deepseek_box #

A DeepSeek AI provider based on ai_box.

Usage #

import 'package:ai_box/ai_box.dart';
import 'package:deepseek_box/deepseek_box.dart';

Future<void> main() async {
  final ai = DeepSeek(apiKey: 'sk-...');

  // One-shot text generation.
  final answer = await ai.generateText(
    model: 'deepseek-chat',
    message: 'Say hello in one short sentence.',
  );
  print(answer);

  // Multi-turn chat.
  final res = await ai.chat(
    model: 'deepseek-chat',
    messages: [
      LLMContent.system('You are concise.'),
      LLMContent.user('What is the capital of Japan?'),
    ],
    maxTokens: 32,
  );
  print(res.text);
  print(res.usage); // token usage
}

Reasoning model #

With deepseek-reasoner, the model's chain of thought is exposed through the common ai_box reasoning API:

final res = await ai.chat(
  model: 'deepseek-reasoner',
  messages: [LLMContent.user('What is 17 * 24?')],
);
print(res.content.reasoning); // thinking
print(res.text); // final answer

Streaming #

True SSE streaming — text arrives incrementally and the final chunk carries the finish reason and token usage:

await for (final text in ai.generateTextStream(
  model: 'deepseek-chat',
  message: 'Write a haiku about Dart.',
)) {
  stdout.write(text);
}

Models and key validation #

final models = await ai.getModels();
final ok = await ai.validateKey();

Tool calling, structured output and the sealed LLMException error hierarchy are all provided by ai_box — see its README for details.