Hey Product Hunt 👋 I built komi-learn out of a small daily frustration: I'd spend the first few minutes of every coding session re-teaching my agent things it already "knew" yesterday — my stack, how I like things done, a fix we'd landed together. It has no memory, so I'm the memory. So I built the loop that Hermes Agent (from Nous Research) demonstrated, but generalized so it plugs into the agents people actually use day to day — Claude Code and Codex — with no commands to learn. It watches a session, distills the durable lessons in the background, and quietly loads the relevant ones next time. Two things I cared about most: Zero friction. If remembering requires you to remember to save things, it won't happen. So it's all automatic, on session hooks. Don't learn the wrong things. A deterministic filter strips secrets, machine-specific paths, and one-off failures before anything is kept — so it doesn't memorize your API key or "this flaky test failed once." It's local by default; there's an optional, opt-in community pool (signed Markdown on GitHub, no server) if you want your agent to benefit from lessons other people's agents figured out. It's early and I'd love your honest feedback — especially if you try it and something feels off. What would make this genuinely useful in your workflow? Repo + install: github.com/kurikomi-labs/komi-learn

komi-learn

Continuous memory and self-improvement for coding agents

Continuous memory and self-improvement for coding agents