Feature / Open Source AI Memory Database

Hybrid Retrieval for AI Memory

Memorose retrieval combines semantic similarity, full-text signals, graph expansion, and time-aware filters so agents can recall relevant memory without relying on vector search alone.

Read the docs Star on GitHub

The problem this solves

Vector-only memory is often too broad, too stale, or too hard to control. Agent recall needs semantic meaning, exact text matches, relationship context, and temporal boundaries.

How Memorose handles it

Memorose exposes retrieval controls through HTTP APIs, including graph depth, organization scope, agent identifiers, namespaces, and time filters.

runtime example

curl -X POST "$BASE_URL/v1/users/dylan/streams/$STREAM_ID/retrieve" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "deployment constraints",
    "graph_depth": 2,
    "agent_id": "ops-agent",
    "top_k": 10
  }'