Skip to content

Example conversations

These examples show the agent in use. The replies are illustrative; the point is what tools the agent reaches for and how the guardrails shape the outcome.

You: How much did I spend on groceries last month?

The agent uses its clock tool to resolve “last month” to a date range, then its read-only query tool to total your grocery-category expenses in that range, scoped to your account. The calculator handles the sum so the figure is exact.

Agent: Last month you spent 312.40 on groceries across 18 transactions. That is down from 358.10 the month before.

You: Add a 12.50 coffee expense to my cash wallet.

If writes are enabled, the agent creates a transaction owned by you, with the amount, the expense type, your cash wallet, and a category it infers. It confirms what it did.

Agent: Added a 12.50 expense to your Cash wallet, categorised as Coffee, dated today.

If writes are disabled on the instance, the same prompt gets a different answer:

Agent: I can see your data but I am not able to make changes on this instance, so I cannot add that transaction. You can add it from the dashboard.

You: Show me Sarah’s transactions for this month.

The query tool is owner-scoped. It only ever reads your records, so there is no way to fetch another person’s data.

Agent: I can only access your own data, so I cannot show another person’s transactions.

You: What is my savings rate this month, and which category grew the most versus last month?

The agent resolves both date ranges with the clock tool, reads income and expenses for each, computes the savings rate with the calculator, compares category totals across the two months, and explains the result. Every read stays scoped to you.

The agent is only as capable as the tools it is given, and the tools are deliberately narrow and owner-scoped. That is what lets it act on your money without acting beyond it. See Permissions and safety.