Support models that expose a thinking/reasoning step (e.g. Claude extended thinking, o1-style models). Display the thinking content in the chat UI, collapsed by default.
Hints:
- In
send_message (llm.py), detect thinking blocks in the response and return them alongside the reply
- Store thinking content in the
Message model (new thinking column or as part of metadata)
- In the chat UI, render a collapsible "Thinking..." block before the assistant reply
- Pass
thinking parameters in the API request when the model supports it (check model name or a flag on ModelConfig)
- Consider a
show_thinking toggle in config.yaml per model or globally
Support models that expose a thinking/reasoning step (e.g. Claude extended thinking, o1-style models). Display the thinking content in the chat UI, collapsed by default.
Hints:
send_message(llm.py), detect thinking blocks in the response and return them alongside the replyMessagemodel (newthinkingcolumn or as part of metadata)thinkingparameters in the API request when the model supports it (check model name or a flag onModelConfig)show_thinkingtoggle inconfig.yamlper model or globally