#C088ZNU7QQ5 Does livekit have integrations with langchain/langgraph or Agno? Has anyone integrates there workflows/Agents of Langchain and Langgraph with Livekit found this langgraph-livekit-plugin but not working. Would be Greatfull if someone can send the eg code/resources of integrating both in workflows.

Thanks for reaching out, Abbas. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

yes thats what i provided but there are no eg code to startup with and the graph i have made is not integrating with this adapter the TTS and the graph are not being comnnected to the main workflwo

Hey there! Let's walk through how to connect your custom LangGraph workflow with LiveKit's

LLMAdapter

to create a smooth voice agent integration. I'll break it down in a friendly, easy-to-understand way.

from langgraph.graph import StateGraph, START
from livekit.agents import Agent, AgentSession, JobContext
from livekit.plugins import langchain, deepgram, silero
from livekit.plugins.turn_detector.multilingual import MultilingualModel

class State(TypedDict):
    messages: Annotated[list[BaseMessage], add_messages]

def create_graph():
    # Define your custom workflow
    builder = StateGraph(State)
    builder.add_node("chatbot", chatbot_node)
    builder.add_edge(START, "chatbot")
    return builder.compile()

async def entrypoint(ctx: JobContext):
    graph = create_graph()
    agent = Agent(
        instructions="",
        llm=langchain.LLMAdapter(graph),
    )
    session = AgentSession(
        vad=silero.VAD.load(),
        stt=deepgram.STT(model="nova-3", language="multi"),
        tts=deepgram.TTS(),
        turn_detection=MultilingualModel(),
    )
    await session.start(
        agent=agent,
        room=ctx.room,
    )

Let me walk you through the key integration points: •

LLMAdapter

is like a friendly translator that connects your LangGraph workflow with the LiveKit agent. It helps route messages smoothly. • Make sure to include compatible STT, TTS, and VAD components in

AgentSession

. Don't worry - non-streaming components get automatically wrapped to play nice together. • Set up turn detection to keep the conversation flowing naturally. • Just double-check that your workflow nodes can handle the message format (LiveKit and LLM messages get converted automatically, so it's pretty seamless). If you're still scratching your head about TTS or graph connections, no worries! Double-check that your workflow is compiled correctly and that

AgentSession

components are playing well together. The official example

voice_agents/langgraph_agent.py

is a great reference - it shows a tried-and-true setup (using

deepgram

and

silero

as plugin solutions, but you can swap these out). Stuck on something specific? Drop me the details about your graph structure or any errors you're seeing, and I'll be happy to help you troubleshoot! Sources: livekit-agents/livekit/agents/voice/agent.py | LLM Integrations | livekit/livekit_composite | DeepWiki

from langgraph.graph import StateGraph, START,END from livekit.agents import Agent, AgentSession, JobContext,cli,WorkerOptions from livekit.plugins import langchain, deepgram, silero from livekit.plugins.turn_detector.multilingual import MultilingualModel from typing import TypedDict, Annotated from langchain_core.messages import BaseMessage, HumanMessage, AIMessage from langgraph.graph import add_messages from langchain_groq import ChatGroq import os from dotenv import load_dotenv load_dotenv() class State(TypedDict): messages: Annotated[list[BaseMessage], add_messages] def chatbot_node(state: State): """ This node acts as a normal LLM call to Groq. """ llm = ChatGroq(model="llama3-8b-8192", temperature=0.7) # You can choose other Groq models response = llm.invoke(state["messages"]) return {"messages": [response]} def create_graph(): # Define your custom workflow builder = StateGraph(State) builder.add_node("chatbot", chatbot_node) builder.add_edge(START, "chatbot") # You might want to add a conditional edge or END here based on your flow # For a simple chat, you could loop back to chatbot or define an END condition. return builder.compile() async def entrypoint(ctx: JobContext): graph = create_graph() agent = Agent( instructions="", llm=langchain.LLMAdapter(graph), ) session = AgentSession( vad=silero.VAD.load(), stt=deepgram.STT(model="nova-3", language="multi"), tts=deepgram.TTS(), turn_detection=MultilingualModel(), ) await session.start( agent=agent, room=ctx.room, ) if name == "__main__": cli.run_app( WorkerOptions( entrypoint_fnc= entrypoint ) ) hey this is a sample code here the stt is working the langgraph graph agent is also working but why i am not getting any output back from the tts can you look into how can i let the tts work for thw ouput

I don't have the answer you're looking for. You could also try asking your question: • in one of the other Slack channels or • to https://deepwiki.com/livekit/livekit_composite which is trained on all LiveKit source code If you find the answer, please post it here to help others!