memoair-livekit 0.3.2

LiveKit Agent integration for MemoAir voice memory

memoair-livekit

LiveKit Agent integration for MemoAir voice memory. Drop-in livekit.agents.Agent subclass that searches memory before every LLM turn and persists each completed turn.

Install

pip install memoair-livekit

Quick start

import asyncio
from livekit.agents import AgentSession
from memoair_livekit import MemoAirLiveKitAgent

async def entrypoint(ctx):
    agent = MemoAirLiveKitAgent(
        api_key="memoair_pk_...",       # account-scoped API key
        project_id="proj_xxx",          # which customer project
        agent_id="agent_xxx",           # which agent identity
        user_id=ctx.room.local_participant.identity,
        instructions="You are a helpful voice assistant.",
    )

    # Or read MEMOAIR_API_KEY / MEMOAIR_PROJECT_ID / MEMOAIR_AGENT_ID /
    # MEMOAIR_USER_ID from the environment:
    # agent = MemoAirLiveKitAgent.from_env(instructions="...")

    session = AgentSession(vad=..., stt=..., llm=..., tts=...)

    # Required: LiveKit's base Agent does NOT call on_agent_response_completed.
    # Without this wiring, search runs every turn but nothing is saved.
    @session.on("conversation_item_added")
    def _on_item_added(event):
        if event.item.role == "assistant":
            asyncio.create_task(
                agent.on_agent_response_completed(session.chat_ctx, event.item)
            )

    await session.start(agent=agent, room=ctx.room)

Full docs: https://docs.memoair.dev/voice/integrations/livekit