Jarso's AI. Safe by design.

It is an AI, and it says so

Jarso's AI helper is built on Claude, made by Anthropic. It is a computer program, not a person, and it can get things wrong.

Before a child's first message it shows a card that says so, and every reply carries a "Jarso AI" label. It goes by the app's name, Jarso, but it has no feelings and no life outside the app, and it never acts as a friend.

On by default, and a parent can turn it off

Jarso's AI is on for each child unless their parent turns it off. When a parent adds a child, the form says so and explains that the child's messages, and the child's own money and learning details Jarso looks up to answer them, are sent to Anthropic to get answers.

A parent can turn it off at any time from the child's page, and back on again. The child is told, before their first message, that their parent can see their chats.

What it helps with

Jarso's money topics: spending within your balance, splitting an allowance between Spend and Save, saving toward a goal, tradeoffs, and changing a plan. It also answers basic money questions in plain words.

In a mission it helps a child think: a guiding question first, then hints. It gives an answer only after the child has tried twice.

What it can see

When a child asks about their own money or learning, Jarso looks it up: their Spend and Save jars, their allowance, the money they received and the purchases they recorded, their savings goal, their requests to take money out of Save or put it in, the money rules their parent set, and their learning progress. It looks only when asked, only at that child's own details, and only what the question needs.

It cannot see other children's money, a parent's planned gifts or anything outside Jarso, and it cannot move money or change anything.

Figures about a child's own money come from the app, not from the AI: the AI writes a placeholder and the app fills in the real amount, and a reply with a figure the app didn't look up is held back. The parent sees what Jarso looked at for every reply.

What it never does

It never asks for personal or account details. It never recommends products, brands, shops, apps or investments, and never tells a child to buy anything. It never judges how much money a child or family has, and never compares it with anyone else's.

Questions about other things get a short, friendly redirect to a money topic.

Three checks on every message

Contact details such as emails, phone numbers and links are removed from a message before it is stored or sent.

A separate safety check reads each message first. Off-topic, personal, unsafe or rule-breaking messages get a fixed, reviewed reply instead of the AI's answer.

Another check reads every AI reply before the child sees it, and holds it back if it breaks any of the rules above. There are limits on how many messages a child can send.

If a child seems unsafe

If a message suggests a child may be hurt, scared or at risk, Jarso does not answer with the AI or look anything up. It shows a kind, fixed message that encourages the child to talk to an adult they trust, lists free helplines (below), and says to call the local emergency number if they are in danger right now.

The chat is flagged for the parent. The parent sees that a safety response was shown, but not the child's message: it is never stored, so a child can reach out safely. If the safety check can't run, a simple keyword check still shows the helplines.

What is kept, and for how long

Chats are stored with the child's account so their parent can read them. Message text is deleted after 30 days. What remains is a record of when a message was sent, how it was handled, which kinds of thing Jarso looked at, and how many words it took, used for safety and cost.

Messages are sent to Anthropic to produce answers and are handled under Anthropic's commercial terms. Nothing is used for advertising.

Helplines Jarso shows

Everywhere else: Find A Helpline, free, confidential support in your country.