OpenNN
OpenNN is a C++20 library for building, training and running neural networks. It supports tabular data, images, time series and text, with CPU and optional CUDA execution. Models can be embedded in C++ applications or exported as standalone source code for supported architectures.
| I want to… | Start here |
|---|---|
| Build and run my first network | Quick start |
| Choose CPU/CUDA backends and build options | Build options |
| Use OpenNN in my C++ application | Install and link |
| Train, evaluate or export a model | Examples |
| Update an 8.x application or model | Migration |
| Understand the source folders | Repository contents |
| Run or review performance comparisons | Benchmarks |
| Contribute or run the tests | Contributing |
| Use OpenNN under the AGPL or a commercial licence | License |
Tutorials are available on opennn.net.
Quick start
You need Git, CMake 3.24+, and a C++20 compiler with std::format and OpenMP:
GCC 13+, Clang 17+, or Visual Studio 2022 C++ tools. On macOS, install the
OpenMP runtime (libomp). CMake retrieves missing dependencies during the
first configuration, so that step needs internet access.
Clone the repository and configure a CPU build:
git clone https://github.com/Artelnics/opennn.git
cd opennn
cmake -S . -B ../opennn-build -DCMAKE_BUILD_TYPE=Release -DOpenNN_DISABLE_CUDA=ON -DOpenNN_BUILD_TESTS=OFF -DOpenNN_BUILD_EXAMPLES=ON
cmake --build ../opennn-build --config Release --target blank --parallel
Run the blank example on Linux/macOS or with a single-configuration generator:
../opennn-build/bin/blank
With Visual Studio on Windows:
../opennn-build/bin/Release/blank.exe
Next, follow the Iris example to train a classifier, evaluate it and export predictions as C or Python code.
CUDA builds
Install a C++20-compatible CUDA toolkit and cuDNN 9+. Use a separate build directory so the CPU and CUDA configurations do not overwrite each other:
cmake -S . -B ../opennn-build/cuda -DCMAKE_BUILD_TYPE=Release -DOpenNN_REQUIRE_CUDA=ON -DOpenNN_BUILD_TESTS=OFF -DOpenNN_BUILD_EXAMPLES=OFF
cmake --build ../opennn-build/cuda --config Release --target opennn --parallel
Require CUDA explicitly to make missing GPU build dependencies an error.
Use OpenNN in your application
Install a completed build to a separate prefix:
cmake --install ../opennn-build --config Release --prefix ../opennn-install
In your application's CMakeLists.txt, consume the installed target:
cmake_minimum_required(VERSION 3.24)
project(MyApplication LANGUAGES CXX)
find_package(OpenNN 26.0 CONFIG REQUIRED)
add_executable(my_application main.cpp)
target_link_libraries(my_application PRIVATE OpenNN::opennn)
target_compile_features(my_application PRIVATE cxx_std_20)
Configure the application with -DCMAKE_PREFIX_PATH=/absolute/path/to/opennn-install.
The exported target supplies OpenNN's include paths and dependency linkage, and
the package still works after moving the install prefix.
The package consumer is a complete working example.
Fetched Eigen and zlib are installed with the library. OpenMP, oneTBB, oneDNN, MKL and NVIDIA runtimes are not copied into the installation; supply those selected by the build. Use a compiler compatible with the one that built the package. Installation packages exclude example datasets and models.
Include the public module headers directly, for example
#include "opennn/network/network.h". Do not include the private precompiled
header opennn/core/pch.h in an application.
Repository contents
| Folder | What it contains |
|---|---|
opennn/ |
The library: public headers, implementations, models and CPU/CUDA backends. See the module map. |
examples/ |
Runnable applications and bundled data. |
tests/ |
C++ unit tests of the library. |
benchmarks/ |
OpenNN versus PyTorch comparisons and their measurement rules. |
tools/ |
Verification scripts, code checkers, fuzzing, package checks and agent skills; see AGENTS.md. |
.github/workflows/ |
Continuous integration. |
Contributing and support
Base contributions on dev, include the validation you ran, and describe the
resulting behavior in the pull request. A pull request can be merged only after
its author has accepted the Contributor License Agreement; see
CONTRIBUTING.md. AGENTS.md holds the
engineering rules, verification commands and release procedure. For a bug
report, include the OpenNN commit, operating system, compiler, CPU/CUDA
configuration, reproduction command and error, and file it in
GitHub Issues.
License
OpenNN is Copyright (C) 2026 Artificial Intelligence Techniques, S.L. (Artelnics) and Roberto López González, and is dual-licensed:
- Open source: the GNU Affero General Public License, version 3 only (LICENSE.txt), with additional permissions for linking with NVIDIA CUDA, cuDNN and Intel oneMKL and for the code and model files that OpenNN generates.
- Commercial: a licence from Artelnics for closed-source products and hosted services that cannot meet the AGPL's terms. Write to artelnics@artelnics.com.
LICENSING.md gives the details. OpenNN 8.x and earlier, and the
development snapshots published before this change, were distributed under the
GNU Lesser General Public License, version 2.1 or later, and copies of them keep
those terms. Dependencies and downloadable pre-trained weights keep their own
terms, listed in THIRD_PARTY_NOTICES.txt. Example data
keeps its publishers' terms, recorded in the SOURCE.md notice of each data folder.