DirectoryChatbots & Conversational AI
PHNTM One logo

PHNTM One

A private AI assistant appliance for your desk. Runs on its own hardware: works offline, no account, no subscription.

Visit PHNTM Onephntmcore.com
from$549United StatesLinux
PHNTM One screenshot, image 11/4

Overview

PHNTM One is a physical AI assistant you own, not an app or a cloud service. It is a small Linux computer (Raspberry Pi 5, 8 GB RAM, 10.1-inch touchscreen, speaker, push-to-talk microphone) that runs Google's open-weights Gemma 3 4B model on the device itself.

What it does: conversation with long-term memory, one-time and recurring reminders, follow-through on open loops, cited answers from your own documents, voice in and out, and a private chat on your phone served directly from the device over your own network.

Privacy you can check: by default nothing you say leaves the device. There is no account, no telemetry and no wake word. The device shows its own live outbound connections, and every claim can be verified with standard tools such as tcpdump. An optional Boosted mode (off by default) uses your own API key for harder tasks and labels every answer that left the box.

Honest limits: a 4B on-device model is far less capable than a frontier cloud model. Answers take about 15-20 seconds when warm, and about 90-106 seconds for the first answer after a long idle spell. It does not control smart-home devices.

$549 once, no subscription. Hand-built to order by a solo maker in the United States, ships in 7-10 business days with free US shipping and 30-day returns. The full bill of materials, measured benchmarks and unedited transcripts are public at phntmcore.com. If the company disappears, every unit keeps working: there are no servers.

Short description

PHNTM One is a private AI assistant appliance: a $549 desk device (Raspberry Pi 5, 10.1-inch touchscreen) that runs the Gemma 3 4B model entirely on its own hardware. Chat, long-term memory, reminders, voice and document search all work with the internet unplugged. No account, no subscription.

Sources behind this listingSee source links, recorded dates and owner corrections.

Observed facts come from public pages. Founder-edited facts are supplied by the verified owner. AI-inferred facts are model interpretations; derived facts are calculated from other data. A source link lets you check the current page. It does not guarantee the fact is still current.

Dates show when a record was saved. Some older records use the listing’s update date; they are not proof of a fresh check. Missing source records are shown explicitly.

TaglineA private AI assistant appliance for your desk. Runs on its own hardware: works offline, no account, no subscription.
Founder-editedDate not recordedSource link unavailable
DescriptionPHNTM One is a physical AI assistant you own, not an app or a cloud service. It is a small Linux computer (Raspberry Pi 5, 8 GB RAM, 10.1-inch touchscreen, speaker, push-to-talk...
Founder-editedDate not recordedSource link unavailable
Starting price$549
Founder-editedDate not recordedSource link unavailable
Free planNo
Founder-editedDate not recordedSource link unavailable
Free trialNo
Founder-editedDate not recordedSource link unavailable
Pricing modellifetime
Founder-editedRecorded Source link unavailable
IntegrationsTelegram
Founder-editedDate not recordedSource link unavailable
PlatformsLinux
Founder-editedRecorded Source link unavailable
Open sourceNo
Founder-editedRecorded Source link unavailable
Company locationUnited States
Founder-editedDate not recordedSource link unavailable
Founded2026
Founder-editedRecorded Source link unavailable
Headquarters countryUS
Founder-editedRecorded Source link unavailable
Team size1
Founder-editedRecorded Source link unavailable
Funding stagebootstrapped
Founder-editedRecorded Source link unavailable
SummaryPHNTM One is a private AI assistant appliance: a $549 desk device (Raspberry Pi 5, 10.1-inch touchscreen) that runs the Gemma 3 4B model entirely on its own hardware. Chat, long...
Founder-editedDate not recordedSource link unavailable

Key Features

  • Runs Gemma 3 4B entirely on-device (Raspberry Pi 5, 8 GB): chat, memory and voice work with the internet unplugged
  • Long-term memory you can inspect, correct and delete, with deletion that removes the searchable copy too
  • One-time and recurring reminders, plus follow-through on open loops
  • Cited answers from your own documents using local retrieval (RAG); files are dropped in over your LAN
  • Push-to-talk voice: on-device speech recognition and text-to-speech, no wake word, never always listening
  • Private phone chat served straight from the device by QR code: no app, no account, no cloud relay
  • Watch the Wire: live view of outbound connections on the device, verifiable with tcpdump
  • Optional Boosted mode with your own API key, off by default, every boosted answer labelled
  • Take Everything: full export of your data; Debian Linux with documented SSH access
  • No servers, no account, no subscription: keeps working if the maker disappears

Who Is It For

Best for

People who want an AI assistant for memory, reminders, voice and their own documents without sending any of it to the cloud or paying a subscription.

–Not for

Anyone needing frontier-level reasoning, instant answers or smart-home control. It is a small on-device model: private and useful, but slower than the cloud.

Strengths & Weaknesses

Strengths
Works offlineGood privacy practicesTransparent pricing
Weaknesses
No mobile appLimited integrationsNo free trialSlow to load

Recent updates

Dates are the source's own. Release and feed entries come straight from PHNTM One's GitHub or changelog; Founder entries are posted by the verified owner.

Classification

Competitors
Google AssistantAmazon AlexaApple SiriChatGPTOllama

Alternatives to PHNTM One

See all alternatives to PHNTM One

Community

Sign in to leave a comment.

No comments yet. Be the first to share your experience.

How we sourced this: Observed fields () were crawled from https://phntmcore.com/ when recorded. Source links and dates are available above. AI-inferred fields () were generated by gpt-4o-mini and are always labelled, never presented as measured fact. Founder-edited fields () have been corrected by the verified owner. Last updated: 30 September 2026. Site availability checked: 30 September 2026.