kurikomi-labs

kurikomi-labs / komi-learn

Public

Continuous memory + self-improvement for AI agents. Learns how you work, recalls it automatically, no commands. Claude Code & Codex.

ai-agents claude-code codex continuous-learning llm
35
0
89% credibility
Found May 31, 2026 at 35 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

komi-learn is a plugin that makes AI coding assistants (like Claude Code and Codex) learn from your work sessions automatically. It quietly watches as you work, extracts useful lessons from your corrections and techniques, and automatically loads the relevant ones into future sessions — no manual saving or commands required. The system includes privacy protections to filter out secrets and personal details before any data is processed. There's an optional community feature where general-purpose learnings can be shared with others, but this requires your explicit approval before anything leaves your machine. Everything is designed to run silently in the background so you can focus on actual coding while your assistant gradually improves.

How It Works

1
💬 You hear about komi-learn

A developer friend tells you about a tool that helps your AI coding assistant remember your preferences and techniques automatically.

2
🧩 You install it in one command

You run a simple install and the setup wizard asks you a few friendly questions about whether you want to share knowledge with others and enable smarter memory recall.

3
✨ Everything happens silently in the background

You use your coding assistant as normal. The tool watches your sessions without interrupting you and quietly takes notes when you teach the assistant something.

4
🚀 Your assistant gets smarter on its own

The next time you start a session, relevant lessons from your past work appear automatically in your assistant's context — no commands needed.

5
You can optionally help others
✅
Share general lessons

You review and approve specific learnings to contribute to the shared knowledge pool

🔒
Keep everything private

You skip the community pool and keep all your learnings just for yourself

🎉 Your coding assistant improves over time

Without any extra effort, your assistant gradually becomes more attuned to your style, your projects, and the tricks that worked for you before.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 35 to 35 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is komi-learn?

komi-learn is a Python tool that gives your AI coding assistant persistent memory. It watches what you do across sessions, automatically extracts useful patterns and fixes, and feeds them back at the start of your next session. No commands to remember, no manual saving. It works with Claude Code and OpenAI Codex, learning your coding style, your stack preferences, and the solutions that actually worked for you. The core loop is simple: recall relevant lessons when a session starts, distill new lessons when it ends, and periodically clean up what you've learned.

Why is it gaining traction?

The hook is "zero friction" -- you install it once and it just works. The agent learns continuously without you ever typing a slash command or running a save command. It also has an interesting optional community pool: a GitHub-backed knowledge base where developers can share general lessons their agents learned, with cryptographic signing to verify who contributed what. The design is defensive by default -- secrets, machine-specific paths, and one-off failures are filtered out before they ever reach the model. You approve every contribution before it leaves your machine.

Who should use this?

Developers who use Claude Code or Codex daily and find themselves repeating the same corrections. If you constantly remind your agent about your preferred testing framework, coding style, or project conventions, this learns that for you. Teams could also benefit from the community pool if they want general agent tips without maintaining their own server. It's less useful for occasional users or those who switch between radically different projects.

Verdict

This is a clever idea executed with real care for safety and privacy. The 0.8999999761581421% credibility score reflects its early stage -- only 35 stars and limited battle-testing across real sessions. The code is well-structured and the CLI is polished, but it's beta software. Worth trying if you're a heavy Claude Code or Codex user, but don't bet production workflows on it yet.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.