recall

Recall memories by meaning using hybrid vector + full-text search. Results are ranked by semantic similarity, recency, and importance.

Parameters

Parameter	Type	Required	Description
query	string	Yes	What to search for
project	string	No	Filter by project (also includes “global”)
topic	string	No	Filter by topic
limit	number	No	Max results (default: 10)
depth	number	No	Graph traversal depth (0-3, default: 0)

Return Value

Returns an array of matching memories with [id:N] prefixes, content, topic, project, importance, and distance score.

Examples

Basic recall

{
  "query": "authentication architecture",
  "project": "my-project"
}

With graph traversal

{
  "query": "auth tokens",
  "project": "my-project",
  "depth": 1
}

Returns matching memories plus their graph neighbors — surfacing related context flat search would miss.

Topic-filtered recall

{
  "query": "deployment process",
  "project": "my-project",
  "topic": "ci-cd"
}

How Ranking Works

Results are ranked by a combination of:

• Semantic similarity — Cosine distance between query and memory embeddings
• Full-text match — SQLite FTS5 term matching
• Recency — More recent memories rank higher
• Importance — Higher importance (1-5) boosts ranking by up to 20%

Related Tools

• remember — Store new memories
• graph — View a memory’s neighborhood
• recall_episode — Recall an entire episode