Memory that knows why things connect. | Clio

Hey Product Hunt 👋 I built Clio because I kept running into the same wall: the more material I fed my AI assistant, the worse it got at reasoning through it. RAG finds text that looks like your question. That's not the same as understanding what's connected and why. Clio takes a different approach. Instead of storing chunks, it builds a typed graph -concepts linked by drives, part_of, enables, measures, and analogy relationships. When you ask a question, the retrieval follows causal chains through that structure. The assistant isn't guessing from similarity anymore — it's reasoning through the shape of the problem. There's a second benefit that surprised me: once the graph exists, a small local model can give precise answers by navigating the structure a smarter model built. The heavy thinking happens once at extraction time. At query time, even a lightweight model has exactly the context it needs — no more, no less. The graph also accumulates. Drop new files into your folder, run clio sync, and the agent merges them into the existing structure — extending concepts, adding facts, discovering new links across old and new material. Knowledge compounds instead of resetting with every session. In my own tests: 289 files, 813K tokens of source material became 5,919 tokens of structured memory. A cross-cutting architecture question dropped from 78K input tokens to ~1,470. It runs entirely on your machine by default. Works as a CLI, or install the skills directly into Claude Code, Cursor, or Windsurf with clio init --ai claude. Happy to answer anything — about the graph architecture, the extraction pipeline, or how it compares to what you're already using.