Recursive self-improvement (RSI) for AI agents
RSI (recursive self-improvement) is how Sno Station gets better with every job: each night the agents review the day's work, keep what worked and learn from what did not.
What is recursive self-improvement (RSI) for AI agents?
The agents' own skills and memory improve from their own work. Every night Sno Station reads the day's sessions, finds the mistakes that repeat, and proposes changes to the AI agents' own skill files. In the morning you accept the ones you like.
What does the nightly review keep, and what does it drop?
After each run the shared memory is distilled: fewer lines, each one earned. A fix counts when the next run proves it: yesterday's change is measured against today's, and the number decides whether it stays.
Does RSI change my AI agents without asking?
No. The nightly review proposes and a human accepts. Nothing changes without that accept, and the agents' own skill changes arrive as pull requests you can read.
What stays on my machine?
Everything the squad learns lives in one workspace on your laptop. The memory store is encrypted on your machine with a key that never leaves it, and Sno never receives your database or your key. Memory, messages and skills are plain files: read them, edit them, export them, delete them.
How do I start?
Install Sno: tell your AI agent "install sno.ai", or run curl -fsSL https://sno.ai/install | sh. The next morning, ask "What did we learn yesterday?"
Last updated