AI assistance and privacy usually pull in opposite directions: the help comes from sending your text somewhere. Writano’s approach is device-first — run the model where the document already is.
What runs in your browser
The clearest example is this documentation’s own assistant: Ask AI downloads its models to your browser once, and every question is searched and answered on your device. Nothing you type leaves it — that’s not a policy promise, it’s how the feature is built; there is no server to send the question to.
The same principle drives Figure Studio’s code modes: your Python, R, and other figure scripts execute in your browser, not on a remote runtime.
What you control
AI features in the editor are opt-in and configurable — the assistant has model settings, and features that would download a model to your device say so before they do (with the size), so heavy downloads are a choice, not a surprise.
Never training data
Your documents, questions, and figures are not used to train models — Writano’s or anyone else’s. The models ship to you; your text doesn’t ship to them.
Honest labelling
Where AI composes text (for example, the optional composed answers in Ask AI), the output is labelled as AI-written and shown with the sources it drew from, so you can verify rather than trust. Anything unlabelled is either your own writing or quoted documentation.