An Xboard compatible chess engine

Introduction

Prophet is a chess playing program written in C. Technically it's a chess engine, meaning there is no graphical user interface. To play a game with Prophet, you'll want to install Winboard (Windows) or Xboard (Linux/Mac). See http://www.gnu.org/software/xboard for details.

Using a Neural Network

By default, Prophet still uses a handcrafted evaluation. You can enable a neural network based evaluation using the 'n' command line parameter.

./prophet -n nn-52-q.txt

You should see a confirmation that the network was loaded:

loading neural network from nn-52-q.txt

The supplied network weights were trained using chess-trainer. chess-trainer is a Pytorch based trainer I wrote, with some influence from David Carteau's excellent Cerebrum library. The inference code within Prophet (particularly the intrinsics) were used almost verbatim. Thank you David!

Configuring Hash Tables

You can control the main hash table and pawn hash table sizes (in MB) using the 'h' and 'p' command line parameters.

./prophet -h 256 -p 32

This sets a 256 MB main hash table and a 32 MB pawn hash table. Defaults are 64 MB for the main hash table and 8 MB for the pawn hash table.

Developer API

Public API documentation lives in docs/API.md.

Building from Source

Binaries are provided for Linux and Windows based systems, but if you want to build Prophet from source, you'll need either a Linux environment with gcc/g++ or Windows with MSVC.

Just follow the standard cmake build process:

mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
cmake --build . --config Release

How Strong is it?

The June 6, 2026 CCRL Blitz list (https://www.computerchess.org.uk/ccrl/404/) shows Prophet 5.2 to be rated at 2753 (ranked 292).

Release Notes

5.2

Improved neural network
Native Windows build

5.1

changed neural network architecture to 2x(768->384)->2
factored quantization error into network training
moved several functions that should be internal out of public headers
lots of code cleanup

5.0

NNUE! Prophet now uses a neural network
more cleanup - public headers reduced to just those needed to support chess4j
Some speed optimizations

4.4

Maintenance release. Updated build system, switching from plain make to cmake. Formatted / cleaned up code.

4.3

Passed pawn by rank (was a single value)
Non-linear mobility (was a single value)
Knight outposts
Trapped bishop penalty

4.2

Evaluation terms were tuned using logistic regression with gradient descent (within chess4j)
Fully implemented tapered evaluation
Added simple mobility terms for bishop and queen
Added a pawn hash table

Roadmap

In the short term, I plan to continue experimenting with neural networks and looking for optimizations to offset the speed hit. Also, I've accumulated a pretty healthy backlog of items to try in the search. So, there will likely be a number of minor releases in the 5.x line.

You can see the combined Prophet / chess4j backlog here: https://trello.com/b/dhcOEaCO/chess4j-board .

Read about the latest development efforts at https://jamesswafford.dev/ .