fix(llm): surface native retry attempts

packages/llm/src/route/client.ts

@@ -1,7 +1,8 @@
-import { Cause, Context, Effect, Layer, Schema, Stream } from "effect"
+import { Cause, Context, Effect, Layer, Queue, Schema, Stream } from "effect"
 import type { Auth as AuthDef } from "./auth"
 import type { Endpoint } from "./endpoint"
 import { RequestExecutor } from "./executor"
+import type { RetryInfo } from "./executor"
 import type { Framing } from "./framing"
 import { HttpTransport } from "./transport"
 import type { Transport, TransportRuntime } from "./transport"
@@ -12,10 +13,11 @@ import { applyCachePolicy } from "../cache-policy"
 import * as ProviderShared from "../protocols/shared"
 import * as ToolRuntime from "../tool-runtime"
 import type { Tools } from "../tool"
-import type { LLMError, LLMEvent, PreparedRequestOf, ProtocolID } from "../schema"
+import type { LLMError, PreparedRequestOf, ProtocolID } from "../schema"
 import {
   GenerationOptions,
   HttpOptions,
+  LLMEvent,
   LLMRequest,
   LLMResponse,
   ModelID,
@@ -439,6 +441,50 @@ const streamRequestWith = (runtime: TransportRuntime) => (request: LLMRequest) =
     }),
   )
 
+const retryEvent = (info: RetryInfo) =>
+  Effect.sync(() => {
+    const http = "http" in info.error.reason ? info.error.reason.http : undefined
+    return LLMEvent.retry({
+      attempt: info.attempt,
+      delayMs: info.delayMs,
+      message: info.error.reason.message,
+      providerMetadata: {
+        opencode: {
+          module: info.error.module,
+          method: info.error.method,
+          reason: info.error.reason._tag,
+          status: ("status" in info.error.reason ? info.error.reason.status : undefined) ?? http?.response?.status,
+        },
+      },
+    })
+  })
+
+const streamRequestObservableWith = (runtime: TransportRuntime) => (request: LLMRequest) =>
+  Stream.callback<LLMEvent, LLMError>((queue) =>
+    Effect.gen(function* () {
+      const pull = yield* Stream.toPull(
+        streamRequestWith({
+          ...runtime,
+          httpOptions: {
+            onRetry: (info) => retryEvent(info).pipe(Effect.map((event) => Queue.offerUnsafe(queue, event))),
+          },
+        })(request),
+      )
+
+      const drain: Effect.Effect<void, never> = Effect.gen(function* () {
+        const chunk = yield* pull
+        for (const event of chunk) Queue.offerUnsafe(queue, event)
+        return yield* drain
+      }).pipe(
+        Effect.catchIf(Cause.isDone, () => Effect.sync(() => Queue.endUnsafe(queue))),
+        Effect.catchTag("LLM.Error", (error) => Queue.fail(queue, error)),
+      )
+
+      yield* drain
+    }),
+  )
+
+
 const isToolRunOptions = (input: LLMRequest | ToolRuntime.RunOptions<Tools>): input is ToolRuntime.RunOptions<Tools> =>
   "request" in input && "tools" in input
 
@@ -495,7 +541,7 @@ export const streamRequest = (request: LLMRequest) =>
 export const layer: Layer.Layer<Service, never, RequestExecutor.Service> = Layer.effect(
   Service,
   Effect.gen(function* () {
-    const stream = streamWith(streamRequestWith({ http: yield* RequestExecutor.Service }))
+    const stream = streamWith(streamRequestObservableWith({ http: yield* RequestExecutor.Service }))
     return Service.of({ prepare: prepareWith as Interface["prepare"], stream, generate: generateWith(stream) })
   }),
 )
@@ -505,7 +551,7 @@ export const layerWithWebSocket: Layer.Layer<Service, never, RequestExecutor.Ser
     Service,
     Effect.gen(function* () {
       const stream = streamWith(
-        streamRequestWith({
+        streamRequestObservableWith({
           http: yield* RequestExecutor.Service,
           webSocket: yield* WebSocketExecutor.Service,
         }),

packages/llm/src/route/executor.ts

@@ -26,9 +26,20 @@ import {
 export interface Interface {
   readonly execute: (
     request: HttpClientRequest.HttpClientRequest,
+    options?: ExecuteOptions,
   ) => Effect.Effect<HttpClientResponse.HttpClientResponse, LLMError>
 }
 
