New Free Tool Runs a Powerful AI on Your Laptop, No Internet Needed

A developer who goes by Niko1221 has released a free program called Strata that runs large AI models directly on your own laptop instead of on a company's servers. In his tests, a mid-range gaming laptop ran a capable open model at 51 tokens per second — about 38 words a second, roughly twice as fast as the standard software most people use. It also read through a 24,000-word document in about 20 seconds, where the usual setup would take more than five minutes.

Why that matters to you: this is AI you actually own. Nothing gets sent to a company's servers, there's no monthly bill, and it keeps working on a plane or when your Wi-Fi dies. That's a real difference if you handle sensitive things — contracts, medical notes, a client's private data — because those words never leave your machine.

The catch is that Strata is a one-person hobby project, and it's picky. It only runs on Nvidia graphics cards (AMD support is still experimental), works with just one family of model files from one research lab, and the laptop in the test had 64GB of memory and a 12GB graphics card — more than a typical office machine has. Expect rough edges.

Still, this points to where things are heading. A model that needed a data center six months ago now fits on a laptop you can buy at a store. Developers build these tools in their spare time, then big companies copy the good ideas. Today's enthusiast experiment tends to become next year's default setting — and the price of running a smart AI keeps falling toward zero.