# Grok Voice · Realtime API

> Realtime variant of the Grok voice agent for streaming, bidirectional audio and text conversations

Voice agent aimed at live applications such as assistants and phone agents, built around streaming audio and text in both directions. Inputs match the standard voice model: `audio_url` for the user's speech, optional `prompt` persona instructions, a `voice` choice and optional `tools` for web, X or MCP access. It returns spoken WAV audio, its duration and a transcript.

- Model id: `xai/grok-voice/realtime`
- Made by: xAI
- Category: Audio
- Output: audio
- Charged by usage (up to $1.494 per run, typical $0.006; the rest of the hold is refunded): $0.00083 / second of the voice session (cheapest run $0.006 (6 API credits))
- Run: `POST https://api.doitong.com/v1/run/xai/grok-voice/realtime`
- Quote: `POST https://api.doitong.com/v1/quote/xai/grok-voice/realtime`
- Docs page: https://doitong.com/developers/models/xai/grok-voice/realtime

## Pricing

Price: **$0.00083 / second of the voice session**. Charged by actual usage at $0.00083 / second of the voice session: up to $1.494 per run (the quote gives the maximum for your input), the rest is refunded automatically after the run. A typical run costs $0.006.

The full example input below holds up to $1.494 = 1494 API credits (up to 1800 s of voice session × $0.00083) when it starts; a typical run costs $0.006 (6 API credits).

Prices are the upstream price in USD, paid from the API balance in API credits (1000 API credits = 1 USD). This model is charged by usage: `POST https://api.doitong.com/v1/quote/xai/grok-voice/realtime` with the same JSON body returns the most the run can cost for that input (`credits`, debited when the run starts); after the run you pay the usage the model reports and the rest is refunded automatically. An input whose maximum would exceed the per-run limit is refused with 422. Invalid input is never charged and failed or canceled requests are refunded automatically.

## Input parameters

Send either the canonical name or the native field name (not both). Canonical names are the same across models.

| Parameter | Canonical name | Type | Required | Default | Allowed values / range | Description |
|---|---|---|---|---|---|---|
| `audio_url` | `audio` | string (audio URL) | yes |  |  | Audio URL for the user's speech, limited to 10 minutes and 50 MB. Most common audio formats can be supplied; before Grok receives the audio, it is converted to mono PCM with 16-bit samples at 24 kHz. The agent responds to the recording with speech. |
| `prompt` |  | string \| null | no |  | ≤ 20000 characters | Optional system instructions that supply conversation context and specify a persona in place of Grok's default; omitting them retains that default persona. |
| `voice` |  | enum (string) \| null | no |  | `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` | Select a built-in xAI voice to override Grok's default for the agent's reply; leaving this unset keeps that default voice. |
| `tools` |  | object {mcp_servers, web_search, x_search} | no | `{"mcp_servers":[],"web_search":{"enabled":false},"x_search":{"enabled":false}}` |  | Agent tool configuration for server-side use, excluding file_search/collections from the typed configuration because of ZDR. All tools start disabled; supported options include remote MCP tools, web tools, and X tools. |
| `tools.mcp_servers` |  | array of object {allowed_tools, authorization, enabled, headers, server_description, server_label, server_url} | no |  |  | Remote MCP servers to which the model provider connects for tool execution. |
| `tools.mcp_servers[].allowed_tools` |  | array of string \| null | no |  | ≥ 1 items | Optional list specifying which tool names this server exposes. |
| `tools.mcp_servers[].authorization` |  | string \| null | no |  |  | Sensitive authorization value sent exclusively to the model provider. |
| `tools.mcp_servers[].enabled` |  | boolean | no | `true` |  | Server inclusion flag; false excludes the server. |
| `tools.mcp_servers[].headers` |  | object \| null | no |  |  | Sensitive headers for MCP requests, sent exclusively to the model provider. |
| `tools.mcp_servers[].server_description` |  | string \| null | no |  | ≤ 2000 characters | Description outlining the capabilities this server makes available. |
| `tools.mcp_servers[].server_label` |  | string | yes |  | 1 – 128 characters | Server label used by the model provider to identify its tools and prefix their names. |
| `tools.mcp_servers[].server_url` |  | string | yes |  | ≥ 1 characters | HTTP(S) address supplied by the caller for an MCP server that Grok connects to remotely over Streaming HTTP/SSE transports. |
| `tools.web_search` |  | object {allowed_domains, enable_image_understanding, enabled, excluded_domains, location} | no | `{"allowed_domains":null,"enable_image_understanding":null,"enabled":false,"excluded_domains":null,"location":null}` |  | Provider-managed web search, disabled unless enabled is true. |
| `tools.web_search.allowed_domains` |  | array of string \| null | no |  | 1 – 5 items | Domain restriction for search results: specify up to five domains, omitting protocols and paths. |
| `tools.web_search.enable_image_understanding` |  | boolean \| null | no |  |  | Permission for the agent to examine images discovered through web searches. |
| `tools.web_search.enabled` |  | boolean | no | `false` |  | Web-search access for the agent: true adds web_search to session.tools and permits web searches; the default leaves it disabled. |
| `tools.web_search.excluded_domains` |  | array of string \| null | no |  | 1 – 5 items | Domains to exclude from search results, limited to five entries with protocols and paths omitted. |
| `tools.web_search.location` |  | object {city, country, region, timezone} \| null | no |  |  | Optional location for tailoring search results geographically. |
| `tools.x_search` |  | object {allowed_x_handles, enable_image_understanding, enable_video_understanding, enabled, excluded_x_handles, from_date, to_date} | no | `{"allowed_x_handles":null,"enable_image_understanding":null,"enable_video_understanding":null,"enabled":false,"excluded_x_handles":null,"from_date":null,"to_date":null}` |  | Provider-managed X search, disabled unless enabled is true. |
| `tools.x_search.allowed_x_handles` |  | array of string \| null | no |  | 1 – 20 items | X handles whose posts may appear in results, limited to 20 handles. |
| `tools.x_search.enable_image_understanding` |  | boolean \| null | no |  |  | Permission for the agent to examine images discovered on X. |
| `tools.x_search.enable_video_understanding` |  | boolean \| null | no |  |  | Permission for the agent to examine videos discovered on X. |
| `tools.x_search.enabled` |  | boolean | no | `false` |  | Access to X posts through search: true adds x_search to session.tools so the agent can search those posts; the default leaves it disabled. |
| `tools.x_search.excluded_x_handles` |  | array of string \| null | no |  | 1 – 20 items | X handles whose posts are excluded, limited to 20 handles. |
| `tools.x_search.from_date` |  | string \| null | no |  |  | Earliest date for posts, expressed in ISO format YYYY-MM-DD. |
| `tools.x_search.to_date` |  | string \| null | no |  |  | Latest date for posts, expressed in ISO format YYYY-MM-DD. |
| `enable_safety_checker` |  | boolean | no | `true` |  | Screening can be relaxed by setting false, provided the caller is authorized. |

