> ## Documentation Index > Fetch the complete documentation index at: https://docs.polymet.ai/llms.txt > Use this file to discover all available pages before exploring further. # Build a dashboard with AI from a CSV file > Attach a CSV to a prompt and get a working dashboard built on its real numbers, with the exact prompt, what it cost and how to check the numbers are right. This guide turns a CSV export into a working dashboard you can publish on a live link. The example file holds eight weeks of daily orders for Fresh Bowl, a lunch delivery app. Use your own export; the steps are the same. ![A dashboard titled Lunch delivery overview with four stat cards for orders, revenue, average delivery time and repeat rate, a line chart of orders per day and a bar chart of revenue per week](https://mintcdn.com/polymet/DrFFR1dAltUlhtUR/images/docs/guides/dashboard-from-csv-result.webp?fit=max\&auto=format\&n=DrFFR1dAltUlhtUR\&q=85\&s=e15fb43199e082feaeebb28da58541ec) What you end with. Every number here comes from the attached file. * **Who it is for:** anyone with a data export who needs a dashboard other people can open. * **What it costs:** the run below used 58 credits and took about three minutes. Every organization gets 500 free credits every 30 days. * **What you need:** a CSV up to 500 KB. To follow along with ours, download [fresh-bowl-orders.csv](https://docs.polymet.ai/files/fresh-bowl-orders.csv): 56 rows and five columns (date, orders, revenue, average delivery minutes, repeat customers). ## 1. Attach the file On the projects page, click the file icon at the bottom left of the prompt input and choose your CSV. You can also drag the file onto the input, which shows **Drop files to attach**. A prompt takes up to 3 PDFs or text files, and each text file can be up to 500 KB. If your export is bigger, cut it to the rows and columns the dashboard needs. ## 2. Say what the dashboard shows Name each card and chart, the period it covers and what to compare it with. The agent reads the file, so use the words your columns use. ```text theme={null} Build a dashboard from the attached CSV. It holds eight weeks of daily numbers for Fresh Bowl, a lunch delivery app: orders, revenue, average delivery time and repeat customers. One page, for a laptop first: - Four stat cards for the last 28 days: orders, revenue, average delivery time, and repeat rate (repeat customers divided by orders). Each shows the change against the 28 days before. - A line chart of orders per day for all eight weeks. - A bar chart of revenue per week. - A table of the last 14 days with every column. Use the real numbers from the file, not sample data. Style: light and calm, white cards on a pale grey page, one deep green accent, tabular numbers. ``` Two lines matter most. "Repeat customers divided by orders" defines a number the file does not hold, so the agent does not have to guess. "Use the real numbers from the file, not sample data" is what stops it from inventing plausible figures. Send the prompt. Polymet creates a project and opens it on the canvas. ## 3. Check the numbers A dashboard that looks right and is wrong is worse than none. Before you share it, check two or three numbers against the file yourself. For this run, we added up the last 28 rows of the file: | Card | The dashboard says | The file says | | - | - | - | | Orders | 3,072 | 3,072 | | Revenue | \$39,894 | \$39,894.09 | | Repeat rate | 41.2% | 41.2% | | Avg delivery time | 23.1 min | 23.2 min as a plain average of the days | Three matched. The fourth was off by a tenth of a minute, and the reason is worth knowing: the agent weighted each day's delivery time by that day's orders, so busy days count for more. Weighted that way the file gives 23.1 too. Both are defensible. If you want one of them, say which in the prompt. ![A table of the most recent days with columns for date, orders, revenue, average delivery, repeat customers and repeat rate](https://mintcdn.com/polymet/DrFFR1dAltUlhtUR/images/docs/guides/dashboard-from-csv-table.webp?fit=max\&auto=format\&n=DrFFR1dAltUlhtUR\&q=85\&s=4a926096cc533437ba9c116844b9ba4a) The table under the charts shows the rows themselves, which makes spot checks quick. ## 4. Adjust it * **One element:** click **Select**, click it and write the change, for example "show this axis in weeks". * **The layout:** say it in a prompt: "Put the two charts side by side on wide screens and stacked on a phone." * **Other widths:** open the page in **Focus**, click **Breakpoints** and check **Mobile view** and **Tablet view**. ## 5. Publish Click **Publish** in the canvas header, then **Publish** in the popover. When it says **Published**, copy the live link. Anyone with the link can open the dashboard. A published dashboard shows the data that was in the file. It does not update when your data changes. To keep live data, [add a backend](https://docs.polymet.ai/backend/overview.md) and ask the agent to read from a table. ## You're done when * Two numbers you checked by hand match the file. * The live link opens in a private browser window. * The page is readable at **Mobile view** width. ## If it goes wrong * **The numbers look made up.** Send: "Recalculate every number from the attached CSV and show me the formula for each card." * **The input says "Attachments are too large to fit the model's context — remove one or use smaller files."** Cut the file down, or split it and attach the part you need. * **The Publish popover says "There are no prototypes yet".** Click **Create a prototype** and send the draft it puts in the prompt input. ## Next * [Prompt the agent](https://docs.polymet.ai/agent/prompt.md): everything you can attach, and the limits * [Compare options side by side](https://docs.polymet.ai/canvas/compare.md): try two chart layouts and pick one * [Add a backend](https://docs.polymet.ai/backend/overview.md): read the numbers from a table that changes * [Publish a prototype](https://docs.polymet.ai/publish/publish.md): update, rename or unpublish the live link This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.