of({
+      provide: (session) => Effect.provide(locations.get(session.location)),
+      provideIfLoaded: (session) => (effect) =>
+        // Scope the borrowed reference without replacing the caller's Scope.
+        Effect.scopedWith((scope) =>
+          Effect.gen(function* () {
+            const context = yield* locations.contextEffectOption(session.location).pipe(Scope.provide(scope))
+            if (Option.isNone(context)) return Option.none()
+            return Option.some(yield* effect.pipe(Effect.provide(context.value)))
+          }),
+        ),
+    })
+  }),
+)
+
+export const byLocationNode = makeGlobalNode({ service: Service, layer, deps: [LocationServiceMap.node] })
@@ -112,7 +112,7 @@ export interface Interface {

 export class Service extends Context.Service<Service, Interface>()("@opencode/PersistentPty") {}

-export const configured = (options: Options = {}) =>
+const makeLayer = (options: Options = {}) =>
  Layer.effect(
    Service,
    Effect.gen(function* () {
@@ -361,8 +361,14 @@ export const configured = (options: Options = {}) =>
    }),
  )

-export const layer = configured()
-export const node = makeGlobalNode({ service: Service, layer, deps: [Bus.node, Global.node] })
+export const layer = makeLayer()
+export const configured = (options?: Options) =>
+  makeGlobalNode({
+    service: Service,
+    layer: options === undefined ? layer : makeLayer(options),
+    deps: [Bus.node, Global.node],
+  })
+export const node = configured()

 const request = (daemon: DaemonTransport, value: object, start = false) =>
  daemon.request(value, start).pipe(Effect.mapError(unavailable))
@@ -80,7 +80,8 @@ export const make = Effect.fn("PluginHost.make")(function* (plugin: Interface, p
  const response = <A, E, R>(effect: Effect.Effect<A, E, R>) =>
    effect.pipe(Effect.map((data) => ({ location: locationInfo(), data })))

-  return {
+  // Keep the instance graph's inferred types independent of Session handles.
+  const context: Plugin.Context = {
    app,
    location: locationInfo(),
    options: {},
@@ -206,7 +207,8 @@ export const make = Effect.fn("PluginHost.make")(function* (plugin: Interface, p
          .subscribe()
          .pipe(
            Stream.filter(
-              (event): event is EventManifest.ServerEvent | RpcEvent => EventManifest.isServer(event) || isRpcEvent(event),
+              (event): event is EventManifest.ServerEvent | RpcEvent =>
+                EventManifest.isServer(event) || isRpcEvent(event),
            ),
          ),
    },
@@ -449,7 +451,8 @@ export const make = Effect.fn("PluginHost.make")(function* (plugin: Interface, p
      wait: (input) => runtime.session.wait(input.sessionID),
      context: (input) => runtime.session.context(input.sessionID),
    },
-  } satisfies Plugin.Context
+  }
+  return context
 })

 export function storage(kv: KV.Interface, pluginID: string): Plugin.Context["storage"] {
@@ -62,7 +62,7 @@ const require = <A, E, R>(cell: Cell, f: (runtime: Interface) => Effect.Effect<A

 const defaultCell = makeCell()

-export const layerWithCell = (cell: Cell) =>
+export const layerWithCell = (cell: Cell): Layer.Layer<Service> =>
  Layer.succeed(
    Service,
    Service.of({
@@ -11,6 +11,9 @@ truth. Follow links from that page when the question needs more detail. Fetch
 <https://opencode.ai/v2/docs/> first when you need to discover the relevant
 documentation page.

+A machine-readable documentation index is available at
+<https://opencode.ai/v2/llms.txt>.
+
 ## Version policy

 Always answer for OpenCode V2 unless the user explicitly asks about V1,
@@ -152,6 +155,8 @@ before answering. Refer to this guide when the user wants to build a plugin. It
 covers hooks, transforms, tools, plugin context capabilities, and package
 entrypoints. Plugins can also extend the TUI; for those, fetch the
 [CLI plugin guide](https://opencode.ai/v2/docs/build/plugins/cli).
+For custom methods and events shared with other plugins or clients, fetch the
+[RPC guide](https://opencode.ai/v2/docs/build/plugins/rpc).

 ## [Service](https://opencode.ai/v2/docs/troubleshooting#check-the-background-service)

@@ -220,6 +225,16 @@ exposes typed Effects, Streams, and decoded OpenCode schema values. Its
 `Service` API can discover, start, stop, and authenticate with the local
 background service from a Node application.

+## [SDK](https://opencode.ai/v2/docs/build/sdk)
+
+For questions about embedding OpenCode directly in an application, fetch the
+full [SDK guide](https://opencode.ai/v2/docs/build/sdk) before answering. The SDK
+hosts OpenCode in the application without opening an HTTP listener.
+
+Use the [Effect SDK guide](https://opencode.ai/v2/docs/build/sdk/effect) for
+Effect applications. For Cloudflare Durable Objects, use the
+[Cloudflare SDK guide](https://opencode.ai/v2/docs/build/sdk/cloudflare).
+
 ## [Troubleshooting](https://opencode.ai/v2/docs/troubleshooting)

 OpenCode runs a client and a background server. Start by determining whether a
@@ -1,7 +1,7 @@
 export * as Session from "./session.js"
 export * from "./session/schema.js"

-import { Cause, Effect, Layer, Schema, Context, RcMap, Stream } from "effect"
+import { Cause, Effect, Layer, Schema, Context, Stream } from "effect"
 import { ListAnchor } from "@opencode-ai/schema/session"
 import { and, desc, eq } from "drizzle-orm"
 import { Project } from "./project.js"
@@ -10,6 +10,7 @@ import { Location } from "./location.js"
 import { SessionMessage } from "./session/message.js"
 import { PromptInput } from "@opencode-ai/schema/prompt-input"
 import { Bus } from "./bus.js"
+import { Instance } from "./instance/service.js"
 import { Database } from "./database/database.js"
 import { SessionProjector } from "./session/projector.js"
 import { SessionMessageTable } from "./session/sql.js"
@@ -238,20 +239,16 @@ const layer = Layer.effect(
    const global = yield* Global.Service
    const execution = yield* SessionExecution.Service
    const store = yield* SessionStore.Service
+    const instances = yield* Instance.Service
    const locations = yield* LocationServiceMap.Service
    const fs = yield* FSUtil.Service
    const jobs = yield* Job.Service
    const environments = yield* SessionEnvironment.Service
-    const sessions = yield* Session.make((ref) => locations.get(ref))
+    const sessions = yield* Session.make()
    const admission = yield* SessionInbox.Service
    const closeTransport = Effect.fn("Session.closeTransport")(function* (session: SessionSchema.Info) {
-      const location = Location.Ref.make({
-        directory: session.location.directory,
-        workspaceID: session.location.workspaceID,
-      })
-      if (!(yield* RcMap.has(locations.rcMap, location))) return
      yield* SessionModelTransport.Service.use((transport) => transport.close(session.id)).pipe(
-        Effect.provide(locations.get(location)),
+        instances.provideIfLoaded(session),
      )
    })
    const isDurableSessionEvent = Schema.is(SessionEvent.Durable)
@@ -403,7 +400,7 @@ const layer = Layer.effect(
      prompt: (input) => sessions.forSession(input.sessionID).prompt(input),
      generate: Effect.fn("Session.generate")(function* (input) {
        const session = yield* result.get(input.sessionID)
-        const generate = yield* SessionGenerate.Service.pipe(Effect.provide(locations.get(session.location)))
+        const generate = yield* SessionGenerate.Service.pipe(instances.provide(session))
        return yield* generate.generate(input)
      }),
      command: Effect.fn("Session.command")(function* (input) {
@@ -412,7 +409,7 @@ const layer = Layer.effect(
          const plugins = yield* PluginSupervisor.Service
          yield* plugins.flush
          return yield* Command.Service
-        }).pipe(Effect.provide(locations.get(session.location)))
+        }).pipe(instances.provide(session))
        const delivery = input.delivery ?? "steer"
        yield* commands.execute({
          name: input.command,
@@ -468,7 +465,8 @@ const layer = Layer.effect(
          Effect.gen(function* () {
            const latest = yield* result.get(input.sessionID)
            const source = yield* fs.stat(latest.location.directory).pipe(Effect.orElseSucceed(() => undefined))
-            if (!source || source.type !== "Directory") {
+            // Active runners must hand off at a step boundary to retain their continuation.
+            if ((!source || source.type !== "Directory") && !(yield* execution.isActive(input.sessionID))) {
              const cancellations = (yield* SessionInbox.moveIDs(db, input.sessionID)).map(
                (item) => [SessionEvent.InboxCancelled, { sessionID: input.sessionID, inboxID: item.id }] as const,
              )
@@ -534,6 +532,7 @@ export const node = makeGlobalNode({
    Project.node,
    SessionExecution.node,
    SessionStore.node,
+    Instance.byLocationNode,
    SessionInbox.node,
    LocationServiceMap.node,
    SessionProjector.node,
@@ -4,7 +4,7 @@ import { Cause, Context, Effect, Exit, Layer } from "effect"
 import { Bus } from "../bus.js"
 import { Database } from "../database/database.js"
 import { Job } from "../job.js"
-import { LocationServiceMap } from "../location-service-map.js"
+import { Instance } from "../instance/service.js"
 import { makeGlobalNode } from "@opencode-ai/util/effect/app-node"
 import { SessionEvent } from "./event.js"
 import { SessionRunCoordinator } from "./run-coordinator.js"
@@ -35,7 +35,7 @@ export interface Interface {
  readonly awaitIdle: (sessionID: SessionSchema.ID) => Effect.Effect<void>
 }

-/** Routes execution from a Session ID to the runner owned by that Session's Location. */
+/** Routes execution from a Session ID to its selected instance's runner. */
 export class Service extends Context.Service<Service, Interface>()("@opencode/SessionExecution") {}

 type InterruptReason = "user" | "shutdown"
@@ -48,12 +48,12 @@ export function terminal(exit: Exit.Exit<void, SessionRunner.RunError>, reason?:
  return { type: "failed" as const, error: toSessionError(failure) }
 }

-/** Process-local execution: drains run in this process, routed through the Session's Location graph. */
+/** Process-local execution: drains run in this process using the selected instance. */
 export const layer = Layer.effect(
  Service,
  Effect.gen(function* () {
    const store = yield* SessionStore.Service
-    const locations = yield* LocationServiceMap.Service
+    const instances = yield* Instance.Service
    const bus = yield* Bus.Service
    const jobs = yield* Job.Service
    const db = (yield* Database.Service).db
@@ -90,7 +90,7 @@ export const layer = Layer.effect(
      const result = yield* SessionRunner.Service.use((runner) =>
        runner.drain({ sessionID, force, continuation, promotable }),
      ).pipe(
-        Effect.provide(locations.get(session.location)),
+        instances.provide(session),
        Effect.tapCause((cause) =>
          Cause.hasInterruptsOnly(cause)
            ? Effect.void
@@ -173,7 +173,7 @@ export const layer = Layer.effect(
 export const node = makeGlobalNode({
  service: Service,
  layer,
-  deps: [SessionStore.node, LocationServiceMap.node, Bus.node, Database.node, Job.node],
+  deps: [SessionStore.node, Instance.byLocationNode, Bus.node, Database.node, Job.node],
 })

 /** Low-level compatibility layer for callers that only need durable Session recording. */
@@ -1,11 +1,11 @@
 export * as Session from "./session.js"

-import { DateTime, Effect, Fiber, Layer, Schema, Scope } from "effect"
+import { DateTime, Effect, Fiber, Schema, Scope } from "effect"
 import type { Agent } from "@opencode-ai/schema/agent"
 import type { Model } from "@opencode-ai/schema/model"
 import { Event } from "@opencode-ai/schema/event"
 import { Bus } from "../bus.js"
-import { Location } from "../location.js"
+import { Instance } from "../instance/service.js"
 import { PluginSupervisor } from "../plugin/supervisor-service.js"
 import { Shell } from "../shell.js"
 import { ShellResult } from "../shell/result.js"
@@ -33,26 +33,19 @@ import { SessionRevert } from "./revert.js"
 import { SessionSchema } from "./schema.js"
 import { SessionStore } from "./store.js"

-export type Services =
-  | PluginSupervisor.Service
-  | Reference.Service
-  | SessionPrompt.Service
-  | SessionRevert.Service
-  | Shell.Service
-  | Skill.Service
-
 type PromptRequest = SessionPrompt.Input & {
  id?: SessionMessage.ID
  resume?: boolean
 }

