RoundUp’s AI turns a teacher’s spoken note into daily-report entries, a rough thought into a message home, and a center message into a family’s own language. Names are replaced before a request leaves RoundUp, and a director can switch AI off for the whole center.
A teacher speaks or types one sentence after nap, and Quick Log splits it into entries for each child. It’s told never to invent children, times or foods, and it flags a name that isn’t in the room instead of guessing.
“Maya and Theo napped from 12:30 to 2. Maya ate all her lunch, Theo had about half.”
Nothing is logged until the teacher confirms. Illustrative example.
AI drafts land in an editable field, and nothing a draft says is saved or sent until a person reviews it. AI is part of Base and Starter, so every plan includes it.
RoundUp’s AI uses Claude Haiku, a model made by Anthropic, through Amazon Bedrock in US AWS regions. Every request takes this path.
The assistant and Smart Search send only what a question needs. “Who’s here?” doesn’t send anyone’s allergies.
Our server swaps children’s names for tokens like [[NAME_1]], and staff and family names too. If children’s names can’t be checked, the request isn’t sent.
Requests are never used to train AI models, by RoundUp or by Amazon Bedrock.
RoundUp puts the names back on our side and shows you the draft. Assistant conversations aren’t saved.
Other details, like a child’s age, allergies, milestones or the text of an incident note, are sent when the task needs them. If a staff or family name can’t be looked up at that moment, it may go through; a child’s name never does. Our privacy policy lists what each service provider receives.
AI is on by default for US centers. For Canadian centers it starts off, and a child is included only if a parent says yes.
Turn AI features off under Settings → Configuration. The assistant, Quick Log, drafting help and message translation stop for your center, and RoundUp refuses AI requests for it. If it can't check the setting, it refuses them too.
You can turn it back on at any time.
A parent can turn off AI for their child in the app, under My child’s data. The assistant and Smart Search then leave that child’s details out, and no AI answer shows the child’s name.
Other features, like Quick Log, may still send a note about the child with the name replaced. To stop AI entirely, ask your center to switch it off.
No. Photos are never sent to the model. A suggested caption is written without seeing the photo, so check that it fits.
Not each one, because translation happens as messages are sent. That’s why each is labelled with the original one tap away. If translation fails, the family gets the original.
Yes. Treat an assistant answer as a lead to check on the screen that holds the real data, and never use it to decide whether a room is in ratio.
Yes, for one job: reading printed text. When someone at the center scans a paper sign-in sheet, a check or a printed menu, Amazon Textract reads the text so nobody retypes it. It runs only on a scan, and the center’s AI switch doesn’t cover it.
RoundUp opens to all centers on January 1, 2027. Until then, a design-partner program of fewer than ten Oregon centers gets RoundUp free through 2026. Bring a real morning’s notes to a demo and watch Quick Log turn them into entries.