Release Status

Salut has not been released, and there is no supported app download yet. The repository can build development artifacts, but those artifacts are not a substitute for the required signed-package evidence.

Planned Packages

macOS

The first menu-bar package targets Apple silicon. Its package task uses the official b11960 arm64 archive locked in llama-runtime.toml, stages the complete 39-file runtime closure, signs every Mach-O file, and seals a final hash and code-signature manifest with the outer app signature. The runtime requires that signature and exact manifest before staging files for launch.

The complete uv run poe package-macos command has passed this process end to end with ad hoc signing and produced an independently re-attested ZIP. Developer ID signing and notarization have not completed. The official b11960 Intel archive is source-locked, but Intel package production and native launch are not implemented yet.

Windows

Windows x64 with CUDA is an active product target. Native process supervision, signed packaging, CUDA0 plan validation, and a physical Salut production run must pass before a Windows package is offered.

The desktop preview uses official Toga WinForms controls and the existing WebView2 chat. Hosted CI runs native GUI lifecycle checks from source and from a Briefcase test-mode build, then produces an unsigned development ZIP. That ZIP does not bundle the CUDA runtime. It does not establish installer behavior, signing or physical GPU acceptance, and is not a supported release.

Linux

The headless daemon supports development of discovery, routing, trust, and control-plane behavior. Native llama.cpp packaging and CUDA acceptance remain in progress.

Development Checkout

Install the source checkout with Python 3.14.10 or newer and uv:

git clone https://github.com/salut-run/salut.git
cd salut
uv pip install -e ".[dev]"
uv run python -m salut

The source checkout starts the interface and control plane. Native inference requires the complete verified v0.8.0/b11960 arm64 closure, bundled by uv run poe package-macos. Source runs can set SALUT_LLAMA_APP to the absolute path of that signed app. The old two-executable overrides are no longer accepted. See Getting Started for the current workflow.

Included Architecture

  • OpenAI-compatible local API on port 7258 by default

  • signed peer HTTPS on port 17258 by default

  • mDNS discovery and mutual pairing

  • supervised pinned llama.cpp processes

  • exact GGUF identity and file leases

  • loopback-only raw RPC with exact-certificate encrypted forwarding

  • signed deployment, status, and stop controls for explicitly allowed operators

Automatic loading supports existing local GGUF files on Apple-silicon Macs. Final packages and arbitrary-N acceptance remain in progress.