Realtime Flow

src/agent_session/main.py

@@ -87,40 +87,45 @@ async def run_agent_session(room_name: str, agent_token: str):
         await room.connect(livekit_url, agent_token)
         print("DEBUG: 2. Connection successful.")
 
-        # 1. Instantiate all plugins
-        vad = VAD.load(min_speech_duration=0.1, min_silence_duration=0.5)
-        tts = TTS(
-            voice=VoiceByName(name="Tiktok Fashion Influencer", provider=VoiceProvider.hume),
-            instant_mode=True,
-        )
-        stt = StreamAdapter(stt=STT(model="whisper-large-v3-turbo"), vad=vad)
-        google_llm = GoogleLLM(model="gemini-2.5-flash")
-        llm_wrapper = LLMStateWrapper(llm=google_llm, room=room)
-
-        # 2. Create the AgentSession, passing the plugins to the constructor
-        session = AgentSession(
-            stt=stt,
-            llm=llm_wrapper,
-            tts=tts,
-            vad=vad,
-        )
-
-        # 3. Create an instance of your agent
-        agent = VoiceAssistant()
-
-        # 4. Start the session. This runs the main loop until disconnection.
-        print("DEBUG: Starting AgentSession...")
-        # This will internally call agent.on_connected after connecting.
-        task = asyncio.create_task(session.start(agent=agent, room=room))
-
-        # 5. Generate the initial greeting after the session has started
-        print("DEBUG: Generating initial greeting...")
-        await send_agent_state(room, "speaking")
-        await session.generate_reply(instructions="hello")
-        print("DEBUG: Initial greeting complete. Agent is now fully operational.")
-        
-        # Await the main session task to keep it running
-        await task
+        # Create HTTP session for TTS
+        async with aiohttp.ClientSession() as session:
+            # 1. Instantiate all plugins
+            vad = VAD.load(min_speech_duration=0.1, min_silence_duration=0.5)
+            tts = TTS(
+                voice=VoiceByName(
+                    name="Tiktok Fashion Influencer", 
+                    provider=VoiceProvider.hume
+                ),
+                instant_mode=True,
+                http_session=session,  # Pass the session here
+            )
+            stt = StreamAdapter(stt=STT(model="whisper-large-v3-turbo"), vad=vad)
+            google_llm = GoogleLLM(model="gemini-2.5-flash")
+            llm_wrapper = LLMStateWrapper(llm=google_llm, room=room)
+
+            # 2. Create the AgentSession
+            session_agent = AgentSession(
+                stt=stt,
+                llm=llm_wrapper,
+                tts=tts,
+                vad=vad,
+            )
+
+            # 3. Create an instance of your agent
+            agent = VoiceAssistant()
+
+            # 4. Start the session
+            print("DEBUG: Starting AgentSession...")
+            task = asyncio.create_task(session_agent.start(agent=agent, room=room))
+
+            # 5. Generate the initial greeting
+            print("DEBUG: Generating initial greeting...")
+            await send_agent_state(room, "speaking")
+            await session_agent.generate_reply(instructions="hello")
+            print("DEBUG: Initial greeting complete. Agent is now fully operational.")
+            
+            # Await the main session task
+            await task
 
     except Exception as e:
         print(f"FATAL ERROR in agent session: {e}")