Grok provider

@anvia/grok connects Anvia agents to xAI’s first-party REST API. It defaults to the Responses API shape and also exposes a Chat Completions adapter for compatible workflows.

Install

pnpm add @anvia/core @anvia/grok

Set your xAI API key in server-side configuration:

XAI_API_KEY=...

Client Setup

import { AgentBuilder } from "@anvia/core/agent";
import { GrokClient } from "@anvia/grok";

const grok = new GrokClient({
  apiKey: process.env.XAI_API_KEY,
});

const model = grok.completionModel("grok-4.3");

export const agent = new AgentBuilder("grok-agent", model)
  .instructions("Answer clearly and concisely.")
  .build();

completionModel() uses the Responses adapter by default. Use the Chat Completions adapter when a workflow needs that API shape:

const grok = new GrokClient({
  apiKey: process.env.XAI_API_KEY,
  completionApi: "chat",
});

const model = grok.completionModel("grok-4.3");

Image Generation

The provider exposes xAI image generation through Anvia’s ImageGenerationModel contract:

const imageModel = grok.imageGenerationModel("grok-imagine-image");

const result = await imageModel.imageGeneration({
  prompt: "A clean product render of a small desk lamp",
  width: 1024,
  height: 1024,
});

The adapter maps the requested dimensions to xAI’s aspect_ratio parameter and requests base64 image output when available. Provider-specific image options can be passed through additionalParams.

Model Listing

GrokClient implements ModelListingClient:

const models = await grok.listModels();

Use model listing for inventory and startup checks. Still smoke test the exact model id and capability your workflow depends on.

Supported Surfaces

Surface	Status
Text completion	supported
Streaming completion	supported
Tools and tool choice	supported through the OpenAI-compatible completion adapters
Structured output schema	supported through the OpenAI-compatible completion adapters
Image input	supported when the selected xAI model supports it
Document input	supported by the Responses adapter when the selected xAI model supports it
Image generation	supported
Model listing	supported
Embeddings, audio, transcription, OCR, video, files, batches	not exposed by this package

Read Capability matrix for a provider-by-provider comparison.