*I wanted DeepSeek streaming in Flutter to feel like Flutter.
*
Not:
final response = await deepseek.createChatCompletion(
ChatCompletionRequest(
model: 'deepseek-chat',
messages: [/* your messages */],
stream: true,
),
);
// Handle stream using Dio's response stream...
But this:
await for (final token in client.chatStream(
messages: [ChatMessage.user('Explain Flutter')],
)) {
setState(() => response += token);
}
Each token reaches your widget the moment DeepSeek generates it.
No waiting for the full response. No manual SSE parsing. Just a Dart Stream that works naturally with Flutter.
So I built flutter_deepseek — the only Flutter DeepSeek package with true Stream, reasoner support, and function calling, all in one.
Features:
→ Real-time SSE streaming with Stream
→ deepseek-reasoner support
→ Reasoning stream and final answer available separately
→ Function calling / tool use
→ Typed exceptions: AuthException · RateLimitException · BadRequestException · ServerException · NetworkException
→ Zero Dio dependency (built on dart:http)
→ Fully injectable http.Client for testing
→ 160 pub points on day one
I built and shipped DeepSeek-powered features in production before writing this package.
That experience shaped the API far more than the documentation did.
Open source. MIT licensed. Actively maintained.
