AI & Privacy · 27 June 2026 · 4 min read
What “runs on your own computers” actually means
“On-device” and “on-premise” get used loosely. Here is what they really mean for an AI assistant — and why the distinction decides who can see your data.
Vendors describe their products as local, private, on-device, or on-premise almost interchangeably. For an AI assistant the words carry real weight, because they decide where the actual thinking happens — and therefore who can see the material it thinks about.
Where the model lives
An AI assistant has two halves: the interface you click, and the model that does the reasoning. Plenty of “apps” are just a thin interface wrapped around a model running in someone else's cloud. Your words still travel; only the buttons are local.
Genuinely local means the model itself runs on your machine. The reasoning happens on the same computer as the document, so nothing has to be transmitted to get an answer.
On-device vs on-premise
On-device means each user's own computer does the work — good for individuals and small teams, with nothing to administer. On-premise means the organization runs the model on a server it controls, and staff connect to it over the internal network — better when you want central models and shared administration.
Sanad supports both: activate directly on a device, or host the model on your own server and let desktop clients check out a seat over the LAN. Either way, the data stays inside your walls.
The test that cuts through the words
When a label is unclear, one test settles it: disconnect from the internet and try to use the assistant. If it keeps answering questions about your documents, the intelligence is genuinely on your side. If it stops, it never was.