What is REA? The open source tool that lets AI take apart any app, explained simply

A new open source project called REA has become one of the fastest-growing projects on GitHub. It went from about 13,000 stars on 8 October 2026 to more than 170,000 five days later, and weekly downloads jumped from around 1,300 to over 130,000. Its name stands for "Reverse Engineer Anything", and it lets AI assistants like Claude Code, Codex and Cursor look inside apps and programs to work out how they do what they do. Here's what that means, in plain words.

The short version

Most apps you use come without their recipe. You get the finished program, but not the source code the developers wrote. Working out how a program works from the finished product is called reverse engineering. It's an old and respected skill, used by security researchers, by people keeping old software and games alive, and by developers who need their software to work with someone else's.

Until now it took experts and specialist tools. REA hands those tools to your AI assistant. You ask a question like "how does the speed-up in Chrome's dinosaur game work?", and the assistant uses REA to look inside, finds the part that matters, explains it, and shows you the evidence it found. It can then write a version of that idea for your own project.

How it works

REA doesn't try to read an app all at once. It's a toolbox that your AI assistant opens through MCP, the common way assistants plug into outside tools.

• Your assistant asks the questions. You describe what you want to understand. The assistant decides which tools to use and in what order.
• REA does the looking. It opens the program, finds the relevant pieces and returns small, focused results instead of pouring the whole app into the conversation.
• Expert tools underneath. For programs that run directly on your computer, REA drives well-known reverse engineering tools: Ghidra (free, from the US National Security Agency), Hopper or IDA. For JavaScript and Electron apps, like many desktop apps today, it needs nothing extra.
• Evidence, not guesses. Each finding comes with where it was found and what's still unknown, so you can check the assistant's work instead of trusting it blindly.
• It runs on your computer. REA doesn't upload the app anywhere. Your AI assistant does see the results, so its provider's privacy rules apply to those.

What it can look at

• Desktop apps built with web technology, like Notion, Slack or VS Code: how screens talk to each other and to the computer.
• Websites: page structure, scripts and what the page sends and receives.
• Programs and games for Windows, Mac and Linux, down to the machine instructions.
• Android apps, and phones or emulators connected to your computer.
• Other formats, including .NET programs, Go programs, device firmware, saved network traffic and SQLite databases.

Real examples

The project shows three worked case studies:

• DX-Ball, a 1990s brick-breaker game: the assistant found how the game decides whether a sound plays from the left or right speaker, and rewrote it in C. The rewrite matched the original in all 3,205 test cases.
• Notion's desktop app: it traced what happens when you copy and paste, from the button through to the computer's clipboard.
• An old Japanese PC game: it recovered the maths behind the rings of bullets enemies fire, and rebuilt it in modern code.

What you need to try it

• An AI assistant that supports MCP, such as Claude Code, Codex, Cursor or Gemini CLI.
• Node.js version 22 or newer. Setup is one command, npx rea-agents setup, which asks before it changes your assistant's settings.
• For looking inside compiled programs, Ghidra (free) or a paid tool like Hopper or IDA. Websites and JavaScript apps work without them.
• A Mac or Linux computer works best. Some features run on Windows, but Windows support for Ghidra is still experimental.

REA itself is free under the MIT licence. You pay only for your AI assistant and any paid tools you choose to use with it.

Use it responsibly

Reverse engineering is legal in many places for things like security research, making software work together, preserving old software and studying programs you own. But it can also break laws or a program's terms of use, depending on where you live and what you do.

REA's own rules are clear. It's meant for security research, malware analysis, education, compatibility work, preservation and software you own or have permission to examine. The project won't help with cracking licence checks or copy protection, making game cheats, getting into systems you aren't allowed in, or copying someone else's code. If you like a feature in another app, use REA to understand the idea, then write your own version rather than copying theirs.

Good to know

• It's brand new and changing fast. It's only six months old, and new versions come out every few days. The developers ask you to update often, because bugs get fixed quickly.
• There's no independent test yet. The case studies come from the project and its contributors. Nobody has measured how often it gets things right across many apps.
• Big programs are slow. Analysing a large program can take several minutes and a lot of memory. 16 GB or more is a good idea.
• You still need judgement. The assistant can misread what it finds. REA shows its evidence so you can check, and it's worth doing.

Should you try it?

If you're curious how the software you use actually works, or you work in security or need to make your app work with someone else's, REA is one of the most interesting open source projects of the year. A gentle place to start is a website or a small JavaScript app you have the right to examine: no extra tools needed, and you can watch your assistant explain it step by step.