For twenty years the bargain has been the same. Software gets better
when it knows you, knowing you requires your data, and handing over your
data meant handing over control of it. So your mail, your calendar, your
documents and your photographs went to companies who wrote the rules for
what happened next — and kept the right to rewrite them.
AI does not extend that bargain so much as complete it. To be genuinely
useful, an assistant needs the whole picture: what you wrote, who you
answer quickly, what you keep putting off, what you said in a meeting
last spring. Handing that over produces the most complete portrait of a
person that has ever existed outside their own head — and under the
usual arrangement, it becomes raw material for someone else’s
product.
We think the premise is wrong, not the technology.
Holding someone’s data does not have to mean owning it. What
happens to it is a choice, and we would rather be the ones able to
make that choice than leave it with a provider we rent from.
So we are building toward that, in the open. The models are
open-source ones anyone can inspect, moving onto our own hardware as
suum-exse grows, while we train models of our own from the smallest
step up, where the Community can watch.