Update starter to use new turn detector model

README.md

@@ -94,7 +94,7 @@ lk app env -w -d .env.local
 
 ## Run the agent
 
-Before your first run, you must download certain models such as [Silero VAD](https://docs.livekit.io/agents/logic/turns/vad/) and the [LiveKit turn detector](https://docs.livekit.io/agents/logic/turns/turn-detector/):
+The [LiveKit turn detector](https://docs.livekit.io/agents/logic/turns/turn-detector/) and its VAD are now built into the Agents SDK, so no model download is required before your first run. If you add a plugin that ships its own model weights (any `@livekit/agents-plugin-*` package), download them with:
 
 ```console
 pnpm run download-files

package.json

@@ -34,9 +34,7 @@
     "vitest": "^4.1.4"
   },
   "dependencies": {
-    "@livekit/agents": "^1.4.3",
-    "@livekit/agents-plugin-livekit": "^1.4.3",
-    "@livekit/agents-plugin-silero": "^1.4.3",
+    "@livekit/agents": "^1.5.0",
     "@livekit/plugins-ai-coustics": "^0.2.14",
     "dotenv": "^17.4.1",
     "zod": "^3.25.76"

src/main.ts

@@ -1,6 +1,4 @@
 import { ServerOptions, cli, defineAgent, inference, voice } from '@livekit/agents';
-import * as livekit from '@livekit/agents-plugin-livekit';
-import * as silero from '@livekit/agents-plugin-silero';
 import { audioEnhancement } from '@livekit/plugins-ai-coustics';
 import dotenv from 'dotenv';
 import { fileURLToPath } from 'node:url';
@@ -11,14 +9,7 @@ import { Agent } from './agent';
 // when running locally or self-hosting your agent server.
 dotenv.config({ path: '.env.local' });
 
-interface ProcessUserData {
-  vad: silero.VAD;
-}
-
-export default defineAgent<ProcessUserData>({
-  prewarm: async (proc) => {
-    proc.userData.vad = await silero.VAD.load();
-  },
+export default defineAgent({
   entry: async (ctx) => {
     // Set up a voice AI pipeline using OpenAI, Cartesia, Deepgram, and the LiveKit turn detector
     const session = new voice.AgentSession({
@@ -36,13 +27,15 @@ export default defineAgent<ProcessUserData>({
         voice: '9626c31c-bec5-4cca-baa8-f8ba9e84c8bc',
       }),
 
-      // VAD and turn detection are used to determine when the user is speaking and when the agent should respond
-      // See more at https://docs.livekit.io/agents/build/turns
-      turnDetection: new livekit.turnDetector.MultilingualModel(),
-      vad: ctx.proc.userData.vad,
-      voiceOptions: {
+      // Turn detection determines when the user is speaking and when the agent should respond.
+      // The LiveKit audio turn detector is a multimodal model that encodes the user's audio
+      // directly to predict end of turn. It's built into the SDK (no extra plugin) and
+      // AgentSession supplies the required VAD automatically.
+      // See more at https://docs.livekit.io/agents/logic/turns/turn-detector/
+      turnHandling: {
+        turnDetection: new inference.AudioTurnDetector(),
         // Allow the LLM to generate a response while waiting for the end of turn
-        preemptiveGeneration: true,
+        preemptiveGeneration: { enabled: true },
       },
     });