Security & Data Practices

Some FocalNest features use a large language model, and we'd rather you read that here than find it in the privacy policy. Six things use one: drafting a project from a plain-language description, writing the small first step for a task, suggesting steps for a task when you ask, turning an email you forward into tasks and events, the one-line observation about your recent momentum that can appear on your Nest, and matching a task to a line in a Play mode adventure.

Those requests go to Anthropic's Claude API. They contain the text you supplied, plus a summary of your household's shape: member roles and age ranges, never names. Play mode matching sends only a task's title and description. Two others send more. Parsing a forwarded email also sends any photos or PDFs attached to it (photos are re-saved first without their location and camera details, and deleted from our side within a day), plus your existing project titles and identifiers, the titles of up to 15 recent open tasks, and the names of your family's places, so the result can be filed against something you already have. The momentum observation is written from your own recent task and routine titles. So while we never send a roster of who is in your family, titles you wrote yourself do leave with the request. Anthropic does not use API requests to train their models.

Nothing else in the product writes text with an LLM. Scheduling, task ranking, capacity calculations, and reminders are all ordinary deterministic code running on our own servers.

AI features are off for a child account until a parent turns them on. That is a separate decision from the consent that created the account. A parent grants it per child from the parental dashboard and can withdraw it at any time, after which every AI feature refuses that child again. And because your own text is sent as you wrote it, don't put sensitive personal, health, or financial details into an AI prompt.

Read-aloud uses ElevenLabs. When a task or reward shows up for a child who is still learning to read, its title is sent to ElevenLabs' text-to-speech service if those exact words have never been spoken on FocalNest before. The audio is saved and reused after that, and only the title's words are sent. For a child, it only happens once a parent has allowed read-aloud for them. That's its own permission, so a parent can allow it without turning on AI features. The name of the routine or event a kitchen screen counts down to is sent the same way, so the screen can say it out loud. Events synced from Google, iCal, or Outlook are never sent.

We also use ElevenLabs to generate narration and the background sounds in Play mode adventures, and OpenAI to generate collectible card art and ambient backgrounds. That happens ahead of time, from text and prompts we write. No customer data is involved.

Every API request is authenticated and scoped to a family. It is not possible to read another family's data through the API. Every query the API makes filters by family, and automated tests check that each endpoint enforces it.

Within a family, the system distinguishes between admin and member roles. Admins can manage family members and settings; members can manage their own tasks and see shared content. Children have a restricted view appropriate to their age.

Staff actions in our admin tools are logged. We don't look at your family's content as part of normal work.