What it is
Production-grade memory engine for LLM agents with chat, working, and coding memory, including hierarchical recall and code diff-chain tracking. Give LLM agents persistent memory across chat, work, and coding: store personal facts, reusable task experience, and exact code revision chains, then retrieve them through one Python SDK.
Key results
- 3 scenes: chat, working, coding, each with a dedicated memory pipeline
- Lite mode does chat writes/search with zero LLM calls
- Coding memory attaches exact per-revision diffs with zero LLM calls
Shows that one memory runtime can unify personal recall, reusable work experience, and exact coding history for long-horizon agents.