File parameters (`audio`) take a public `https://` URL, a `data:` URI up to 10 MB, or the `file_url` of an upload from `POST /files`.

## Example input

Required parameters only:

```json
{
  "audio": "https://content.doitong.com/audio/showcase/v3/tools/dub_es.mp3"
}
```

A fuller example:

```json
{
  "audio_url": "https://content.doitong.com/audio/showcase/v3/tools/dub_es.mp3"
}
```

## Run it

Set `DOITONG_API_KEY` to a key from https://doitong.com/developers/dashboard#keys. `?wait=20` returns the finished request in the same call when it is ready within 20 seconds; otherwise poll `GET /requests/{id}` (it also accepts `?wait=`), or pass `webhook_url`.

### curl

```bash
# Run the model; wait up to 20 s for the finished result in the same call
curl -X POST "https://api.doitong.com/v1/run/xai/grok-voice/realtime?wait=20" \
  -H "Authorization: Bearer $DOITONG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "audio_url": "https://content.doitong.com/audio/showcase/v3/tools/dub_es.mp3"
  }'

# Still "queued" or "running"? Poll with the id from the response:
curl "https://api.doitong.com/v1/requests/REQUEST_ID?wait=20" \
  -H "Authorization: Bearer $DOITONG_API_KEY"
```

### JavaScript

```javascript
const API = "https://api.doitong.com/v1";
const headers = {
  "Authorization": "Bearer " + process.env.DOITONG_API_KEY,
  "Content-Type": "application/json",
};

// Run the model; wait up to 20 s for the finished result in the same call
let res = await fetch(API + "/run/xai/grok-voice/realtime?wait=20", {
  method: "POST",
  headers,
  body: JSON.stringify({
    "audio_url": "https://content.doitong.com/audio/showcase/v3/tools/dub_es.mp3"
  }),
});
let request = await res.json();
if (!res.ok) throw new Error(request.error.code + ": " + request.error.message);

// Not finished yet: long-poll until it is
while (request.status === "queued" || request.status === "running") {
  res = await fetch(API + "/requests/" + request.id + "?wait=20", { headers });
  request = await res.json();
}
if (request.status !== "succeeded") throw new Error(JSON.stringify(request.error));
console.log(request.outputs.map((o) => o.url), request.cost.credits + " API credits ($" + request.cost.usd + ")");
```

### Python

```python
import os
import requests

API = "https://api.doitong.com/v1"
headers = {"Authorization": "Bearer " + os.environ["DOITONG_API_KEY"]}

payload = {
    "audio_url": "https://content.doitong.com/audio/showcase/v3/tools/dub_es.mp3",
}

# Run the model; wait up to 20 s for the finished result in the same call
r = requests.post(API + "/run/xai/grok-voice/realtime", params={"wait": 20}, json=payload, headers=headers)
request = r.json()
if not r.ok:
    raise RuntimeError(request["error"]["code"] + ": " + request["error"]["message"])

# Not finished yet: long-poll until it is
while request["status"] in ("queued", "running"):
    request = requests.get(API + "/requests/" + request["id"], params={"wait": 20}, headers=headers).json()
if request["status"] != "succeeded":
    raise RuntimeError(request["error"])
print([o["url"] for o in request["outputs"]], request["cost"]["credits"], "API credits", "$" + request["cost"]["usd"])
```

