Get started

Install the adapter with Core:

pnpm add @anvia/core @anvia/mistral

Create the provider client on the server:

import { Agent } from '@anvia/core'
import { MistralClient } from '@anvia/mistral'

const mistral = new MistralClient({
  apiKey: process.env.MISTRAL_API_KEY!,
})

const agent = new Agent({
  id: 'assistant',
  model: mistral.completionModel({
      modelId: 'mistral-large-latest'
  }),
})

const result = await agent.generate({
    prompt: 'Extract the launch risks.'
})

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

Create embeddings

const embeddings = mistral.embeddingModel({
    modelId: 'mistral-embed',
    maxBatchSize: 32
})

const vectors = await embeddings.embedTexts([
  'Password reset links expire after thirty minutes.',
])

Process a document with OCR

const ocr = mistral.ocrModel({ modelId: 'mistral-ocr-latest' })
const result = await ocr.ocr({
  source: {
    type: 'document_url',
    url: 'https://example.com/report.pdf',
    documentName: 'report.pdf',
  },
  tableFormat: 'markdown',
})

console.log(result.markdown)

See OCR before accepting untrusted URLs or uploading bytes.

Before production

• Keep provider keys and document access server-side.
• Bound model turns and tool execution.
• Validate OCR source authorization and size before provider upload.
• Keep model IDs explicit and allowlisted.
• Test malformed tool arguments and stream termination.
• Rebuild vector indexes after changing embedding dimensions.