TikTok creator open-sources Meta Ray-Ban glasses chess assistance workflow
TikTok creator sourgrape02 has open-sourced the chess assistance system built on their Meta Ray-Ban smart glasses, with the related code and instructions published at reverse.dev/pjchess. The tool is aimed at macOS. Users send the glasses camera feed in real time to a Mac through a WhatsApp video call, a vision model identifies the chessboard, Stockfish calculates the best move, and then a virtual microphone sends the move back into the call as voice, which the wearer can hear directly as prompts.
The system workflow includes: capturing the call window roughly once per second (about 1 frame/second while waiting for the opponent, about 1 frame/3 seconds while it is the user’s turn to move); using Jev to perform low-cost filtering on the frames to determine whether the board is readable, whether hands are blocking the view, or whether pieces have moved, and skipping subsequent calls if there is no change; the vision model reading the board state only when necessary and comparing it with the existing position, accepting only legal moves; Stockfish thinking for about half a second before returning the best move and an evaluation (such as +1.5 or “mate in 3”); and macOS injecting the move as speech into the call through virtual audio devices such as BlackHole. In addition, the system supports prompting “Lost track of the board” when the position is lost and allowing manual correction, automatically restarting once if Stockfish behaves abnormally, saving games as PGN, and overlaying suggested-move arrows on the image for use by OBS and others. The reverse.dev page describes the project as “Video-call your Mac on WhatsApp from Meta glasses, look at a chess game, and know the best move”.
This open-source move is related to the park chess video that sourgrape02 previously released. Their earlier video, which was reportedly viewed about 13 million times, showed the process of using the glasses camera, sending it to a Mac through a video call, having AI identify the pieces and Stockfish calculate the move, and then sending it back by voice. GothamChess (Levy Rozman) once described a similar demonstration in a related short video as “chess LARP” and pointed out that as early as January 2025, Eddy Xu and others had already publicly shared similar assistance ideas based on Ray-Ban Meta glasses. This open-source release does not disclose independent third-party test data on the system’s accuracy, latency, or full-game performance in real games, nor does it clearly identify the specific provider and version of the vision model.
Original sources:
Title: Chess hacks for Meta Glasses V3 Publisher: reverse.dev (picklerpeej project) Publication date: unknown (the page shows updated 1 day ago, accessed around 2026-10-27) URL: https://www.reverse.dev/pjchess Source type: project page / open-source code description Verification status: the page’s main content was read in full for this report (about 96 visible lines), verification date 2026-10-27 Supported content: the tool is a macOS command-line tool; it gets frames from Meta glasses through a WhatsApp video call; Jev filters the frames; a vision model reads the board; Stockfish calculates moves; voice is injected into the call through BlackHole and similar tools; overall workflow description
Title: Ragebaiting a French dude in chess #metaglasses #chess Publisher: Ridiculous Marketplaces (YouTube) Publication date: 2026-10-24 URL: https://www.youtube.com/shorts/HAh8GOCHkTA Source type: video transcript and description Verification status: the title, description (including the open-source link), and the full auto-generated transcript were read for this report, verification date 2026-10-27 Supported content: the creator said they decided to open-source the project after their previous video received about 13 million views; the reverse.dev/pjchess link is explicitly provided; the demonstration uses Meta glasses
Title: GothamChess blasts TikToker’s viral Meta glasses chess cheating stunt: “Not impressed” Publisher: Dexerto (Michael Gwilliam) Publication date: 2026-10-23 URL: https://www.dexerto.com/youtube/gothamchess-blasts-tiktokers-viral-meta-glasses-chess-cheating-stunt-not-impressed-3421387/ Source type: media report Verification status: the relevant reported passages were read for this report, verification date 2026-10-27 Supported content: GothamChess’s comments on a similar glasses-assisted demonstration; background on Eddy Xu’s 2025 case; this does not prove that the current open-source project has been independently verified or has reached a specific Elo
ABAB AI Insight
sourgrape02’s open-source release turns the glasses-assisted workflow that previously appeared mainly in individual demo videos into a reproducible macOS tool. Compared with the Ray-Ban Meta glasses assistance idea that Eddy Xu publicly shared during TeenHacks LI in January 2025, this version places more emphasis on a complete closed loop of capturing frames from a WhatsApp call window, using Jev filtering to reduce model calls, and sending audio back through BlackHole virtual audio, rather than relying on Instagram Live mirroring. At the time, Eddy Xu mainly demonstrated combining image recognition with Stockfish and returning the result through the glasses speaker or text; this version instead injects speech directly into the call, making the prompts more concealed, while also adding legality checks for positions and exception recovery mechanisms.
On the capital and business path, the project itself does not involve financing or commercialization disclosure. It relies on existing open-source components: Stockfish (a free strong engine), chess.js (rules validation), Jev (frame filtering), and macOS system capabilities (screen capture, speech, Keychain). Vision model calls incur API costs, but the page and the post do not disclose the specific model name, unit price, or actual call volume. reverse.dev provides a clone-and-edit entry point, and users must configure API keys and the environment themselves. Unlike a pure software chess engine, the key resource here is the glasses camera as a continuous visual input, and the call channel as a low-latency two-way channel, shifting the computational burden to the local Mac.
Compared with GothamChess’s “chess LARP” comment and Eddy Xu’s earlier demonstration, the biggest similarity in this release is that all of them use the combination of Ray-Ban Meta series glasses camera + external computation + Stockfish; the most important difference is the degree of openness and engineering completeness—earlier examples were mostly one-off demos or limited sharing, while this release provides a complete toolchain that can be set up with one command (npx chesshacker setup) and includes filtering, validation, recovery, and PGN recording. GothamChess’s view that “you do not need such a complicated setup to cheat” is aimed at the credibility and necessity of the demo video itself, rather than denying that the technical path exists.
This open-source release lowers the technical barrier to copying a similar assistance system, allowing users with Meta Ray-Ban glasses and a Mac to avoid writing frame capture, filtering, and audio return logic from scratch. What it changes is the cost and time for individual users to obtain the combination of “glasses camera + chess engine,” not the upper limit of Stockfish’s playing strength, and it does not provide evidence of usability in official competitions or in scenarios with anti-cheating measures. The point to watch is whether users later publish complete game recordings and system logs, or whether platforms such as chess.com issue new detection or ban notices targeting this type of glasses-plus-external-computation combination.
ABAB News · Cognitive Law
Open-sourcing turns a one-off demo into a reusable toolchain, changing the replication cost from “write it yourself” to “configure the environment.”
The glasses camera provides continuous vision, while computation and validation still happen on the Mac side; the concealment depends on audio return rather than the glasses themselves.
Stockfish’s high Elo does not automatically mean that the full system is usable in real games; position recognition and latency remain separate components.