 /**
- * Build once in the host Scope: `const sessions = yield* Session.make(servicesFor)`.
+ * Build once in the host Scope: `const sessions = yield* Session.make()`.
 * Use `sessions.forSession(id)` for handles that share host services and reload current state.
 */
-export const make = Effect.fn("Session.make")(function* (servicesFor: (ref: Location.Ref) => Layer.Layer<Services>) {
+export const make = Effect.fn("Session.make")(function* () {
  const bus = yield* Bus.Service
  const store = yield* SessionStore.Service
+  const instances = yield* Instance.Service
  const execution = yield* SessionExecution.Service
  const admission = yield* SessionInbox.Service
  const scope = yield* Scope.Scope
@@ -174,7 +167,7 @@ export const make = Effect.fn("Session.make")(function* (servicesFor: (ref: Loca
              const preparation = yield* SessionPrompt.Service
              const references = yield* Reference.Service
              return { item: yield* preparation.prepare({ sessionID, messageID, input }), references }
-            }).pipe(Effect.provide(servicesFor(session.location))),
+            }).pipe(instances.provide(session)),
          )
          // Commit a staged revert only after preparation succeeds, before admitting new work.
          if (session.revert) yield* SessionRevert.commit(bus, session)
@@ -205,7 +198,7 @@ export const make = Effect.fn("Session.make")(function* (servicesFor: (ref: Loca
        const plugins = yield* PluginSupervisor.Service
        yield* plugins.flush
        return yield* Shell.Service
-      }).pipe(Effect.provide(servicesFor(session.location)))
+      }).pipe(instances.provide(session))
      const started = yield* shell
        .create({
          command: input.command,
@@ -256,7 +249,7 @@ export const make = Effect.fn("Session.make")(function* (servicesFor: (ref: Loca
    input: { id?: SessionMessage.ID; skill: Skill.ID; resume?: boolean },
  ) {
    const session = yield* get(sessionID)
-    const skills = yield* Skill.Service.pipe(Effect.provide(servicesFor(session.location)))
+    const skills = yield* Skill.Service.pipe(instances.provide(session))
    const skill = yield* skills.get(input.skill)
    if (!skill) return yield* new SkillNotFoundError({ skill: input.skill })
    yield* bus.publish(
@@ -355,14 +348,12 @@ export const make = Effect.fn("Session.make")(function* (servicesFor: (ref: Loca
    if (yield* execution.isActive(sessionID)) return yield* new BusyError({ sessionID })
    return yield* SessionRevert.Service.use((revert) =>
      revert.stage({ session, messageID: input.messageID, files: input.files }),
-    ).pipe(Effect.provide(servicesFor(session.location)))
+    ).pipe(instances.provide(session))
  })
  const clear = Effect.fn("Session.revert.clear")(function* (sessionID: SessionSchema.ID) {
    const session = yield* get(sessionID)
    if (yield* execution.isActive(sessionID)) return yield* new BusyError({ sessionID })
-    yield* SessionRevert.Service.use((revert) => revert.clear(session)).pipe(
-      Effect.provide(servicesFor(session.location)),
-    )
+    yield* SessionRevert.Service.use((revert) => revert.clear(session)).pipe(instances.provide(session))
    return yield* execution.wake(sessionID)
  })
  const commit = Effect.fn("Session.revert.commit")(function* (sessionID: SessionSchema.ID) {
@@ -95,7 +95,7 @@ export function convertHTMLToMarkdown(html: string) {
    const remaining = limit - outputBytes
    const next = bytes.byteLength <= remaining ? value : sliceBytes(value, remaining)
    output.push(next)
-    outputBytes += encoder.encode(next).byteLength
+    outputBytes += bytes.byteLength <= remaining ? bytes.byteLength : encoder.encode(next).byteLength
    last = next.at(-1) ?? last
  }
  const appendRaw = (value: string) => {
@@ -22,8 +22,8 @@ const globalLayer = Layer.succeed(Global.Service, Global.Service.of(global))

 const it = testEffect(
  AppNodeBuilder.build(LayerNode.group([Agent.node, Bus.node, Location.node]), [
-    [Global.node, globalLayer],
-    [Location.node, locationLayer],
+    Global.node.replace(globalLayer),
+    Location.node.replace(locationLayer),
  ]) as unknown as Layer.Layer<unknown, never>,
 )

@@ -10,8 +10,6 @@ import { Provider } from "@opencode-ai/core/provider"
 import {
  LLM,
  AIError,
-  CompactionPart,
-  ProviderID,
  HttpContext,
  LLMEvent,
  Message,
@@ -70,31 +68,6 @@ const client = LLMClient.layer.pipe(
  ),
 )

-it.effect("rejects native provider compaction rather than silently dropping replay state", () =>
-  Effect.gen(function* () {
-    const aisdk = yield* AISDK.Service
-    yield* aisdk.hook.sdk((event) => {
-      event.sdk = { languageModel: () => streamModel([]) }
-    })
-    const resolved = yield* aisdk.model(model("@ai-sdk/openai"))
-    const error = yield* compileRequest(
-      LLM.request({
-        model: resolved,
-        messages: [
-          Message.assistant(
-            CompactionPart.make({
-              provider: ProviderID.make("test-provider"),
-              encrypted: "opaque",
-            }),
-          ),
-        ],
-      }),
-    ).pipe(Effect.flip)
-    expect(error.reason._tag).toBe("InvalidRequest")
-    expect(error.message).toContain("cannot replay")
-  }),
-)
-
 it.effect("keys language models by package and flattened overlays", () =>
  Effect.gen(function* () {
    const aisdk = yield* AISDK.Service
@@ -21,7 +21,7 @@ import { testEffect } from "./lib/effect"

 const it = testEffect(
  AppNodeBuilder.build(LayerNode.group([Database.node, Bus.node, SessionProjector.node]), [
-    [Bus.node, Bus.configured({ persist: true })],
+    Bus.node.replace(Bus.configured({ persist: true })),
  ]),
 )
 const a = Location.Ref.make({ directory: AbsolutePath.make("/a") })
@@ -100,12 +100,14 @@ const tail = (bus: Bus.Interface, input: { aggregateID: string; after?: number }

 const it = testEffect(
  AppNodeBuilder.build(LayerNode.group([Database.node, Bus.node, Location.node]), [
-    [Location.node, locationLayer],
-    [Bus.node, Bus.configured({ persist: true })],
+    Location.node.replace(locationLayer),
+    Bus.node.replace(Bus.configured({ persist: true })),
  ]),
 )
 const itWithoutLocation = testEffect(
-  AppNodeBuilder.build(LayerNode.group([Database.node, Bus.node]), [[Bus.node, Bus.configured({ persist: true })]]),
+  AppNodeBuilder.build(LayerNode.group([Database.node, Bus.node]), [
+    Bus.node.replace(Bus.configured({ persist: true })),
+  ]),
 )
 const itWithoutPersistence = testEffect(AppNodeBuilder.build(LayerNode.group([Database.node, Bus.node])))

@@ -631,8 +633,7 @@ describe("Bus", () => {
      const continueRead = yield* Deferred.make<void>()
      let pause = true
      const eventLayer = AppNodeBuilder.build(LayerNode.group([Database.node, Bus.node]), [
-        [
-          Bus.node,
+        Bus.node.replace(
          Bus.configured({
            persist: true,
            beforeAggregateRead: () =>
@@ -640,7 +641,7 @@ describe("Bus", () => {
                ? Deferred.succeed(readStarted, undefined).pipe(Effect.andThen(Deferred.await(continueRead)))
                : Effect.void,
          }),
-        ],
+        ),
      ])

      yield* Effect.gen(function* () {
@@ -1318,7 +1319,7 @@ describe("Bus", () => {
  it.effect("log replays across configured read pages", () =>
    Effect.gen(function* () {
      const eventLayer = AppNodeBuilder.build(LayerNode.group([Database.node, Bus.node]), [
-        [Bus.node, Bus.configured({ persist: true, logReadPageSize: 2 })],
+        Bus.node.replace(Bus.configured({ persist: true, logReadPageSize: 2 })),
      ])

      yield* Effect.gen(function* () {
@@ -1351,8 +1352,7 @@ describe("Bus", () => {
      const releaseRead = yield* Deferred.make<void>()
      const firstRead = yield* Ref.make(true)
      const eventLayer = AppNodeBuilder.build(LayerNode.group([Database.node, Bus.node]), [
-        [
-          Bus.node,
+        Bus.node.replace(
          Bus.configured({
            persist: true,
            beforeAggregateRead: () =>
@@ -1363,7 +1363,7 @@ describe("Bus", () => {
                }),
              ),
          }),
-        ],
+        ),
      ])

      yield* Effect.gen(function* () {
@@ -25,7 +25,7 @@ const locationLayer = Layer.succeed(
 )
 const catalogLayer = AppNodeBuilder.build(
  LayerNode.group([Catalog.node, Bus.node, Credential.node, Integration.node]),
-  [[Location.node, locationLayer]],
+  [Location.node.replace(locationLayer)],
 )
 const it = testEffect(catalogLayer)

@@ -48,7 +48,7 @@ describe("Catalog", () => {
  it.effect("derives availability from active credentials without changing provider state", () => {
    const integrationID = Integration.ID.make("test")
    const localCatalogLayer = Layer.fresh(
-      AppNodeBuilder.build(LayerNode.group([Catalog.node, Credential.node]), [[Location.node, locationLayer]]),
+      AppNodeBuilder.build(LayerNode.group([Catalog.node, Credential.node]), [Location.node.replace(locationLayer)]),
    )

    return Effect.gen(function* () {
@@ -78,7 +78,7 @@ describe("Catalog", () => {
    const providerID = Provider.ID.make("remote")
    const localCatalogLayer = Layer.fresh(
      AppNodeBuilder.build(LayerNode.group([Catalog.node, Credential.node, Integration.node]), [
-        [Location.node, locationLayer],
+        Location.node.replace(locationLayer),
      ]),
    )

@@ -108,7 +108,7 @@ describe("Catalog", () => {
    const providerID = Provider.ID.make("remote")
    const localCatalogLayer = Layer.fresh(
      AppNodeBuilder.build(LayerNode.group([Catalog.node, Credential.node, Integration.node]), [
-        [Location.node, locationLayer],
+        Location.node.replace(locationLayer),
      ]),
    )

@@ -35,7 +35,7 @@ describe("CodeMode", () => {
      Effect.scoped,
      Effect.provide(
        AppNodeBuilder.build(Tool.node, [
-          [Location.node, Location.boundNode({ directory: AbsolutePath.make("/project") })],
+          Location.node.replace(Location.boundNode({ directory: AbsolutePath.make("/project") })),
        ]),
      ),
    ),
@@ -85,7 +85,7 @@ describe("CodeModeInstructions", () => {
      execute: () => Effect.succeed({ output: "zeta" }),
    }
    const layer = AppNodeBuilder.build(Tool.node, [
-      [Location.node, Location.boundNode({ directory: AbsolutePath.make("/project") })],
+      Location.node.replace(Location.boundNode({ directory: AbsolutePath.make("/project") })),
    ])

    return Effect.gen(function* () {
@@ -43,10 +43,10 @@ const it = testEffect(
  AppNodeBuilder.build(
    LayerNode.group([Command.node, Bus.node, FSUtil.node, AppProcess.node, Location.node, ShellSelect.node]),
    [
-      [Mcp.node, emptyMcpLayer],
-      [Config.node, emptyConfigLayer],
-      [Location.node, testLocationLayer],
-      [ShellSelect.node, shellLayer],
+      Mcp.node.replace(emptyMcpLayer),
+      Config.node.replace(emptyConfigLayer),
+      Location.node.replace(testLocationLayer),
+      ShellSelect.node.replace(shellLayer),
    ],
  ),
 )
@@ -340,17 +340,16 @@ describeNative("ConfigCommandPlugin native watcher", () => {
              ShellSelect.node,
            ]),
            [
-              [
-                Location.node,
+              Location.node.replace(
                Layer.succeed(
                  Location.Service,
                  Location.Service.of(location({ directory: AbsolutePath.make(path.join(tmp, "project")) })),
                ),
-              ],
-              [Global.node, Global.layerWith({ config: global, home: path.join(global, "home") })],
-              [ShellSelect.node, shellLayer],
-              [Credential.node, emptyCredentialNode],
-              [WellKnown.node, emptyWellknownNode],
+              ),
+              Global.node.replace(Global.layerWith({ config: global, home: path.join(global, "home") })),
+              ShellSelect.node.replace(shellLayer),
+              Credential.node.replace(emptyCredentialNode),
+              WellKnown.node.replace(emptyWellknownNode),
            ],
          ),
        ),
@@ -40,13 +40,12 @@ const it = testEffect(
  Layer.merge(
    config,
    AppNodeBuilder.build(LayerNode.group([SessionCompaction.node, SessionModelRequest.node, Config.node, Bus.node]), [
-      [
-        llmClient,
+      llmClient.replace(
        Layer.mock(LLMClient.Service)({
          stream: () => Stream.make(LLMEvent.textDelta({ id: "summary", text: "summary" })),
        }),
-      ],
-      [Config.node, config],
+      ),
+      Config.node.replace(config),
    ]),
  ),
 )
@@ -55,12 +55,12 @@ function testLayer(
    ),
  )
  const built = AppNodeBuilder.build(LayerNode.group([Config.node, Bus.node]), [
-    [Config.node, Config.configured(options)],
-    [Location.node, locationLayer],
-    [Global.node, Global.layerWith({ config: globalDirectory, home: path.join(globalDirectory, "home") })],
-    [Credential.node, credentialNode],
-    [WellKnown.node, wellknownNode],
-    [Watcher.node, watcher],
+    Config.node.replace(Config.configured(options)),
+    Location.node.replace(locationLayer),
+    Global.node.replace(Global.layerWith({ config: globalDirectory, home: path.join(globalDirectory, "home") })),
+    Credential.node.replace(credentialNode),
+    WellKnown.node.replace(wellknownNode),
+    Watcher.node.replace(watcher),
  ])
  // Merge the watcher layer by reference so Watcher.Test resolves to the same
  // memoized instance the built graph uses.
@@ -311,16 +311,15 @@ describe("Config", () => {
          }).pipe(
            Effect.provide(
              AppNodeBuilder.build(LayerNode.group([Config.node, Bus.node]), [
-                [
-                  Location.node,
+                Location.node.replace(
                  Layer.succeed(
                    Location.Service,
                    Location.Service.of(location({ directory: AbsolutePath.make(project) })),
                  ),
-                ],
-                [Global.node, Global.layerWith({ config: global, home: path.join(global, "home") })],
-                [Credential.node, emptyCredentialNode],
-                [WellKnown.node, emptyWellknownNode],
+                ),
+                Global.node.replace(Global.layerWith({ config: global, home: path.join(global, "home") })),
+                Credential.node.replace(emptyCredentialNode),
+                WellKnown.node.replace(emptyWellknownNode),
              ]),
            ),
          )
@@ -28,13 +28,13 @@ import { testEffect } from "../lib/effect"

 const it = testEffect(
  AppNodeBuilder.build(LayerNode.group([Database.node, Bus.node, SdkPlugins.node, LocationServiceMap.node]), [
-    [Global.node, tempGlobalLayer],
+    Global.node.replace(tempGlobalLayer),
  ]),
 )
 const staticIt = testEffect(
  AppNodeBuilder.build(LayerNode.group([Database.node, Bus.node, SdkPlugins.node, LocationServiceMap.node]), [
-    [ConfigPluginSource.node, ConfigPluginSource.empty],
-    [Global.node, tempGlobalLayer],
+    ConfigPluginSource.node.replace(ConfigPluginSource.empty),
+    Global.node.replace(tempGlobalLayer),
  ]),
 )
 const refreshNpm = makeGlobalNode({
@@ -65,10 +65,7 @@ const refreshNpm = makeGlobalNode({
 const refreshIt = testEffect(
  AppNodeBuilder.build(
    LayerNode.group([Database.node, Bus.node, SdkPlugins.node, LocationServiceMap.node, Global.node]),
-    [
-      [Global.node, tempGlobalLayer],
-      [Npm.node, refreshNpm],
-    ],
+    [Global.node.replace(tempGlobalLayer), Npm.node.replace(refreshNpm)],
  ),
 )

