Search memories
Recall the most relevant learned context and source passages from a namespace. Hybrid search combines memories with document chunks by default, with optional query rewriting, reranking, and supporting context attachments.
Authorizations
Bearer authentication header of the form Bearer <token>, where <token> is your auth token.
Path Parameters
The isolated namespace to search. This can be an ID for your user, a project ID, or any other identifier you wish to use to scope memories.
100^[a-zA-Z0-9_:-]+$"user_alex"
Body
Natural-language question, topic, or phrase to retrieve relevant context for. Replaces the v4 q field.
1"what are the API rate limits"
Type-safe metadata conditions applied before ranking results. Replaces the v4 filters field and unifies both legacy filter formats into one expression type.
- Option 1
- Option 2
- Option 3
- Option 4
- Option 5
- Option 6
- Option 7
Search surface. "hybrid" combines learned memories with source chunks, "memories" returns learned context, and "chunks" returns source passages.
hybrid, memories, chunks "hybrid"
Maximum number of results to return
1 <= x <= 10010
Optional context to include alongside each matching result
Minimum relevance score from 0 to 1. Raise it for precision (fewer, accurate results) or lower it for broader recall (more results).
0 <= x <= 10.5
Post-retrieval ranking. "order" improves result ordering; "aggregate" also combines overlapping context into cleaner answers. This is helpful if you want to ensure the most relevant results are returned.
none, order, aggregate "order"
Expand and clarify the query before retrieval to improve recall for conversational or underspecified prompts. This increases the latency by about 400ms.
false