One binary. Your platform is the easy case.

Haybox is in private alpha. The platform builds below are real and tested; none of them is downloadable yet. Join the waitlist and you get the first public build and a savings estimate on the way in.

PlatformStatusNotes
Windowsin private alphaNative: real GPU detection, no WSL, no container sidecar. If your 4090 lives in a gaming rig, you’re who this build was tested on.
macOSin private alphaApple silicon first-class; unified memory sized into recommendations.
Linuxin private alphaOf course. Package-manager instructions land with the public build.

Checksums and signing information publish with the first public build. There is deliberately no placeholder here: a checksum you cannot verify is worse than none.

What happens on first run

  1. Hardware detection: GPU, VRAM, RAM, OS, free disk.
  2. Choose a door: the native agent, or point the tool you already use at the gateway.
  3. Shadow Mode starts building your report.

Sixty seconds, no account required for the free tier.

System requirements

Full local experience: 16 GB+ VRAM, or Apple silicon with 32 GB+ unified memory. Shadow Mode and the Ledger run on anything that runs the binary.

Model Manager, illustrative: detected RTX 4090 with 24 GB VRAM on Windows 11; recommended 30B-class primary at an expected 41 tok/s and 8B-class utility at 96 tok/s; one-click install via a local runtime.
Model Manager, illustrative: detected RTX 4090 with 24 GB VRAM on Windows 11; recommended 30B-class primary at an expected 41 tok/s and 8B-class utility at 96 tok/s; one-click install via a local runtime.

Detected hardware and a recommended setup. Illustrative.

Check your hardware fit → · Quickstarts per tool →