@@ -86,14 +86,13 @@ const discover = (directory: string, global: string) =>
  }).pipe(
    Effect.provide(
      AppNodeBuilder.build(LayerNode.group([Config.node, Bus.node]), [
-        [
-          Location.node,
+        Location.node.replace(
          Layer.succeed(Location.Service, Location.Service.of(location({ directory: AbsolutePath.make(directory) }))),
-        ],
-        [Global.node, Global.layerWith({ config: global, home: path.join(global, "home") })],
-        [Credential.node, emptyCredentialNode],
-        [WellKnown.node, emptyWellknownNode],
-        [Watcher.node, Watcher.testLayer],
+        ),
+        Global.node.replace(Global.layerWith({ config: global, home: path.join(global, "home") })),
+        Credential.node.replace(emptyCredentialNode),
+        WellKnown.node.replace(emptyWellknownNode),
+        Watcher.node.replace(Watcher.testLayer),
      ]),
    ),
  )
@@ -51,8 +51,8 @@ describe("ConfigSnapshotPlugin.Plugin", () => {
          }).pipe(
            Effect.provide(
              AppNodeBuilder.build(Snapshot.node, [
-                [Location.node, Location.boundNode(Location.Ref.make({ directory: AbsolutePath.make(project) }))],
-                [Global.node, Global.layerWith({ data: tmp.path, config: path.join(tmp.path, "config") })],
+                Location.node.replace(Location.boundNode(Location.Ref.make({ directory: AbsolutePath.make(project) }))),
+                Global.node.replace(Global.layerWith({ data: tmp.path, config: path.join(tmp.path, "config") })),
              ]),
            ),
          )
@@ -44,7 +44,9 @@ describe("ConfigToolOutputPlugin.Plugin", () => {
          }
          yield* Effect.die(new Error("Timed out waiting for tool output config reload"))
        }).pipe(
-          Effect.provide(AppNodeBuilder.build(ToolOutput.node, [[Global.node, Global.layerWith({ data: tmp.path })]])),
+          Effect.provide(
+            AppNodeBuilder.build(ToolOutput.node, [Global.node.replace(Global.layerWith({ data: tmp.path }))]),
+          ),
        ),
      (tmp) => Effect.promise(() => tmp[Symbol.asyncDispose]()),
    ).pipe(
@@ -180,6 +180,37 @@ describe("cross-spawn spawner", () => {
  })

  describe("combined output (all)", () => {
+    for (const output of ["stdout", "stderr", "all"] as const) {
+      fx.live(
+        `captures ${output} when reading starts after process exit`,
+        Effect.gen(function* () {
+          const handle = yield* js('process.stdout.write("stdout\\n"); process.stderr.write("stderr\\n")')
+          expect(yield* handle.exitCode).toBe(ChildProcessSpawner.ExitCode(0))
+          // Let exit callbacks finish before attaching a reader; the handle scope remains open.
+          yield* Effect.promise(() => new Promise<void>((resolve) => setImmediate(resolve)))
+          expect((yield* decodeByteStream(handle[output])).split("\n").toSorted()).toEqual(
+            output === "all" ? ["stderr", "stdout"] : [output],
+          )
+        }).pipe(Effect.timeout("3 seconds")),
+      )
+    }
+
+    fx.live(
+      "drains output larger than the capture buffers",
+      Effect.gen(function* () {
+        const text = "x".repeat(1024 * 1024)
+        const handle = yield* js(
+          `const text = "x".repeat(${text.length}); process.stdout.write(text); process.stderr.write(text)`,
+        )
+        const [stdout, stderr] = yield* Effect.all([decodeByteStream(handle.stdout), decodeByteStream(handle.stderr)], {
+          concurrency: 2,
+        })
+        expect(stdout).toBe(text)
+        expect(stderr).toBe(text)
+        expect(yield* handle.exitCode).toBe(ChildProcessSpawner.ExitCode(0))
+      }).pipe(Effect.timeout("3 seconds")),
+    )
+
    fx.effect(
      "captures stdout via .all when no stderr",
      Effect.gen(function* () {
@@ -217,6 +248,63 @@ describe("cross-spawn spawner", () => {
  })

  describe("process control", () => {
+    fx.live(
+      "reports exit without waiting for unread stdout",
+      Effect.gen(function* () {
+        const handle = yield* js("process.stdout.write(Buffer.alloc(1024 * 1024)); process.exit(0)")
+        expect(yield* Effect.promise(() => gone(Number(handle.pid)))).toBe(true)
+        expect(yield* handle.exitCode.pipe(Effect.timeout("500 millis"))).toBe(ChildProcessSpawner.ExitCode(0))
+        expect(yield* handle.isRunning).toBe(false)
+      }),
+    )
+
+    fx.live(
+      "releases a process with unread buffered stdout",
+      Effect.gen(function* () {
+        const pid = yield* Effect.scoped(
+          Effect.gen(function* () {
+            const handle = yield* js(
+              'process.stdout.write("x".repeat(1024 * 1024)); process.stderr.write("ready"); setInterval(() => {}, 10_000)',
+              { forceKillAfter: 100 },
+            )
+            expect(yield* decodeByteStream(handle.stderr.pipe(Stream.take(1)))).toBe("ready")
+            return Number(handle.pid)
+          }),
+        )
+        expect(yield* Effect.promise(() => gone(pid))).toBe(true)
+      }).pipe(Effect.timeout("3 seconds")),
+    )
+
+    // Node puts non-detached Windows children in a kill-on-parent-exit job; this guards POSIX group cleanup.
+    const groupTest = process.platform === "win32" ? fx.live.skip : fx.live
+    groupTest(
+      "preserves successful descendants when an exit-only scope closes",
+      Effect.gen(function* () {
+        const tmp = yield* Effect.acquireRelease(
+          Effect.promise(() => tmpdir()),
+          (tmp) => Effect.promise(() => tmp[Symbol.asyncDispose]()),
+        )
+        const pidFile = path.join(tmp.path, "child.pid")
+        yield* Effect.addFinalizer(() =>
+          Effect.tryPromise(async () => process.kill(Number(await fs.readFile(pidFile, "utf8")), "SIGKILL")).pipe(
+            Effect.ignore,
+          ),
+        )
+        yield* Effect.scoped(
+          Effect.gen(function* () {
+            // This fixture's child shares the process group and holds stdio after the parent exits on stdin EOF.
+            const handle = yield* ChildProcess.make(
+              "node",
+              [path.join(import.meta.dir, "../fixture/held-stdio.cjs"), "mcp", pidFile],
+              { stdin: "ignore", forceKillAfter: 100 },
+            )
+            expect(yield* handle.exitCode).toBe(ChildProcessSpawner.ExitCode(0))
+          }),
+        )
+        expect(alive(Number(yield* Effect.promise(() => fs.readFile(pidFile, "utf8"))))).toBe(true)
+      }).pipe(Effect.timeout("3 seconds")),
+    )
+
    for (const mode of ["exit", "SIGKILL"] as const) {
      const test = mode === "SIGKILL" && process.platform === "win32" ? fx.live.skip : fx.live
      test(
@@ -13,131 +13,218 @@ class OtherError {
  readonly _tag = "OtherError"
 }

-const tags = LayerNode.tags({ app: [] })
-const make = tags.make("app")
-const build = <A, E>(root: LayerNode.Node<A, E, any>) => LayerNode.compile(root) as Layer.Layer<A, E>
-const aLayer = Layer.succeed(A, A.of({}))
-const bLayer = Layer.effect(B, Effect.as(A, B.of({})))
-const cLayer = Layer.effect(
-  C,
-  Effect.gen(function* () {
-    yield* A
-    yield* B
-    return C.of({})
-  }),
-)
-const failingA = Layer.effect(A, Effect.fail(new LayerError()))
-const a = make({ service: A, layer: aLayer, deps: [] })
-const b = make({ service: B, layer: bLayer, deps: [a] })
-const c = make({ service: C, layer: cLayer, deps: [a, b] })
-const failing = make({ service: A, layer: failingA, deps: [] })
-const dependent = make({ service: B, layer: bLayer, deps: [failing] })
-const inputA = LayerNode.unbound(A, tags.values.app)
-const inputDependent = make({ service: B, layer: bLayer, deps: [inputA] })
+// Keep intentionally invalid expressions out of runtime execution.
+const contracts = (tag: LayerNode.Tag<"app"> | LayerNode.Tag<"other">, flag: boolean) => {
+  const tags = LayerNode.tags({ app: [] })
+  const make = tags.make("app")
+  const aLayer = Layer.succeed(A, A.of({}))
+  const bLayer = Layer.effect(B, Effect.as(A, B.of({})))
+  const cLayer = Layer.effect(
+    C,
+    Effect.gen(function* () {
+      yield* A
+      yield* B
+      return C.of({})
+    }),
+  )
+  const a = make({ service: A, layer: aLayer, deps: [] })
+  const b = make({ service: B, layer: bLayer, deps: [a] })
+  const c = make({ service: C, layer: cLayer, deps: [a, b] })
+  const ab = make({ name: "a-and-b", layer: Layer.mergeAll(aLayer, Layer.succeed(B, {})), deps: [] })
+  const failing = make({ service: A, layer: Layer.effect(A, Effect.fail(new LayerError())), deps: [] })
+  const dependent = make({ service: B, layer: bLayer, deps: [failing] })
+  const inputA = LayerNode.unbound(A, tags.values.app)
+  const group = LayerNode.group([a, b])

-make({ name: "manual-a", layer: aLayer, deps: [] })
+  make({ name: "manual-a", layer: aLayer, deps: [] })
+  // @ts-expect-error A node must have a service or name
+  make({ layer: aLayer, deps: [] })
+  // @ts-expect-error Service and name are mutually exclusive
+  make({ service: A, name: "a", layer: aLayer, deps: [] })
+  // @ts-expect-error An explicit tagged contract requires a corresponding runtime tag
+  LayerNode.make<typeof aLayer, readonly [], typeof tags.values.app>({ service: A, layer: aLayer, deps: [] })
+  // @ts-expect-error B requires A
+  make({ service: B, layer: bLayer, deps: [] })
+  // @ts-expect-error C requires A and B
+  make({ service: C, layer: cLayer, deps: [a] })
+  const erasedLayer: Layer.Any = bLayer
+  // @ts-expect-error Erasing a Layer's contract cannot hide its inputs and errors
+  make({ service: B, layer: erasedLayer, deps: [] })

-// @ts-expect-error A node must have a service or name
-make({ layer: aLayer, deps: [] })
+  LayerNode.compile(c) satisfies Layer.Layer<C, never, never>
+  LayerNode.compile(dependent) satisfies Layer.Layer<B, LayerError, never>
+  LayerNode.compile(group) satisfies Layer.Layer<A | B, never, never>
+  LayerNode.compile(LayerNode.group([])) satisfies Layer.Layer<never>
+  // @ts-expect-error An empty graph cannot supply arbitrary services
+  LayerNode.compile(LayerNode.group([])) satisfies Layer.Layer<A>
+  LayerNode.compile(inputA, { replacements: [inputA.replace(a)] }) satisfies Layer.Layer<A, never, never>
+  // @ts-expect-error A is a private dependency, not a root output
+  LayerNode.compile(c) satisfies Layer.Layer<A | C>
+  // @ts-expect-error Dependency failures are not erased
+  LayerNode.compile(dependent) satisfies Layer.Layer<B>

-// @ts-expect-error Service and name are mutually exclusive
-make({ service: A, name: "a", layer: aLayer, deps: [] })
-
-// @ts-expect-error B requires A
-make({ service: B, layer: bLayer, deps: [] })
-
-// @ts-expect-error C requires A and B
-make({ service: C, layer: cLayer, deps: [a] })
-
-const closed = build(LayerNode.group([c]))
-const closedWithError = build(LayerNode.group([dependent]))
-const checkClosed: Layer.Layer<C, never, never> = closed
-const checkError: Layer.Layer<B, LayerError, never> = closedWithError
-void checkClosed
-void checkError
-
-LayerNode.compile(a, [[a, Layer.succeed(A, A.of({}))]])
-LayerNode.compile(a, [[a, make({ service: A, layer: Layer.succeed(A, A.of({})), deps: [] })]])
-
-// @ts-expect-error Replacement must provide A
-LayerNode.compile(a, [[a, Layer.succeed(B, B.of({}))]])
-
-// @ts-expect-error Node replacement must provide A
-const invalidNodeReplacement = () => LayerNode.compile(a, [[a, b]])
-void invalidNodeReplacement
-
-// @ts-expect-error Replacement cannot introduce a new error
-LayerNode.compile(a, [[a, Layer.effect(A, Effect.fail(new OtherError()))]])
-
-const invalidNodeErrorReplacement = () =>
+  const replacements: LayerNode.Replacements = [a.replace(aLayer), a.replace(ab), failing.replace(a)]
+  const replacement: LayerNode.Replacement = a.replace(Layer.mergeAll(aLayer, Layer.succeed(B, {})))
+  LayerNode.compile(a, { replacements: [...replacements, replacement] })
+  inputA.replace(a)
+  a.replace(a)
+  // @ts-expect-error Closed layer replacements must provide every source output
+  ab.replace(aLayer)
+  // @ts-expect-error Node replacements must provide every source output
+  ab.replace(a)
+  // @ts-expect-error Replacement must provide A
+  a.replace(Layer.succeed(B, {}))
+  // @ts-expect-error Node replacement must provide A
+  a.replace(b)
+  // @ts-expect-error Raw layers with inputs are not closed
+  a.replace(Layer.effect(A, Effect.as(B, A.of({}))))
+  // @ts-expect-error Replacement cannot introduce a new error
+  a.replace(Layer.effect(A, Effect.fail(new OtherError())))
  // @ts-expect-error Node replacement cannot introduce a new error
-  LayerNode.compile(a, [[a, make({ service: A, layer: Layer.effect(A, Effect.fail(new OtherError())), deps: [] })]])
-void invalidNodeErrorReplacement
+  a.replace(failing)
+  // @ts-expect-error Existing errors do not authorize unrelated replacement errors
+  failing.replace(Layer.effect(A, Effect.fail(new OtherError())))
+  // @ts-expect-error Every alternative of a node replacement must supply A
+  a.replace(flag ? a : b)
+  // @ts-expect-error Every alternative of a raw-layer replacement must supply A
+  a.replace(flag ? aLayer : Layer.succeed(B, {}))
+  // @ts-expect-error A valid alternative cannot hide a new error in another alternative
+  a.replace(flag ? a : failing)
+  a.replace(flag ? a : ab)
+  failing.replace(flag ? a : failing)
+  // @ts-expect-error Storing replacements must not erase their validation
+  const invalidStored: LayerNode.Replacements = [a.replace(b)]
+  // @ts-expect-error Raw tuples cannot be stored as opaque replacements
+  const rawStored: LayerNode.Replacements = [[a, aLayer]]
+  // @ts-expect-error Raw tuples cannot be supplied to compile
+  LayerNode.compile(a, { replacements: [[a, aLayer]] })
+  // @ts-expect-error Replacements are not structurally forgeable
+  const forged: LayerNode.Replacement = { source: a, target: a }
+  // @ts-expect-error Groups are not replaceable nodes
+  group.replace(a)
+  // @ts-expect-error Groups cannot be replacement targets
+  a.replace(group)
+  // @ts-expect-error Groups cannot be widened to nodes
+  const groupNode: LayerNode.Node<A | B, never, typeof tags.values.app> = group
+  // @ts-expect-error Graphs are opaque
+  const forgedGraph: LayerNode.Graph<A> = { name: "a" }

