Komi-learn, an open‑source tool from kurikomi‑labs, adds continuous memory and self‑improvement to AI coding agents. It watches a developer session, distills durable lessons—style, stack choices, successful fixes—and automatically recalls the most relevant ones when the next session starts. No slash commands or manual saves are required, and it works with Claude Code and Codex.

Installation runs via a single pip command; the script guides users through an interactive setup, then activates recall and background learning. Commands such as komi‑learn status, config, sync, and queue let users inspect learned material, adjust settings, or contribute to an optional community pool hosted on GitHub. Contributions are signed, content‑addressed and filtered to exclude secrets before reaching the LLM.

The tool stores lessons locally, merging overlapping entries and archiving stale ones, while deterministic checks prevent leaking sensitive data. Advanced installs can enable cryptographic signing and semantic recall via optional extras. Because it relies on the developer’s existing Claude Code login or an API key, Komi-learn adds memory without additional hosting costs, delivering a more fluid coding assistant today immediately.