AI API reverse engineering: private APIs and protocols decoded with AI
API and protocol reverse engineering is about working out how two pieces of software talk when there is no public documentation. The projects on this page used a language model to decode undocumented interfaces: the messages a lamp or home battery sends over Bluetooth, the binary frames a web app sends over a WebSocket, and private system frameworks on macOS and Windows. A common approach is to give the model a decompiled companion app, or a capture of real traffic, and ask it to describe the message format and write a small client. The model is quick at spotting structure in hex dumps and following code paths in decompiled Java or JavaScript. The person tests each command against the real device or service and corrects the model when it guesses. Most authors did this for interoperability, such as adding a device to Home Assistant. Each entry links the source and lists the model, tools and outcome.
- Projects
- 7
- Outcomes
- Works6Partial1
- Models used
- Claude7
The projects
Share the writeup or repo, and what the AI got right and wrong.
Related guides
Common questions
The model reads a decompiled client app or a traffic capture, describes the message format, and writes a client that can be tested against the real service. The person runs those tests and feeds back what failed.
A decompiled companion app gives the model the exact code that builds each message. Traffic captures help confirm it. Projects here used JADX for Android apps and browser or Bluetooth logging tools for traffic.
Laws differ by country, and many allow it for interoperability. Terms of service may also apply. This is not legal advice; check what applies to you before you publish a client.