-class TagA extends Context.Service<TagA, {}>()("test/TagA") {}
-class TagB extends Context.Service<TagB, {}>()("test/TagB") {}
-class TagC extends Context.Service<TagC, {}>()("test/TagC") {}
+  const aContract: LayerNode.Node<A, never, typeof tags.values.app> = a
+  aContract.replace(aLayer)
+  // @ts-expect-error A method cannot be rebound to a declaration with a stronger contract
+  a.replace.call(ab, aLayer)
+  const detached = a.replace
+  // @ts-expect-error Replacement authority requires its checked receiver
+  detached(aLayer)
+  // @ts-expect-error Output narrowing cannot forget B before replacement
+  const narrowedOutput: LayerNode.Node<A, never, typeof tags.values.app> = ab
+  // @ts-expect-error Output widening cannot add B before replacement
+  const widenedOutput: LayerNode.Node<A | B, never, typeof tags.values.app> = a
+  // @ts-expect-error Error widening cannot authorize a new replacement error
+  const widenedError: LayerNode.Node<A, LayerError, typeof tags.values.app> = a
+  // @ts-expect-error Error narrowing cannot forget an existing failure
+  const narrowedError: LayerNode.Node<A, never, typeof tags.values.app> = failing
+  // @ts-expect-error Tag widening cannot authorize replacement across tags
+  const widenedTag: LayerNode.Node<A, never, LayerNode.Tag | undefined> = a
+  const unionTag = LayerNode.unbound(A, tag)
+  // @ts-expect-error Tag narrowing cannot forget a possible tag
+  const narrowedTag: LayerNode.Node<A, never, typeof tags.values.app> = unionTag

-const scopedTags = LayerNode.tags({ request: ["global"], global: [] })
-const request = scopedTags.make("request")
-const global = scopedTags.make("global")
-const globalA = global({ service: TagA, layer: Layer.succeed(TagA, TagA.of({})), deps: [] })
-const requestA = request({ service: TagA, layer: Layer.succeed(TagA, TagA.of({})), deps: [] })
-const requestB = request({ service: TagB, layer: Layer.succeed(TagB, TagB.of({})), deps: [] })
-const tagBLayer = Layer.effect(TagB, Effect.as(TagA, TagB.of({})))
-const tagCLayer = Layer.effect(
-  TagC,
-  Effect.gen(function* () {
-    yield* TagA
-    yield* TagB
-    return TagC.of({})
-  }),
-)
+  const outputProjection: LayerNode.Graph<A, never, typeof tags.values.app> = group
+  // @ts-expect-error Graph output projection cannot invent a service
+  const widenedGraph: LayerNode.Graph<A | B, never, typeof tags.values.app> = a
+  // @ts-expect-error A projected Graph has no replacement authority
+  outputProjection.replace(aLayer)

-request({ service: TagB, layer: tagBLayer, deps: [globalA] })
-request({ service: TagC, layer: tagCLayer, deps: [globalA, requestB] })
-request({ service: TagC, layer: tagCLayer, deps: [LayerNode.group([globalA, requestB])] })
+  const choice = flag ? a : b
+  // @ts-expect-error Choosing one dependency does not provide both services
+  make({ service: C, layer: cLayer, deps: [choice] })
+  // @ts-expect-error A conditional root promises only outputs present in every alternative
+  LayerNode.compile(LayerNode.group([choice])) satisfies Layer.Layer<A | B>
+  const conditional = make({ name: "conditional", layer: flag ? aLayer : Layer.succeed(B, {}), deps: [] })
+  LayerNode.compile(conditional) satisfies Layer.Layer<never>
+  // @ts-expect-error A conditional implementation does not acquire both branches
+  LayerNode.compile(conditional) satisfies Layer.Layer<A | B>
+  LayerNode.compile(LayerNode.group([flag ? a : ab])) satisfies Layer.Layer<A>
+  const dynamic: Array<typeof a> = []
+  // @ts-expect-error An unbounded array may contain no roots
+  LayerNode.compile(LayerNode.group(dynamic)) satisfies Layer.Layer<A>

-// @ts-expect-error Tag configuration can only reference declared tags
-LayerNode.tags({ request: ["missing"], global: [] })
+  const decorated = b.mapLayer((layer) => layer.pipe(Layer.tap(() => Effect.void)))
+  LayerNode.compile(decorated) satisfies Layer.Layer<B>
+  b.replace(decorated)
+  // @ts-expect-error A layer mapper cannot be rebound to a weaker declaration
+  ab.mapLayer.call(a, (layer) => layer)
+  // @ts-expect-error mapLayer cannot add an input requirement
+  b.mapLayer((layer) => layer.pipe(Layer.tap(() => C)))
+  // @ts-expect-error mapLayer cannot grow the error channel
+  b.mapLayer((layer) => layer.pipe(Layer.tap(() => Effect.fail(new OtherError()))))
+  // @ts-expect-error mapLayer cannot drop an output
+  ab.mapLayer(() => aLayer)
+  // @ts-expect-error Unbound declarations have no implementation to map
+  inputA.mapLayer((layer: Layer.Layer<A>) => layer)

-// @ts-expect-error An unrelated dependency cannot satisfy TagA
-request({ service: TagB, layer: tagBLayer, deps: [requestB] })
+  const scopedTags = LayerNode.tags({ request: ["global"], global: [] })
+  const request = scopedTags.make("request")
+  const global = scopedTags.make("global")
+  const globalA = global({ service: A, layer: aLayer, deps: [] })
+  const requestA = request({ service: A, layer: aLayer, deps: [] })
+  const requestB = request({ service: B, layer: Layer.succeed(B, {}), deps: [] })
+  request({ service: B, layer: bLayer, deps: [globalA] })
+  request({ service: C, layer: cLayer, deps: [globalA, requestB] })
+  request({ service: C, layer: cLayer, deps: [LayerNode.group([globalA, requestB])] })
+  LayerNode.compile(LayerNode.group([globalA, requestB]), { shared: scopedTags.values.global }) satisfies Layer.Layer<
+    A | B
+  >
+  // @ts-expect-error Tag configuration can only reference declared tags
+  LayerNode.tags({ request: ["missing"], global: [] })
+  // @ts-expect-error Shared tags must be branded
+  LayerNode.compile(globalA, { shared: "global" })
+  // @ts-expect-error Replacement targets must keep the source tag
+  globalA.replace(requestA)
+  // @ts-expect-error Replacement targets must keep the source tag in either direction
+  requestA.replace(globalA)
+  // @ts-expect-error Every alternative must keep the source tag
+  globalA.replace(flag ? globalA : requestA)
+  // @ts-expect-error Providing only A leaves B missing
+  request({ service: C, layer: cLayer, deps: [globalA] })
+  // @ts-expect-error Providing only B leaves A missing
+  request({ service: C, layer: cLayer, deps: [requestB] })
+  // @ts-expect-error Duplicate A providers still leave B missing
+  request({ service: C, layer: cLayer, deps: [globalA, requestA] })
+  // @ts-expect-error A group with only A still leaves B missing
+  request({ service: C, layer: cLayer, deps: [LayerNode.group([globalA])] })
+  // @ts-expect-error Global cannot depend on request
+  global({ service: B, layer: bLayer, deps: [requestA] })
+  // @ts-expect-error Groups preserve their child tags
+  global({ service: B, layer: bLayer, deps: [LayerNode.group([requestA])] })

-// @ts-expect-error Providing only TagA leaves TagB missing
-request({ service: TagC, layer: tagCLayer, deps: [globalA] })
+  const globalScopedA = makeGlobalNode({ service: A, layer: aLayer, deps: [] })
+  const locationScopedA = makeLocationNode({ service: A, layer: aLayer, deps: [] })
+  makeGlobalNode({ service: B, layer: bLayer, deps: [globalScopedA] })
+  makeLocationNode({ service: B, layer: bLayer, deps: [globalScopedA] })
+  makeLocationNode({ service: B, layer: bLayer, deps: [locationScopedA] })
+  // @ts-expect-error Global nodes cannot depend on location nodes
+  makeGlobalNode({ service: B, layer: bLayer, deps: [locationScopedA] })
+  // @ts-expect-error B requires A
+  makeLocationNode({ service: B, layer: bLayer, deps: [] })

-// @ts-expect-error Providing only TagB leaves TagA missing
-request({ service: TagC, layer: tagCLayer, deps: [requestB] })
+  void [
+    invalidStored,
+    rawStored,
+    forged,
+    groupNode,
+    forgedGraph,
+    narrowedOutput,
+    widenedOutput,
+    widenedError,
+    narrowedError,
+    widenedTag,
+    narrowedTag,
+    widenedGraph,
+  ]
+}

-// @ts-expect-error Duplicate TagA providers still leave TagB missing
-request({ service: TagC, layer: tagCLayer, deps: [globalA, requestA] })
-
-// @ts-expect-error A group with only TagA still leaves TagB missing
-request({ service: TagC, layer: tagCLayer, deps: [LayerNode.group([globalA])] })
-
-// @ts-expect-error Global cannot depend on request
-global({ service: TagB, layer: tagBLayer, deps: [requestA] })
-
-// @ts-expect-error Groups preserve their child tags
-global({ service: TagB, layer: tagBLayer, deps: [LayerNode.group([requestA])] })
-
-class ScopedA extends Context.Service<ScopedA, {}>()("test/ScopedA") {}
-class ScopedB extends Context.Service<ScopedB, {}>()("test/ScopedB") {}
-
-const scopedA = Layer.succeed(ScopedA, ScopedA.of({}))
-const scopedB = Layer.effect(ScopedB, Effect.as(ScopedA, ScopedB.of({})))
-const globalScopedA = makeGlobalNode({ service: ScopedA, layer: scopedA, deps: [] })
-const locationScopedA = makeLocationNode({ service: ScopedA, layer: scopedA, deps: [] })
-
-makeGlobalNode({ service: ScopedB, layer: scopedB, deps: [globalScopedA] })
-makeLocationNode({ service: ScopedB, layer: scopedB, deps: [globalScopedA] })
-makeLocationNode({ service: ScopedB, layer: scopedB, deps: [locationScopedA] })
-
-// @ts-expect-error Global nodes cannot depend on location nodes
-makeGlobalNode({ service: ScopedB, layer: scopedB, deps: [locationScopedA] })
-
-// @ts-expect-error ScopedB requires ScopedA
-makeLocationNode({ service: ScopedB, layer: scopedB, deps: [] })
-
-test("type exploration compiles", () => {})
+test("layer node type contracts compile", () => {
+  void contracts
+})
@@ -1,19 +1,21 @@
 import { describe, expect, test } from "bun:test"
-import { Context, Effect, Layer } from "effect"
+import { Context, Deferred, Duration, Effect, Fiber, Layer, LayerMap, Option } from "effect"
 import { LayerNode } from "@opencode-ai/util/effect/layer-node"
+import { testEffect } from "../../lib/effect"

 class Value extends Context.Service<Value, { readonly value: string }>()("test/LayerNodeValue") {}
 class Greeting extends Context.Service<Greeting, { readonly value: string }>()("test/LayerNodeGreeting") {}
 class Left extends Context.Service<Left, { readonly value: string }>()("test/LayerNodeLeft") {}
 class Right extends Context.Service<Right, { readonly value: string }>()("test/LayerNodeRight") {}
-class Database extends Context.Service<Database, { readonly name: string }>()("test/GraphDatabase") {}
-class Users extends Context.Service<Users, { readonly list: Effect.Effect<string[]> }>()("test/GraphUsers") {}
-class App extends Context.Service<App, { readonly run: Effect.Effect<string[]> }>()("test/GraphApp") {}
+class Memo extends Context.Service<Memo, Layer.MemoMap>()("test/LayerNodeMemo") {}
+class Support extends Context.Service<Support, {}>()("test/LayerNodeSupport") {}
+class Locations extends Context.Service<Locations, LayerMap.LayerMap<string, Value | Right, "failed location">>()(
+  "test/LayerNodeLocations",
+) {}

