Open Source & Corresponding Source

CheesCoach · last updated 2026-07-21

CheesCoach is built on open-source software. This page provides the required attribution for the components we use and, for the GPLv3-licensed chess engines that the Android app distributes on-device, the Corresponding Source as required by section 6 of the GNU General Public License, version 3.

How to get the source. For each GPL engine below we give the exact upstream version tag plus the complete build instructions we used to produce the binary shipped in the app. Together these constitute the Corresponding Source. This page is kept available for as long as CheesCoach distributes the corresponding binary. If any upstream link is unavailable, contact us (see the bottom of this page) and we will provide the source directly.

1. GPLv3 engines shipped on Android (on-device)

Stockfish 18 — GPL-3.0-or-later

The Android app bundles the official, unmodified Stockfish 18 binary for arm64-v8a (release tag sf_18), placed at app/src/main/jniLibs/arm64-v8a/libstockfish.so. Because it is the official release binary with no modifications, its Corresponding Source is the upstream source at the same tag:

Copyright © The Stockfish developers (see the AUTHORS file in the source).

Leela Chess Zero (lc0) v0.31.2 — GPL-3.0-or-later

The "human-like bots" (Maia) run on an lc0 binary that we cross-compile ourselves for arm64-v8a from tag v0.31.2, placed at app/src/main/jniLibs/arm64-v8a/liblc0.so (no embedded network). Because this is a self-built binary, the Corresponding Source comprises the upstream lc0 source, the statically-linked OpenBLAS source, and the exact build configuration below.

Exact build steps (the complete build script is documentos/maia/build-lc0-android.sh in our source tree; its essential commands are reproduced here so the binary can be regenerated from upstream alone):

# 1) OpenBLAS v0.3.27, static, android-aarch64
git clone --depth 1 -b v0.3.27 https://github.com/OpenMathLib/OpenBLAS.git
make -C OpenBLAS TARGET=ARMV8 ONLY_CBLAS=1 NO_LAPACK=1 NO_SHARED=1 \
     USE_OPENMP=0 USE_THREAD=0 HOSTCC=cc \
     CC=aarch64-linux-android21-clang AR=llvm-ar RANLIB=llvm-ranlib
make -C OpenBLAS PREFIX=<install> NO_SHARED=1 install

# 2) lc0 v0.31.2 with submodules
git clone --depth 1 -b v0.31.2 --recurse-submodules \
     https://github.com/LeelaChessZero/lc0.git

# 3) meson cross-file (NDK 26, 16 KB page alignment for Android 15+)
#   [host_machine] system=android cpu_family=aarch64 cpu=aarch64 endian=little
#   [properties]  cpp_link_args = ['-llog','-static-libstdc++','-Wl,-z,max-page-size=16384']
#   [binaries]    c/cpp = aarch64-linux-android21-clang(++), ar/strip/ranlib = llvm-*

# 4) configure + build
meson setup arm64-v8a --buildtype release \
     -Dgtest=false -Dembed=false -Ddefault_library=static \
     -Dopenblas=true -Dopenblas_include=<install>/include \
     -Dopenblas_libdirs=<install>/lib --cross-file crossfile-aarch64
ninja -C arm64-v8a lc0
llvm-strip arm64-v8a/lc0   # → liblc0.so

Copyright © The LCZero Authors (lc0) and © The OpenBLAS contributors (OpenBLAS, BSD-3-Clause).

Maia networks (weights) — data

The nine Maia strength networks (maia-1100 … maia-1900) are shipped as data (neural-network weights loaded by lc0 at runtime), not as executable code. They originate from the Maia Chess project: github.com/CSSLab/maia-chess (CSSLab, University of Toronto). Maia-1 networks are distributed under GPL-3.0.

2. iOS build

The iOS app does not ship any GPL binary on-device. Stockfish runs server-side (accessed over the network, which is not "conveying" under the GPLv3); on-device play and analysis use Caissa (MIT), and the human-like play uses a clean-room ONNX Runtime (MIT) with MIT-licensed models. Section 6 therefore does not apply to the iOS build; only attribution applies (section 3 below).

3. Other components & attributions

4. Full license texts

Contact

For any question about these licenses or to request the Corresponding Source directly, email juanfranj80@gmail.com.

← Back to CheesCoach