Everything it does, and where it does it.
PHNTM One is a full assistant, not a chat box. Here is the whole surface: what runs on the device, what stays there, and the optional paths that can leave.
Eight systems at the core.
The core of PHNTM One 1.0, on the hardware on your desk. Pick one, or let it walk you through. Everything built on top of them is on the features page.
A real model, on the device
A Gemma 3 4B model answers from the box itself — with the internet unplugged. A smaller local model stands by as an automatic fallback, so the assistant keeps working even while the main model reloads.
It remembers what matters
It builds a personal, on-device memory of your names, projects and preferences — and recalls them by meaning, not keyword. Correct it and the old fact is superseded; ask it to forget and it does. Everything lives in one place: the box on your desk.
Give it your documents
Hand it your files and it indexes them locally for retrieval-augmented answers grounded in your own material — and it tells you when it is answering from a document versus from the model. Delete a document and it leaves the index too.
Talk to it, out loud
Tap the mic or press Alt+M for a local voice conversation. Manual Talk stops on silence, another tap, or after 15 seconds. Optional local wake detection is a separate setting, off by default.
Your machine, from your pocket
Scan the on-screen QR once and a private chat opens in your phone’s browser — no app store, no account, no cloud relay in between. PHNTM Drop lets you send text and files straight from the phone into the assistant, all across your own network.
It closes the loop
Set one-off and recurring reminders, pause or skip them, and let it track open loops so nothing falls through. Reminders survive restarts — the schedule lives on the device, not in a browser tab.
Built to be questioned
Watch the Wire counts what your conversation sends (zero in Private mode unless you've connected Telegram or email) next to every byte the whole machine sends to the internet — verify it yourself with tcpdump. The box watches its own disk, memory and services, heals stuck services, and reports what it did in plain language.
You own the whole thing
It is a Debian Linux computer. SSH is documented and never removed. Take Everything exports your data, and owner-initiated updates roll themselves back if a self-test fails — nothing changes underneath you.
And the everyday things, built in.
The same private core, doing the jobs you'd expect from a home assistant, without the cloud those usually need.
Timers, alarms and lists
Planned · not release-verified Planned 1.0 capability, not yet release-verified: several named kitchen timers with on-screen countdowns and completion sounds. Restart timing has not been established. Documented · local Notes is available on the current device. Shopping-list packages use the local skills framework; a preinstalled pack of shopping, to-do and everyday tools remains planned 1.0 scope, not yet release-verified.
Photos and scans
A receipt, a letter, a whiteboard: it reads the text on the device and cites the photo in its answers.
Knows things offline
Planned · not release-verified Planned 1.0 capability, not yet release-verified: local Wikipedia and reference packs with source-linked answers. The current local model can answer general questions, but that is separate from a reference library.
Your home, locally
Control through your own Home Assistant over your local network is planned for 1.0 and is not yet release-verified. The planned scope covers the lights, plugs, scenes and thermostat you allow.
Dictation and read-aloud
Type a thought or record clips of up to 60 seconds in Notes. Review the transcript, then choose Clean up, Develop idea, or Project checkpoint. Originals are saved separately from AI drafts; processing stays local in every mode. Suggestions need your review. Document read-aloud and multilingual speech remain planned 1.0 additions, not yet release-verified.
Yours to run
The Skills screen, on-screen Update Center, account setup screens and paired-phone list remain planned 1.0 scope, not yet release-verified. Existing CLI/API and signed-update paths are separate from these proposed screens.
Private by default. Boosted or Autonomous when you choose.
Boosted and Autonomous stay off until you choose them. See it as a diagram on the privacy page.
Local inference, no required subscription
- Local inference stays on the device. Connected Telegram, email and calendar accounts have their own network paths.
- No account, no subscription, no cloud.
- The reason the product exists, and where most people stay.
A frontier model, your key
- Dramatically more capable answers on a hard task.
- Off by default, on your own credit.
- Not private: your prompt leaves the box, and every such answer is labelled. Local is always the fallback.
It acts, after you say yes
- Email and calendar, from your own private Telegram chat.
- Every send or change is shown in full and waits for your yes.
- Not private: what you ask goes to Anthropic and through Telegram. Off by default.
The fine print, in large print.
Deliberate trade-offs, stated plainly, so nothing surprises you after it arrives.
Capable, not magic
A 4B local model is genuinely useful, not a frontier model. It can be wrong, and the device always shows which brain answered.
A beat slower
The local model writes about two words a second, so a short answer usually takes tens of seconds and a long one can run past a minute. The first answer after an idle spell is slower, while the model reloads.
Voice on your terms
Tap-to-talk is the default. Optional local wake detection is off until you enable it; Notes clips can run for up to 60 seconds.
Two ways out, both off by default
Boosted and Autonomous are the only settings that send your words off the box (plus Telegram or email, if you connect them). Both are off by default and use your own key. Boosted is labelled every time it applies.
8 GB budget
It runs within 8 GB on a Raspberry Pi 5. Bigger models don't fit, and that constraint is the point.
1.0 first, more later
This page describes current local functions and planned 1.0 additions; their statuses differ. The vision is labelled separately.
Your AI. On your desk.
Order yours. Built by hand, burned in and tested before it ships.
Order PHNTM One$799 · one-time · no subscription · made to order · ships within 7–10 business days after payment