+const it = testEffect(Layer.empty)
 const tags = LayerNode.tags({ app: [] })
 const make = tags.make("app")
-const build = <A, E>(root: LayerNode.Node<A, E, any>, replacements?: readonly LayerNode.Replacement[]) =>
-  LayerNode.compile(root, replacements) as Layer.Layer<A, E>
 const valueLayer = Layer.succeed(Value, Value.of({ value: "production" }))
 const greetingLayer = Layer.effect(
  Greeting,
@@ -23,240 +25,443 @@ const value = make({ service: Value, layer: valueLayer, deps: [] })
 const greeting = make({ service: Greeting, layer: greetingLayer, deps: [value] })

 describe("layer node", () => {
-  test("builds an untagged graph", async () => {
-    const value = LayerNode.make({ service: Value, layer: valueLayer, deps: [] })
-    const greeting = LayerNode.make({ service: Greeting, layer: greetingLayer, deps: [value] })
-    const program = Effect.map(Greeting, (item) => item.value).pipe(
-      Effect.provide(LayerNode.compile(LayerNode.group([greeting]))),
+  it.effect("builds an untagged graph", () =>
+    Effect.gen(function* () {
+      const value = LayerNode.make({ service: Value, layer: valueLayer, deps: [] })
+      const greeting = LayerNode.make({ service: Greeting, layer: greetingLayer, deps: [value] })
+      const result = yield* Greeting.pipe(Effect.provide(LayerNode.compile(LayerNode.group([greeting]))))
+      expect(result.value).toBe("hello production")
+    }),
+  )
+
+  it.effect("exposes roots but hides transitive dependencies", () =>
+    Effect.gen(function* () {
+      const context = yield* Layer.build(LayerNode.compile(LayerNode.group([greeting])))
+      expect(Context.get(context, Greeting).value).toBe("hello production")
+      expect(Option.isNone(Context.getOption(context, Value))).toBe(true)
+    }),
+  )
+
+  it.effect("replaces exact declarations, not sibling names or native layer identities", () =>
+    Effect.gen(function* () {
+      const sibling = make({ service: Value, layer: valueLayer, deps: [] })
+      const target = make({ name: "different-name", layer: Layer.succeed(Value, { value: "replaced" }), deps: [] })
+      const left = make({
+        service: Left,
+        layer: Layer.effect(
+          Left,
+          Effect.map(Value, (item) => Left.of({ value: item.value })),
+        ),
+        deps: [value],
+      })
+      const right = make({
+        service: Right,
+        layer: Layer.effect(
+          Right,
+          Effect.map(Value, (item) => Right.of({ value: item.value })),
+        ),
+        deps: [sibling],
+      })
+      const context = yield* Layer.build(
+        LayerNode.compile(LayerNode.group([left, right]), { replacements: [value.replace(target)] }),
+      )
+      expect(Context.get(context, Left).value).toBe("replaced")
+      expect(Context.get(context, Right).value).toBe("production")
+    }),
+  )
+
+  it.effect("requires reachable unbound nodes to be replaced", () =>
+    Effect.gen(function* () {
+      const unbound = LayerNode.unbound(Value, tags.values.app)
+      const root = make({ service: Greeting, layer: greetingLayer, deps: [unbound] })
+      expect(() => LayerNode.compile(root)).toThrow("Unbound layer node: test/LayerNodeValue")
+      const result = yield* Greeting.pipe(
+        Effect.provide(LayerNode.compile(root, { replacements: [unbound.replace(value)] })),
+      )
+      expect(result.value).toBe("hello production")
+    }),
+  )
+
+  it.effect("replaces every use of a declaration with a stored closed-layer replacement", () =>
+    Effect.gen(function* () {
+      const replacements: LayerNode.Replacements = [value.replace(Layer.succeed(Value, { value: "replacement" }))]
+      const right = make({
+        service: Right,
+        layer: Layer.effect(
+          Right,
+          Effect.map(Value, (item) => Right.of({ value: item.value })),
+        ),
+        deps: [value],
+      })
+      const context = yield* Layer.build(LayerNode.compile(LayerNode.group([greeting, right]), { replacements }))
+      expect(Context.get(context, Greeting).value).toBe("hello replacement")
+      expect(Context.get(context, Right).value).toBe("replacement")
+    }),
+  )
+
+  it.effect("uses the last replacement and ignores unreachable unbound defaults and cycles", () =>
+    Effect.gen(function* () {
+      const unbound = LayerNode.unbound(Value, tags.values.app)
+      const unused = make({ service: Value, layer: valueLayer, deps: [] })
+      const result = yield* Greeting.pipe(
+        Effect.provide(
+          LayerNode.compile(greeting, {
+            replacements: [
+              value.replace(unbound),
+              unbound.replace(unused),
+              unused.replace(unbound),
+              value.replace(Layer.succeed(Value, { value: "last" })),
+            ],
+          }),
+        ),
+      )
+      expect(result.value).toBe("hello last")
+    }),
+  )
+
+  it.effect("resolves target chains independently of replacement order and treats self-replacement as identity", () =>
+    Effect.gen(function* () {
+      const middle = make({ service: Value, layer: Layer.succeed(Value, { value: "middle" }), deps: [] })
+      const target = make({ service: Value, layer: Layer.succeed(Value, { value: "target" }), deps: [] })
+      const result = yield* Greeting.pipe(
+        Effect.provide(
+          LayerNode.compile(greeting, {
+            replacements: [target.replace(target), middle.replace(target), value.replace(middle)],
+          }),
+        ),
+      )
+      expect(result.value).toBe("hello target")
+    }),
+  )
+
+  test("rejects reachable replacement and dependency cycles", () => {
+    const other = make({ service: Value, layer: valueLayer, deps: [] })
+    expect(() => LayerNode.compile(greeting, { replacements: [value.replace(other), other.replace(value)] })).toThrow(
+      "Cycle detected in layer graph",
    )
-    expect(await Effect.runPromise(program)).toBe("hello production")
-  })
-
-  test("builds a dependency graph", async () => {
-    const program = Effect.map(Greeting, (item) => item.value).pipe(Effect.provide(build(LayerNode.group([greeting]))))
-    expect(await Effect.runPromise(program)).toBe("hello production")
-  })
-
-  test("exposes roots but hides transitive dependencies", () => {
-    const layer = build(LayerNode.group([greeting]))
-    const check: Layer.Layer<Greeting> = layer
-    void check
-  })
-
-  test("preserves branch-specific implementations across roots", async () => {
-    const firstValue = make({ service: Value, layer: Layer.succeed(Value, Value.of({ value: "first" })), deps: [] })
-    const secondValue = make({ service: Value, layer: Layer.succeed(Value, Value.of({ value: "second" })), deps: [] })
-    const leftLayer = Layer.effect(
-      Left,
-      Effect.map(Value, (item) => Left.of({ value: item.value })),
-    )
-    const rightLayer = Layer.effect(
-      Right,
-      Effect.map(Value, (item) => Right.of({ value: item.value })),
-    )
-    const left = make({ service: Left, layer: leftLayer, deps: [firstValue] })
-    const right = make({ service: Right, layer: rightLayer, deps: [secondValue] })
-    const layer = build(LayerNode.group([left, right]))
-    const program = Effect.gen(function* () {
-      return [(yield* Left).value, (yield* Right).value]
-    }).pipe(Effect.provide(layer))
-    expect(await Effect.runPromise(program)).toEqual(["first", "second"])
-  })
-
-  test("requires unbound nodes to be replaced before compilation", async () => {
-    const unbound = LayerNode.unbound(Value, tags.values.app)
-    const greeting = make({ service: Greeting, layer: greetingLayer, deps: [unbound] })
-    const tree = LayerNode.group([greeting])
-    expect(() => LayerNode.compile(tree)).toThrow("Unbound layer node: test/LayerNodeValue")
-    const layer = LayerNode.compile(tree, [[unbound, value]]) as Layer.Layer<Greeting>
-    const program = Effect.map(Greeting, (item) => item.value).pipe(Effect.provide(layer))
-    expect(await Effect.runPromise(program)).toBe("hello production")
-  })
-
-  test("replaces a node with a closed layer", async () => {
-    const replacement = Layer.succeed(Value, Value.of({ value: "simulation" }))
-    const program = Effect.map(Greeting, (item) => item.value).pipe(
-      Effect.provide(build(LayerNode.group([greeting]), [[value, replacement]])),
-    )
-    expect(await Effect.runPromise(program)).toBe("hello simulation")
-  })
-
-  test("replaces every use of the same layer", async () => {
-    const leftLayer = Layer.effect(
-      Left,
-      Effect.map(Value, (item) => Left.of({ value: item.value })),
-    )
-    const rightLayer = Layer.effect(
-      Right,
-      Effect.map(Value, (item) => Right.of({ value: item.value })),
-    )
-    const left = make({ service: Left, layer: leftLayer, deps: [value] })
-    const right = make({ service: Right, layer: rightLayer, deps: [value] })
-    const replacement = Layer.succeed(Value, Value.of({ value: "replaced" }))
-    const layer = build(LayerNode.group([left, right]), [[value, replacement]])
-    const program = Effect.gen(function* () {
-      return [(yield* Left).value, (yield* Right).value]
-    }).pipe(Effect.provide(layer))
-    expect(await Effect.runPromise(program)).toEqual(["replaced", "replaced"])
-  })
-
-  test("does not acquire an unused replacement", async () => {
-    let acquisitions = 0
-    const other = make({ service: Left, layer: Layer.succeed(Left, Left.of({ value: "other" })), deps: [] })
-    const replacement = Layer.effect(
-      Left,
-      Effect.sync(() => {
-        acquisitions++
-        return Left.of({ value: "replacement" })
-      }),
-    )
-    await Effect.runPromise(
-      Effect.map(Greeting, (item) => item.value).pipe(
-        Effect.provide(build(LayerNode.group([greeting]), [[other, replacement]])),
-      ),
-    )
-    expect(acquisitions).toBe(0)
-  })
-
-  test("replaces a node without acquiring its dependencies", async () => {
-    let acquisitions = 0
-    const dependencyLayer = Layer.effect(
-      Value,
-      Effect.sync(() => {
-        acquisitions++
-        return Value.of({ value: "dependency" })
-      }),
-    )
-    const dependency = make({ service: Value, layer: dependencyLayer, deps: [] })
-    const original = make({ service: Greeting, layer: greetingLayer, deps: [dependency] })
-    const replacement = make({
-      service: Greeting,
-      layer: Layer.succeed(Greeting, Greeting.of({ value: "replacement" })),
-      deps: [],
-    })
-
-    const program = Effect.map(Greeting, (item) => item.value).pipe(
-      Effect.provide(build(LayerNode.group([original]), [[original, replacement]])),
-    )
-
-    expect(await Effect.runPromise(program)).toBe("replacement")
-    expect(acquisitions).toBe(0)
-  })
-
-  test("applies later replacements inside earlier replacement nodes", async () => {
-    const original = make({ service: Greeting, layer: greetingLayer, deps: [value] })
-    const replacement = make({ service: Greeting, layer: greetingLayer, deps: [value] })
-    const program = Effect.map(Greeting, (item) => item.value).pipe(
-      Effect.provide(
-        build(LayerNode.group([original]), [
-          [original, replacement],
-          [value, Layer.succeed(Value, Value.of({ value: "replacement dependency" }))],
-        ]),
-      ),
-    )
-
-    expect(await Effect.runPromise(program)).toBe("hello replacement dependency")
-  })
-
-  test("hoists and compiles tagged graphs", async () => {
-    const tags = LayerNode.tags({ location: ["global"], global: [] })
-    const global = tags.make("global")
-    const location = tags.make("location")
-    const database = global({
-      service: Database,
-      layer: Layer.succeed(Database, Database.of({ name: "Alice" })),
-      deps: [],
-    })
-    const users = location({
-      service: Users,
+    const dependent = make({
+      service: Value,
      layer: Layer.effect(
-        Users,
-        Effect.gen(function* () {
-          const db = yield* Database
-          return Users.of({ list: Effect.succeed([db.name]) })
-        }),
+        Value,
+        Effect.map(Greeting, (item) => Value.of({ value: item.value })),
      ),
-      deps: [database],
+      deps: [greeting],
    })
-    const app = location({
-      service: App,
-      layer: Layer.effect(
-        App,
-        Effect.gen(function* () {
-          const service = yield* Users
-          return App.of({ run: service.list })
-        }),
-      ),
-      deps: [users],
-    })
-
-    const result = LayerNode.hoist(LayerNode.group([app]), tags.values.global)
-    expect(result.node.dependencies[0]?.dependencies[0]?.dependencies[0]).toMatchObject({
-      kind: "group",
-      dependencies: [],
-    })
-    expect(result.hoisted.dependencies).toEqual([database])
-
-    const layer = LayerNode.compile(result.node).pipe(
-      Layer.provide(LayerNode.compile(result.hoisted)),
-    ) as unknown as Layer.Layer<App>
-    const program = Effect.gen(function* () {
-      const app = yield* App
-      return yield* app.run
-    }).pipe(Effect.provide(layer))
-
-    expect(await Effect.runPromise(program)).toEqual(["Alice"])
-  })
-
-  test("rejects conflicting hoisted implementations", () => {
-    const tags = LayerNode.tags({ location: ["global"], global: [] })
-    const global = tags.make("global")
-    const location = tags.make("location")
-    const first = global({
-      service: Database,
-      layer: Layer.succeed(Database, Database.of({ name: "first" })),
-      deps: [],
-    })
-    const second = global({
-      service: Database,
-      layer: Layer.succeed(Database, Database.of({ name: "second" })),
-      deps: [],
-    })
-    const left = location({
-      service: Users,
-      layer: Layer.effect(Users, Effect.as(Database, Users.of({ list: Effect.succeed([]) }))),
-      deps: [first],
-    })
-    const right = location({
-      service: App,
-      layer: Layer.effect(App, Effect.as(Database, App.of({ run: Effect.succeed([]) }))),
-      deps: [second],
-    })
-
-    expect(() => LayerNode.hoist(LayerNode.group([left, right]), tags.values.global)).toThrow(
-      "Tag global has conflicting implementations for test/GraphDatabase",
+    expect(() => LayerNode.compile(greeting, { replacements: [value.replace(dependent)] })).toThrow(
+      "Cycle detected in layer graph",
    )
  })

-  test("treats dependency groups as transparent while hoisting", () => {
-    const tags = LayerNode.tags({ location: ["global"], global: [] })
-    const global = tags.make("global")
-    const location = tags.make("location")
-    const database = global({
-      service: Database,
-      layer: Layer.succeed(Database, Database.of({ name: "Alice" })),
-      deps: [],
-    })
-    const users = location({
-      service: Users,
-      layer: Layer.effect(Users, Effect.as(Database, Users.of({ list: Effect.succeed([]) }))),
-      deps: [LayerNode.group([database])],
-    })
-    const result = LayerNode.hoist(LayerNode.group([users]), tags.values.global)
+  it.effect("does not acquire replaced dependencies or unused replacement targets", () =>
+    Effect.gen(function* () {
+      const acquired: string[] = []
+      const dependency = make({
+        service: Value,
+        layer: Layer.effect(
+          Value,
+          Effect.sync(() => {
+            acquired.push("old dependency")
+            return Value.of({ value: "dependency" })
+          }),
+        ),
+        deps: [],
+      })
+      const original = make({ service: Greeting, layer: greetingLayer, deps: [dependency] })
+      const result = yield* Greeting.pipe(
+        Effect.provide(
+          LayerNode.compile(original, {
+            replacements: [
+              original.replace(Layer.succeed(Greeting, { value: "replacement" })),
+              value.replace(
+                Layer.effect(
+                  Value,
+                  Effect.sync(() => {
+                    acquired.push("unused target")
+                    return Value.of({ value: "unused" })
+                  }),
+                ),
+              ),
+            ],
+          }),
+        ),
+      )
+      expect(result.value).toBe("replacement")
+      expect(acquired).toEqual([])
+    }),
+  )

