feat(llm): add response reducer (#34417)

packages/llm/test/adapter.test.ts

@@ -115,6 +115,7 @@ describe("llm route", () => {
 
       expect(events.map((event) => event.type)).toEqual(["text-delta", "finish"])
       expect(response.events.map((event) => event.type)).toEqual(["text-delta", "finish"])
+      expect(response.message.content).toEqual([{ type: "text", text: 'echo:{"body":"hello"}' }])
     }),
   )
 

packages/llm/test/fixtures/recordings/gemini-cache/reports-cachedcontenttokencount-on-identical-second-call.json

@@ -21,7 +21,7 @@
         "headers": {
           "content-type": "text/event-stream"
         },
-        "body": ""
+        "body": "data: {\"candidates\":[{\"content\":{\"role\":\"model\",\"parts\":[{\"text\":\"Hi.\"}]},\"finishReason\":\"STOP\"}],\"usageMetadata\":{\"promptTokenCount\":1200,\"candidatesTokenCount\":2,\"totalTokenCount\":1202}}\n\n"
       }
     },
     {
@@ -39,7 +39,7 @@
         "headers": {
           "content-type": "text/event-stream"
         },
-        "body": ""
+        "body": "data: {\"candidates\":[{\"content\":{\"role\":\"model\",\"parts\":[{\"text\":\"Hi.\"}]},\"finishReason\":\"STOP\"}],\"usageMetadata\":{\"cachedContentTokenCount\":1100,\"promptTokenCount\":1200,\"candidatesTokenCount\":2,\"totalTokenCount\":1202}}\n\n"
       }
     }
   ]

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

@@ -1,7 +1,7 @@
 import { describe, expect } from "bun:test"
 import { Effect, Schema, Stream } from "effect"
 import { HttpClientRequest } from "effect/unstable/http"
-import { LLM, LLMError, Message, Model, ToolCallPart, Usage } from "../../src"
+import { LLM, LLMError, LLMEvent, Message, Model, ToolCallPart, Usage } from "../../src"
 import * as Azure from "../../src/providers/azure"
 import * as OpenAI from "../../src/providers/openai"
 import * as OpenAIChat from "../../src/protocols/openai-chat"
@@ -597,19 +597,22 @@ describe("OpenAI Chat route", () => {
         }),
         deltaChunk({ tool_calls: [{ index: 0, function: { arguments: ':"weather"}' } }] }),
       )
-      const response = yield* LLMClient.generate(
-        LLM.updateRequest(request, {
-          tools: [{ name: "lookup", description: "Lookup data", inputSchema: { type: "object" } }],
-        }),
-      ).pipe(Effect.provide(fixedResponse(body)))
+      const input = LLM.updateRequest(request, {
+        tools: [{ name: "lookup", description: "Lookup data", inputSchema: { type: "object" } }],
+      })
+      const events = Array.from(
+        yield* LLMClient.stream(input).pipe(Stream.runCollect, Effect.provide(fixedResponse(body))),
+      )
+      const error = yield* LLMClient.generate(input).pipe(Effect.provide(fixedResponse(body)), Effect.flip)
 
-      expect(response.events).toEqual([
+      expect(events).toEqual([
         { type: "step-start", index: 0 },
         { type: "tool-input-start", id: "call_1", name: "lookup", providerMetadata: undefined },
         { type: "tool-input-delta", id: "call_1", name: "lookup", text: '{"query"' },
         { type: "tool-input-delta", id: "call_1", name: "lookup", text: ':"weather"}' },
       ])
-      expect(response.toolCalls).toEqual([])
+      expect(events.filter(LLMEvent.is.toolCall)).toEqual([])
+      expect(error.message).toContain("Provider stream ended without a terminal finish event")
     }),
   )
 

packages/llm/test/response.test.ts

@@ -0,0 +1,60 @@
+import { describe, expect, test } from "bun:test"
+import { LLMEvent, LLMResponse } from "../src"
+
+const reduce = (events: ReadonlyArray<LLMEvent>) => events.reduce(LLMResponse.reduce, LLMResponse.empty())
+
+describe("LLMResponse reducer", () => {
+  test("assembles interleaved reasoning and text with end metadata", () => {
+    const response = LLMResponse.fromEvents([
+      LLMEvent.reasoningStart({ id: "r1" }),
+      LLMEvent.reasoningDelta({ id: "r1", text: "I should " }),
+      LLMEvent.textStart({ id: "t1" }),
+      LLMEvent.reasoningDelta({ id: "r1", text: "compare..." }),
+      LLMEvent.reasoningEnd({ id: "r1", providerMetadata: { anthropic: { signature: "sig" } } }),
+      LLMEvent.textDelta({ id: "t1", text: "Answer" }),
+      LLMEvent.textEnd({ id: "t1" }),
+      LLMEvent.finish({ reason: "stop", usage: { outputTokens: 5 } }),
+    ])
+
+    expect(response?.finishReason).toBe("stop")
+    expect(response?.usage).toMatchObject({ outputTokens: 5 })
+    expect(response?.message.content).toEqual([
+      {
+        type: "reasoning",
+        text: "I should compare...",
+        providerMetadata: { anthropic: { signature: "sig" } },
+      },
+      { type: "text", text: "Answer" },
+    ])
+    expect(response?.events).toHaveLength(8)
+  })
+
+  test("preserves partial content without completing a failed stream", () => {
+    const state = reduce([LLMEvent.textStart({ id: "t1" }), LLMEvent.textDelta({ id: "t1", text: "partial" })])
+
+    expect(LLMResponse.complete(state)).toBeUndefined()
+    expect(state.message.content).toEqual([{ type: "text", text: "partial" }])
+  })
+
+  test("assembles tool-call content only after the completed tool call event", () => {
+    const pending = reduce([
+      LLMEvent.toolInputStart({ id: "call_1", name: "lookup" }),
+      LLMEvent.toolInputDelta({ id: "call_1", name: "lookup", text: '{"query"' }),
+    ])
+
+    expect(pending.message.content).toEqual([])
+    expect(pending.toolInputs.call_1?.text).toBe('{"query"')
+
+    const response = LLMResponse.fromEvents([
+      ...pending.events,
+      LLMEvent.toolInputDelta({ id: "call_1", name: "lookup", text: ':"weather"}' }),
+      LLMEvent.toolInputEnd({ id: "call_1", name: "lookup" }),
+      LLMEvent.toolCall({ id: "call_1", name: "lookup", input: { query: "weather" } }),
+      LLMEvent.finish({ reason: "tool-calls" }),
+    ])
+
+    expect(response?.message.content).toEqual([
+      { type: "tool-call", id: "call_1", name: "lookup", input: { query: "weather" } },
+    ])
+  })
+})