What is an AI family assistant? Boundaries, privacy and examples

What does an AI family assistant do?

An AI family assistant has one job: take the unstructured paperwork of family life (an email from school, a PDF schedule, a photo of a flyer on the fridge) and turn it into something the calendar already understands, without a parent doing the data entry by hand. That translation step is the whole value of the thing. A school office sends a PDF with eleven dates buried across three paragraphs, not eleven calendar invites, and someone still has to read it, work out which dates matter to their child, and type each one in by hand. An assistant that reads the PDF and proposes the events removes that chore because it’s faster than a parent at the mechanical part: parsing text, matching a date to the right person and filing the result where the household can already see it. It’s a narrow tool with a narrow job, not a general-purpose AI you chat with for its own sake.

What can you hand it?

You can hand it three kinds of input, each with its own path into the household’s calendar, tasks and lists: things you forward, things you share in from another app, and things you ask it directly. Every household gets its own private email address, set once in Settings → General; forwarding an invite, a newsletter, a PDF schedule or a photo of a flyer to that address gets it read and filed under the right person and time, attachments included, without touching anyone’s personal inbox (see how the family inbox works). Sharing a photo, PDF or link from another app’s share sheet opens Magic Import instead: add a short instruction (“assign to Vera”, “this recipe is in Swedish”), pick which member and which kind of item it should become, and it processes in the background, finishing with a toast and a push notification rather than a spinner you wait out (see how Magic Import works). Asking it directly in chat is the third path: “what’s for dinner this week?” or “is everyone free Saturday?” gets answered from the household’s own calendar, meals and lists on the spot, nothing new created (see meeting Plannie Assistant). A recipe link is a small special case: it feeds the weekly meal-planning system as a planned meal, not the calendar directly, and its ingredients can flow on into a shared grocery list.

What does it create, and how is it attributed?

It creates the same four kinds of object a person could create by hand, events, tasks, list items and planned meals, never anything a family member couldn’t add themselves. What differs by input is where the result lands and how it’s tied back to what triggered it:

What you send it	What it creates	Where it lands	How it’s attributed
Forwarded school email	One event per date mentioned (concert, half day, deadline)	The family calendar, under the right child	Sparkle badge on the event, activity feed entry naming the sender
Photo of a PDF term calendar	One event per listed date, created as a batch	The family calendar	Sparkle badge, one activity feed entry per import
Photo of a flyer on the fridge	A single event with whatever date and time it can read	The family calendar	Sparkle badge, activity feed entry
Recipe link	A planned meal with a photo and ingredient list	The weekly meal plan, then optionally the grocery list	Sparkle badge, activity feed entry
A chat question (“is everyone free Saturday?”)	Nothing new, just an answer drawn from existing data	Nowhere, it’s read only	Nothing to attribute, no item was created

It also resolves the obvious kind of clash, two events with the same title landing on the same afternoon, and can re-plan a week’s meals around one swapped-out dinner, but it always works inside the household’s existing calendar and lists rather than replacing them.

What can it never see?

It can never see anything you haven’t explicitly forwarded, shared, or granted through a connected account. It has no standing access to a parent’s personal email, photos or files, and it doesn’t monitor an inbox waiting for something relevant to appear; it only acts on the specific message, photo or question it was actually given. Connected calendars (Google, Outlook, an ICS feed) and any agent API key are the one category of standing access it can have, and even those are opt-in grants made explicitly in Settings. Revoking one cuts off what it can see immediately, not after the next sync cycle, so a family that changes its mind isn’t stuck waiting it out.

How do you check what it did?

You check it the same way you’d check anything another family member did: by looking at the activity feed. Every event, task, list item or planned meal it creates carries a small sparkle badge on the item itself, plus a named entry in the household’s shared activity feed, so a parent scrolling through the week sees “Assistant added: School concert, Thu 6pm” next to entries a person added by hand, not folded away in a separate log nobody opens. That visibility is deliberate: an assistant that can write to a family calendar only stays trustworthy if what it did is as visible, and as easy to undo, as what a person did.

What happens when it isn’t sure?

When it can’t confidently resolve something, it leaves the decision for a person instead of guessing. A flyer photo with a smudged or half-cropped date becomes a task asking someone to confirm the date, not a wrong event on the calendar; an email that doesn’t make clear which child it’s about goes in as unassigned rather than a guess that’s right half the time. The trade-off is on purpose: a slightly less automatic result that a parent still has to glance at beats a fully automatic one that’s sometimes wrong without anyone noticing.

How is this different from a general chatbot?

A general chatbot answers from its training data and whatever you paste into the conversation, and none of that is connected to your family’s actual calendar unless you copy the result over by hand. This kind of assistant is grounded in data the household already controls, its own calendar, meals, lists and whatever’s been forwarded or shared with it, and what it produces is a structured object with a name and a badge attached, not prose you still have to turn into an event later. A general chatbot gives you an answer to type in somewhere. A family assistant does the typing.

What can a developer wire into it?

Beyond the built-in Assistant, developers can connect their own agents to Plannie over the open REST API or the Model Context Protocol (MCP), an open standard for connecting external tools and data sources to AI systems that Plannie’s own developer docs build on (modelcontextprotocol.io/specification). Both paths use the same scoped, attributed model as the built-in Assistant: a key created in Settings → Integrations, limited to exactly the scopes it needs, with every item it creates carrying its own sparkle badge and activity feed entry back to that key, not a generic “Assistant” label. A revoked key stops working immediately, the same way disconnecting a calendar does, so a household controls what a connected agent can reach and doesn’t have to rely on it behaving well.