livekit-plugins-mistralai 1.5.4

LiveKit Agents Plugin for services from Mistral AI

Mistral AI Plugin for LiveKit Agents

Support for Mistral AI STT, TTS, and LLM services.

Installation

pip install livekit-plugins-mistralai

For streaming STT (Voxtral Realtime), also install silero plugin.

pip install livekit-plugins-silero

Pre-requisites

You'll need an API key from Mistral AI. It can be set as an environment variable:

export MISTRAL_API_KEY=your_api_key_here

Usage

Speech-to-Text (STT)

Offline transcription

from livekit.plugins import mistralai

stt = mistralai.STT()

# With context biasing
stt = mistralai.STT(
    model="voxtral-mini-latest",
    context_bias=["LiveKit", "Voxtral", "Mistral"]
)

Realtime streaming transcription

Voxtral Realtime streams interim transcripts over a WebSocket connection. Since this model has no server-side endpointing, the plugin runs an internal Silero VAD to detect when the user stops speaking and flush the audio — producing final transcripts and driving the end-of-turn pipeline.

from livekit.plugins import mistralai
from livekit.plugins.silero import VAD

# Using Silero VAD with default settings (550ms silence threshold)
stt = mistralai.STT(model="voxtral-mini-transcribe-realtime-2602")

# Using custom VAD settings (e.g. shorter silence threshold for faster responses)
stt = mistralai.STT(
    model="voxtral-mini-transcribe-realtime-2602",
    vad=VAD.load(min_silence_duration=0.3),
)

Text-to-Speech (TTS)

from livekit.plugins import mistralai

# Using a built-in voice
tts = mistralai.TTS(voice="en_paul_neutral")

# Using zero-shot voice cloning
import base64
ref_audio_b64 = base64.b64encode(open("sample.mp3", "rb").read()).decode()
tts = mistralai.TTS(ref_audio=ref_audio_b64)

LLM

from livekit.plugins import mistralai

llm = mistralai.LLM()

# With custom temperature/max. tokens
llm = mistralai.LLM(
    model="mistral-large-latest",
    temperature=0.7,
    max_completion_tokens=150
)