Search & Discovery
Recall
Run semantic search across an agent’s stored memories using natural language.
POST
Overview
Run semantic search across stored memories. This retrieves contextually relevant memories for the active agent based on semantic similarity to the query. Filters are sent as JSON in the request body only (no query-string parameters).Authentication
API clients do not send an API key orAuthorization header.
string
required
Session token from Activate Agent. Must match
agent_id.string
required
Must be
application/jsonPath Parameters
string
required
The unique identifier of the agent.
Body
string
required
Natural-language search text matched against the agent’s memories (max 1000 characters).
integer
Maximum number of results to return. Range
1–100. If omitted, the server default applies (RECALL_LIMIT).number
Minimum similarity score in the range
0.0–1.0 to filter out less relevant memories.array
Optional list of memory type filters (e.g.
["fact", "preference"]).ITS Scoring System
Memories are scored using Information Theoretic Similarity (ITS), providing nuanced relevance measurements. Use themin_similarity parameter to filter out results below a certain threshold:
Next Steps
- Generate AI Answer to have an LLM synthesize the returned memories into a response
- Recall Recent to fetch the most recently stored memories
- Recall As Of for point-in-time queries
- Recall Changed Since for differential retrieval