-    expect(result.node.dependencies[0]?.dependencies[0]?.dependencies[0]).toMatchObject({
-      kind: "group",
-      dependencies: [],
-    })
+  it.effect("mapLayer preserves dependency wiring and replacement traversal", () =>
+    Effect.gen(function* () {
+      const acquired: string[] = []
+      const decorated = greeting.mapLayer((layer) =>
+        layer.pipe(
+          Layer.tap((context) =>
+            Effect.sync(() => {
+              acquired.push(Context.get(context, Greeting).value)
+            }),
+          ),
+        ),
+      )
+      const result = yield* Greeting.pipe(
+        Effect.provide(
+          LayerNode.compile(greeting, {
+            replacements: [
+              greeting.replace(decorated),
+              value.replace(Layer.succeed(Value, { value: "mapped dependency" })),
+            ],
+          }),
+        ),
+      )
+      expect(result.value).toBe("hello mapped dependency")
+      expect(acquired).toEqual(["hello mapped dependency"])
+    }),
+  )
+
+  it.effect("memoizes shared wiring instead of expanding a diamond into a tree", () =>
+    Effect.gen(function* () {
+      const acquisitions: string[] = []
+      const shared = value.mapLayer((layer) =>
+        layer.pipe(Layer.tap(() => Effect.sync(() => acquisitions.push("shared")))),
+      )
+      const left = make({ name: "left", layer: Layer.empty, deps: [shared] })
+      const right = make({ name: "right", layer: Layer.empty, deps: [shared] })
+      yield* Layer.build(LayerNode.compile(LayerNode.group([left, right])))
+      expect(acquisitions).toEqual(["shared"])
+    }),
+  )
+
+  it.effect("preserves declared memo-service outputs rather than filtering them as build metadata", () =>
+    Effect.gen(function* () {
+      const supplied = yield* Layer.makeMemoMap
+      const memo = make({
+        service: Layer.CurrentMemoMap,
+        layer: Layer.succeed(Layer.CurrentMemoMap, supplied),
+        deps: [],
+      })
+      const observer = make({ service: Memo, layer: Layer.effect(Memo, Layer.CurrentMemoMap), deps: [memo] })
+      expect(yield* Memo.pipe(Effect.provide(LayerNode.compile(observer)))).toBe(supplied)
+    }),
+  )
+
+  it.effect("rejects one implementation wired to different effective dependencies in either memo domain", () =>
+    Effect.gen(function* () {
+      const other = make({ service: Value, layer: Layer.succeed(Value, { value: "other" }), deps: [] })
+      const sibling = make({ service: Greeting, layer: greetingLayer, deps: [other] })
+      const root = LayerNode.group([greeting, sibling])
+      expect(() => LayerNode.compile(root)).toThrow("wired to different dependencies")
+      expect(() => LayerNode.compile(root, { shared: tags.values.app })).toThrow("wired to different dependencies")
+      const result = yield* Greeting.pipe(
+        Effect.provide(LayerNode.compile(root, { replacements: [value.replace(other)] })),
+      )
+      expect(result.value).toBe("hello other")
+    }),
+  )
+
+  it.effect("starts dependencies in parallel and nested group roots in order", () =>
+    Effect.gen(function* () {
+      const valueStarted = yield* Deferred.make<void>()
+      const greetingStarted = yield* Deferred.make<void>()
+      const firstStarted = yield* Deferred.make<void>()
+      const releaseFirst = yield* Deferred.make<void>()
+      const events: string[] = []
+      const value = make({
+        service: Value,
+        layer: Layer.effect(
+          Value,
+          Effect.gen(function* () {
+            yield* Deferred.succeed(valueStarted, undefined)
+            yield* Deferred.await(greetingStarted)
+            return Value.of({ value: "value" })
+          }),
+        ),
+        deps: [],
+      })
+      const greeting = make({
+        service: Greeting,
+        layer: Layer.effect(
+          Greeting,
+          Effect.gen(function* () {
+            yield* Deferred.succeed(greetingStarted, undefined)
+            yield* Deferred.await(valueStarted)
+            return Greeting.of({ value: "greeting" })
+          }),
+        ),
+        deps: [],
+      })
+      const first = make({
+        service: Left,
+        layer: Layer.effect(
+          Left,
+          Effect.gen(function* () {
+            yield* Value
+            yield* Greeting
+            events.push("first started")
+            yield* Deferred.succeed(firstStarted, undefined)
+            yield* Deferred.await(releaseFirst)
+            events.push("first finished")
+            return Left.of({ value: "first" })
+          }),
+        ),
+        deps: [value, greeting],
+      })
+      const second = make({
+        service: Right,
+        layer: Layer.effect(
+          Right,
+          Effect.sync(() => {
+            expect(events).toEqual(["first started", "first finished"])
+            events.push("second started")
+            return Right.of({ value: "second" })
+          }),
+        ),
+        deps: [],
+      })
+      const fiber = yield* Layer.build(LayerNode.compile(LayerNode.group([LayerNode.group([first]), second]))).pipe(
+        Effect.forkChild,
+      )
+      yield* Deferred.await(firstStarted)
+      expect(events).toEqual(["first started"])
+      yield* Deferred.succeed(releaseFirst, undefined)
+      const context = yield* Fiber.join(fiber)
+      expect(events).toEqual(["first started", "first finished", "second started"])
+      expect(Context.get(context, Left).value).toBe("first")
+      expect(Context.get(context, Right).value).toBe("second")
+    }),
+  )
+  ;[false, true].forEach((topLevel) => {
+    it.effect(
+      `LayerMap isolates builds and retains resources ${topLevel ? "with" : "without"} a top-level global owner`,
+      () =>
+        Effect.gen(function* () {
+          const acquired = { global: 0, local: 0, support: 0 }
+          const released: string[] = []
+          const startup: string[] = []
+          yield* Effect.gen(function* () {
+            const memoMap = yield* Layer.makeMemoMap
+            const tags = LayerNode.tags({ location: ["global"], global: [] })
+            const global = tags.make("global")
+            const location = tags.make("location")
+            const support = LayerNode.make({
+              service: Support,
+              layer: Layer.effect(
+                Support,
+                Effect.acquireRelease(
+                  Effect.sync(() => {
+                    acquired.support++
+                    return Support.of({})
+                  }),
+                  () =>
+                    Effect.sync(() => {
+                      released.push("support")
+                    }),
+                ),
+              ),
+              deps: [],
+            })
+            const value = global({
+              service: Value,
+              layer: Layer.effect(
+                Value,
+                Effect.andThen(
+                  Support,
+                  Effect.acquireRelease(
+                    Effect.sync(() => {
+                      startup.push("global")
+                      return Value.of({ value: `global-${++acquired.global}` })
+                    }),
+                    (value) =>
+                      Effect.sync(() => {
+                        released.push(value.value)
+                      }),
+                  ),
+                ),
+              ),
+              deps: [support],
+            })
+            const local = location({
+              service: Greeting,
+              layer: Layer.effect(
+                Greeting,
+                Effect.gen(function* () {
+                  yield* Value
+                  return yield* Effect.acquireRelease(
+                    Effect.sync(() => Greeting.of({ value: `local-${++acquired.local}` })),
+                    (value) =>
+                      Effect.sync(() => {
+                        released.push(value.value)
+                      }),
+                  )
+                }),
+              ),
+              deps: [LayerNode.group([value])],
+            })
+            const root = location({
+              service: Right,
+              layer: Layer.effect(
+                Right,
+                Effect.gen(function* () {
+                  const local = yield* Greeting
+                  if (local.value === "local-2") return yield* Effect.fail("failed location" as const)
+                  return Right.of(local)
+                }),
+              ),
+              deps: [local],
+            })
+            // Every key builds the same compiled Layer, not a new graph per lookup.
+            const compiled = LayerNode.compile(LayerNode.group([value, root]), { shared: tags.values.global })
+            const locations = location({
+              service: Locations,
+              layer: Layer.effect(
+                Locations,
+                Effect.gen(function* () {
+                  startup.push("map")
+                  expect(Option.getOrUndefined(yield* Effect.serviceOption(Layer.CurrentMemoMap))).toBe(memoMap)
+                  return yield* LayerMap.make((_: string) => compiled, { idleTimeToLive: Duration.infinity })
+                }),
+              ),
+              deps: [],
+            })
+            const scope = yield* Effect.scope
+            const context = yield* Layer.buildWithMemoMap(
+              LayerNode.compile(LayerNode.group([locations, ...(topLevel ? [value] : [])]), {
+                shared: tags.values.global,
+              }),
+              memoMap,
+              scope,
+            )
+            expect(startup).toEqual(topLevel ? ["map", "global"] : ["map"])
+            const map = Context.get(context, Locations)
+            const first = yield* map.contextEffect("first").pipe(Effect.scoped)
+            expect(Option.getOrUndefined(Context.getOption(context, Value))).toBe(
+              topLevel ? Context.get(first, Value) : undefined,
+            )
+            expect(Option.isNone(Context.getOption(first, Greeting))).toBe(true)
+            expect(Context.get(first, Right).value).toBe("local-1")
+
+            expect(yield* map.contextEffect("failed").pipe(Effect.scoped, Effect.flip)).toBe("failed location")
+            expect(released).toEqual(["local-2"])
+            expect(Context.get(yield* map.contextEffect("first").pipe(Effect.scoped), Right)).toBe(
+              Context.get(first, Right),
+            )
+
+            const second = yield* map.contextEffect("second").pipe(Effect.scoped)
+            expect(Context.get(second, Value)).toBe(Context.get(first, Value))
+            expect(Context.get(second, Right)).not.toBe(Context.get(first, Right))
+            expect(acquired).toEqual({ global: 1, local: 3, support: 1 })
+
+            yield* map.invalidate("first")
+            expect(released).toEqual(["local-2", "local-1"])
+            expect(Context.get(yield* map.contextEffect("second").pipe(Effect.scoped), Right)).toBe(
+              Context.get(second, Right),
+            )
+            const rebuilt = yield* map.contextEffect("first").pipe(Effect.scoped)
+            expect(Context.get(rebuilt, Right).value).toBe("local-4")
+            expect(Context.get(rebuilt, Value)).toBe(Context.get(first, Value))
+            expect(acquired).toEqual({ global: 1, local: 4, support: 1 })
+            expect(released).not.toContain("global-1")
+          }).pipe(Effect.scoped)
+          expect(released.toSorted()).toEqual(["global-1", "local-1", "local-2", "local-3", "local-4", "support"])
+        }),
+    )
  })
 })
@@ -1,20 +1,23 @@
 import { describe, expect, test } from "bun:test"
-import { Context, Effect, Layer, LayerMap, Option } from "effect"
+import { Context, Effect, Layer, Option } from "effect"
 import { Node } from "@opencode-ai/util/effect/app-node"
 import { AppNodeBuilder } from "@opencode-ai/core/effect/app-node-builder"
 import { LayerNode } from "@opencode-ai/util/effect/layer-node"
 import { Location } from "@opencode-ai/core/location"
 import { LocationServiceMap } from "@opencode-ai/core/location-service-map"
-import type { LocationError, LocationServices } from "@opencode-ai/core/location-services"
+import { buildLocationServiceMap } from "@opencode-ai/core/location-services"
 import { Project } from "@opencode-ai/core/project"
 import { AbsolutePath } from "@opencode-ai/core/schema"
 import { tmpdir } from "../../fixture/tmpdir"
+import { testEffect } from "../../lib/effect"

 class Value extends Context.Service<Value, { readonly value: string }>()("test/TagValue") {}
 class Result extends Context.Service<Result, { readonly value: string }>()("test/TagResult") {}
 class CycleA extends Context.Service<CycleA, {}>()("test/NodeBuildA") {}
 class CycleB extends Context.Service<CycleB, { readonly directory: AbsolutePath }>()("test/NodeBuildB") {}

+const it = testEffect(Layer.empty)
+
 describe("node build", () => {
  test("does not build a location service map when the graph does not require it", async () => {
    const result = Node.makeGlobalNode({
@@ -31,7 +34,7 @@ describe("node build", () => {
    expect(await Effect.runPromise(program)).toBe("plain")
  })

-  test("detects cycles through a replaced location service map", async () => {
+  test("detects cycles through a replaced location service map", () => {
    const a = Node.makeGlobalNode({
      service: CycleA,
      layer: Layer.effect(CycleA, Effect.as(LocationServiceMap.Service, CycleA.of({}))),
@@ -45,31 +48,49 @@ describe("node build", () => {
      ),
      deps: [a],
    })
-    const mapLayer = Layer.effect(
-      LocationServiceMap.Service,
-      Effect.gen(function* () {
-        const service = yield* CycleB
-        return yield* LayerMap.make(
-          (ref: Location.Ref) =>
-            Layer.succeed(
-              Location.Service,
-              Location.Service.of({
-                directory: ref.directory,
-                workspaceID: ref.workspaceID,
-                project: { id: Project.ID.global, directory: service.directory, canonical: service.directory },
-              }),
-            ),
-          { idleTimeToLive: "1 minute" },
-        )
-      }) as unknown as Effect.Effect<LayerMap.LayerMap<Location.Ref, LocationServices, LocationError>, never, CycleB>,
-    )
+    const mapLayer = Layer.unwrap(Effect.as(CycleB, buildLocationServiceMap()))
    const map = Node.makeGlobalNode({ service: LocationServiceMap.Service, layer: mapLayer, deps: [b] })
-    expect(() => AppNodeBuilder.build(LayerNode.group([a]), [[LocationServiceMap.node, map]])).toThrow(
-      "Cycle detected in layer tree",
+    expect(() => AppNodeBuilder.build(LayerNode.group([a]), [LocationServiceMap.node.replace(map)])).toThrow(
+      "Cycle detected in layer graph",
    )
  })

