# REA: helping AI look under the hood of software

- Source: MechNext — https://mechnext.eu/ru/article/46-rea-helping-ai-look-under-the-hood-of-software
- Published: 2026-10-21
- Updated: 2026-10-21
- Category: AI News
- Language: ru (Markdown in other languages: ?lang=en, ?lang=de, ?lang=sk, ?lang=pl, ?lang=zh)

> REA gives AI assistants tools to investigate how software works. Here is the idea in plain language, with examples, useful limits and a link to the project.

Ever wondered what happens inside an application when you click a button? REA helps a coding assistant investigate that question using evidence from the program. **[Visit REA on GitHub — morluto/rea.](https://github.com/morluto/rea)**

### Start with one question

Reverse engineering studies a finished program to understand its rules. Think of a mechanic following the parts of a machine. REA supplies inspection tools; the assistant connects findings and explains them. MCP provides the connection between the assistant and those tools. REA also works from a terminal. [Project FAQ.](https://rea.tools/faq/)

A useful first question is: “Where does this app write exported notes?” The project offers a small example that follows an export button through the application to its CSV writer. Browser analysis can also connect a click with a network request and the script behind it. [First investigation](https://rea.tools/first-investigation/); [application guide](https://rea.tools/guides/javascript/); [browser guide](https://rea.tools/guides/browser/).

### From calculator percentages to game audio

The project explains why a calculator treats 200 + 10% as 220, while 200 × 10% gives 20: the percentage operation uses different rules after addition and multiplication. The investigation links those rules to program instructions. [Calculator example.](https://rea.tools/)

In a DX-Ball case study, the team recovered the calculation that maps a brick’s horizontal position to stereo sound panning. The authors report 3,205 comparisons with the original function and a match of its 63 compiled bytes. This checks a particular function; it does not establish automatic reconstruction of an entire game. [DX-Ball investigation.](https://rea.tools/showcase/dx-ball/)

![](/uploads/rea-dxball-stereo-20261021.webp)_AI-generated illustration: a hit position and stereo sound placement. Not a DX-Ball or REA screenshot._

### Useful findings still need checking

Native analysis can return instructions and pseudocode: a tool’s attempt to express machine instructions as more readable code. That attempt may miss details. The DX-Ball example shows why further inspection and testing matter. [Native analysis guide.](https://rea.tools/guides/native/)

Required tools depend on the target and operating system. Deep native analysis can use Ghidra, Hopper or IDA. REA analyzes targets locally, but tool results reach the assistant, whose model provider’s data policy applies. [Setup documentation](https://github.com/morluto/rea/blob/main/docs/installation.md); [data FAQ](https://rea.tools/faq/#data-and-sharing).

REA is intended for authorized software investigation. Its MIT license does not grant rights to target software or licenses for commercial analysis tools. [Project license information.](https://github.com/morluto/rea#license)

Editorial note: based on the authors’ public documentation, checked on 21 October 2026. We did not install or independently test REA. Both images are AI-generated conceptual illustrations, not screenshots.