Docs · Search

Search records within a project

Full-text search uses SQLite FTS5 over record titles and bodies and ranks matches by relevance. Search is performed locally within the selected project.

Use concise search terms

Enter words related to the decision, file, or task. The current search parser tokenizes words and treats them as alternatives, using up to 50 terms. It returns up to 30 results; an empty query lists recent project records.

Review source and record type

Results include the record title, kind, source, and body excerpt. Open a result to see its full contents and revision history. Search is not a substitute for checking the original source.

Optional semantic search

Semantic search uses locally stored vectors from an Ollama embedding model that you install and configure. It accepts only a loopback endpoint. If no model is configured or the local model is unavailable, use FTS5 full-text search.

Explore AgentMemory