### MCP

The MCP server at `https://api.doitong.com/v1/mcp` exposes the tools `search_models`, `get_model`, `quote`, `run`, `get_request`, `cancel_request`, `list_requests`, `create_upload` and `get_account`. A `run` call for this model:

```json
{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "run",
    "arguments": {
      "model_id": "xai/grok-voice/realtime",
      "input": {
        "audio_url": "https://content.doitong.com/audio/showcase/v3/tools/dub_es.mp3"
      },
      "wait_seconds": 20
    }
  }
}
```

## Response

Every call returns a request object. `status` is one of `queued`, `running`, `succeeded`, `failed`, `canceled`. `outputs` lists the generated files with URLs on our storage (kept forever, or 7 days with `?storage=temporary`; `storage` says which); `output` is the model's own result object with the same URLs.

```json
{
  "id": "req_8k2m4q7x1z9c3v5b6n0p2r4t",
  "object": "request",
  "model": "xai/grok-voice/realtime",
  "status": "succeeded",
  "created_at": "2026-10-20T12:00:00Z",
  "started_at": "2026-10-20T12:00:02Z",
  "completed_at": "2026-10-20T12:00:09Z",
  "queue_position": null,
  "cost": {
    "credits": 1494,
    "usd": "1.494",
    "charged": 1494,
    "refunded": 0
  },
  "input": {
    "audio_url": "https://content.doitong.com/audio/showcase/v3/tools/dub_es.mp3"
  },
  "output": {
    "audio": {
      "url": "https://content.doitong.com/dev-api/2026/10/20/req_8k2m4q7x1z9c3v5b6n0p2r4t/0-output.mp3",
      "content_type": "audio/mpeg",
      "file_name": "0-output.mp3",
      "file_size": 1843200
    },
    "text": "",
    "duration": 5
  },
  "outputs": [
    {
      "type": "audio",
      "url": "https://content.doitong.com/dev-api/2026/10/20/req_8k2m4q7x1z9c3v5b6n0p2r4t/0-output.mp3",
      "content_type": "audio/mpeg",
      "file_size": 1843200,
      "duration": 12,
      "stored": true
    }
  ],
  "error": null,
  "metrics": {
    "inference_seconds": 6.8
  },
  "urls": {
    "self": "https://api.doitong.com/v1/requests/req_8k2m4q7x1z9c3v5b6n0p2r4t",
    "cancel": "https://api.doitong.com/v1/requests/req_8k2m4q7x1z9c3v5b6n0p2r4t/cancel"
  }
}
```

### Output fields (`output`)

| Field | Type | Description |
|---|---|---|
| `audio` | object {url, content_type, file_name, file_size} | Spoken reply from the Grok agent, delivered as mono WAV at 24 kHz. |
| `text` | string | Text accompanying the agent's speech: an empty string indicates that Grok did not supply a transcript; otherwise, it contains the transcript. |
| `duration` | number | Length of the generated audio, measured in seconds. |

## Related models

- [Grok Voice](https://doitong.com/developers/models/xai/grok-voice.md) (`xai/grok-voice`): $0.00083 / second of the voice session
- [MMAudio V2](https://doitong.com/developers/models/mmaudio-v2.md) (`mmaudio-v2`): $0.001 / second
- [DeepFilterNet 3](https://doitong.com/developers/models/deepfilternet3.md) (`deepfilternet3`): $0.001 / second of input audio
- [Demucs](https://doitong.com/developers/models/demucs.md) (`demucs`): $0.0007 / second of input audio
- [ElevenLabs Audio Isolation](https://doitong.com/developers/models/elevenlabs/audio-isolation.md) (`elevenlabs/audio-isolation`): $0.10 / minute of input file
- [Elevenlabs Sound Effects V2](https://doitong.com/developers/models/elevenlabs/sound-effects/v2.md) (`elevenlabs/sound-effects/v2`): $0.002 / second
- [ElevenLabs Voice Changer](https://doitong.com/developers/models/elevenlabs/voice-changer.md) (`elevenlabs/voice-changer`): $0.30 / minute of input audio
- [FFmpeg API \[Merge Audios\]](https://doitong.com/developers/models/ffmpeg-api/merge-audios.md) (`ffmpeg-api/merge-audios`): $0.00017 / second of processing time

## Links

- HTML page: https://doitong.com/developers/models/xai/grok-voice/realtime
- All models and guides for AI agents: https://doitong.com/developers/llms.txt
- OpenAPI 3.1 with this model's typed request body: https://doitong.com/developers/openapi.json?model=xai/grok-voice/realtime
- Errors: https://doitong.com/developers/docs/errors
- Model details as JSON: https://api.doitong.com/v1/models/xai/grok-voice/realtime