A modern UCI chess engine written from scratch in Rust.
Currently rated at 2682 Elo (CCRL).
Roxie is a hobby chess engine written entirely from scratch in Rust, focusing on modern search techniques, efficient evaluation, and clean engineering.
The project began as a way to learn how chess engines work. Over time, it evolved into a serious long-term project through continuous experimentation, profiling, tuning, and thousands of self-play and engine-vs-engine games.
Today, Roxie has achieved an official CCRL rating of 2682 Elo for version 0.6.3 and continues to improve with every release.
- ♟️ UCI compatible
- ⚡ Written entirely in Rust
- 🧠 Quantized HalfKP NNUE evaluation
- 🚀 Magic Bitboard move generation
- 🔍 Modern alpha-beta search with advanced pruning and move ordering
- 📈 Continuously tuned through thousands of self-play and engine matches
- 💻 Optimized builds for Native, AVX2 and AVX-512 processors
Roxie is continuously tested against established engines using automated tournaments and self-play to measure playing strength and evaluate new ideas.
One memorable game is shown below, where Roxie managed to defeat Lambergar (~3500 Elo) in a 10+0.1 rapid game.
The full game is available in assets/roxie_vs_lambergar.pgn.
Roxie uses a modern alpha-beta search framework designed to search deeper while minimizing unnecessary work.
- Principal Variation Search (PVS)
- Iterative Deepening
- Aspiration Windows
- Quiescence Search
- Transposition Tables
- Advanced pruning and reduction techniques
- Efficient move ordering heuristics
- Quantized HalfKP NNUE evaluation
- Efficient neural inference for positional evaluation
- Classical evaluation terms blended with neural evaluation
- Magic Bitboards
- Optimized make/unmake move implementation
- Zobrist hashing for transposition and repetition detection
- Native CPU optimizations
- AVX2 and AVX-512 optimized builds
- Profile Guided Optimization (PGO) support
- Rust (stable toolchain)
- Cargo
- GNU Make
Clone the repository:
git clone https://github.com/sham-404/roxie.git
cd roxieBuild Roxie using one of the following targets:
# Optimized for your current CPU
make native
# Optimized for AVX2 + BMI2 + POPCNT capable processors
make avx2
# Optimized for AVX-512 capable processors
make avx512
# Build using Profile-Guided Optimization
make pgo-avx512To remove build artifacts:
make cleanAfter building, the engine executable will be available in the project directory (the filename depends on the selected build target).
./roxie*
Roxie implements the Universal Chess Interface (UCI) protocol.
./roxie*Example session:
uci
isready
ucinewgame
position startpos
go depth 12
Or search a custom position:
position startpos moves e2e4 e7e5 g1f3
go movetime 5000
Roxie works with any UCI-compatible GUI, including:
- Arena
- Cute Chess
- Banksia GUI
- Nibbler
Simply add the compiled executable as a new UCI engine.
Roxie can also be used with tournament managers such as cutechess-cli for engine matches and automated self-play testing.
Current focus:
- Improve NNUE playing strength using larger and higher-quality training datasets
- Continue search tuning and evaluation improvements
- Stronger endgame play
- Better time management
- Multi-threaded search (SMP)
Long-term goal:
Reach 3500 Elo while remaining a clean, educational, and highly optimized open-source chess engine.
Roxie was developed independently from scratch, but several resources and projects encouraged and inspired me throughout its development.
Special thanks to:
- Sebastian Lague for the Coding Adventure: Chess series, which inspired me to begin writing my own chess engine.
- The Chess Programming Wiki for providing an excellent introduction to many chess programming concepts and serving as a handy reference during development.
- The many authors of open-source chess engines whose work and discussions continue to advance the chess programming community.
- Lambergar, whose impressive playing strength and steady progress motivated me to keep pushing Roxie further whenever I felt like stopping.
Roxie is actively developed, with every significant improvement documented through version history, testing, and benchmarking.
For additional information, see:
- CHANGELOG.md for the complete development history.
- docs/nnue.md for details about Roxie's NNUE architecture and implementation.
This project is licensed under the MIT License.
See the LICENSE file for details.