fix bugs

[xprilion]

This pull request adds support for streaming and displaying "thinking" or reasoning content from advanced LLMs (such as Claude 3.7+ and OpenAI's reasoning models) before the main assistant reply. It introduces a new ThinkingChunk type, updates backend streaming and event handling, and enhances the frontend UI to show and manage these thinking segments, including duration and collapsing/expanding the block.

Backend: Extended "thinking" support for LLMs

• Added a ThinkingChunk type and logic to detect and stream reasoning/thinking content from compatible models (Claude 3.7+, Claude 4+, OpenAI reasoning models). (backend/openmlr/agent/types.py [1] backend/openmlr/agent/llm.py [2] [3] [4] [5] [6] [7] [8] [9] [10]
• Updated streaming and retry functions to yield ThinkingChunk objects as appropriate. (backend/openmlr/agent/llm.py [1] [2] [3] [4]
• Modified the agent loop to emit thinking_chunk and thinking_end events, including timing the duration of the thinking phase and handling transitions to assistant output or tool calls. (backend/openmlr/agent/loop.py [1] [2] [3] [4]

Frontend: Thinking content UI and event handling

• Enhanced the chat UI to display streamed thinking content as a special block, collapse it when the assistant reply or tool call starts, and show the duration of the thinking phase. Users can expand/collapse the block to review the reasoning. (frontend/src/App.tsx [1] [2] [3] frontend/src/components/MessageList.tsx [4] [5] [6] [7]
• Updated the stop button logic to only show during active processing, not while waiting for input/approval. (frontend/src/components/InputArea.tsx frontend/src/components/InputArea.tsxL253-R254)

These changes improve transparency and user experience by surfacing the model's intermediate reasoning steps in real time.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud