Duplicate Detection
Before writing new knowledge to a vault, the system checks for semantically similar existing entries. This prevents the knowledge base from filling with duplicate or near-duplicate content. When duplicates are detected, the system offers to update the existing entry or merge the new information.
How it works
When create_knowledge or /knowz save is called, the system vectorizes the new content and searches for similar entries in the target vault. If an entry with similarity above a configurable threshold is found, the system presents the match: 'Similar entry found: [title]. Update existing, save as new, or skip?' For automatic captures (from the Closer agent), high-confidence duplicates are silently merged while ambiguous matches are queued for human review.
Duplicate detection showing similar entry found and merge options
Screenshot coming soon
Why it matters
Knowledge bases without deduplication become noisy and unreliable. When you search for 'caching strategy', you do not want five slightly different versions of the same decision. Duplicate detection keeps the knowledge base clean and authoritative. It also handles knowledge evolution gracefully -- when a decision is updated, the existing entry is enriched rather than duplicated.
Clean vault with deduplicated entries vs noisy vault with duplicates
Screenshot coming soon