Borrowed intelligence, on borrowed terms.
The assistant is powered by third-party models — today from Anthropic, OpenAI, and Google — reached through our own proxy. When you ask it something, the context needed to answer leaves your device for whichever provider serves the call; the rest of your archive stays local, the way the privacy page describes.
Because we ride on their models, their rules ride along with you. Your use of the assistant is also governed by the acceptable-use and usage policies of whichever provider answers — you agree not to do through Seriesly anything those policies forbid, including generating prohibited content or using the outputs to build a model that competes with them. Where our terms and a provider's disagree about what is allowed, read the stricter one as binding.
The output is generated, not authored. It can be wrong, dated, or confidently mistaken, and it is not legal, financial, medical, or professional advice. Treat it as a fast first draft you remain responsible for checking — especially before you act on a summary, a decision record, or anything the app drafts in your name.