Read the book at thersibook.com.
The RSI Book is an open, practical guide to recursive self-improvement in AI agents: how an agent learns from failures, changes its own system, and proves that the change actually helped.
- Step-by-step chapters that build the core improvement loop.
- Clear reading notes on important RSI and self-evolving-agent research.
- Companion Python examples for experimenting with the ideas.
The goal is to make recursive self-improvement concrete, testable, and easier to reason about without hiding the hard questions around evaluation and safety.
The website is plain static HTML with no build step. Open index.html locally to read it, or start with the published version at thersibook.com.
To add a research note, follow ADDING-PAPERS.md.