VANTAGE distribution

niryukti packages the independently implemented C++ optimization engine. Python wheels contain the CPU executable and native C ABI library; no existing optimization solver is used. Python 3.10+ is required. Source installations need CMake 3.24+ and a C++20 compiler. Default wheels use FP64 and do not require CUDA.

from niryukti import Model, NativeSession
m = Model()
m.add_var("x", ub=10)
m.add_constraint({"x": 1}, ">=", 2)
m.set_objective({"x": 3})
with NativeSession(m) as session:
    print(session.solve())

Build: python -m build. Install a generated wheel with pip install dist/*.whl. The installed niryukti command exposes the engine CLI. CUDA builds can be used through VANTAGE_BINARY and VANTAGE_LIBRARY; CUDA is not bundled into CPU wheels.

License: AGPL-3.0-only. Modified covered redistributions must retain AGPL; modified network deployments must offer corresponding source to their users. Copying and commercial use remain permitted. Third-party licenses are retained. See LICENSE and NOTICE in each distribution.

Registry publication needs account ownership and trusted-publisher setup. Only tested host platforms should be advertised. The release workflow builds repaired manylinux x86_64 wheels plus a corresponding-source tarball; other platforms may build from source. CUDA builds are not bundled in the initial CPU release.