By name: memo-search description: Search the current project's memo database using hybrid semantic + keyword search. Use it before processing any question from user
Memo Search
Search the current project's memo database using hybrid semantic + keyword search (BM25 + vector similarity with RRF scoring).
Workflow
Step 1: Determine What to Search For
Extract the key topic or question from the user's request. If the request is broad (e.g., "what do we know about the API?"), you may need to run multiple searches with different terms.
Step 2: Search
Run a hybrid search:
memo search "<query>" --limit 10
The search combines:
- Vector similarity — finds conceptually related memories (synonyms, related concepts)
- BM25 keyword search — finds exact word matches via full-text search
- RRF scoring — combines both without requiring exact keywords
Adjust --limit based on how broad the search is. Use a higher limit for broad topics, lower for specific lookups.
Listing All Memories
If the user asks to "show everything" or "list all memories", use:
memo list --all
This returns all memories without semantic ranking. Use it for browsing rather than targeted lookup.
Important Notes
- Hybrid search finds both semantic and keyword matches. A query for "database" will find memories about "PostgreSQL" (semantic) AND memories containing the word "database" (keyword). Exact keyword matches typically score higher (closer to 1.0).
- Scores are 0-1 scale. 1.0 = exact match in both vector and keyword search. Lower scores still indicate relevance through semantic similarity.
- No results doesn't mean no information. The user may simply not have stored that information yet — suggest adding it.