1 Commits
Author SHA1 Message Date
youssefvdel 1d0a4e6be8 feat: add <think>/<thinking> tag extraction with no-leak buffering
- StreamingToolParser now extracts <think> and <thinking> blocks before
  tool call parsing, buffering content until closing tag (same pattern as
  <tool_call>) — nothing leaks to client text
- Partial think tag detection prevents fragments like '<thi' from flushing
- flush() emits remaining thinking as reasoning_content
- chat.ts non-streaming path strips residual think tags from content
- Move Python debug/fix scripts to scripts/ directory
- Add integration, largeBuffer, and leak test files
- Improve JSON parsing robustness and tool call guard validation
- Update README and .gitignore
2026-05-23 21:26:00 +03:00