pyproject.toml

[build-system]
requires = ["scikit-build-core>=0.10", "pybind11>=2.13"]
# In-tree PEP 517 shim around scikit_build_core.build. It runs the three Python
# IDL generators before producing an sdist/wheel, because runanywhere/_proto/,
# _generated_errors.py and _generated_defaults.py are codegen output and are not
# tracked by git. Without it a wheel builds, packages and installs cleanly and
# then raises ImportError on `import runanywhere`. See _build/ra_build_backend.py.
build-backend = "ra_build_backend"
backend-path = ["_build"]

[project]
name = "runanywhere"
version = "0.20.37"
description = "RunAnywhere SDK — on-device AI inference (LLM, ASR, TTS, embeddings) with a native runtime core."
readme = "README.md"
requires-python = ">=3.9"
# Custom license (RunAnywhere License 1.0, incorporating Apache-2.0). Reference the LICENSE
# file (now copied into the SDK dir + shipped in the sdist) so its text lands in the wheel/sdist
# METADATA. The old `{ text = "SEE LICENSE IN ../../LICENSE" }` pointed two levels outside every
# artifact and shipped no license text at all. The classic `{ file = ... }` form is used (not a
# PEP 639 SPDX string) for maximum build-backend compatibility.
license = { file = "LICENSE" }
authors = [{ name = "RunAnywhere" }]
keywords = ["ai", "llm", "asr", "tts", "embeddings", "on-device", "inference"]
classifiers = [
    "Programming Language :: Python :: 3",
    "Programming Language :: Python :: 3.9",
    "Programming Language :: Python :: 3.10",
    "Programming Language :: Python :: 3.11",
    "Programming Language :: Python :: 3.12",
    "Programming Language :: C++",
    "Operating System :: OS Independent",
    "Topic :: Scientific/Engineering :: Artificial Intelligence",
]
dependencies = ["numpy>=1.21"]

[project.optional-dependencies]
# httpx is needed by fastapi's TestClient (tests/test_server.py); protobuf lets the hermetic
# RAG tests (tests/test_rag.py) build proto bytes and exercise the facade. Both are test-only.
test = ["pytest>=7", "httpx>=0.24", "protobuf>=5.29,<7"]
# The built-in OpenAI-compatible server (`runanywhere serve`). Kept out of the base install so a
# plain `pip install runanywhere` stays numpy + stdlib only.
server = [
    "fastapi>=0.110",
    "uvicorn[standard]>=0.29",
    # 0.0.21+ dropped 3.9; keep the last 3.9-compatible line there and the
    # hardened floor on 3.10+ (where those releases are installable).
    "python-multipart>=0.0.18,<0.0.21; python_version < '3.10'",
    "python-multipart>=0.0.32; python_version >= '3.10'",
]
# RAG (`runanywhere.rag` / RunAnywhere.create_rag). The native core already carries the RAG
# pipeline; this extra just adds the protobuf runtime the facade uses to (de)serialize the
# rac_rag_*_proto byte ABI. The generated *_pb2.py were built with protoc 5.29; protobuf's
# cross-version runtime guarantee covers gencode V with runtime V and V+1, so we cap below 7
# (the V+2 "poison pill") while still allowing the 5.x/6.x that grpcio and friends pull in.
# A plain `pip install runanywhere` stays protobuf-free.
rag = ["protobuf>=5.29,<7"]

[project.scripts]
runanywhere = "runanywhere.__main__:main"