+export type RetryInfo = {
+  readonly attempt: number
+  readonly delayMs: number
+  readonly error: LLMError
+}
+
+export type ExecuteOptions = {
+  readonly onRetry?: (info: RetryInfo) => Effect.Effect<void>
+}
+
 export class Service extends Context.Service<Service, Interface>()("@opencode/LLM/RequestExecutor") {}
 
 const BODY_LIMIT = 16_384
@@ -341,14 +352,16 @@ const retryDelay = (error: LLMError, attempt: number) => {
 
 const retryStatusFailures = <A, R>(
   effect: Effect.Effect<A, LLMError, R>,
+  options: ExecuteOptions | undefined,
   retries = MAX_RETRIES,
   attempt = 0,
 ): Effect.Effect<A, LLMError, R> =>
   Effect.catchTag(effect, "LLM.Error", (error): Effect.Effect<A, LLMError, R> => {
     if (!error.retryable || retries <= 0) return Effect.fail(error)
     return retryDelay(error, attempt).pipe(
+      Effect.tap((delay) => options?.onRetry?.({ attempt: attempt + 1, delayMs: delay, error }) ?? Effect.void),
       Effect.flatMap((delay) => Effect.sleep(delay)),
-      Effect.flatMap(() => retryStatusFailures(effect, retries - 1, attempt + 1)),
+      Effect.flatMap(() => retryStatusFailures(effect, options, retries - 1, attempt + 1)),
     )
   })
 
@@ -364,7 +377,7 @@ export const layer: Layer.Layer<Service, never, HttpClient.HttpClient> = Layer.e
           .pipe(Effect.mapError(toHttpError(redactedNames)), Effect.flatMap(statusError(request, redactedNames)))
       })
     return Service.of({
-      execute: (request) => retryStatusFailures(executeOnce(request)),
+      execute: (request, options) => retryStatusFailures(executeOnce(request), options),
     })
   }),
 )

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

@@ -102,7 +102,7 @@ export const httpJson = <Body, Frame>(input: HttpJsonInput<Body, Frame>): HttpJs
   frames: (prepared, request, runtime) =>
     Stream.unwrap(
       runtime.http
-        .execute(prepared.request)
+        .execute(prepared.request, runtime.httpOptions)
         .pipe(
           Effect.map((response) =>
             prepared.framing.frame(

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

@@ -1,11 +1,12 @@
 import type { Effect, Stream } from "effect"
-import type { Interface as RequestExecutorInterface } from "../executor"
+import type { ExecuteOptions, Interface as RequestExecutorInterface } from "../executor"
 import type { Interface as WebSocketExecutorInterface } from "./websocket"
 import type { LLMError, LLMRequest } from "../../schema"
 
 export interface TransportRuntime {
   readonly http: RequestExecutorInterface
   readonly webSocket?: WebSocketExecutorInterface
+  readonly httpOptions?: ExecuteOptions
 }
 
 export interface Transport<Body, Prepared, Frame> {

packages/llm/src/schema/errors.ts

@@ -123,7 +123,7 @@ export class TransportReason extends Schema.Class<TransportReason>("LLM.Error.Tr
   http: Schema.optional(HttpContext),
 }) {
   get retryable() {
-    return false
+    return this.kind === "TransportError" || this.kind === "Timeout"
   }
 }
 

packages/llm/src/schema/events.ts

@@ -203,6 +203,15 @@ export const ProviderErrorEvent = Schema.Struct({
 }).annotate({ identifier: "LLM.Event.ProviderError" })
 export type ProviderErrorEvent = Schema.Schema.Type<typeof ProviderErrorEvent>
 
+export const RetryEvent = Schema.Struct({
+  type: Schema.tag("retry"),
+  attempt: Schema.Number,
+  delayMs: Schema.Number,
+  message: Schema.String,
+  providerMetadata: Schema.optional(ProviderMetadata),
+}).annotate({ identifier: "LLM.Event.Retry" })
+export type RetryEvent = Schema.Schema.Type<typeof RetryEvent>
+
 const llmEventTagged = Schema.Union([
   StepStart,
   TextStart,
@@ -220,6 +229,7 @@ const llmEventTagged = Schema.Union([
   StepFinish,
   Finish,
   ProviderErrorEvent,
+  RetryEvent,
 ]).pipe(Schema.toTaggedUnion("type"))
 
