xAI TTS

Overview

xAI TTS is available in LiveKit Agents through LiveKit Inference and the xAI plugin. With LiveKit Inference, your agent runs on LiveKit's infrastructure to minimize latency. No separate provider API key is required, and usage and rate limits are managed through LiveKit Cloud. Use the plugin instead if you want to manage your own billing and rate limits. Pricing for LiveKit Inference is available on the pricing page .

LiveKit Inference

Use LiveKit Inference to access xAI TTS without a separate xAI API key.

Usage

To use xAI, use the TTS class from the inference module:

from livekit.agents import AgentSession, inference

session = AgentSession(
    tts=inference.TTS(
        model="xai/tts-1", 
        voice="ara", 
        language="en"
    ),
    # ... stt, llm, vad, turn_handling, etc.
)

import { AgentSession, inference } from '@livekit/agents';
const session = new AgentSession({
    tts: new inference.TTS({
        model: "xai/tts-1",
        voice: "ara",
    }),
    // ... stt, llm, vad, turnHandling, etc.
});

Parameters

Voices

The following is a sample of the xAI voices available in LiveKit Inference. For a complete list of voices, see the xAI TTS documentation .

String descriptors

As a shortcut, you can also pass a descriptor with the model ID and voice directly to the tts argument in your AgentSession:

from livekit.agents import AgentSession

session = AgentSession(
    tts="xai/tts-1:ara",
    # ... llm, stt, vad, turn_handling, etc.
)

import { AgentSession } from '@livekit/agents';

session = new AgentSession({
    tts: "xai/tts-1:ara",
    // ... llm, stt, vad, turnHandling, etc.
});

Plugin

Use the xAI plugin to connect directly to xAI's TTS API with your own API key. For Node.js, use LiveKit Inference instead.

Installation

Install the plugin from PyPI:

uv add "livekit-agents[xai]~=1.5"

Authentication

The xAI plugin requires an xAI API key .

Set XAI_API_KEY in your .env file.

Usage

Use xAI TTS in an AgentSession or as a standalone speech generator. For example, you can use this TTS in the Voice AI quickstart.

from livekit.plugins import xai

session = AgentSession(
  tts = xai.TTS(
    voice="ara",
  ),
  # ... llm, stt, etc.
)

Parameters

This section describes some of the available parameters. See the plugin reference links in the Additional resources section for a complete list of all available parameters.

Speech tags

xAI TTS supports speech tags that add expressive delivery to synthesized speech. There are two categories of tags:

• Inline tags insert a vocal expression at a specific point in the text. For example, a [pause] tag can create a short pause, while a [laugh] tag can add laughter.
• Wrapping tags modify how a section of text is delivered. For example, <whisper>a secret</whisper> delivers the text in a whispered voice, while <slow>text</slow> is spoken at a slower pace.

Tags can be combined for layered effects:

So I walked in and [pause] there it was. [laugh]

I need to tell you something. <whisper>It is a secret.</whisper>

<slow><soft>Goodnight, sleep well.</soft></slow>

For a complete list of tags and examples, see the xAI speech tags documentation .

Expressive mode

xAI TTS works with expressive mode. Set expressive=True on your AgentSession and the LLM adds the speech tags above automatically, shaping delivery through prosody, non-verbal sounds, and pauses. To adjust the delivery, see Customization.

xAI's expressive delivery comes from prosody, non-verbal sounds, and pauses rather than a mood label, so it doesn't publish an lk.expression mood. Its speech is expressive, but it can't drive the mood UI. For prompting techniques you can apply by hand, see Voice realism.

Additional resources

The following resources provide more information about using xAI with LiveKit Agents.

Expressive mode

Let the LLM shape xAI delivery with prosody, sounds, and pauses automatically.

Python plugin

ReferenceGitHubPyPI

xAI docs

xAI TTS docs.

Voice AI quickstart

Get started with LiveKit Agents and xAI TTS.

Prompting voice agents to sound more realistic

Tips for using emotion tagging and prompt design to make voice agents sound natural.