Get started

Install the adapter with Core:

pnpm add @anvia/core @anvia/grok

Create a client for xAI’s API:

import { Agent } from '@anvia/core'
import { GrokClient, tools as grokTools } from '@anvia/grok'

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

const agent = new Agent({
  id: 'researcher',
  model: grok.completionModel({ modelId: 'grok-4.5', api: 'responses' }),
  maxTurns: 5,
  tools: [grokTools.webSearch({ allowedDomains: ['x.ai'] }), grokTools.codeInterpreter()],
})

const result = await agent.generate({
    prompt: 'Summarize recent xAI product updates.'
})

if (result.type === 'response') {
  console.log(result.output)
  console.log(result.sources)
}

The example selects Responses and grok-4.5 explicitly. Provider tools require the Responses adapter.

Add media

const image = grok.imageGenerationModel({ modelId: 'grok-imagine-image-quality' })
const speech = grok.speechGenerationModel()
const transcription = grok.transcriptionModel()

These are batch media APIs. Realtime voice and streaming speech are not included.

Before production

• Keep xAI credentials server-side.
• Allowlist provider tools and their domains, handles, stores, or MCP servers.
• Treat remote MCP headers and authorization as secrets.
• Bound turns and server-tool activity.
• Pass fetch explicitly in runtimes without a global implementation.
• Test image ratios and audio formats against the current xAI API.