mistral_client

A Dart client library for the Mistral AI API, covering chat completions and audio transcription.

Features

• Chat completions — chat() with system/user/assistant messages, multimodal content parts, JSON mode (response_format), temperature, max tokens, top-p, and seed.
• Audio transcription — transcribeAudio() with multipart file upload, file URLs, or file IDs; language, temperature, and diarization options; timestamp granularity for per-segment and per-word timings; progress callbacks; confidence scores.
• Structured errors — MistralApiException (API errors with status code and message), MistralNetworkException (connection failures), and MistralException (validation errors).
• Testable — accepts an injected http.Client and overridable baseUrl.

Installation

Add the dependency to your pubspec.yaml:

dependencies:
  mistral_client:
    git:
      url: https://github.com/HolotypePteLtd/mistral_voice_client.git

The package is not yet published to pub.dev; once it is, you'll be able to use mistral_client: ^0.1.0.

Usage

Chat completions

import 'package:mistral_client/mistral_client.dart';

final client = MistralClient(apiKey: 'YOUR_API_KEY');

final response = await client.chat(ChatRequest.simple(
  model: 'mistral-small-latest',
  systemPrompt: 'You are a helpful assistant.',
  userMessage: 'What is the capital of France?',
));

print(response.choices.first.message.content);
client.dispose();

For structured output, enable JSON mode:

final response = await client.chat(ChatRequest.simple(
  model: 'mistral-small-latest',
  systemPrompt: 'Return JSON.',
  userMessage: 'List three colors as a JSON array.',
  jsonMode: true,
));

Audio transcription

final client = MistralClient(apiKey: 'YOUR_API_KEY');

final file = File('recording.mp3');
final response = await client.transcribeAudio(
  TranscriptionRequest(
    model: 'voxtral-mini-latest',
    fileBytes: await file.readAsBytes(),
    fileName: file.path,
  ),
  onProgress: (progress) => print('${(progress * 100).toStringAsFixed(0)}%'),
);

print(response.text);

// Timed segments:
for (final segment in response.segments) {
  print('[${segment.start}s - ${segment.end}s] ${segment.text}');
}

// Per-word timings (request with timestampGranularities: ['word']):
for (final word in response.words) {
  print('${word.text} ${word.start}-${word.end}s');
}

client.dispose();

Command-line transcription

A small CLI is included for quick testing:

dart run bin/transcribe.dart <api-key> <audio-file-path>
dart run bin/transcribe.dart <api-key> --url <audio-url>

Development

dart pub get
dart test

License

MIT