Someone Turned Meta Glasses Into a Chess Cheating Device: Elo 3000+, Opus 5.5 Reads the Board, Fully Open Source

A developer reverse-engineered his Meta Ray-Bans 3 and, pairing them with Opus 5.5 + Stockfish, built AI chess-cheating glasses. Elo 3000+, and the entire project is open source.

The project is called chesshacker. Put on the glasses and connect to a Mac via a WhatsApp video call. The glasses look at the board, the program analyzes the position, and the best move is spoken into your ear through a virtual microphone.

The video shows the complete demo: while the glasses wearer plays, the system gives suggested moves in real time.

Workflow

1. Setup: Completed with a single command, npx chesshacker setup. It automatically downloads Stockfish and verifies the checksum.
2. Connection: The glasses initiate a WhatsApp video call to the Mac. Whatever the glasses see, the Mac sees.
3. Frame capture: The Mac captures frames from the call window. While waiting for the opponent, it captures 1 frame per second; on your turn, 1 frame every 3 seconds.
4. Cheap filtering: Jev first asks three questions: Can the board be read? Is a hand blocking it? Did a piece move? If nothing changed, skip it and avoid calling the expensive model.
5. Board reading: Frames that pass the filter are sent to a vision model. The default is gpt-6.1-sol, and Opus 5.5 also works. The model only reports which squares changed; it does not make decisions.
6. Matching legal moves: chess.js matches the “changes” to legal moves, and only confirms after two consecutive frames agree.
7. Stockfish calculates the move: It thinks for half a second and returns the best reply, along with an evaluation score (+1.5 or "mate in 3").
8. Voice playback: macOS speaks the move into the call through the BlackHole 2ch virtual microphone: "Rook a to d1"—the sound goes straight into the glasses.

Key design

The vision model only handles seeing, Stockfish only handles thinking, and the code turns the two into legal moves. The language model does not participate in decision-making.

The details were considered too:

• Losing track of the board: the glasses show the prompt "Lost track of the board"; you can undo or enter moves manually.
• Stockfish crashes: it automatically restarts once and shows "Something went wrong. Still trying".
• Every game is saved as PGN, and API keys are stored in the macOS Keychain.
• An overlay can draw arrows on the livestream feed and connects directly to OBS.
• The default spending cap is $10.

Cost and speed

From the opponent making a move to you hearing the reply is about 10 seconds.

An observation

The project documentation is quite honest: end-to-end testing passed on video recorded by the glasses, but it has not yet been run in a real WhatsApp call.

There is no new technology in this system. Stockfish is an open-source engine, the vision model is off the shelf, and anyone can use a WhatsApp call. The innovation is in piecing these together so the entire process happens at your ear.

A developer in the comments said he wrote his own chess engine that searched twice as many nodes as Stockfish and still lost 0:92 to Stockfish 18. The one doing the real work is Stockfish, not the glasses.

3000+ Elo is normal for Stockfish. The glasses just make “cheating with Stockfish” hard to notice.

Tournament organizers are probably not ready for this yet. But “cheating by wearing glasses” is no longer a movie plot.

Project URL: reverse.dev/pjchess

To learn how to reverse-engineer any software with Opus 5.5, check out Movez's reverse-engineering course.