Offline AI Knowledge Phone — Prepper Edition
The entire offline internet, plus an AI you can talk to — in your hand. No signal. No laptop. No subscription.
A standalone touchscreen Android device — not a USB stick you plug into a computer — factory-imaged with a quantized open-source language model, a full offline Wikipedia dump, and offline survival references, all running with the radios off.
One device. Everything already on it.
This is a generic, unbranded Android handset that boots straight into a chat screen. You type or talk; it answers — from a quantized open-source LLM running locally on the phone itself. No account, no cloud, no data connection. Alongside the model sits a complete offline Wikipedia dump and a set of offline survival references and maps, all imaged onto the device before it ships. It is not an app you install, not a stick you plug into a laptop, and not a service you subscribe to. It is a finished thing you turn on.
The datasheet, not adjectives
Every unit ships with a spec sheet, and its exact values are printed on-device and published in that batch's changelog. We name the base handset and the model version rather than describing them vaguely — the same way you'd read a component list before trusting a tool in the field. We build the disk image and flash the units in-house; we do not manufacture the hardware and never claim to.
Airplane mode on, response left uncut
The demo clip on this page is a single unedited take: airplane mode visibly enabled, a real question typed in, and the model's answer generated on-device with the response time left in. We don't trim the wait. A phone-class device running a quantized model answers in seconds, not instantly, and the clip shows exactly what that feels like so you can decide if that trade is right for how you'd use it. What you see in the clip is what the device does with every radio off.
Why this exists
I got tired of survival gadgets that need a laptop in the field — reference libraries locked inside a computer, AI tools that quietly phone home, 'offline' kits that assume you'll be somewhere with a desk and a cable. This is the version that doesn't. Buy the white-label hardware in bulk, build one careful disk image, flash it across the run, and hand you a device that already knows things when nothing else has signal. That's the whole idea, and the spec sheet above is me showing my work.
First flash batch — dated on-device
Every unit in this run is imaged from one dated disk image in a single bounded production run. Each batch republishes a short changelog: the model version, the dataset date, and the known limitations honestly listed. The exact base-handset model name and your unit's flash-batch date are printed on the device and recorded in that batch's changelog at the time it's imaged. A demand deposit reserves your place in that run before it ships.
First flash batch: every unit imaged from one dated disk image in a single bounded production run, with the flash-batch date printed on-device and published in that batch's changelog.
What you can check
- Exact base handset, RAM/storage/battery, model family, parameter count and quantization level named on every batch's spec sheet
- Runs fully offline — Wikipedia dump, references and the LLM are imaged onto the device, not streamed
- Unedited on-device Q&A demo with airplane mode on and the real response time left in
- Disk image built and flashed in-house; hardware is white-label and never claimed as domestically made
This is a demand deposit, not an order — no shipping info collected now. The $2 is fully refundable: it's credited toward your $249 if you keep your reservation, and refunded instantly the moment you don't.