Sync from internal repo (YYYY/MM/DD)

package.json

@@ -1,6 +1,6 @@
 {
   "name": "assemblyai",
-  "version": "4.26.0",
+  "version": "4.26.1",
   "description": "The AssemblyAI JavaScript SDK provides an easy-to-use interface for interacting with the AssemblyAI API, which supports async and real-time transcription, as well as the latest LeMUR models.",
   "engines": {
     "node": ">=18"

src/services/streaming/service.ts

@@ -13,6 +13,7 @@ import {
   BeginEvent,
   StreamingEventMessage,
   TurnEvent,
+  LLMGatewayResponseEvent,
   StreamingUpdateConfiguration,
   StreamingForceEndpoint,
 } from "../..";
@@ -171,13 +172,21 @@ export class StreamingTranscriber {
       searchParams.set("max_speakers", this.params.maxSpeakers.toString());
     }
 
+    if (this.params.llmGateway !== undefined) {
+      searchParams.set("llm_gateway", JSON.stringify(this.params.llmGateway));
+    }
+
     url.search = searchParams.toString();
 
     return url;
   }
 
   on(event: "open", listener: (event: BeginEvent) => void): void;
   on(event: "turn", listener: (event: TurnEvent) => void): void;
+  on(
+    event: "llmGatewayResponse",
+    listener: (event: LLMGatewayResponseEvent) => void,
+  ): void;
   on(event: "error", listener: (error: Error) => void): void;
   on(event: "close", listener: (code: number, reason: string) => void): void;
   // eslint-disable-next-line @typescript-eslint/no-explicit-any
@@ -248,6 +257,10 @@ Learn more at https://github.com/AssemblyAI/assemblyai-node-sdk/blob/main/docs/c
             this.listeners.speechStarted?.(message);
             break;
           }
+          case "LLMGatewayResponse": {
+            this.listeners.llmGatewayResponse?.(message);
+            break;
+          }
           case "Termination": {
             this.sessionTerminatedResolve?.();
             break;

src/types/streaming/index.ts

@@ -1,5 +1,16 @@
 import { AudioEncoding } from "..";
 
+export type LLMGatewayMessage = {
+  role: string;
+  content: string;
+};
+
+export type LLMGatewayConfig = {
+  model: string;
+  messages: LLMGatewayMessage[];
+  max_tokens: number;
+};
+
 export type StreamingTranscriberParams = {
   websocketBaseUrl?: string;
   apiKey?: string;
@@ -24,20 +35,23 @@ export type StreamingTranscriberParams = {
   inactivityTimeout?: number;
   speakerLabels?: boolean;
   maxSpeakers?: number;
+  llmGateway?: LLMGatewayConfig;
 };
 
 export type StreamingEvents =
   | "open"
   | "close"
   | "turn"
   | "speechStarted"
+  | "llmGatewayResponse"
   | "error";
 
 export type StreamingListeners = {
   open?: (event: BeginEvent) => void;
   close?: (code: number, reason: string) => void;
   turn?: (event: TurnEvent) => void;
   speechStarted?: (event: SpeechStartedEvent) => void;
+  llmGatewayResponse?: (event: LLMGatewayResponseEvent) => void;
   error?: (error: Error) => void;
 };
 
@@ -125,11 +139,19 @@ export type ErrorEvent = {
   error: string;
 };
 
+export type LLMGatewayResponseEvent = {
+  type: "LLMGatewayResponse";
+  turn_order: number;
+  transcript: string;
+  data: unknown;
+};
+
 export type StreamingEventMessage =
   | BeginEvent
   | TurnEvent
   | SpeechStartedEvent
   | TerminationEvent
+  | LLMGatewayResponseEvent
   | ErrorEvent;
 
 export type StreamingOperationMessage =

tests/integration/streaming.test.ts

@@ -73,6 +73,7 @@ async function createStreamingTranscriber(useToken: boolean) {
   const serviceParams: StreamingTranscriberParams = {
     websocketBaseUrl: process.env.ASSEMBLYAI_STREAMING_WS_API_HOST,
     sampleRate: 16_000,
+    speechModel: "universal-streaming-english",
     apiKey: useToken ? undefined : process.env.ASSEMBLYAI_API_KEY,
     token: useToken
       ? await client.streaming.createTemporaryToken({

tests/unit/streaming.test.ts

@@ -134,4 +134,60 @@ describe("streaming", () => {
     const turn = await turnPromise;
     expect(turn.speaker_label).toBe("A");
   });
+
+  it("should include llm_gateway in connection URL", async () => {
+    await cleanup();
+    WS.clean();
+
+    const llmGatewayConfig = {
+      model: "claude-3-5-sonnet",
+      messages: [
+        { role: "system", content: "You are a helpful assistant." },
+        { role: "user", content: "Hello" },
+      ],
+      max_tokens: 100,
+    };
+
+    const wsUrl = `${websocketBaseUrl}?token=123&sample_rate=16000&speech_model=universal-streaming-english&llm_gateway=${encodeURIComponent(JSON.stringify(llmGatewayConfig))}`;
+    server = new WS(wsUrl);
+    rt = new StreamingTranscriber({
+      websocketBaseUrl,
+      token: "123",
+      sampleRate: 16_000,
+      speechModel: "universal-streaming-english",
+      llmGateway: llmGatewayConfig,
+    });
+    onOpen = jest.fn();
+    rt.on("open", onOpen);
+    await connect(rt, server);
+  });
+
+  it("should parse LLMGatewayResponse event", async () => {
+    const llmResponsePromise = new Promise<{
+      turn_order: number;
+      transcript: string;
+      data: unknown;
+    }>((resolve) => {
+      rt.on("llmGatewayResponse", (event) => resolve(event));
+    });
+
+    const llmResponseData = {
+      type: "LLMGatewayResponse",
+      turn_order: 1,
+      transcript: "hello world",
+      data: {
+        response: "This is an LLM response",
+        model: "claude-3-5-sonnet",
+      },
+    };
+
+    server.send(JSON.stringify(llmResponseData));
+    const response = await llmResponsePromise;
+    expect(response.turn_order).toBe(1);
+    expect(response.transcript).toBe("hello world");
+    expect(response.data).toEqual({
+      response: "This is an LLM response",
+      model: "claude-3-5-sonnet",
+    });
+  });
 });