mnemon

Why we included this project

LLM agents lose everything between sessions, and that is the problem Mnemon is built around. It is a single local binary that keeps a graph-based knowledge store, with your host LLM acting as the supervisor: the binary handles storage, indexing, and search, while the model decides what to remember, how to link it, and when to let it decay. That split keeps recall fast without adding a second inference bill, and it works with Claude Code, DeepSeek Harness, OpenClaw, or any other agent runtime. If context compaction keeps dropping decisions your agent made last week, this gives it durable, intent-aware memory across sessions without standing up a service or managing another API key.