Files
matrix-ai-agent/agent.py
Christian Gick fa65fbeb3d feat: Matrix AI voice agent (LiveKit + LiteLLM)
Bot @ai:agiliton.eu accepts room invites, dispatches LiveKit agent.
Agent joins call with STT (Groq Whisper) → LLM (Sonnet) → TTS (ElevenLabs)
pipeline, all routed through LiteLLM.

CF-1147

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-15 07:31:52 +02:00

59 lines
1.6 KiB
Python

import os
import logging
from livekit.agents import Agent, AgentSession, AgentServer, JobContext, JobProcess, cli
from livekit.plugins import openai as lk_openai, elevenlabs, silero
logger = logging.getLogger("matrix-ai-agent")
LITELLM_URL = os.environ["LITELLM_BASE_URL"]
LITELLM_KEY = os.environ.get("LITELLM_API_KEY", "not-needed")
SYSTEM_PROMPT = """You are a helpful voice assistant in a Matrix call.
Rules:
- Keep answers SHORT — 1-3 sentences max
- Be direct, no filler words
- If the user wants more detail, they will ask
- Speak naturally as in a conversation"""
server = AgentServer()
def prewarm(proc: JobProcess):
proc.userdata["vad"] = silero.VAD.load()
server.setup_fnc = prewarm
@server.rtc_session(agent_name="matrix-ai")
async def entrypoint(ctx: JobContext):
model = os.environ.get("LITELLM_MODEL", "claude-sonnet")
voice_id = os.environ.get("ELEVENLABS_VOICE_ID", "21m00Tcm4TlvDq8ikWAM")
session = AgentSession(
stt=lk_openai.STT(
base_url=LITELLM_URL,
api_key=LITELLM_KEY,
model="whisper",
),
llm=lk_openai.LLM(
base_url=LITELLM_URL,
api_key=LITELLM_KEY,
model=model,
),
tts=elevenlabs.TTS(
voice=voice_id,
model="eleven_multilingual_v2",
),
vad=ctx.proc.userdata["vad"],
)
agent = Agent(instructions=SYSTEM_PROMPT)
await session.start(agent=agent, room=ctx.room)
await session.generate_reply(instructions="Greet the user briefly.")
if __name__ == "__main__":
cli.run_app(server)