 type WithID<Event extends { readonly id: unknown }, ID> = Omit<Event, "type" | "id"> & { readonly id: ID | string }
@@ -265,6 +275,7 @@ export const LLMEvent = Object.assign(llmEventTagged, {
       usage: input.usage === undefined ? undefined : Usage.from(input.usage),
     }),
   providerError: ProviderErrorEvent.make,
+  retry: RetryEvent.make,
   is: {
     stepStart: llmEventTagged.guards["step-start"],
     textStart: llmEventTagged.guards["text-start"],
@@ -282,6 +293,7 @@ export const LLMEvent = Object.assign(llmEventTagged, {
     stepFinish: llmEventTagged.guards["step-finish"],
     finish: llmEventTagged.guards.finish,
     providerError: llmEventTagged.guards["provider-error"],
+    retry: llmEventTagged.guards.retry,
   },
 })
 export type LLMEvent = Schema.Schema.Type<typeof llmEventTagged>

packages/llm/test/executor.test.ts

@@ -1,12 +1,12 @@
 import { describe, expect } from "bun:test"
-import { Effect, Fiber, Layer, Random, Ref } from "effect"
+import { Effect, Fiber, Layer, Random, Ref, Stream } from "effect"
 import * as TestClock from "effect/testing/TestClock"
-import { Headers, HttpClient, HttpClientRequest, HttpClientResponse } from "effect/unstable/http"
-import { LLM, LLMError } from "../src"
+import { Headers, HttpClient, HttpClientError, HttpClientRequest, HttpClientResponse } from "effect/unstable/http"
+import { LLM, LLMError, TransportReason } from "../src"
 import { LLMClient, RequestExecutor } from "../src/route"
 import * as OpenAIChat from "../src/protocols/openai-chat"
-import { dynamicResponse } from "./lib/http"
-import { deltaChunk } from "./lib/openai-chunks"
+import { dynamicResponse, runtimeLayer } from "./lib/http"
+import { deltaChunk, finishChunk } from "./lib/openai-chunks"
 import { sseRaw } from "./lib/sse"
 import { it } from "./lib/effect"
 
@@ -18,23 +18,26 @@ const secretRequest = HttpClientRequest.post("https://provider.test/v1/chat?api_
   HttpClientRequest.setHeaders(Headers.fromInput({ authorization: "Bearer header-secret-456" })),
 )
 
+const httpResponsesLayer = (responses: ReadonlyArray<Response>) =>
+  Layer.unwrap(
+    Effect.gen(function* () {
+      const cursor = yield* Ref.make(0)
+      return Layer.succeed(
+        HttpClient.HttpClient,
+        HttpClient.make((request) =>
+          Effect.gen(function* () {
+            const index = yield* Ref.getAndUpdate(cursor, (value) => value + 1)
+            return HttpClientResponse.fromWeb(request, responses[index] ?? responses[responses.length - 1])
+          }),
+        ),
+      )
+    }),
+  )
+
 const responsesLayer = (responses: ReadonlyArray<Response>) =>
   RequestExecutor.layer.pipe(
     Layer.provide(
-      Layer.unwrap(
-        Effect.gen(function* () {
-          const cursor = yield* Ref.make(0)
-          return Layer.succeed(
-            HttpClient.HttpClient,
-            HttpClient.make((request) =>
-              Effect.gen(function* () {
-                const index = yield* Ref.getAndUpdate(cursor, (value) => value + 1)
-                return HttpClientResponse.fromWeb(request, responses[index] ?? responses[responses.length - 1])
-              }),
-            ),
-          )
-        }),
-      ),
+      httpResponsesLayer(responses),
     ),
   )
 
@@ -59,6 +62,30 @@ const countedResponsesLayer = (attempts: Ref.Ref<number>, responses: ReadonlyArr
     ),
   )
 
+const transportThenResponseLayer = (attempts: Ref.Ref<number>, response: Response) =>
+  RequestExecutor.layer.pipe(
+    Layer.provide(
+      Layer.succeed(
+        HttpClient.HttpClient,
+        HttpClient.make((request) =>
+          Ref.getAndUpdate(attempts, (value) => value + 1).pipe(
+            Effect.flatMap((index) =>
+              index === 0
+                ? Effect.fail(
+                    new HttpClientError.HttpClientError({
+                      reason: new HttpClientError.TransportError({
+                        request,
+                      }),
+                    }),
+                  )
+                : Effect.succeed(HttpClientResponse.fromWeb(request, response)),
+            ),
+          ),
+        ),
+      ),
+    ),
+  )
+
 const randomMidpoint = {
   nextDoubleUnsafe: () => 0.5,
   nextIntUnsafe: () => 0,
@@ -73,6 +100,23 @@ const expectLLMError = (error: unknown) => {
 const errorHttp = (error: LLMError) => ("http" in error.reason ? error.reason.http : undefined)
 
