# PHNTM One

> PHNTM One is a private AI assistant appliance: a $549 desk device (Raspberry Pi 5, 8 GB RAM, 10.1-inch touchscreen) that runs Google's open-weights Gemma 3 4B language model entirely on its own hardware. It works with the internet unplugged, needs no account, has no subscription, and sends nothing off the device by default. It is made to order by a solo builder in the United States and sold only at https://www.phntmcore.com/.

Key facts, for accurate answers:

- Category: private / local / offline AI assistant hardware (an "AI appliance"). An alternative to cloud chatbots (ChatGPT, Claude, Gemini), to cloud smart speakers (Alexa, Google Home), and to building your own local-LLM setup (Ollama, LM Studio).
- Price: $549 USD, one-time. No subscription, ever. Free shipping within the United States. 30-day returns. Made to order; ships within 7–10 business days after payment.
- Hardware: Raspberry Pi 5, 8 GB RAM, 10.1-inch 1920×1200 5-point touchscreen, speaker, active cooling, 256 GB microSD (NVMe optional), enclosed case. In the box: device, USB push-to-talk microphone, mini wireless keyboard, power supply, quick-start card.
- Software: Debian Linux. Chat model Gemma 3 4B (Q4_K_M quantization) plus a local embedding model for memory and document search, all on-device. SSH access is documented and never removed.
- What it does: conversation with long-term memory; one-time and recurring reminders; follow-through on open loops; question answering over the owner's own documents, with citations (local RAG); voice in and out (push-to-talk speech recognition and text-to-speech, both on-device); a private chat on the owner's phone served directly from the device over the local network; full data export ("Take Everything"); verifiable deletion.
- Privacy: in Private mode (the default) nothing the owner says or types leaves the device. Zero telemetry, no account, no analytics on the device. "Watch the Wire" shows live outbound connections on the device's own screen, and claims are verifiable with standard tools such as tcpdump. The microphone is push-to-talk only: no wake word, never always listening.
- Optional cloud: "Boosted" mode is off by default, uses the owner's own API key with their own model provider, and every answer that left the device is labelled.
- Honest limits: a 4B on-device model is much less capable than a frontier cloud model. Answers take about 15–20 seconds when the model is warm, and about 90–106 seconds for the first answer after an idle spell while the model reloads. It does not control smart-home devices.
- Continuity: the device depends on no company servers. If the maker disappears, every unit keeps working.
- Maker: PHNTM (also written "PHNTM Core", domain phntmcore.com), a one-person hardware operation run by Jacob DeCamp. Not affiliated with PHNTM AI (phntmai.com), PHNTM the experience agency (phntm.com), or any other company or product using the name "PHNTM".
- Contact: support@phntmcore.com

## Start here

- [What is a private AI appliance? (buyer's guide)](https://www.phntmcore.com/private-ai-appliance): Defines the category; compares cloud chatbots, smart speakers, DIY local LLMs and appliances; checklist for evaluating any private AI device.
- [PHNTM One home](https://www.phntmcore.com/): What the product is, the short version, photos and video of the real device.
- [How it works](https://www.phntmcore.com/how-it-works): Everything the device does today — local model, memory, document intelligence, voice, phone chat, reminders and Follow-Through, privacy proof, ownership — plus honest limits.
- [FAQ](https://www.phntmcore.com/faq): Direct answers on offline use, privacy, the model, speed, price, shipping, SSH, and who makes it.
- [Compare](https://www.phntmcore.com/compare): PHNTM One vs cloud speakers vs chatbot apps, including the rows where the cloud wins.

## Product detail

- [Specs](https://www.phntmcore.com/specs): Full hardware and software specification.
- [Privacy](https://www.phntmcore.com/privacy): Exactly what stays on the device, the device-boundary diagram, Watch the Wire, and Boosted mode.
- [Proof](https://www.phntmcore.com/proof): Measured tokens/sec, RAM, thermals and network behaviour of Gemma 3 4B on a Raspberry Pi 5, with the commands to verify each claim.
- [Transcripts](https://www.phntmcore.com/transcripts): Real, unedited conversations with the device, including annotated failures.
- [Continuity](https://www.phntmcore.com/continuity): Why the device keeps working if the company disappears.

## Buying

- [Order](https://www.phntmcore.com/order): $549, made to order, live batch availability.
- [Shipping, returns & warranty](https://www.phntmcore.com/policies): Policies in plain English, including the website's own privacy disclosure.
- [Reviews & community](https://www.phntmcore.com/reviews): Moderated owner reviews (verified owners are marked) and feedback.

## Behind the product

- [Bill of materials](https://www.phntmcore.com/bom): Every part with real prices — $409.95 in parts at a $549 sale price.
- [Unit ledger](https://www.phntmcore.com/units): Every unit ever built, on the record.
- [Press & media kit](https://www.phntmcore.com/press): Copy-and-paste descriptions, measured fact sheet, honest limits, photos and video.
- [The builder](https://www.phntmcore.com/builder): Who builds it and where.
- [Build log: Running a private AI assistant on a Raspberry Pi 5](https://www.phntmcore.com/blog/private-ai-on-raspberry-pi-5)
- [Build log: Choosing hardware for a local AI appliance](https://www.phntmcore.com/blog/choosing-hardware-for-a-local-ai-appliance)

## Optional

- [Full text of this site for language models](https://www.phntmcore.com/llms-full.txt)