FindSimilarConcepts

Discovers semantically similar concepts using 384-dimensional vector embeddings and cosine similarity.

Parameters

• conceptName (string, required): Concept to find similar concepts for
• maxResults (int, optional): Maximum results to return (default: 10)

Returns

Example

Similarity Scores

• 0.90-1.00: Likely synonyms - consider consolidation
• 0.75-0.90: Strong relationship - good for WikiLinks
• 0.60-0.75: Related concepts - explore connections
• < 0.60: Weak relationship

Notes:

• conceptName must be non-empty; otherwise the tool returns an error.
• Scores can appear saturated (e.g., many results at 1.000) depending on the current embedding distribution and vector distance values. Treat this tool primarily as a ranking signal, and validate via BuildContext / SearchMemories.
• Very large maxResults values can produce extremely long output.

Semantic vs Graph

FindSimilarConcepts (meaning-based):

• Finds concepts that mean similar things
• Discovers implicit relationships
• Use for: synonym detection, exploration

BuildContext (connection-based):

• Finds concepts explicitly linked in files
• Respects intentional connections
• Use for: following established structure

Decay Weighting

Similarity scores are weighted by recency using time-based decay:

• Each concept's decay weight is derived from its source files (the files containing that [[WikiLink]])
• The maximum decay weight across all source files is used (concept freshness = most recent occurrence)
• Final score: similarity * decayWeight

This means concepts appearing only in old, unaccessed files will rank lower than equally similar concepts in recent files.

Grace periods and half-life follow the same configuration as SearchMemories (see searchmemories.md for details).

Integration

• Sync: Run first to populate vec_concepts embeddings (otherwise you may get few/no results)
• AnalyzeConceptCorruption: Identify duplicates from high-similarity results
• RepairConcepts: Consolidate similar concepts that are true duplicates
• BuildContext: Explore graph relationships of similar concepts (decay-weighted)
• SearchMemories: Find files containing discovered concepts (decay-weighted)