g1nation
1.4 KiB
1.4 KiB
Feature Plan: Second Brain Trace Mode
1. Feature Name
Second Brain Trace Mode
2. Reason
The user wants to verify whether AI answers actually use Second Brain notes, which documents were searched, and how retrieved knowledge influenced the answer.
3. Original User Request
Add a way to see if the AI searched Second Brain, which notes it referenced, whether each note was used in the final answer, and why unused notes were excluded.
4. Interpreted User Intent
The user does not want to rely on vague claims that the AI used memory. They want visible grounding evidence and a developer debug trace.
5. Scope
- Add a trace mode toggle in the sidebar.
- Add a debug JSON toggle.
- Retrieve relevant Markdown notes from the active Second Brain.
- Inject selected notes into the answer context.
- Append a user-visible 2nd Brain usage section to answers.
- Show searched-but-unused notes and basic grounding metrics.
6. Out Of Scope
- Vector database integration
- Full semantic embedding search
- Perfect factual attribution at sentence level
- External observability tools
7. Development Direction
Use a low-dependency local Markdown search first. Keep vector search and richer grounding analysis as later expansion.
8. Success Criteria
- Users can see whether Second Brain was used.
- Users can see referenced Markdown paths.
- Debug mode exposes retrieval JSON.
- General questions can explicitly show that Second Brain was not used.