Eschess
see how engine thinks
A chess engine built across Python, C++, and Rust, from raw bitboards and alpha-beta search to neural networks and an LLM coach. Every part is made visible.
Play or watch the engines. C++/Rust compile to WebAssembly and search client-side at near-native speed; Python/ML stream over a socket. Learning mode overlays the searched branches and live evaluations; Competition mode lets you tune depth, time, and style.
Open moduleAn interactive engineering blog in six chapters — one per chess piece, Pawn to King — walking from bitboards and alpha-beta up to ML and orchestration, with live widgets embedded between the paragraphs.
Open moduleA telemetry room streaming a persistent self-play loop: moving averages of policy loss, value loss, evolving win/loss ratios, and NPS — graphed in real time over a WebSocket while a background server does the training.
Open moduleA chat beside a board. Every move is scored by the engine — top-3 principal variations and centipawn shifts — then packaged into a strict prompt so an LLM can explain, in plain English, why a move was good or bad.
Open module