Local-first memory for AI agents. Extract structured knowledge, recall it across tools, consolidate it over time.
# Extract knowledge from a conversation $ agenr extract session.jsonl --json | agenr store stored 12 entries (3 merged, 1 contradiction) # Later - in any tool, any session $ agenr recall "what did we decide about auth?" fact (0.94): We switched to OAuth2 with PKCE flow decision (0.88): API keys for server-to-server only
LLM extracts typed entries — facts, decisions, preferences, todos, events, lessons — from any conversation. Not just summaries. Real knowledge.
Entries strengthen when recalled, decay when stale, and get penalized when contradicted. FSRS-style forgetting curve keeps your knowledge fresh.
One local DB shared via MCP. OpenClaw and any MCP-compatible tool share the same brain. Your knowledge follows you everywhere.
Near-duplicates merge. Stale entries expire. The DB gets healthier over time, not just bigger.
Your memory stays on your machine. SQLite database. No cloud dependency for storage. Your data, your control.
Works with any tool that speaks MCP. Three tools: recall, store, extract. That's the entire API surface.
Three steps to give your AI agents memory.
AGENR is AGPL-3.0 licensed and built in the open. Your AI's memory should belong to you, not a cloud provider. Contribute, fork, or just remember everything.