hal0 docs

hal0 is a homelab AI inference platform. It manages model slots as systemd units, exposes an OpenAI-compatible /v1/* API on port 8080, and ships with a Vue dashboard plus a prewired OpenWebUI chat tab on port 3001. It’s happy bare-metal, in a VM, or as a tenant in a privileged LXC on your Proxmox node (AppArmor unconfined + GPU/NPU passthrough).

Status: v0.1.0-alpha. Cosign-signed releases are wired end-to-end and the install command actually installs. APIs may shift. Full FLM NPU coverage and ROCm/CUDA toolbox parity are still landing. Pages marked “Coming soon” are stubs.

Start here

Install The one-line installer, the pre-flight checks, and what `hal0 status` should say after.

Strix Halo The reference platform. Unified memory, iGPU + XDNA NPU, the tuning details.

Slot architecture The lifecycle state machine and how single-flight dispatch works.

API reference OpenAI-compatible endpoints, model registry, dispatcher.