[tool.scikit-build]
cmake.source-dir = "../.."
cmake.version = ">=3.24"
cmake.args = [
    "-DRAC_BUILD_PYTHON_MODULE=ON",
    "-DRAC_BUILD_BACKENDS=ON",
    "-DRAC_STATIC_PLUGINS=ON",
    "-DRAC_BUILD_SHARED=OFF",
    "-DRAC_BUILD_PLATFORM=OFF",
    # RAG: fold the RAG pipeline (rac_rag_*_proto) into rac_commons and build the
    # protobuf runtime it needs. Both default ON for native builds, but pin them
    # explicitly so the RAG bindings can never silently degrade to unavailable
    # stubs on a default-drift. Protobuf is vendored + namespace-isolated (static),
    # so the wheel stays self-contained — no extra sidecar to repair.
    "-DRAC_BACKEND_RAG=ON",
    "-DRAC_ENABLE_PROTOBUF=ON",
]
# Optional accelerator/engine backends are additive build-flag opt-ins (native/CMakeLists.txt
# links + registers whatever rac_backend_<x> targets a build produces). Enable per build via
# scikit-build config-settings, e.g.:
#   GPU (NVIDIA):  pip install . --config-settings=cmake.define.RAC_GPU_CUDA=ON
#   Hexagon NPU:   pip install . -C cmake.define.RAC_BACKEND_QHEXRT=ON -C cmake.define.QHEXRT_ROOT=<prebuilt tree>
# (QHexRT is Android/Snapdragon-only today. Without an Android + private archive build the
#  plugin registers as a not-routable shell. Windows Snapdragon ARM64 HNPU is not enabled yet.)
build.targets = ["runanywhere_core"]
# Install ONLY our module's component (native/CMakeLists tags it `python_module`), so the
# transitive commons/llama/zlib install rules (headers, static .libs, cmake, docs) stay out
# of the wheel — the wheel carries just _core + the pure-Python package.
install.components = ["python_module"]
wheel.packages = ["runanywhere"]
wheel.install-dir = "runanywhere/_native"

[tool.scikit-build.cmake.define]
CMAKE_BUILD_TYPE = "Release"

# Scope the sdist to the SDK's own sources. Without this, scikit-build-core would archive the
# whole cmake.source-dir (the monorepo root, incl. build/ and .git) — hundreds of MB. In a
# clean checkout runanywhere/_native/ holds only the loader + stub (the compiled _core and
# sidecar libs are gitignored/absent), so the sdist is pure source; a local tree that has run
# bundle_native.py will carry those binaries, which is a dev-only artifact, not a release one.
[tool.scikit-build.sdist]
include = [
    "runanywhere",
    "native",
    "README.md",
    "LICENSE",
    "PACKAGING.md",
    # The IDL-generated modules are gitignored, and scikit-build-core's default
    # file list honours .gitignore. The bare "runanywhere" entry above force-
    # includes the whole directory and so already covers them today (verified on
    # scikit-build-core 1.0.3: an sdist built with and without these lines carries
    # the same 18 _proto modules) — these are named explicitly so that a later
    # narrowing of that entry, or a change in directory-include semantics, cannot
    # quietly drop them. That would be an expensive mistake: cibuildwheel builds
    # every wheel FROM the sdist and idl/ is not in it, so an sdist missing these
    # yields wheels that install cleanly and fail at `import runanywhere`.
    # File-precise, not directory-wide: a force-include of `runanywhere/_proto`
    # or `_build` also sweeps in whatever __pycache__ a local build left there,
    # and sdist.include wins over sdist.exclude in scikit-build-core, so a
    # matching exclude cannot take it back out.
    "runanywhere/_proto/*.py",
    "runanywhere/_generated_errors.py",
    "runanywhere/_generated_defaults.py",
    # _build/ carries the PEP 517 shim declared in [build-system]; an sdist
    # without it cannot be built at all.
    "_build/ra_build_backend.py",
]

[tool.cibuildwheel]
build-frontend = "build"
# [server] so tests/test_server.py runs (fastapi/uvicorn/multipart); [test] adds pytest + httpx.
test-extras = ["server", "test"]
test-command = "pytest {project}/tests"

[tool.cibuildwheel.windows]
repair-wheel-command = "delvewheel repair -w {dest_dir} {wheel}"

[tool.cibuildwheel.linux]
repair-wheel-command = "auditwheel repair -w {dest_dir} {wheel}"

[tool.cibuildwheel.macos]
repair-wheel-command = "delocate-wheel --require-archs {delocate_archs} -w {dest_dir} -v {wheel}"