Screenshots 5
Phntm-ONE
Summary
Every AI assistant you've tried so far has one thing in common: your words leave your machine the moment you type them. Phntm-ONE is a physical Linux appliance that runs its model on the device itself, so the conversation never reaches a data center.
The device ships pre-loaded with Gemma 3 4B (Q4_K_M), a quantized local model that runs with the network cable unplugged. It handles reminders, project notes, and document QA against files stored on-device — no upload, no account, zero telemetry by default. The touchscreen and paired wireless keyboard mean it sits on a desk and stays there. The ceiling is real: a 4B quantized model is capable at focused tasks and noticeably limited on anything requiring deep reasoning or broad knowledge recall. An optional Boosted mode routes queries through a user-supplied API key, but every answer that left the box is labeled as such.
Bottom line: The right buy for someone who wants a private, always-on desk assistant they own outright and can interrogate with SSH; the wrong buy the moment they need a model that can handle nuanced, multi-step reasoning or follow them across devices.
Pricing Plans
- Price
- $549 one-time
Standard
One-time purchase of the physical appliance
- Raspberry Pi 5 hardware
- Gemma 3 4B on-device
- Free US shipping
View full pricing on phntmcore.com →
Pricing may have changed since last verified. Check the official site for current plans.
Community Performance Report Card
No community ratings yet. Be the first to rate this tool!
Pros
Sign in to edit- Runs fully air-gapped by default, so documents you query against never leave the physical device — directly addressing the data-residency concern that rules out every cloud-based assistant for regulated or sensitive contexts.
- One-time purchase with no subscription, which means the assistant keeps working and keeps its memory of your projects whether or not you keep paying anyone anything.
- SSH access is documented and never removed, so you can inspect every process running on the device and verify the privacy claims rather than taking them on faith.
- Boosted mode routes a query to a frontier model via your own API key when the local model hits its ceiling, and the device marks the response so you always know which path it took — no silent cloud escalation.
- Functions without internet connectivity, so it remains available in environments where network reliability is not guaranteed and cloud-dependent tools would simply stop.
Cons
Sign in to edit- Gemma 3 4B at Q4_K_M quantization answers focused, single-topic queries adequately and loses coherence on multi-document synthesis, nuanced argument analysis, or tasks that require a wide knowledge base — users who hit that ceiling regularly will find themselves defaulting to Boosted mode, at which point the privacy guarantee they bought the device for only holds when they remember to stay out of Boosted mode.
- There is no software-only path: the assistant is tied to the hardware unit, so it cannot follow a user to a laptop, phone, or second desk without purchasing another device — teams or individuals who need cross-device continuity will abandon this for a self-hosted open-source stack like Ollama paired with a synced memory layer.
- No API surface is exposed, which means the assistant cannot be embedded into a broader workflow, triggered by another tool, or queried programmatically — anyone who wants their private assistant to participate in an automation chain hits a hard wall and needs a different solution entirely.
About
- Platforms
- Raspberry Pi 5, Debian Linux
- API Available
- No
- Self-Hosted
- Yes
- Last Updated
- 2026-09-09T05:02:51.252Z
Best For
Who it's for
- Users prioritizing data privacy and ownership
- Environments without reliable internet
- Buy-once hardware rather than subscriptions
What it does well
- Private desk-side assistant for reminders and project notes
- Local document QA without cloud upload
- Always-available on-device chat via touchscreen or paired phone
Add notes, reviews, and benchmarks so the next visitor gets a clearer picture.
Spotted incorrect or missing data? Join our community of contributors.
Sign Up to ContributeFrequently Asked Questions
- Is Phntm-ONE free?
- Phntm-ONE is a paid tool ($549 one-time). No permanent free tier is offered.
- Is Phntm-ONE open source?
- No — Phntm-ONE is a closed-source tool. Source code is not publicly available.
- Can I self-host Phntm-ONE?
- Yes. Phntm-ONE supports self-hosting on your own infrastructure.
- What platforms does Phntm-ONE support?
- Phntm-ONE is available on: Raspberry Pi 5, Debian Linux.
Curated lists that include this category
The data-leak problem with cloud assistants
Every AI assistant you’ve tried so far has one thing in common: your words leave your machine the moment you type them. Phntm-ONE is a physical Linux appliance that runs its model on the device itself, so the conversation never reaches a data center.
How it works
The device ships pre-loaded with Gemma 3 4B (Q4_K_M), a quantized local model that runs with the network cable unplugged. It handles reminders, project notes, and document QA against files stored on-device — no upload, no account, zero telemetry by default. The touchscreen and paired wireless keyboard mean it sits on a desk and stays there. The vendor states the ceiling is real: a 4B quantized model is capable at focused tasks and noticeably limited on anything requiring deep reasoning or broad knowledge recall. An optional Boosted mode routes queries through a user-supplied API key, but every answer that left the device is no longer private. Price is $549 one-time. Platforms are Raspberry Pi 5 running Debian Linux.
Use cases and trade-offs
Best for private desk-side assistant work, local document QA without cloud upload, and always-available on-device chat via touchscreen or paired phone. Pros include air-gapped operation by default so documents never leave the physical device, one-time purchase with no subscription, and documented SSH access for inspection. Cons are that Gemma 3 4B at Q4_K_M loses coherence on multi-document synthesis or wide knowledge tasks, and there is no software-only path — the assistant is tied to the hardware unit.
Who it is for / who should skip it
Users prioritizing data privacy and ownership, environments without reliable internet, or those who prefer buy-once hardware rather than subscriptions will find the fit. Skip it if you need deep reasoning or broad recall without switching to Boosted mode and losing the privacy guarantee.