CrankGPT is an unusual response to modern AI: a private chatbot that only works when a user powers it by hand. Instead of relying on a battery or conventional wall power, the device depends entirely on a hand crank to generate electricity for each conversation.
That design choice places privacy and local processing ahead of speed or scale. Built by Squeez Labs, the concept aims to make generative AI cheaper to run, less energy intensive, and more controlled by the person using it.
A local chatbot with a very physical power source
Inside the small box is a 20W hand-crank generator paired with a standard Raspberry Pi 5 with 8GB of RAM. The entire system runs locally, which means the chatbot does not need to send requests to a cloud service to function.
CrankGPT can listen to a user’s speech, process the request on the device itself, and then speak an answer using the Piper voice model. In practice, that makes it feel like a conventional chatbot, except the energy for every step comes from the person turning the crank.
Small models, limited workload
Because the hardware and power budget are narrow, CrankGPT is meant for lightweight models rather than large ones. Its creator recommends Liquid AI LFM2 variants at 350M and 1.2B, along with Gemma 1B.
The emphasis on compact models shows the broader idea behind the project. CrankGPT is not trying to compete with large data-center chatbots, but to show that simpler AI tasks can still be handled with far less compute.
Privacy is the core selling point
Running everything locally keeps control of the data and the interaction in the user’s hands. There is no need to hand conversations over to a cloud platform, which gives the concept a clear appeal for users concerned about privacy.
The approach also reduces dependence on the large-scale infrastructure that powers most modern chatbots. Squeez Labs has criticized the industry’s habit of spending huge amounts of compute on tasks that smaller models can already handle more efficiently.
There are clear trade-offs
CrankGPT is not a silent or frictionless experience. Demo video footage shows the device working in a way that feels more mechanical than a typical digital chatbot, which makes the interaction noticeably different.
A cooling fan is also installed to manage heat, underscoring that even a compact AI system still has thermal challenges when it runs locally. The company says the crank can become much harder to turn when speech synthesis and language-model inference happen at the same time.
That physical effort makes the device better suited to short everyday questions than longer or more demanding tasks. The makers have experimented with image generation, poetry, and code, but the system appears most practical for quick, simple exchanges.
Still a concept, not a retail product
For now, CrankGPT is not available for purchase. Its website does not list a wider launch or a price, and Squeez Labs has only opened requests for demonstrations.
That leaves the project closer to a public concept than a consumer device. Even so, it offers a striking reminder that AI does not have to mean giant servers and heavy energy use.
It also comes with an important limitation. While a hand-cranked device may lower operating impact during use, the training of AI models still carries a major burden on the planet’s resources.
CrankGPT therefore does not solve the environmental debate around AI, but it does show a different path forward. The concept suggests that AI can be smaller, more private, and, quite literally, powered by human effort.