From REA hype to your first verified case.
realearn is a structured learning path for using AI reverse-engineering agents — like REA — on binaries you are actually allowed to inspect. Real commands, real output, every guide lab verified on a real machine.
bash — you@localhost
$ npm install rea-agents@latest added 35 packages · rea-agents 6.4.0 · 0 vulnerabilities $ npx rea inspect-binary-layout ./hello provider: pwntools-elf @ pwntools 4.15.0;pyelftools 0.33;unicorn 2.1.2 [✓] evidence_id ev_6e479c83… · 31 sections · entry 0x400320
pick your path
Beginners Track
Three guided lessons: what reverse engineering is, choosing a legal first target, and reading your first disassembly. No prior RE experience needed — every exercise verified on our lab machine.
REA Setup Guide
Install and configure the REA CLI, wire it to your AI agent over MCP, light up a zero-cost analysis engine, and run your first evidence-backed analysis.
Tools & Comparison
How AI agents compare with classic tools like Ghidra, Hopper, and Binary Ninja — and which one fits your workflow.
why "legal first" is not a disclaimer — it is the method
Reverse engineering has a perception problem. Headlines about the REA agent boom and high-profile security incidents make it sound like AI-powered binary analysis is a gray-market skill. It is not. In most jurisdictions, reverse engineering for interoperability, security research, and education is lawful — and it has been a foundation of the software industry for decades. What determines legality is not the tool but the target: analyzing software you own, open-source binaries, official demo files, or systems you are explicitly authorized to test.
That principle shapes everything on this site. Every tutorial here declares its target software and the authorization basis for analyzing it. Every terminal block reflects commands actually run on our lab machine, with dated lab records behind them — see the verified: ran badges on each guide. If we have not run it, we do not claim it.
what an AI reverse-engineering agent actually does
REA (ReverseEngineeringAgent) is an orchestration layer, not a decompiler — we verified this on a real install. It wraps classic analysis engines (Ghidra, Hopper, IDA — or the free pwntools-based ELF provider) behind an agent loop, and refuses an operation with a precise diagnosis when no engine is available. You describe what you want to know in natural language, and the agent plans a sequence of steps: parse headers, locate entry points, follow cross-references, and summarize findings with a verifiable evidence trail — every result carries a content-addressed evidence ID, the target's sha256, and the exact provider versions that produced it.
The skill that separates productive users from frustrated ones is not prompt tricks. It is understanding what the agent's primitives mean: what a disassembly listing actually shows, why symbols matter, what a cross-reference tells you, and how to spot when the agent is confidently wrong. That is a learnable discipline, and it is what our beginners track teaches before you touch any agent.
who this site is for
Developers who installed an AI coding agent and wondered "can this thing read binaries too". Security students who want a structured first path instead of scattered forum posts. Engineers curious what actually happens inside the software they ship every day. If that is you, start with the first lesson — it takes about fifteen minutes.
One honest caveat: local analysis is not the same as private analysis. When an agent backed by a cloud model inspects a binary, parts of that analysis may flow to the model provider. If you work with employer or client software, check your agreements before running anything. We flag this in every relevant guide.
the learning path, in order
- What reverse engineering actually is — mental model, legality in plain language, a real system binary under inspection.
- Pick a legal first target — compile your own, then reverse a hidden constant out of it.
- Read your first disassembly — stripped binaries, anchor constants, and string evidence.
- Set up REA — install, doctor, MCP wiring, zero-cost engine, first evidence-backed run.
- Choose your toolset — where AI agents fit next to Ghidra, Hopper, and Binary Ninja.