Visualizations
An answer is words first — a plain sentence computed from your vault. When the shape of that answer suits it, millfolio draws a picture below the sentence: a headline number, a table, a chart, or a map. You never ask for a chart type. The model computes the exact figures and emits them with a structure; the client picks the view from that structure.
How the view is chosen
Each question is answered by a small program the model writes and runs in a sandbox over your real data (see Chat). Alongside its text sentence, that program may emit typed result data — a single number, a labeled table, one or more ordered series, a share-of-whole split, or a set of geographic places. The app reads the shape of that data and renders the matching view. A single figure becomes a KPI tile; a per-month breakdown becomes a line chart; a per-category one becomes a bar chart; and so on. The words always arrive first, and the visual — if any — appears beneath them.
The result types
Text answer
The default. A plain, exact sentence with no visual — a lookup, a date, a yes/no, a single fact read from a document. Most questions that aren't about totals or trends land here.
When does my car insurance renew?
KPI tile — a big number
One headline figure, shown as a labeled tile: a total, a count, a maximum. A question with a single-number answer produces a KPI. Several headline figures produce several tiles side by side.
How much did I spend last month?
Table
A ranked or labeled list — several rows, one or more columns, dollar columns right-aligned. Top merchants, spend per tag, a breakdown too long or too flat for a chart. The table is also the graceful fallback for any data that doesn't fit a chart.
What are my top merchants by total spend?
Line chart — a trend over time
An ordered breakdown whose x-axis is a date: spending per month, a running trend. Ask for a per-merchant or per-category breakdown across several months and each series is drawn as its own line.
Show my monthly spending per merchant over the last 6 months, with line graphs for the top 10.
Bar chart — a breakdown by category
An ordered breakdown whose x-axis is a set of labels rather than dates: spend per category, per tag, per merchant at a single point in time.
How much did I spend in each category?
Grouped bar chart — two categories side by side
When an answer has two category breakdowns that share the same labels — this month versus last month, debits versus credits per category — they're drawn as a grouped (side-by-side) bar chart.
Compare my spending per category this month versus last month.
Pie chart — share of a whole
A split of one total across a small number of named parts, when the question is about the share, fraction, or percentage each part takes. The client computes the percentages and draws the pie. Many parts fall back to a table instead.
What share of my spending is each category?
Map — by state
A geographic breakdown drawn on a US map, from the location stamped on each transaction at index time. Spending grouped by US state renders as bubbles across the states.
Which states did I spend the most in?
Map — by country
The same geographic view at country level, on a world map — useful for travel and foreign charges. Transactions with no country on the descriptor (transfers, online charges, PayPal) simply don't appear.
How much did I spend abroad, by country?
Map — by city
The finest geographic view, placing a bubble at each city. This one is US-only: a city is mapped by the zip parsed from its transactions, so a city seen only without a zip falls back to a table row rather than a bubble.
Show a map of my spending by city.
Locations come from your statements. The state, country, city, and zip behind every map are parsed from each transaction's descriptor when your vault is indexed — nothing is sent anywhere. Descriptors that carry no location (transfers, PayPal, many online charges) are left off the map, not guessed.