Read memory

Two reads. Both exist on REST and MCP.

recall

Search for facts. Send at least one of query, predicate, or related_to.

{ "query": "user:maya", "limit": 20 }

{
  "query": "programming languages Maya knows",
  "limit": 20
}

{
  "predicate": "has_skill",
  "related_to": "user:maya"
}

A type:id query looks up that entity.
Natural language uses semantic search.
predicate / predicates filter relationship type.
entity_type keeps results to one prefix (user, company).
related_to finds facts connected to an entity.
fuzzy_match helps with typos in names.
depth expands neighbors. Default 2. Start small.
limit defaults to 20 (max 200). offset pages.
sort_by orders a listing: created_at, predicate, subject, object, confidence, id. object is the entity name, or the literal value when there is no entity.
sort_order is asc or desc. Default asc. Pages are stable: the tie-break is id.
Omit both to keep the old order. Semantic search stays ranked by relevance. Set sort_by when you want a sorted page instead of a relevance rank.

Field reference: recall.

Walk the graph from one entity.

{ "entity": "user:maya", "depth": 2 }

Default depth is 2. limit defaults to 20 (max 500).

Use this when you already know the node. Use recall when you are searching.