Highlights
Added visible reasoning support for Ollama chat and OpenAI Responses API reasoning summary flow
Fixed workflow progression for mixed code/no-code turns and improved startup prompt retention with retry
Hardened auto-fix behavior with retry handling, stricter follow-up instruction, and lifecycle logging
Fixed Python fenced block parsing for responses containing comment lines
Fixed hover accessibility for user-message copy/delete action buttons
Added sidebar disclaimer under prompt input: AI can make mistakes. Please verify responses.
Added
- Added visible reasoning/thinking support for Ollama chat in Conversational Notebook, including streamed thinking metadata handling.
- Upgraded OpenAI provider reasoning path to Responses API with reasoning summary support, so thinking can be surfaced in the same UI flow.
- Added sidebar chat safety disclaimer under the prompt input using centered helper text: "AI can make mistakes. Please verify responses."
Improved
- Improved workflow startup robustness: initial workflow prompt is now retained and retried until send acknowledgment succeeds, instead of being dropped during slow notebook/chat initialization.
- Hardened auto-fix behavior:
- no-code model replies are now treated as retryable attempts (up to max attempts),
- stricter follow-up instruction is applied after format failures,
- explicit auto-fix attempt lifecycle logging was added for diagnosis.
Fixed
- Fixed workflow prompt progression for mixed code/no-code turns so workflows can complete correctly when a final step returns text-only output.
- Fixed auto-fix Python block parsing to correctly accept valid fenced Python responses that include comment lines.
- Fixed user-message action button accessibility in Conversational Notebook so copy/delete controls remain reachable on hover without changing their visual position.