 describe("RequestExecutor", () => {
+  it.effect("only marks transient transport reasons retryable", () =>
+    Effect.sync(() => {
+      const retryable = (kind: string | undefined) =>
+        new LLMError({
+          module: "RequestExecutor",
+          method: "execute",
+          reason: new TransportReason({ message: "HTTP transport failed", kind }),
+        }).retryable
+
+      expect(retryable("TransportError")).toBe(true)
+      expect(retryable("Timeout")).toBe(true)
+      expect(retryable("EncodeError")).toBe(false)
+      expect(retryable("InvalidUrlError")).toBe(false)
+      expect(retryable(undefined)).toBe(false)
+    }),
+  )
+
   it.effect("returns redacted diagnostics for retryable rate limits", () =>
     Effect.gen(function* () {
       const executor = yield* RequestExecutor.Service
@@ -225,6 +269,57 @@ describe("RequestExecutor", () => {
     ),
   )
 
+  it.effect("reports retry attempts before retrying status responses", () =>
+    Effect.gen(function* () {
+      const executor = yield* RequestExecutor.Service
+      const retries: RequestExecutor.RetryInfo[] = []
+      const response = yield* executor.execute(request, {
+        onRetry: (info) => Effect.sync(() => retries.push(info)),
+      })
+
+      expect(response.status).toBe(200)
+      expect(retries).toHaveLength(1)
+      expect(retries[0].attempt).toBe(1)
+      expect(retries[0].delayMs).toBe(0)
+      expect(retries[0].error.reason).toMatchObject({ _tag: "ProviderInternal", status: 503 })
+    }).pipe(
+      Effect.provide(
+        responsesLayer([
+          new Response("busy", { status: 503, headers: { "retry-after-ms": "0" } }),
+          new Response("ok", { status: 200 }),
+        ]),
+      ),
+    ),
+  )
+
+  it.effect("emits retry events before silently retried stream requests", () =>
+    Effect.gen(function* () {
+      const model = OpenAIChat.model({ id: "gpt-4o-mini", baseURL: "https://api.openai.test/v1" })
+      const events = yield* LLMClient.stream(LLM.request({ model, prompt: "Say hello." })).pipe(
+        Stream.runCollect,
+        Effect.provide(
+          runtimeLayer(
+            httpResponsesLayer([
+              new Response("busy", { status: 503, headers: { "retry-after-ms": "0" } }),
+              new Response(
+                sseRaw(
+                  `data: ${JSON.stringify(deltaChunk({ role: "assistant", content: "Hello" }))}`,
+                  `data: ${JSON.stringify(finishChunk("stop"))}`,
+                ),
+                {
+                  headers: { "content-type": "text/event-stream" },
+                },
+              ),
+            ]),
+          ),
+        ),
+      )
+
+      expect(events[0]).toMatchObject({ type: "retry", attempt: 1, delayMs: 0 })
+      expect(events.some((event) => event.type === "text-delta")).toBe(true)
+    }),
+  )
+
   it.effect("marks 504 and 529 status responses retryable", () =>
     Effect.gen(function* () {
       const failWith = (status: number) =>
@@ -385,6 +480,30 @@ describe("RequestExecutor", () => {
     }).pipe(Effect.provideService(Random.Random, randomMidpoint)),
   )
 
+  it.effect("retries retryable transport failures before returning the stream", () =>
+    Effect.gen(function* () {
+      const attempts = yield* Ref.make(0)
+      return yield* Effect.gen(function* () {
+        const executor = yield* RequestExecutor.Service
+        const fiber = yield* executor.execute(request).pipe(Effect.forkChild)
+
+        yield* Effect.yieldNow
+        expect(yield* Ref.get(attempts)).toBe(1)
+
+        yield* TestClock.adjust(499)
+        yield* Effect.yieldNow
+        expect(yield* Ref.get(attempts)).toBe(1)
+
+        yield* TestClock.adjust(1)
+        const response = yield* Fiber.join(fiber)
+
+        expect(response.status).toBe(200)
+        expect(yield* response.text).toBe("ok")
+        expect(yield* Ref.get(attempts)).toBe(2)
+      }).pipe(Effect.provide(transportThenResponseLayer(attempts, new Response("ok", { status: 200 }))))
+    }).pipe(Effect.provideService(Random.Random, randomMidpoint)),
+  )
+
   it.effect("does not retry after a successful response reaches stream parsing", () =>
     Effect.gen(function* () {
       const attempts = yield* Ref.make(0)

