Uptime Industries just launched Lemony AI, a pocket-sized “AI-in-a-box” device that runs large language models locally. The gadget is about the size of a sandwich and needs only 65 watts of power — roughly what it takes to charge a laptop.
Each Lemony node can handle LLMs up to 75 billion parameters. It supports open-source models and adapted closed ones. Users can link multiple Lemony units to create AI clusters, each running different models.
The startup is teaming up with IBM and JetBrains. These partnerships give customers easy access to AI models, including IBM’s proprietary ones. 
Uptime’s CEO Sascha Buehrle shared with TechCrunch the origin story: co-founders first tested AI on Raspberry Pis. Realizing on-prem AI could boost adoption among privacy-conscious companies, they built Lemony small and scalable.
“We need to build something small which can go easily into the teams, and which [does] not require any organization-wide decision making — bringing on-premise generative AI solutions into the business teams, basically,” Buehrle said.
Clusters of small and powerful devices can help the system grow according to a customer’s needs, he added.
Demand is strong from regulated sectors like finance, health, and law.
“Everything stays in your box,” Buehrle said.
“So your documents, your files, your emails — the models are hosted in the box, the agents are running locally in the box, and nothing is leaving the box.”
Uptime closed a $2 million seed round led by True Ventures, with JetBrains and others joining. Funds will speed up device development.
The company also plans to bring its Lemony OS software to third-party hardware like Nvidia’s DGX Spark. They aim to expand usage from single users to teams.
Lemony AI costs $499 per month and supports up to five users.
