feat: Show Thinking Text for Thinking Models #46
Conversation
|
@kwsantiago is attempting to deploy a commit to the evanyang's Team Team on Vercel. A member of the Team first needs to authorize it. |
sh1ftred
left a comment
sh1ftred
left a comment
There was a problem hiding this comment.
This works well for OpenAI and Anthropic API streaming, but doesn't work for OpenRouter as they output reasoning is a separate keyword {'reasoning': 'reasons'}.
Merging this as it's still useful for us as it has thinking UI and the parser.
I'm modifying the UI to work for OpenRouter.
I suppose this also doesn't work for displaying the stream for thinking as you don't update UI with accumulatedThinking. Will add that feature as well.
Thanks for the PR :). Really appreciate it.
|
Awesome, glad it's of value @sh1ftred . If you'd like me to work on a follow up PR or any other related help, would love to as I think Routstr is a great idea and want to contribute! |
|
Hey @kwsantiago, would love it if you can work on #48. It should be similar to how cashu.me works. If not, you can pick any issue from the TODO section in this view: Freedom AI Tech (view) |
|
@sh1ftred 48 sounds great, will do! Thanks |
2 participants
Closes #42
Summary
Adds display of intermediate thinking steps from models that support it (o1, Claude 3.5 Sonnet, etc.) to improve transparency in AI responses.
Changes
utils/thinkingParser.ts) - Detects thinking-capable models and extracts<thinking>contentcomponents/ui/ThinkingSection.tsx) - Collapsible display with copy functionalitycomponents/chat/ChatMessages.tsx) - Shows thinking sections for assistant messagesutils/apiUtils.ts) - Extracts thinking during response streamingFeatures
Testing
Tested with thinking-capable model selection and response streaming. Thinking content appears in collapsible sections, preserves in conversation history, and degrades gracefully for regular models.