livekit/python-sdks

LiveKit Python SDK

What it is – A pair of Python packages that let you connect a Python program to a LiveKit server (cloud‑hosted or self‑hosted) and act as a real‑time participant in a video/audio room. One package (livekit) provides the real‑time client API; the other (livekit‑api) offers server‑side helpers such as token creation and REST‑style room management.

Why it matters – LiveKit is a platform for low‑latency, multi‑party video, audio and data streams. By exposing these capabilities in Python, the SDK makes it easy to build:

• AI‑powered voice agents (speech‑to‑text, LLM‑driven dialogue, text‑to‑speech, turn detection) that run in a live call.
• Real‑time streaming or broadcasting apps that need to publish or consume media from Python code.
• Custom back‑ends that orchestrate rooms, egress/ingress, SIP dialing, or agent dispatch via the server API.

Key components

Component	Purpose
livekit	Connects to a LiveKit room, publishes/receives audio‑video tracks, handles events, and supports RPC between participants.
livekit-api	Generates JWT access tokens, calls server endpoints (create/list rooms, egress, SIP, agent dispatch, etc.) using async aiohttp.
MediaDevices (in livekit.rtc)	High‑level wrapper around the sounddevice library for microphone capture and speaker playback, with built‑in acoustic echo cancellation, noise suppression, etc.
RPC system	Register a Python coroutine as a remote method and invoke it from another participant; useful for forwarding LLM function calls.
Agents integration	The SDK can be combined with the separate LiveKit Agents framework to run voice‑AI bots that internally use STT, LLM, TTS, turn detection, and tool usage without managing provider keys.

Typical workflow

1. Install – pip install livekit livekit-api.
2. Create a token (backend) with api.AccessToken() and appropriate grants.
3. Connect – room = rtc.Room(); await room.connect(url, token).
4. Publish media – use MediaDevices to open a microphone or speaker, wrap the source in LocalAudioTrack/LocalVideoTrack, and publish it.
5. Handle events – register callbacks like room.on("track_subscribed", ...) to receive remote streams.
6. Optional RPC – register a method with @room.local_participant.register_rpc_method and call it from another participant via perform_rpc.
7. Server‑side management – use LiveKitAPI to create rooms, list participants, start egress recordings, dial SIP participants, etc.

Example snippets

# generate a token (backend)
from livekit import api
token = api.AccessToken().with_identity("bot").with_grants(
    api.VideoGrants(room_join=True, room="demo")
).to_jwt()

# connect and publish microphone
from livekit import rtc
room = rtc.Room()
await room.connect(LIVEKIT_URL, token)

devices = rtc.MediaDevices()
mic = devices.open_input(enable_aec=True, noise_suppression=True)
track = rtc.LocalAudioTrack.create_audio_track("mic", mic.source)
await room.local_participant.publish_track(track)

Where to go next

• Browse the examples/ folder (basic room, face‑landmark, hue video, full‑duplex audio).
• Read the LiveKit Agents docs if you want a voice‑AI bot that bundles STT/LLM/TTS.
• Join the LiveKit Slack community for help and to contribute.

License – The SDK is released under the MIT License (as indicated in the repository).