Trust
Honest limitations
Running AI on your own machines has real benefits and real limits. Here are the limits, stated plainly.
Local models are smaller
Models that run on a laptop or a modest server are slower and less capable than the largest models offered by cloud AI companies. They are good at many everyday tasks, not all of them.
Speed depends on your hardware
More memory and a GPU mean larger and faster models. A small CPU-only server can complete small tasks but will feel slow with large models.
Private is not the same as offline
A server you reach over the internet keeps your data under your control, but it is online. Only a machine with no network connection is offline.
"No token quota" is not infinite capacity
We never meter your usage, but your machine can only handle so many requests at once. Heavy use needs bigger hardware.
Limited systems at first
At launch the server supports Ubuntu 22.04 and 24.04 on x86_64. Other systems come later, and we will list them here when they are tested.
Model licences vary
Each model has its own licence, and some restrict commercial use. We show the licence before every download; it is your responsibility to follow it.
Benchmarks are coming