Use case / Open Source AI Memory Database

Open Source AI Agent Memory

Memorose gives long-lived agents persistent memory that survives sessions, grows through real work, and stays queryable through hybrid retrieval, graph context, and time-aware controls.

Read the docs Star on GitHub

The problem this solves

Most agent memory starts as a prompt appendix or vector-only recall layer. That works for demos, but breaks down when agents need durable facts, procedural traces, shared knowledge, task state, and controlled forgetting.

How Memorose handles it

Memorose treats memory as a runtime surface. Events become L1 durable memories, L2 graph-linked insights, and L3 task-aware state that agents can retrieve and maintain through HTTP APIs.

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": "What should the coding agent remember about this repo?",
    "agent_id": "coding-assistant",
    "graph_depth": 1,
    "top_k": 8
  }'