-  test("shares top-level project with location services", async () => {
+  it.effect("supplies the lazy map when only a replacement introduces the dependency", () =>
+    Effect.gen(function* () {
+      const original = Node.makeGlobalNode({
+        service: Result,
+        layer: Layer.succeed(Result, { value: "original" }),
+        deps: [],
+      })
+      const replacement = Node.makeGlobalNode({
+        service: Result,
+        layer: Layer.effect(Result, Effect.as(LocationServiceMap.Service, Result.of({ value: "has map" }))),
+        deps: [LocationServiceMap.node],
+      })
+      const result = yield* Result.pipe(Effect.provide(AppNodeBuilder.build(original, [original.replace(replacement)])))
+      expect(result.value).toBe("has map")
+    }),
+  )
+
+  it.effect("caller replacements override the lazy default without building any locations", () =>
+    Effect.gen(function* () {
+      const acquisitions: string[] = []
+      const override = buildLocationServiceMap().pipe(
+        Layer.tap(() =>
+          Effect.sync(() => {
+            acquisitions.push("caller map")
+          }),
+        ),
+      )
+      const context = yield* Layer.build(
+        AppNodeBuilder.build(LocationServiceMap.node, [LocationServiceMap.node.replace(override)]),
+      )
+      expect(Context.get(context, LocationServiceMap.Service)).toBeDefined()
+      expect(acquisitions).toEqual(["caller map"])
+    }),
+  )
+
+  test("shares top-level project even when the location service map is built first", async () => {
    await using tmp = await tmpdir()
    let acquisitions = 0
    const projectLayer = Layer.effect(
@@ -84,8 +105,8 @@ describe("node build", () => {
      }),
    )
    const ref = Location.Ref.make({ directory: AbsolutePath.make(tmp.path) })
-    const layer = AppNodeBuilder.build(LayerNode.group([Project.node, LocationServiceMap.node]), [
-      [Project.node, projectLayer],
+    const layer = AppNodeBuilder.build(LayerNode.group([LocationServiceMap.node, Project.node]), [
+      Project.node.replace(projectLayer),
    ])
    const program = Effect.gen(function* () {
      yield* Project.Service
@@ -21,8 +21,8 @@ function provide(directory: string, transformFiles: EnvironmentFilesTransform =
  )
  return Effect.provide(
    AppNodeBuilder.build(LayerNode.group([LocationMutation.node, FileMutation.node]), [
-      [Location.node, activeLocation],
-      [Environment.node, transformEnvironmentFiles(transformFiles)],
+      Location.node.replace(activeLocation),
+      Environment.node.replace(transformEnvironmentFiles(transformFiles)),
    ]),
  )
 }
@@ -77,16 +77,15 @@ describe("FileSystemSearch", () => {
        workspaceID: Workspace.ID.make("wrk_test"),
      })
      const layer = AppNodeBuilder.build(FileSystemSearch.node, [
-        [
-          Location.node,
+        Location.node.replace(
          Layer.succeed(
            Location.Service,
            Location.Service.of(
              location(ref, { vcs: { type: "git", store: AbsolutePath.make(path.join(directory, ".git")) } }),
            ),
          ),
-        ],
-        [Ripgrep.node, ripgrepStub("remote.ts", (input) => (observed = input))],
+        ),
+        Ripgrep.node.replace(ripgrepStub("remote.ts", (input) => (observed = input))),
      ])

      yield* Effect.gen(function* () {
@@ -103,8 +102,7 @@ describe("FileSystemSearch", () => {
      let observed: Ripgrep.FindInput | undefined
      const home = AbsolutePath.make(os.homedir())
      const layer = AppNodeBuilder.build(FileSystemSearch.node, [
-        [
-          Location.node,
+        Location.node.replace(
          Layer.succeed(
            Location.Service,
            Location.Service.of(
@@ -114,8 +112,8 @@ describe("FileSystemSearch", () => {
              ),
            ),
          ),
-        ],
-        [Ripgrep.node, ripgrepStub("src/index.ts", (input) => (observed = input))],
+        ),
+        Ripgrep.node.replace(ripgrepStub("src/index.ts", (input) => (observed = input))),
      ])
      yield* Effect.gen(function* () {
        const search = yield* FileSystemSearch.Service
@@ -137,17 +135,15 @@ describe("FileSystemSearch", () => {
      const started = yield* Deferred.make<void>()
      const release = yield* Deferred.make<void>()
      const layer = AppNodeBuilder.build(FileSystemSearch.node, [
-        [
-          Location.node,
+        Location.node.replace(
          Layer.succeed(
            Location.Service,
            Location.Service.of(
              location({ directory: AbsolutePath.make(path.join(os.tmpdir(), "opencode-search-atomic")) }),
            ),
          ),
-        ],
-        [
-          Ripgrep.node,
+        ),
+        Ripgrep.node.replace(
          Layer.succeed(
            Ripgrep.Service,
            Ripgrep.Service.of({
@@ -169,7 +165,7 @@ describe("FileSystemSearch", () => {
              grep: () => Effect.succeed([]),
            }),
          ),
-        ],
+        ),
      ])

      yield* Effect.gen(function* () {
@@ -208,17 +204,15 @@ describe("FileSystemSearch", () => {
        (value) => Effect.sync(() => value.mockRestore()),
      )
      const layer = AppNodeBuilder.build(FileSystemSearch.node, [
-        [
-          Location.node,
+        Location.node.replace(
          Layer.succeed(
            Location.Service,
            Location.Service.of(
              location({ directory: AbsolutePath.make(path.join(os.tmpdir(), "opencode-search-cache")) }),
            ),
          ),
-        ],
-        [
-          Ripgrep.node,
+        ),
+        Ripgrep.node.replace(
          Layer.succeed(
            Ripgrep.Service,
            Ripgrep.Service.of({
@@ -234,7 +228,7 @@ describe("FileSystemSearch", () => {
              grep: () => Effect.succeed([]),
            }),
          ),
-        ],
+        ),
      ])

      yield* Effect.gen(function* () {
@@ -6,7 +6,7 @@ import { Deferred, Duration, Effect, Fiber, Layer, Option, Schedule, Stream } fr
 import { Config } from "@opencode-ai/core/config"
 import { ConfigLocationWatcherPlugin } from "@opencode-ai/core/config/plugin/location-watcher"
 import { AppNodeBuilder } from "@opencode-ai/core/effect/app-node-builder"
-import { makeLocationNode, type LocationNode } from "@opencode-ai/util/effect/app-node"
+import { makeLocationNode } from "@opencode-ai/util/effect/app-node"
 import { LayerNode } from "@opencode-ai/util/effect/layer-node"
 import { Bus } from "@opencode-ai/core/bus"
 import { FSUtil } from "@opencode-ai/util/fs-util"
@@ -129,7 +129,7 @@ function provide(
  vcs?: Location.Interface["vcs"],
  watcher?: Layer.Layer<Watcher.Service>,
  config: Layer.Layer<Config.Service> = configLayer,
-  plugins: LocationNode<PluginSupervisor.Service> = pluginNode,
+  plugins: typeof pluginNode = pluginNode,
 ) {
  const locationLayer = Layer.succeed(
    Location.Service,
@@ -138,10 +138,10 @@ function provide(
  const built = AppNodeBuilder.build(
    LayerNode.group([LocationWatcher.node, LocationWatcherPolicy.node, Bus.node, Config.node]),
    [
-      [Config.node, config],
-      [Location.node, locationLayer],
-      [PluginSupervisor.node, plugins],
-      ...(watcher ? ([[Watcher.node, watcher]] as const) : []),
+      Config.node.replace(config),
+      Location.node.replace(locationLayer),
+      PluginSupervisor.node.replace(plugins),
+      ...(watcher ? ([Watcher.node.replace(watcher)] as const) : []),
    ],
  )
  return Effect.provide(built)
@@ -154,7 +154,7 @@ function withTmp<A, E, R>(
    init?: (directory: string) => Promise<void>
    watcher?: Layer.Layer<Watcher.Service>
    config?: Layer.Layer<Config.Service>
-    plugins?: LocationNode<PluginSupervisor.Service>
+    plugins?: typeof pluginNode
  },
 ) {
  return Effect.acquireRelease(
@@ -30,7 +30,7 @@ const testGlobal = Global.layerWith({
  log: os.tmpdir(),
 })

-const testLayer = LayerNode.compile(EffectFlock.node, [[Global.node, testGlobal]])
+const testLayer = LayerNode.compile(EffectFlock.node, { replacements: [Global.node.replace(testGlobal)] })

 async function job() {
  if (msg.ready) await fs.writeFile(msg.ready, String(process.pid))
@@ -26,9 +26,9 @@ export const promptLocationNode = makeGlobalNode({
          SessionPrompt.layer.pipe(
            Layer.provideMerge(
              Layer.mergeAll(
-                LayerNode.compile(LayerNode.group([PluginHooks.node, Image.node, Skill.node]), [
-                  [Bus.node, Layer.succeed(Bus.Service, bus)],
-                ]),
+                LayerNode.compile(LayerNode.group([PluginHooks.node, Image.node, Skill.node]), {
+                  replacements: [Bus.node.replace(Layer.succeed(Bus.Service, bus))],
+                }),
                Layer.succeed(FSUtil.Service, fs),
                Layer.succeed(PluginSupervisor.Service, { flush: Effect.void }),
                Layer.mock(Reference.Service, { refresh: () => Effect.void }),
@@ -20,7 +20,7 @@ import { testEffect } from "./lib/effect"

 const it = testEffect(
  AppNodeBuilder.build(LayerNode.group([Database.node, Bus.node, SdkPlugins.node, LocationServiceMap.node]), [
-    [Global.node, tempGlobalLayer],
+    Global.node.replace(tempGlobalLayer),
  ]),
 )
 type ConfigInput = typeof Info.Encoded
@@ -58,6 +58,34 @@ function withFormatter<A, E, R>(
 }

 describe("Formatter", () => {
+  ;[
+    { file: "test.match", extension: ".match", matches: true },
+    { file: "test.other", extension: ".match", matches: false },
+    { file: "test.MATCH", extension: ".match", matches: false },
+    { file: "test.MATCH", extension: ".MATCH", matches: true },
+    { file: ".match", extension: ".match", matches: false },
+    { file: ".match", extension: "", matches: true },
+    { file: "README", extension: ".match", matches: false },
+    { file: "README", extension: "", matches: true },
+    { file: "test.part.match", extension: ".match", matches: true },
+    { file: "test.part.match", extension: ".part.match", matches: false },
+  ].forEach((entry) =>
+    it.live(`matches ${entry.file} against ${JSON.stringify(entry.extension)}: ${entry.matches}`, () =>
+      withFormatter(
+        {
+          matching: {
+            command: [process.execPath, "-e", "process.exit(0)", "$FILE"],
+            extensions: [entry.extension],
+          },
+        },
+        (formatter, directory) =>
+          Effect.gen(function* () {
+            expect(yield* formatter.file(path.join(directory, entry.file))).toBe(entry.matches)
+          }),
+      ),
+    ),
+  )
+
  it.live("does not run formatters marked as disabled in config", () =>
    withFormatter(
      {
@@ -34,7 +34,7 @@ const instances = Layer.effect(
    (ref: Location.Ref) =>
      Instance.layer(ref, {
        plugins: path.basename(ref.directory) === "thread-a" ? [agentPlugin("thread-a-plugin", "thread-a-agent")] : [],
-        replacements: [[Global.node, tempGlobalLayer]],
+        replacements: [Global.node.replace(tempGlobalLayer)],
      }),
    { idleTimeToLive: Duration.infinity },
  ),
@@ -42,8 +42,8 @@ const instances = Layer.effect(

 const it = testEffect(
  AppNodeBuilder.build(LayerNode.group([Database.node, Bus.node, SdkPlugins.node, LocationServiceMap.node]), [
-    [Global.node, tempGlobalLayer],
-    [LocationServiceMap.node, instances],
+    Global.node.replace(tempGlobalLayer),
+    LocationServiceMap.node.replace(instances),
  ]),
 )

@@ -23,14 +23,13 @@ import { Bus } from "../src/bus"
 // Config the host hands the vanilla instance explicitly: a value and an
 // explicit plugin removal, both of which must survive discovery: false.
 const hostConfig: LayerNode.Replacements = [
-  [
-    Config.node,
+  Config.node.replace(
    Config.configured({
      project: false,
      global: false,
      content: JSON.stringify({ shell: "vanilla-host", plugins: ["-opencode.tool.shell"] }),
    }),
-  ],
+  ),
 ]

 // Same directory contents, two instances: one vanilla, one with discovery.
@@ -43,7 +42,7 @@ const instances = Layer.effect(
        // "bare" exercises the vanilla defaults themselves: no caller Config.
        discovery: name !== "vanilla" && name !== "bare",
        // Caller replacements win over the vanilla defaults.
-        replacements: [[Global.node, tempGlobalLayer], ...(name === "vanilla" ? hostConfig : [])],
+        replacements: [Global.node.replace(tempGlobalLayer), ...(name === "vanilla" ? hostConfig : [])],
      })
    },
    { idleTimeToLive: Duration.infinity },
@@ -52,8 +51,8 @@ const instances = Layer.effect(

 const it = testEffect(
  AppNodeBuilder.build(LayerNode.group([Database.node, Bus.node, SdkPlugins.node, LocationServiceMap.node]), [
-    [Global.node, tempGlobalLayer],
-    [LocationServiceMap.node, instances],
+    Global.node.replace(tempGlobalLayer),
+    LocationServiceMap.node.replace(instances),
  ]),
 )

@@ -33,19 +33,18 @@ const instructionLayer = (input: {
    AppNodeBuilder.build(
      LayerNode.group([InstructionDiscovery.node, Bus.node, FSUtil.node, Global.node, Location.node, Watcher.node]),
      [
-        [InstructionDiscovery.node, InstructionDiscovery.configured({ project: input.project })],
-        [
-          Global.node,
+        InstructionDiscovery.node.replace(InstructionDiscovery.configured({ project: input.project })),
+        Global.node.replace(
          input.config || input.home
            ? Global.layerWith({
                ...(input.config ? { config: input.config } : {}),
                ...(input.home ? { home: input.home } : {}),
              })
            : tempGlobalLayer,
-        ],
-        [Location.node, input.locationServiceLayer],
-        [Watcher.node, watcher],
-        ...(input.filesystemLayer ? [[FSUtil.node, input.filesystemLayer] as const] : []),
+        ),
+        Location.node.replace(input.locationServiceLayer),
+        Watcher.node.replace(watcher),
+        ...(input.filesystemLayer ? [FSUtil.node.replace(input.filesystemLayer)] : []),
      ],
    ),
    watcher,
@@ -24,7 +24,7 @@ import { testEffect } from "./lib/effect"

 const it = testEffect(
  AppNodeBuilder.build(LayerNode.group([Database.node, Bus.node, SessionProjector.node]), [
-    [Bus.node, Bus.configured({ persist: true })],
+    Bus.node.replace(Bus.configured({ persist: true })),
  ]),
 )

