/search

Search the knowledge graph with structured output and similarity scores. Use this when you need entry IDs, JSON output, or score-based evaluation.

For general retrieval, use /known:recall instead — it uses hybrid search by default and returns plain text optimized for LLM context.

When to Use

• Need entry IDs to update, delete, link, or inspect entries
• Need JSON output for programmatic processing (known --json search)
• Need similarity scores to evaluate result quality
• Pure vector search without the hybrid text+graph expansion that recall adds

Instructions

1. Run the search command:

known search '<query>' [flags]

2. Present the results with their IDs and similarity scores.

3. Suggest follow-up actions based on the results:

   • /known:forget to delete an unwanted entry
   • known show <id> to see full details and relationships
   • known update <id> --content '...' to correct an entry

Flag Reference

Flag	Default	Purpose
--scope	auto (from cwd)	Scope to search within
--limit	5	Maximum results
--threshold	0.4	Minimum similarity (raise for precision)
--recency	0	Recency weight (0=pure similarity, 1=pure recency)
--text	false	Use FTS5 full-text search instead of vector search
--hybrid	false	Enable vector + graph expansion (recall does this by default)
--expand-depth	1	Graph expansion hops (only with --hybrid)
--label	all	Filter by label (repeatable)

Global flags go before the subcommand:

known --json search '<query>'     # JSON output
known --quiet search '<query>'    # suppress non-essential output

Examples

known search 'deployment process' --limit 5
known search 'API rate limits' --hybrid --expand-depth 2
known --json search 'database schema' --scope backend
known search 'auth' --threshold 0.5 --recency 0.3
known search --text 'rate-limit'