LiveKit Agents Adapter

Bridge any external AI agent framework to the LiveKit Agent runtime.

Overview

This package provides AgentAdapter, a livekit.agents.voice.Agent subclass that delegates LLM inference to an external framework (LangChain, LangGraph, Strands, ADK, custom code, etc.) via a pluggable Translator interface.

The adapter is a pure pass-through:

1. Converts LiveKit's ChatContext to the external framework's native format
2. Forwards text and tool calls streamed from the external framework into LiveKit's llm_node
3. Syncs conversation history back to the LiveKit ChatContext after each turn

Tool calls are handled entirely within the external framework. The adapter does not execute tools. When the external framework streams a ToolCall, it is forwarded to LiveKit's ChatContext for transcript/logging purposes only. The external framework decides when and how to stream its final text response.

Quick Start

from livekit_agents_adapter import AgentAdapter, AgentAdapterConfig, Translator, TextChunk, Done


class MyTranslator(Translator):
    async def invoke(self, chat_ctx, tools):
        yield TextChunk(content="Hello!", is_final=True)
        yield Done(metadata={})

    # ... implement to_external_chat_ctx, to_external_tools, sync_back_messages


agent = AgentAdapter(
    config=AgentAdapterConfig(translator=MyTranslator(my_agent)),
    instructions="You are a helpful voice assistant.",
)

# Use like any LiveKit Agent in AgentSession...
await session.start(agent=agent, room=room)

Writing a Translator

Subclass Translator (an abc.ABC) and implement four required methods:

from livekit_agents_adapter import Translator, TextChunk, ToolCall, Done


class MyTranslator(Translator):
    async def to_external_chat_ctx(self, chat_ctx):
        # Convert LiveKit ChatContext → your framework's format
        return my_framework_messages

    async def to_external_tools(self, tools):
        # Convert LiveKit llm.Tool → your framework's tool format
        return my_framework_tools

    async def invoke(self, chat_ctx, tools):
        # Stream responses from your framework.
        # The adapter forwards TextChunk to TTS, ToolCall to ChatContext.
        # Yield Done to signal end of turn.
        async for chunk in my_framework_stream(chat_ctx, tools):
            yield TextChunk(content=chunk)  # or ToolCall(...)

        yield Done(metadata={})

    async def sync_back_messages(self, chat_ctx):
        # Return new ChatItems for the adapter to insert into ChatContext.
        from livekit_agents_adapter.translators.chat_context import sync_from_list_to_items

        return sync_from_list_to_items(self._messages[self._last_synced_count :])

Installation

# Core package
pip install livekit-agents-adapter

# With LangChain support
pip install livekit-agents-adapter-langchain

More

• Source: https://github.com/guttume/livekit_agents_adapter
• Examples: see the examples/ directory in the source repo