Skip to content

About

AIRunner native launcher, crash handler, and runtime sidecar build tooling

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

72 Commits

Folders and files

Repository files navigation

airunner-native

Pinned native runtime sidecar (llama.cpp / whisper.cpp) build tooling for Capsize-Games/airunner (AI Runner).

This repository builds and packages the native sidecars with CMake and Ninja, using MinGW for Windows cross-compilation. Its scripts and tests include Python code, but its runtime archives do not include the Python application or an interpreter. Sidecar source pins, patches and toolchains are released separately from the consuming application.

An earlier version of this repository also carried the Python launcher that starts the desktop app (airunner_native.launcher, crash_handler, repo_paths). That code turned out not to belong here: it's tightly coupled to airunner/airunner-services (it exists only to bootstrap them) and changes in lockstep with the rest of the application, not on an independent cadence -- splitting it added cross-repo coordination for no real benefit. It moved back into Capsize-Games/airunner's own native/ directory.

What's here

  • runtime_sidecars/runtime_pins.env -- exact upstream commits for llama.cpp and whisper.cpp, so bundled runtime binaries don't drift with either project's own master branch.
  • scripts/build_runtime_sidecars.sh -- builds both binaries for a given target platform.
  • .github/workflows/native-runtime-sidecars.yml -- builds the linux/windows matrix on workflow dispatch and checks both archives on their native operating systems. Release promotion builds from an annotated tag, attaches the verified assets to a draft, and publishes it as immutable. Published-release events verify the existing release.
  • scripts/build_cuda_sm120_candidate.sh and scripts/package_cuda_sm120_candidate.py -- build and hash-lock a separate Linux x86_64 CUDA SM 120a candidate. The package inputs are listed in runtime_sidecars/cuda-sm120-package-lock.json.

The CUDA candidate workflow builds an artifact without publishing a release. Release promotion requires a separately validated CUDA profile for the same annotated tag, uploaded to its draft release first. The native release workflow checks those three asset digests and the Linux/Windows CPU archives, then publishes the complete nine-asset set as one immutable release. A missing CUDA profile prevents publication. Existing published tags and CPU bundle paths remain unchanged.

Build locally

./scripts/build_runtime_sidecars.sh --target-platform linux

Requires cmake, mingw-w64, and ninja. Output lands in build/runtime-sidecars/<platform>/{bin/,share/airunner/}.

Consuming a built bundle

Capsize-Games/airunner's own release pipeline downloads a pinned release from here rather than building sidecars itself -- see that repository's .github/native-sidecar-version for which tag.

Provenance

Extracted from Capsize-Games/airunner (issue #2196, part of the repository-split tracker #2185). History for the moved files is preserved (git filter-repo).

Licensing

GPL-3.0-only, matching the application this builds for. See LICENSE.

About

AIRunner native launcher, crash handler, and runtime sidecar build tooling

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages