xai/grok-voice/realtime

Build real-time voice applications powered by Grok. Stream audio and text bidirectionally via WebSocket for voice assistants, phone agents, and interactive voice systems.

Inference

Commercial use

Partner

When the request is deemed to be in violation of xAI terms the generation of the request will still be charged.

About

Bidirectional proxy for Grok's realtime speech-to-speech agent.

Relays xAI's realtime events verbatim in both directions between the fal client and wss://api.x.ai/v1/realtime?model=grok-voice-latest. Connect with a fal realtime client (fal_client.realtime_async in Python, fal.realtime.connect in JS) and exchange the event dicts documented by xAI — the client library owns the wire framing, so payloads are plain dicts, not JSON strings. The upstream connection is opened as soon as the session starts, so xAI's session.created reaches clients that connect and wait without sending. The XAI_KEY credential is injected server-side on the upstream handshake and never exposed to the caller.

Session parameters:

session_timeout=3600 — matches xAI's 1-hour maximum session length.
buffering=1 — voice agents are latency-sensitive; do not accumulate input frames.

1. Calling the API#

Install the client#

The client provides a convenient way to interact with the model API.

npm install --save @fal-ai/client

Migrate to @fal-ai/client

The @fal-ai/serverless-client package has been deprecated in favor of @fal-ai/client. Install the new package and update your imports — see client setup.

Setup your API Key#

Set FAL_KEY as an environment variable in your runtime.

export FAL_KEY="YOUR_API_KEY"

Real-time via WebSockets#

This model has a real-time mode via WebSockets, supported by fal.realtime.connect.

import { fal } from "@fal-ai/client";

const connection = fal.realtime.connect("xai/grok-voice/realtime", {
  onResult: (result) => {
    console.log(result);
  },
  onError: (error) => {
    console.error(error);
  },
  // Fetch short-lived JWT token from your backend
  tokenProvider: async (app) => {
    const response = await fetch("/api/fal/realtime-token", {
      method: "POST",
      headers: { "Content-Type": "application/json" },
      body: JSON.stringify({ app }),
    });
    return response.text();
  },
  tokenExpirationSeconds: 10,
});

connection.send({
  type: "session.update"
});

2. Authentication#

The API uses an API Key for authentication. It is recommended you set the FAL_KEY environment variable in your runtime when possible.

API Key#

In an environment where you cannot set environment variables, you can configure the API key manually on the client.

import { fal } from "@fal-ai/client";

fal.config({
  credentials: "YOUR_FAL_KEY"
});

Protect your API Key

When running code on the client-side (e.g. in a browser, mobile app or GUI applications), make sure to not expose your FAL_KEY. Instead, use a server-side proxy to make requests to the API. For more information, check out our server-side integration guide.

4. Files#

Some attributes in the API accept file URLs as input. Whenever that's the case you can pass your own URL or a Base64 data URI.

Data URI (base64)#

You can pass a Base64 data URI as a file input. The API will handle the file decoding for you. Keep in mind that for large files, this alternative although convenient can impact the request performance.

Hosted files (URL)#

You can also pass your own URLs as long as they are publicly accessible. Be aware that some hosts might block cross-site requests, rate-limit, or consider the request as a bot.

Uploading files#

We provide a convenient file storage that allows you to upload files and use them in your requests. You can upload files using the client API and use the returned URL in your requests.

import { fal } from "@fal-ai/client";

const file = new File(["Hello, World!"], "hello.txt", { type: "text/plain" });
const url = await fal.storage.upload(file);

Auto uploads

The client will auto-upload the file for you if you pass a binary object (e.g. File, Data).

Read more about file handling in our file upload guide.

5. Schema#

Input#

prompt string

System instructions describing the agent's persona and conversation context. Uses Grok's default persona when omitted. Default value: "null"

voice Enum

Built-in xAI voice for the agent's speech. Uses Grok's default voice when omitted.

Possible enum values: carina, zagan, helix, orion, luna, iris, altair, zenith, perseus, helios, lux, kepler, rigel, cosmo, celeste, ursa, sirius, lumen, castor, naksh, atlas, aurora, liora, ara, eve, leo, rex, sal

turn_detection GrokTurnDetectionConfiguration

Automatic turn detection: any configured object (even empty) lets the agent detect when you stop speaking and reply on its own. Set null for manual turns; omit to keep the session default.

type string* required

xAI realtime event type (e.g. 'session.update', 'input_audio_buffer.append', 'response.create', 'response.output_audio.delta'). Events are relayed verbatim in both directions.

tools GrokToolsConfiguration

Server-side tools available to the agent. Web, X, and remote MCP tools are supported; every tool is disabled by default. An explicitly supplied object always reaches the session, so previously enabled tools can be cleared.

