A private AI appliance for tinkerers and home-labbers

You could build a local AI box yourself. PHNTM One is the version where the tuning, the burn-in and the privacy plumbing are done, and nothing is locked. It's your machine, and it's meant to be opened up.

What's yours

• Debian 13 on a Raspberry Pi 5 with 8 GB, with Ollama bound to localhost and run with --offline.
• SSH, documented and never removed. The username and password you choose during setup are the device's login.
• Readable source on the device under the PHNTM One Source License, so you can check what it does.
• A local HTTPS API on port 8843, the same API the phone app uses (the screen reaches it locally), gated by your pairing token and reachable only from your LAN.
• Skills you can write. A skill is a short manifest and a Python file, run in a sandbox with no network and no credentials. How skills work →
• Home Assistant, over your LAN with your own token: lights, plugs, scenes and climate from a box you can SSH into. Smart home →
• The full bill of materials, every part and every dollar. See the BOM →

Why not just run Ollama yourself?

You can, and the buyer's guide compares both honestly. PHNTM One adds the parts that take weeks: a boot-time RAM budget for the model, local memory and document search, tap-to-talk voice, phone pairing with no relay, a firewall that stops the model server reaching the internet, fifteen bounded self-repairs, sandboxed skills, a Home Assistant bridge, an offline library, and signed updates that roll back if a self-test fails.

Questions