With a rented model, every question and document goes to someone else's servers, for good. Here you choose: keep everything on your own machine, or let us finetune it for you - and even then, we keep nothing.
The difference isn't a setting. It's who ends up holding your data.
Private by default. Honest about where the data goes in each path.
Send the data, we finetune the model on our infra, hand it back, and delete the data. We never keep it and never reuse it.
You get a model you run on-prem: it and your data stay on your hardware. Nothing leaves, works offline - the honest fit for regulated data.
A public library of open models and self-serve training are on the way. For now, we build your model together with you.
For the work where data legally can't leave, the local option isn't a nicety - it's the only way to use AI at all.
Patient records, case files, financial data. Train and run on your own hardware and you can use AI without breaking the rules that bind you.
A clinic, a ship, a remote site. Once the model is on the machine, it runs fully offline.
On the local path, your conversations never reach anyone. On the hosted path, we delete your data after training - either way it isn't mined.
Plain promises, not fine print.
Private by default. Your data and the resulting model are yours. Nothing is public - a public library will be strictly opt-in when it launches.
When we build it, we keep nothing. Data sent for hosted training is used only to build your model, then deleted. It never trains anything else.
Runs on your hardware. You receive a model you run on-prem: prompts, answers, and inference all stay on your own infrastructure.
You own the result. The model is a file you keep and run - online or off, on your terms.
For the formal data-protection terms, see our Privacy Policy.
We finetune it from your data, hand it back, and keep nothing. You own the model and run it on your own infrastructure.
See how it works →