Skip to Content
Compatible APIsLiveKit

LiveKit agents (Cartesia plugin)

Use Revolab voices in a livekit-agents voice agent through the stock livekit-plugins-cartesia plugin — no custom plugin needed, just a base_url.

What the plugin talks to

WS   wss://api.revolab.ai/tts/websocket
POST https://api.revolab.ai/tts/bytes

Agent configuration

# pip install livekit-agents livekit-plugins-cartesia
from livekit.agents import AgentSession
from livekit.plugins import cartesia
 
session = AgentSession(
    # ... your STT / LLM / VAD config ...
    tts=cartesia.TTS(
        api_key="rvl_live_...",              # your Revolab API key
        base_url="https://api.revolab.ai",   # point the plugin at Revolab
        model="nada-1.0-flash",
        voice="<your-voice-id>",
        sample_rate=24000,
    ),
)

Behaviour notes

AreaBehaviour
How it connectsThe plugin drives the Cartesia-compatible websocket (wss …/tts/websocket) with your key in the X-API-Key header. base_url is all you change — the plugin derives the websocket URL from it.
Per-sentence latencyThe plugin streams one sentence per message with max_buffer_delay_ms=0; Revolab synthesizes each sentence as it arrives, so agent speech starts after the first sentence — not after the whole LLM turn.
One-shot pathtts.synthesize(...) uses POST /tts/bytes with the raw PCM container — also supported.
Word timestampsThe plugin requests word timestamps for synced transcripts; Revolab produces none yet, which the plugin tolerates — audio and transcripts work, per-word alignment is simply absent.
Models & voicesPass a Revolab model (nada-1.0-flash / nada-1.0-pro) and a voice ID from your voice library — the plugin defaults to sonic-3, which returns a 400 listing valid models.

The full websocket contract lives on the Cartesia compatibility page.

Cartesia