{
  "type": "session.update"
}

Output#

prompt string

System instructions describing the agent's persona and conversation context. Uses Grok's default persona when omitted. Default value: "null"

voice Enum

Built-in xAI voice for the agent's speech. Uses Grok's default voice when omitted.

turn_detection GrokTurnDetectionConfiguration

Automatic turn detection: any configured object (even empty) lets the agent detect when you stop speaking and reply on its own. Set null for manual turns; omit to keep the session default.

type string* required

xAI realtime event type (e.g. 'session.update', 'input_audio_buffer.append', 'response.create', 'response.output_audio.delta'). Events are relayed verbatim in both directions.

tools GrokToolsConfiguration

{
  "type": "session.update"
}

Other types#

GrokSearchLocation#

timezone string

IANA timezone name, for example America/New_York. Default value: "null"

city string

City used to localize web-search results. Default value: "null"

country string

Country used to localize web-search results. Default value: "null"

region string

Region used to localize web-search results. Default value: "null"

GrokRemoteMCPTool#

authorization string

Sensitive authorization value forwarded only to xAI. Default value: "null"

server_label string* required

Label xAI uses to identify and prefix this server's tools.

server_description string

Description of the capabilities exposed by this server. Default value: "null"

server_url string* required

Caller-selected HTTP(S) MCP endpoint. xAI makes the remote connection and supports Streaming HTTP/SSE transports. Use at your own risk; fal does not connect to or SSRF-fetch this URL.

allowed_tools list<string>

Optional allow-list of tool names exposed from this server.

headers object

Sensitive MCP request headers forwarded only to xAI.

enabled boolean

False omits this server. Default value: true

File#

url string* required

The URL where the file can be downloaded from.

content_type string

The mime type of the file.

file_name string

The name of the file. It will be auto-generated if not provided.

file_size integer

The size of the file in bytes.

GrokTurnDetectionConfiguration#

silence_duration_ms integer

Silence duration that ends a detected turn, in milliseconds.

idle_timeout_ms integer

Idle time after an assistant response before xAI starts a proactive turn, in milliseconds.

threshold float

VAD activation threshold; higher values require louder audio.

prefix_padding_ms integer

Audio retained before detected speech, in milliseconds.

GrokXSearchTool#

excluded_x_handles list<string>

Exclude posts from at most 20 X handles.

enable_image_understanding boolean

Allow the agent to inspect images found on X.

enable_video_understanding boolean

Allow the agent to inspect videos found on X.

allowed_x_handles list<string>

Restrict results to posts from at most 20 X handles.

to_date string

Latest post date, as ISO YYYY-MM-DD. Default value: "null"

from_date string

Earliest post date, as ISO YYYY-MM-DD. Default value: "null"

enabled boolean

Whether to include x_search in session.tools. Disabled by default; set true to let the agent search X posts.

GrokToolsConfiguration#

x_search GrokXSearchTool

xAI-managed X search; off unless enabled is set true.

mcp_servers list<GrokRemoteMCPTool>

Remote MCP servers that xAI connects to and executes.

web_search GrokWebSearchTool

xAI-managed web search; off unless enabled is set true.

GrokWebSearchTool#

excluded_domains list<string>

Exclude results from at most five domains, without protocol or path.

allowed_domains list<string>

Restrict results to at most five domains, without protocol or path.

location GrokSearchLocation

Optional location used to localize search results.

enable_image_understanding boolean

Allow the agent to inspect images found during web search.

enabled boolean

Whether to include web_search in session.tools. Disabled by default; set true to let the agent search the web.

xai/grok-voice/realtime

When the request is deemed to be in violation of xAI terms the generation of the request will still be charged.

Table of contents

1. Calling the API

2. Authentication

4. Files

5. Schema

About

1. Calling the API#

Install the client#

Migrate to @fal-ai/client

Setup your API Key#

Real-time via WebSockets#

2. Authentication#

API Key#

Protect your API Key

4. Files#

Data URI (base64)#

Hosted files (URL)#

Uploading files#

Auto uploads

5. Schema#

Input#

Output#

Other types#

GrokSearchLocation#

GrokRemoteMCPTool#

File#

GrokTurnDetectionConfiguration#

GrokXSearchTool#

GrokToolsConfiguration#

GrokWebSearchTool#

Related Models