Asking Questions (Query)

Query (/app/query) lets you ask about household records in everyday English instead of stacking filters.

Nobleway uses AI to understand what you ask. The ledger and calculations stay deterministic: totals always come from saved transactions, not from the model guessing a number.

Your household shares a daily Nobleway AI allowance for Capture and Ask. Every member uses the same allowance. Filter-only queries do not use it. If you use your own AI key, that work does not use the household allowance. The allowance resets at UTC midnight. Your own key can still be limited by burst protection and by your provider.

How It Works: The Five-Step Pipeline

The AI does not add up the numbers. Language models can misread a question or invent a total. Nobleway uses AI only to understand the words. The database does the arithmetic.

1. You ask, for example: How much did we spend on dining in August?
2. Nobleway translates that into filters (expense, Dining, August dates).
3. The ledger calculates the exact sum from saved transactions.
4. Evidence is listed: the total plus each supporting row.
5. A short explanation is written from those calculated facts only.

What You Can Ask

Category spending

• "How much did we spend on groceries last month?"
• "What was our total spending on utilities this year?"
• "Show all dining expenses in July."

People

• "How much did Sarah spend on transport this month?"
• "Compare expenses by person for August."
• "What did Alex spend on groceries last week?"

Accounts

• "What came out of Main Checking in the last 30 days?"
• "Show all transfers into Emergency Savings."
• "List expenses paid with Sapphire Card in September."

Amounts

• "Show all expenses greater than 200 this year."
• "List transactions between 50 and 100 on groceries."

Understanding Query Results

The primary answer appears at the top (for example, $482.50).

• If you asked to group (for example, "by category" or "by person"), you get each group’s total, count, and share.

Under the summary are the transaction rows that produced the total.

• Open any row to see details, history, or corrections.
• If a total looks wrong, check which purchases were included.

The short narrative uses the date range, matching row count, and highlights from the calculated result. It does not invent extra figures.

The Structured Query Builder (Fallback)

If you prefer filters, or you want to tweak what the question produced:

1. On Query, open Query Builder.
2. Adjust:
   • Aggregation: Sum, count, or average.
   • Group by: None, category, account, or person.
   • Date / month.
   • Type: Expense, income, or transfer.
   • Accounts & categories.
3. Click Run query.

The builder uses the same results and evidence table, without calling the language model.

Best Practices & Limitations

• Use the names you already set: Checking is clearer than bank if you have accounts named Checking and Savings.
• Read the filters and evidence before acting on a big decision. The math is from the ledger; the interpretation of dates or category words can still miss an ambiguous phrase.
• Historical totals only: Query does not forecast markets, estimate tax bills, or run hypothetical budgets.

Next Steps

• Record new activity in everyday language: Natural-Language Capture.
• Review how balances are calculated: Accounts & Opening Balances.
• Review the transactions behind an answer: Recording Transactions.