@@ -30,8 +30,8 @@ const locationLayer = Layer.succeed(
 )
 const it = testEffect(
  AppNodeBuilder.build(InstructionBuiltIns.node, [
-    [Location.node, locationLayer],
-    [Global.node, Global.layerWith({ config: temporary, tmp: temporary })],
+    Location.node.replace(locationLayer),
+    Global.node.replace(Global.layerWith({ config: temporary, tmp: temporary })),
  ]),
 )

@@ -27,7 +27,7 @@ const failingCredentialNode = makeGlobalNode({
  deps: [],
 })
 const failingIt = testEffect(
-  AppNodeBuilder.build(LayerNode.group([Integration.node, Bus.node]), [[Credential.node, failingCredentialNode]]),
+  AppNodeBuilder.build(LayerNode.group([Integration.node, Bus.node]), [Credential.node.replace(failingCredentialNode)]),
 )

 function eventually<A, E, R>(
@@ -13,15 +13,16 @@ import { it } from "./lib/effect"

 const provide = (directory: string, workspaceID?: Workspace.ID) =>
  Effect.provide(
-    LayerNode.compile(FileSystem.node, [
-      [
-        Location.node,
-        Layer.succeed(
-          Location.Service,
-          Location.Service.of(location({ directory: AbsolutePath.make(directory), workspaceID })),
+    LayerNode.compile(FileSystem.node, {
+      replacements: [
+        Location.node.replace(
+          Layer.succeed(
+            Location.Service,
+            Location.Service.of(location({ directory: AbsolutePath.make(directory), workspaceID })),
+          ),
        ),
      ],
-    ]),
+    }),
  )

 const withTmp = <A, E, R>(f: (directory: string) => Effect.Effect<A, E, R>) =>
@@ -51,12 +51,12 @@ import { Tool } from "../src/tool"

 const it = testEffect(
  AppNodeBuilder.build(LayerNode.group([Database.node, Bus.node, LocationServiceMap.node]), [
-    [Global.node, tempGlobalLayer],
+    Global.node.replace(tempGlobalLayer),
  ]),
 )
 const itWithSdk = testEffect(
  AppNodeBuilder.build(LayerNode.group([Database.node, Bus.node, SdkPlugins.node, LocationServiceMap.node]), [
-    [Global.node, tempGlobalLayer],
+    Global.node.replace(tempGlobalLayer),
  ]),
 )
 const activityLocations = Layer.effect(
@@ -77,7 +77,7 @@ const activityLocations = Layer.effect(
 )
 const itWithActivity = testEffect(
  AppNodeBuilder.build(LayerNode.group([Database.node, Bus.node, LocationServiceMap.node, LocationActivity.node]), [
-    [LocationServiceMap.node, activityLocations],
+    LocationServiceMap.node.replace(activityLocations),
  ]),
 )

@@ -13,20 +13,21 @@ import { it } from "./lib/effect"

 function provide(directory: string, projectDirectory = directory) {
  return Effect.provide(
-    LayerNode.compile(LocationMutation.node, [
-      [
-        Location.node,
-        Layer.succeed(
-          Location.Service,
-          Location.Service.of(
-            location(
-              { directory: AbsolutePath.make(directory) },
-              { projectDirectory: AbsolutePath.make(projectDirectory) },
+    LayerNode.compile(LocationMutation.node, {
+      replacements: [
+        Location.node.replace(
+          Layer.succeed(
+            Location.Service,
+            Location.Service.of(
+              location(
+                { directory: AbsolutePath.make(directory) },
+                { projectDirectory: AbsolutePath.make(projectDirectory) },
+              ),
            ),
          ),
        ),
      ],
-    ]),
+    }),
  )
 }

@@ -23,7 +23,7 @@ const projectLayer = Layer.succeed(
      }),
  }),
 )
-const it = testEffect(AppNodeBuilder.build(Location.boundNode(ref), [[Project.node, projectLayer]]))
+const it = testEffect(AppNodeBuilder.build(Location.boundNode(ref), [Project.node.replace(projectLayer)]))

 describe("Location", () => {
  it.effect("resolves the current project and vcs information", () =>
@@ -23,13 +23,12 @@ const tool = (server: string, name = "search") => new Mcp.Tool({ server: Mcp.Ser

 const layer = (catalog: () => Mcp.ServerInstructions[], tools: () => Mcp.Tool[]) =>
  AppNodeBuilder.build(McpInstructions.node, [
-    [
-      Mcp.node,
+    Mcp.node.replace(
      Layer.mock(Mcp.Service, {
        instructions: () => Effect.succeed(catalog()),
        tools: () => Effect.succeed(tools()),
      }),
-    ],
+    ),
  ])

 describe("McpInstructions", () => {
@@ -378,10 +378,10 @@ const permissions = Layer.mock(Permission.Service, {
 const events = Layer.mock(Bus.Service, { subscribe: () => Stream.never })
 const it = testEffect(
  AppNodeBuilder.build(LayerNode.group([Tool.node, McpTool.node]), [
-    [Mcp.node, mcp],
-    [Permission.node, permissions],
-    [Bus.node, events],
-    [Image.node, imagePassthrough],
+    Mcp.node.replace(mcp),
+    Permission.node.replace(permissions),
+    Bus.node.replace(events),
+    Image.node.replace(imagePassthrough),
  ]),
 )

@@ -1688,8 +1688,7 @@ testEffect(Layer.empty).live("isolates invalid MCP tools and preserves plugin tr
      Effect.provide(
        Layer.fresh(
          AppNodeBuilder.build(LayerNode.group([Tool.node, McpTool.node, Bus.node]), [
-            [
-              Mcp.node,
+            Mcp.node.replace(
              Layer.mock(Mcp.Service, {
                tools: () => Ref.get(catalog),
                callTool: (input) =>
@@ -1702,9 +1701,9 @@ testEffect(Layer.empty).live("isolates invalid MCP tools and preserves plugin tr
                    }),
                  ),
              }),
-            ],
-            [Permission.node, Layer.mock(Permission.Service, { assert: () => Effect.void })],
-            [Image.node, imagePassthrough],
+            ),
+            Permission.node.replace(Layer.mock(Permission.Service, { assert: () => Effect.void })),
+            Image.node.replace(imagePassthrough),
          ]),
        ),
      ),
@@ -1731,8 +1730,7 @@ testEffect(Layer.empty).effect("coalesces queued MCP tool notifications after in
  }).pipe(
    Effect.provide(
      AppNodeBuilder.build(LayerNode.group([Tool.node, McpTool.node, Bus.node]), [
-        [
-          Mcp.node,
+        Mcp.node.replace(
          Layer.mock(Mcp.Service, {
            tools: () =>
              Effect.sync(() => [
@@ -1744,9 +1742,9 @@ testEffect(Layer.empty).effect("coalesces queued MCP tool notifications after in
                }),
              ]),
          }),
-        ],
-        [Permission.node, Layer.mock(Permission.Service, { assert: () => Effect.void })],
-        [Image.node, imagePassthrough],
+        ),
+        Permission.node.replace(Layer.mock(Permission.Service, { assert: () => Effect.void })),
+        Image.node.replace(imagePassthrough),
      ]),
    ),
  )
@@ -182,9 +182,9 @@ const buildLayer = (state: Ref.Ref<MockState>, cache: MockCache, options: Models
  // every test would reuse the cachedInvalidateWithTTL state from the first run.
  Layer.fresh(
    AppNodeBuilder.build(LayerNode.group([ModelsDev.node, Bus.node]), [
-      [ModelsDev.node, ModelsDev.configured(options)],
-      [LayerNodePlatform.httpClient, Layer.succeed(HttpClient.HttpClient, makeMockClient(state))],
-      [KV.node, makeMockKV(cache)],
+      ModelsDev.node.replace(ModelsDev.configured(options)),
+      LayerNodePlatform.httpClient.replace(Layer.succeed(HttpClient.HttpClient, makeMockClient(state))),
+      KV.node.replace(makeMockKV(cache)),
    ]),
  )

@@ -312,9 +312,9 @@ describe("ModelsDev Service", () => {
      const state = yield* Ref.make({ ...initialState, body: JSON.stringify(fixture2) })
      const layer = Layer.fresh(
        AppNodeBuilder.build(ModelsDev.node, [
-          [ModelsDev.node, ModelsDev.configured({ fetch: true, snapshot: false })],
-          [LayerNodePlatform.httpClient, Layer.succeed(HttpClient.HttpClient, makeMockClient(state))],
-          [KV.node, makeFailingWriteKV(cache)],
+          ModelsDev.node.replace(ModelsDev.configured({ fetch: true, snapshot: false })),
+          LayerNodePlatform.httpClient.replace(Layer.succeed(HttpClient.HttpClient, makeMockClient(state))),
+          KV.node.replace(makeFailingWriteKV(cache)),
        ]),
      )
      const result = yield* ModelsDev.Service.use((s) => s.get()).pipe(Effect.provide(layer))
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Kit Langton	583103f2dd	test(core): limit process group survival check to POSIX Node assigns non-detached Windows children to a kill-on-parent-exit job, so the held-stdio mcp fixture cannot assert POSIX process-group survival there. Keep detached descendant, capture deadline, and success-policy coverage enabled on Windows.	2026-08-31 15:35:32 -04:00
Kit Langton	acaca2cc01	fix(util): separate process exit from capture completion Report exit and running state from the child exit signal, independently of buffered output. On scope release, discard abandoned capture and retain the existing pipe-close or capture-deadline wait before applying process-group cleanup policy. Cover unread output after confirmed process exit, successful descendant survival, and the approved policy that a parent exiting successfully before its invocation timeout retains success while the bounded capture grace finishes.	2026-08-31 15:12:01 -04:00
Kit Langton	ef8b8c1b8d	fix(util): preserve process output for late readers Buffer child stdout and stderr before Effect consumers attach, retaining stream backpressure and scoped cleanup. Detach capture buffers before the existing post-exit discard deadline drains inherited pipes. Cover post-exit readers, output larger than the buffers, and teardown with unread stdout through the real process spawner.	2026-08-31 14:55:04 -04:00
Dax Raad	5df9cecf03	fix(tui): remove plugin current marker	2026-08-31 14:31:03 -04:00
Dax Raad	a68fe8a97d	fix(tui): toggle plugin on dialog submit	2026-08-31 14:28:19 -04:00
Dax Raad	c17c104827	fix(tui): toggle internal plugin controls	2026-08-31 14:25:06 -04:00
Dax Raad	5d4cc4a804	feat(tui): hide internal plugins by default	2026-08-31 14:25:06 -04:00
Kit Langton	1f04baa684	test: migrate fixture layer replacements (#46458 ) Update the Core compile options and Server replacement values to the current LayerNode API. Preserve test expectations, replacement targets, and layer lifetimes.	2026-08-31 14:16:40 -04:00
Kit Langton	3e9b009642	feat(core): add session-aware instance selection (#46442 )	2026-08-31 13:46:33 -04:00
Kit Langton	36ac35a7c8	refactor(util): make layer graphs opaque and composable Replace exposed layer graph assembly with opaque declarations, checked substitutions, and lifetime-aware compilation. Preserve deep replacement, ordered startup, and Effect-owned resource lifetimes; migrate callers and verify source and published package contracts.	2026-08-31 13:46:27 -04:00
Kit Langton	197d28e033	fix(tui): pin sidebar headings without scrollbar flashes (#46449 ) Keep the title and workspace label above scrollable sidebar details. Disable the unused horizontal scrollbar and place the automatic vertical scrollbar in the reserved gutter so tab changes do not flash or shift the sidebar.	2026-08-31 13:45:55 -04:00
opencode-agent[bot]andkitlangton	fcce2d7cc9	test(tui): await dialog text selection (#46143 ) Co-authored-by: kitlangton <7587245+kitlangton@users.noreply.github.com>	2026-08-31 13:44:00 -04:00
Dax Raad	ec0dcb3da9	docs: improve build documentation discovery	2026-08-31 12:52:59 -04:00
Kit Langton	afd7492018	fix(tui): reduce cached transcript remount work (#46145 ) Configure custom Markdown renderers before assigning content and share a reactive message-position index across assistant footers. Preserve completion ordering and historical footer metrics, with regression coverage for prepend, same-length refresh, and revert.	2026-08-31 12:15:39 -04:00
Kit Langton	9517ff1054	fix(core): preserve active session continuation when moving	2026-08-31 12:14:45 -04:00
Kit Langton	1ced747051	fix(ai): handle message-less Gemini errors (#46069 )	2026-08-31 12:14:31 -04:00
opencode-agent[bot]andDavid	43819dc376	fix(app): restore maskable pwa icons (#46434 ) Co-authored-by: David <1879069+iamdavidhill@users.noreply.github.com>	2026-09-01 00:11:07 +08:00
Kit Langton	e15dd8ecd3	fix(ai): require Bedrock message stop for finish (#46065 )	2026-08-31 12:03:22 -04:00
Dax	6a38cacc1d	docs: improve plugin guide readability (#46342 )	2026-08-31 11:48:04 -04:00
Kit Langton	d609752891	refactor(codemode): avoid merging root definitions twice (#46081 )	2026-08-31 11:40:17 -04:00
Brendan Allan	5894e46688	fix(app): improve touch controls and standalone PWA relaunch (#46391 )	2026-08-31 23:35:40 +08:00
Kit Langton	327dc809c5	refactor(core): reuse formatter file extension (#46080 )	2026-08-31 11:30:57 -04:00
Kit Langton	e9f7331516	refactor(core): reuse Markdown chunk byte counts (#46079 )	2026-08-31 11:30:49 -04:00
Kit Langton	8be3ce8b6c	refactor(util): reuse BOM-stripped text (#46078 )	2026-08-31 11:30:40 -04:00
Kit Langton	30721b8b5d	fix(server): await providers before catalog reads (#46066 )	2026-08-31 11:24:56 -04:00
				`@@ -1 +0,0 @@`
				`export { messagesModel as model } from "../amazon-bedrock.js"`