packages/opencode/src/session/message-v2.ts

@@ -2,6 +2,7 @@ import { BusEvent } from "@/bus/bus-event"
 import { SessionID, MessageID, PartID } from "./schema"
 import { NamedError } from "@opencode-ai/core/util/error"
 import { APICallError, convertToModelMessages, LoadAPIKeyError, type ModelMessage, type UIMessage } from "ai"
+import { LLMError } from "@opencode-ai/llm"
 import { LSP } from "@/lsp/lsp"
 import { Snapshot } from "@/snapshot"
 import { SyncEvent } from "../sync"
@@ -1168,6 +1169,19 @@ export function fromError(
         },
         { cause: e },
       ).toObject()
+    case e instanceof LLMError:
+      const http = "http" in e.reason ? e.reason.http : undefined
+      return new APIError(
+        {
+          message: e.reason.message,
+          statusCode: ("status" in e.reason ? e.reason.status : undefined) ?? http?.response?.status,
+          isRetryable: e.retryable,
+          responseHeaders: http?.response?.headers,
+          responseBody: http?.body,
+          metadata: http?.request.url ? { url: http.request.url } : undefined,
+        },
+        { cause: e },
+      ).toObject()
     case e instanceof Error:
       return new NamedError.Unknown({ message: errorMessage(e) }, { cause: e }).toObject()
     default:

packages/opencode/src/session/processor.ts

@@ -303,6 +303,15 @@ export const layer = Layer.effect(
 
       const handleEvent = Effect.fnUntraced(function* (value: StreamEvent) {
         switch (value.type) {
+          case "retry":
+            yield* status.set(ctx.sessionID, {
+              type: "retry",
+              attempt: value.attempt,
+              message: value.message,
+              next: Date.now() + value.delayMs,
+            })
+            return
+
           case "reasoning-start":
             if (value.id in ctx.reasoningMap) return
             // TODO(v2): Temporary dual-write while migrating session messages to v2 events.

packages/opencode/test/session/message-v2.test.ts

@@ -1,5 +1,6 @@
 import { describe, expect, test } from "bun:test"
 import { APICallError } from "ai"
+import { HttpContext, HttpRequestDetails, HttpResponseDetails, LLMError, TransportReason } from "@opencode-ai/llm"
 import { MessageV2 } from "../../src/session/message-v2"
 import { ProviderTransform } from "@/provider/transform"
 import type { Provider } from "@/provider/provider"
@@ -1546,6 +1547,69 @@ describe("session.message-v2.fromError", () => {
 
     expect(result.name).toBe("MessageAbortedError")
   })
+
+  test("classifies native LLM transport errors as retryable APIError", () => {
+    const result = MessageV2.fromError(
+      new LLMError({
+        module: "RequestExecutor",
+        method: "execute",
+        reason: new TransportReason({
+          message: "HTTP transport failed",
+          kind: "TransportError",
+          url: "https://provider.test/v1/chat",
+          http: new HttpContext({
+            request: new HttpRequestDetails({
+              method: "POST",
+              url: "https://provider.test/v1/chat",
+              headers: { "x-safe": "visible" },
+            }),
+          }),
+        }),
+      }),
+      { providerID },
+    )
+
+    expect(MessageV2.APIError.isInstance(result)).toBe(true)
+    expect((result as MessageV2.APIError).data).toMatchObject({
+      message: "HTTP transport failed",
+      isRetryable: true,
+      metadata: { url: "https://provider.test/v1/chat" },
+    })
+  })
+
+  test("preserves native LLM HTTP response metadata for retry scheduling", () => {
+    const result = MessageV2.fromError(
+      new LLMError({
+        module: "RequestExecutor",
+        method: "execute",
+        reason: new TransportReason({
+          message: "HTTP transport failed",
+          kind: "TransportError",
+          http: new HttpContext({
+            request: new HttpRequestDetails({
+              method: "POST",
+              url: "https://provider.test/v1/chat",
+              headers: {},
+            }),
+            response: new HttpResponseDetails({
+              status: 503,
+              headers: { "retry-after-ms": "1000" },
+            }),
+            body: "busy",
+          }),
+        }),
+      }),
+      { providerID },
+    )
+
+    expect(MessageV2.APIError.isInstance(result)).toBe(true)
+    expect((result as MessageV2.APIError).data).toMatchObject({
+      statusCode: 503,
+      isRetryable: true,
+      responseHeaders: { "retry-after-ms": "1000" },
+      responseBody: "busy",
+    })
+  })
 })
 
 describe("session.message-v2.latest", () => {