What's Changed

• Fix real-time streaming with proper HTTP flushing by @TianYi0217 in #8
• Add unit tests for streaming functionality

Changes

Bug Fixes

Real-time Streaming with Proper HTTP Flushing (#8)

Fixed streaming responses that were buffered and only appeared at the end instead of showing tokens in real-time.

Technical Details:

• Replaced io.Copy() with manual read/write loop with immediate Flush() calls
• Implemented http.Flusher interface on responseWrapper
• Use 512-byte buffer optimized for Anthropic SSE events (typically 100-200 bytes)
• Proper error handling for write vs read errors
• Graceful degradation when underlying ResponseWriter doesn't support flushing

Impact:

• Streaming responses now deliver content in real-time instead of buffering until completion
• Significantly improved user experience for streaming requests
• Lower perceived latency for SSE-based responses

Testing

• Added unit tests for responseWrapper.Flush() implementation
• Added test coverage for graceful degradation scenarios

Contributors

Thanks to @TianYi0217 for this excellent contribution!

Full Changelog: v1.0.2...v1.0.4