CrankGPT is a fully offline AI box that you power by turning a crank. There's no battery and no internet connection: just a single-board computer (Raspberry or Orange Pi) that runs speech recognition, a small language model, and text-to-speech locally. Crank, speak, and get a spoken answer back.
The standard build pairs a stock Raspberry Pi 5 with 8 gigabytes of RAM and a cheap 20-watt emergency hand generator, with a custom capacitor board smoothing the output and holding about 20 seconds of reserve so the Pi doesn't brown out when inference spikes the draw. The language model runs on llama.cpp, with Liquid AI's LFM2 in 350-million and 1.2-billion-parameter forms and Google's Gemma 3 at 1 billion parameters. Moonshine handles speech recognition and Piper does the voice synthesis. It boots in about 30 seconds from the first crank, and the time to first response runs from under a second on the smallest model to about three seconds on Gemma.
It even makes bad poetry, says Squeez Labs, which took a week to build a proof-of-concept and several months of board revisions and kernel work to make a conversation "feel real" on hardware this small.
Motivations
For something to have "smarts" currently assumes a wall socket and a data center. CrankGPT is a small argument that neither has to be true.
Local models are private models. Why give away what we don't have to?
It offended our European small-practical-car sensibilities to see people around us throwing kilowatts and thousands of tokens at tasks small models could accomplish just as well as huge ones, for a fraction of the cost and energy.
Everyone is busy making things bigger. We figured opportunities abound to make things smaller.
It's an argument that a functional AI assistant does not require a data center, and that plenty of useful work fits on a device a person can power by hand. And it gives us something to worship after the bombs fall.
Previously:
• A website that only works offline
• Earleaf: an offline audiobook app built out of frustration
