See a feature you love in someone else's app? Find out how it works — with proof.

You open a popular app and its search finds your notes instantly. You want that in your own project. But the app is a sealed box.

So you guess. You poke at it for a weekend, then ship a copy of what you think it does.

You could use special tools to look inside. But each kind of program needs its own tool, and none of them share notes. Ask a chatbot instead, and you get a confident paragraph that points at nothing: no place, no file, no line.

That is a drawing with no measurements.

The problem

The painA guess you can't check is worse than no answer. You build on it, and find the mistake later.

REA gives your AI helper a set of tools for looking inside programs. It works on apps, web code, websites, phone apps and more.

You ask a plain question, like “how does search work in this app?” The helper picks the right tool and looks. Then it writes back what it found, where it found it, and what it could not reach.

The helper can then write your own version and test it. Everything runs on your computer.

The big idea

Picture a drawing tool with two arms. One arm traces the real thing. The other arm copies each line onto your page. The copy arm cannot move unless the tracing arm touched something.

REA works the same way. A finding is only kept if the tool really touched it, and the finding carries the place. A claim with no touch stays a red question mark.

The aha

Oh — so the gaps are part of the answer. The page shows what is known and what is not, and nothing in between is made up.

Under the hood, REA is a program with two front doors and a swappable set of tool tips. The code is built in rings. Outer rings may use inner rings, never the other way round.

The front doors are an MCP server (the standard plug that lets chatbots call tools) and a terminal command. Both run the same workflows.

The tool tips are adapters for Hopper, Ghidra and IDA (programs that read compiled code), a browser, JavaScript, .NET, Android, firmware and more. Pick one and the rest are untouched.

At the center sits the evidence rules: what was seen, what was worked out, and what is still unknown.

Architecture

Data flow

This is a real screenshot from the project: REA starting its helper inside Hopper while it inspects a program. Your agent drives this; you read the answers.

In the real world

Understand how search works in the Notes app,
show me the evidence, and build a similar
feature for my project.

You need Node.js 22, 24 or 26+ and npm. For compiled programs you also need Hopper, Ghidra or IDA. JavaScript apps need nothing more.

npx rea-agents setup

1. Run the command above. It lists your agents, such as Claude Code, Codex or Cursor, and shows exactly what it will change.
2. Approve the changes. Setup adds REA to your agent and saves a backup of every file it touches.
3. Restart your agent. Then ask a question about a program, like the example above.
4. Or try the terminal first. Run npx -y rea-agents@latest analyze-javascript-application /path/to/app --json. You will see a list of modules, imports and where each came from.
5. What's next. Read the guides at rea.tools, and use rea update to stay current.

This explainer ships with a knowledge pack: the whole repo, indexed so any AI helper can search it and answer questions about REA. Humans get this page; your AI gets the pack.