Anvia

Appearance

Completions ​

Use completionModel(...) for agents, direct model calls, extractors, and model-driven pipeline stages.

ts
import { AgentBuilder } from '@anvia/core'
import { MistralClient } from '@anvia/mistral'

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

const model = mistral.completionModel('mistral-large-latest')

export const supportAgent = new AgentBuilder('support', model)
  .instructions('Answer support questions clearly and concisely.')
  .build()

The returned model implements Anvia's streaming completion contract, so it can power both .send() and .stream() agent runs.

Send one direct request ​

Call the model directly when the application owns the flow around a single generation:

ts
import { createCompletion } from '@anvia/core'

const result = await createCompletion(model, {
  instructions: 'Write a concise internal incident summary.',
  input: 'Checkout requests timed out for 12 minutes.',
  maxTokens: 180,
})

console.log(result.text)
console.log(result.usage.totalTokens)

Use an agent when the run needs tools, memory, dynamic context, hooks, or multiple turns.

Stream a run ​

An agent stream executes only while it is consumed:

ts
for await (const event of supportAgent
  .prompt('Draft a response to this support ticket.')
  .stream()) {
  if (event.type === 'text_delta') {
    process.stdout.write(event.delta)
  }
}

Consume through the final event so usage, tool results, observers, and run completion can settle. For browser transport, expose the run through @anvia/server rather than sending provider credentials to the client.

Supported request features ​

The adapter maps these completion features to Mistral chat completions:

  • text instructions and message history
  • temperature and maximum output tokens
  • streaming text and tool-call deltas
  • tools and tool choice
  • JSON output schemas
  • provider-specific request parameters

It rejects chat image inputs and document-file inputs before making the provider request. Mistral OCR remains available as a separate model; see OCR.

For model-driven application functions and JSON results, continue to Tools and schemas.

Provider-specific parameters ​

Pass Mistral-specific values through the completion helper's params option only at the narrow call site that needs them:

ts
const result = await createCompletion(model, {
  input: 'Give this release note a short title.',
  params: {
    randomSeed: 42,
  },
})

The adapter keeps its selected model and normalized messages; provider parameters cannot replace those request-identity fields. Verify provider option names against the Mistral API version used by the deployment.

Built for Anvia.