Skip to main content
Searches memories using a combination of semantic vector search and BM25 full-text search. Returns results ranked by a combined score (semantic similarity, keyword match, recency, salience). Always call this before add_memory to check for duplicates.

​
Parameters

​
query
string
required
Search query. Can be natural language ("how does authentication work?") or keywords ("jwt token validation"). Semantic search finds conceptually similar content even without exact keyword matches.
​
limit
number
default:"10"
Maximum results to return (1-100). Default 10 is usually sufficient. Increase to 20-50 for comprehensive knowledge review.
​
layers
string[]
Filter by specific layers. Omit to search all layers. Example: ["rule"] for rules only, ["working", "long_term"] to exclude rules.
​
scope
string
default:"project"
Search scope.
ValueDescription
projectCurrent project + global memories (default)
globalGlobal memories only
allAll projects without filtering

​
Example

{
  "query": "authentication setup",
  "limit": 5,
  "layers": ["rule", "long_term"],
  "scope": "project"
}

​
Scoring

Each result includes a hybrid score and breakdown: 
score = semantic × 0.4 + bm25 × 0.3 + recency × 0.2 + salience × 0.1
See Search & Scoring for details.

​
Returns

Sorted list of results with score breakdown: 
1. [memory] (score: 0.847) This project uses React 19 with TypeScript
   semantic: 0.92, bm25: 0.85, recency: 0.78, salience: 0.65