Uses the xAI Responses path. The recommended path is SuperGrok/X Premium OAuth; API keys still work via `XAI_API_KEY` or plugin config, and Grok `web_search` reuses the same auth profile before API-key fallback. `grok-4.3` is the bundled public API default chat model, while `grok-composer-2.5-fast` and `grok-build-0.1` are selectable for Grok OAuth / Grok Build coding workloads. `/fast` or `params.fastMode: true` rewrites `grok-3`, `grok-3-mini`, `grok-4`, and `grok-4-0709` to their `*-fast` variants. `tool_stream` defaults on; disable via `agents.defaults.models["xai/<model>"].params.tool_stream=false`.

- `realtime.provider`: selects the active browser/server realtime voice provider. Use `openai` for WebRTC, `google` for provider WebSocket, `xai` for Grok Voice through provider WebSocket or Gateway relay, or another bridge-only provider through Gateway relay.

- `realtime.providers.xai.vadThreshold`, `silenceDurationMs`, and `prefixPaddingMs`: forwarded to xAI server VAD for provider WebSocket and Gateway relay sessions.

- `realtime.providers.xai.baseUrl`: custom xAI-compatible Realtime base URL for `gateway-relay` only. Browser-owned xAI `provider-websocket` sessions require the native `https://api.x.ai/v1` endpoint.

sendUserMessage: () => {},

speakText: () => {},

truncating or clearing the active audio response. Implement `speakText`

when the provider can schedule assistant speech from OpenClaw-owned text;

`RealtimeVoiceSpeakTextOptions.mode: "exact"` means the provider should

preserve wording, while `"natural"` allows provider-side verbal polish.

Providers that omit `speakText` should implement `sendUserMessage` so

Gateway relay sessions can fall back to normal provider response

generation.

| Family | Model ids |

| -------------------- | ------------------------------------------------------------------------ |

| Composer 2.5 (OAuth) | `grok-composer-2.5-fast` |

| Grok Build 0.1 | `grok-build-0.1` |

| Grok 4.3 | `grok-4.3` |

| Grok 4.20 Beta | `grok-4.20-beta-latest-reasoning`, `grok-4.20-beta-latest-non-reasoning` |

Use `grok-4.3` for the public xAI API default. Use `grok-composer-2.5-fast`

or `grok-build-0.1` for Grok OAuth / Grok Build coding workloads unless you

explicitly need a Grok 4.20 beta alias.

browser WebSocket and Gateway relay sessions.

- Transports: `provider-websocket`, `gateway-relay`

- Browser auth: OpenClaw creates an ephemeral xAI Realtime client secret

with `POST /v1/realtime/client_secrets`; the browser connects with the

`xai-client-secret.<token>` WebSocket subprotocol.

- Browser transport origin: `provider-websocket` is limited to the native

`https://api.x.ai/v1` Realtime endpoint. Use `gateway-relay` for custom

`baseUrl` or proxy endpoints.

transport: "provider-websocket",

`XAI_API_KEY` and then the configured xAI auth profile. `baseUrl` is a

server-side option for `gateway-relay`; browser-owned `provider-websocket`

sessions require the native xAI Realtime endpoint.

Browser-owned xAI Realtime sessions use ephemeral client secrets. Gateway

relay remains available when you need the vendor WebSocket to stay

server-side, such as native clients, backend-owned meeting capture, or

custom/proxy `baseUrl` deployments.

- xAI Realtime voice supports Talk Gateway relay and browser-owned

provider-WebSocket sessions. Custom or proxy `baseUrl` values are

Gateway relay only; browser-owned provider-WebSocket sessions require

xAI's native `https://api.x.ai/v1` endpoint.

Talk mode uses a registered realtime voice provider. Configure OpenAI with `talk.realtime.provider: "openai"` plus an `openai` API-key auth profile, `talk.realtime.providers.openai.apiKey`, or `OPENAI_API_KEY`; OpenAI OAuth profiles do not configure Realtime voice. Configure Google with `talk.realtime.provider: "google"` plus `talk.realtime.providers.google.apiKey`. Configure xAI Grok Voice with `talk.realtime.provider: "xai"` plus either `talk.realtime.providers.xai.apiKey`, `XAI_API_KEY`, or an xAI auth profile. The browser never receives a standard provider API key. OpenAI receives an ephemeral Realtime client secret for WebRTC. Google Live receives a one-use constrained Live API auth token for a browser WebSocket session, with instructions and tool declarations locked into the token by the Gateway. xAI receives an ephemeral Realtime client secret from `/v1/realtime/client_secrets` and connects to the Grok Voice WebSocket with the `xai-client-secret.<token>` subprotocol. Browser-owned xAI sessions require the native `https://api.x.ai/v1` endpoint; use Gateway relay for custom xAI `baseUrl` or proxy endpoints. Providers that only expose a backend realtime bridge run through the Gateway relay transport, so credentials and vendor sockets stay server-side while browser audio moves through authenticated Gateway RPCs. The Realtime session prompt is assembled by the Gateway; `talk.client.create` does not accept caller-provided instruction overrides.

Provider WebSocket and Gateway relay providers can emit inbound transcript updates while audio is still streaming. For xAI, the bridge maps the cumulative `conversation.item.input_audio_transcription.updated` event to user transcript updates so Talk and meeting transcription surfaces can consume live inbound speech content.

expect(result.provider.models.map((model) => model.id)).toContain("grok-composer-2.5-fast");

export const XAI_GROK_OAUTH_BASE_URL = "https://cli-chat-proxy.grok.com/v1";

const XAI_COMPOSER_CONTEXT_WINDOW = 200_000;

const XAI_COMPOSER_MAX_TOKENS = 30_000;

api?: ModelDefinitionConfig["api"];

baseUrl?: string;

const XAI_SUBSCRIPTION_COST_UNKNOWN = {

input: 0,

output: 0,

cacheRead: 0,

cacheWrite: 0,

} satisfies XaiCost;

const XAI_OAUTH_MODEL_CATALOG = [

{

id: "grok-composer-2.5-fast",

name: "Composer 2.5",

api: "openai-responses",

baseUrl: XAI_GROK_OAUTH_BASE_URL,

reasoning: false,

input: ["text"],

contextWindow: XAI_COMPOSER_CONTEXT_WINDOW,

maxTokens: XAI_COMPOSER_MAX_TOKENS,

cost: XAI_SUBSCRIPTION_COST_UNKNOWN,

},

] as const satisfies readonly XaiCatalogEntry[];

...(entry.api ? { api: entry.api } : {}),

...(entry.baseUrl ? { baseUrl: entry.baseUrl } : {}),

export function buildXaiOAuthCatalogModels(): ModelDefinitionConfig[] {

return [

...XAI_OAUTH_MODEL_CATALOG.map((entry) => toModelDefinition(entry)),

...buildXaiCatalogModels(),

];

const exact = [...XAI_OAUTH_MODEL_CATALOG, ...XAI_MODEL_CATALOG].find(