gr.Markdown visible toggling is unreliable in Gradio streaming generators.
Switched to gr.HTML with inline styles — empty string hides the element,
HTML string shows the styled box. No visibility state needed.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Both __meta (LangGraph/search progress) and __thinking (LLM reasoning)
tokens now stream into the thinking box instead of the chatbot.
Chatbot shows only the final answer. Thinking box shows the full
analysis pipeline: [LangGraph → ...], 문서 검색 중, thinking content.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Show thinking progress in a separate animated box above chatbot
(🤔 사고 중... while streaming, 💭 사고 완료 when answer starts)
- Fix ValueError: add missing 5th yield value (thinking_box) to all
respond() yield statements
- Fix [Reset] and other sync handlers: replace asyncio.get_event_loop()
.run_until_complete() with asyncio.run() for AnyIO thread compatibility
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Introduced "thinking box" UI to display intermediate thought processes.
- Added CSS styling for the thinking box with scrollable and formatted design.
- Updated response handling to show thinking progress and completion dynamically.
- Enhanced Gradio outputs to include the new thinking box component.
Filter __meta dict tokens from TTS accumulator so progress messages
([LangGraph], thinking blocks, source references) are displayed in chat
but not read aloud. Answer tokens continue to accumulate in tts_text.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
shinalok