How it works
Your data, described. Not invented.
It describes your data. It does not fabricate it.
Every chart in a Plotwell dashboard is built by aggregating your actual rows in code. If the dashboard shows that your Q3 revenue was $84,000, that figure came from summing or averaging your cells, not from a model guessing. The model's only job on the numbers side is to look at the column names and aggregated results and decide which chart types fit.
Every insight and takeaway is marked as generated, because it was produced by a language model reading the computed results. You can see which charts it is describing.
Why it cannot fabricate numbers
The model receives the column names and the aggregated results. It does not receive your raw rows, it does not run its own calculations, and it has no way to produce a figure that is not already in the aggregated output. The architecture makes fabrication structurally impossible at the chart level.
For takeaways, Plotwell runs a verification step: any takeaway that cites a specific figure is checked against the aggregated data before you see it. If a figure in the takeaway cannot be traced to the computed results, that takeaway is dropped. You never see a claim Plotwell cannot back up with the numbers on screen.
What the model does
- Reads your column names and infers which are dates, categories and measures.
- Looks at the aggregated output and selects which chart types suit the data structure.
- Writes the takeaways in plain language: what went up, what looks off, what to look at next.
- Is checked against the aggregated results before any takeaway with a number reaches you.
What the model does not do
- Read your raw rows.
- Calculate any figure in any chart.
- Produce a takeaway citing a number that is not in the computed output.
- Make recommendations about decisions, investments or plans.
Plotwell does and does not
| Plotwell does | Plotwell does not |
|---|---|
| Aggregate your rows in code to produce every number in every chart | Invent or estimate numbers that are not in your data |
| Let the model choose which chart types fit your columns | Let the model calculate or supply any figure |
| Verify takeaways against computed results before showing them | Show a takeaway citing a figure that cannot be traced to your data |
| Mark every chart and takeaway as generated | Present generated text as editorial fact |
| Process your data only to build your dashboard | Use your data to train or fine-tune any model |
| Let you export and delete your dashboards at any time | Sell your data to anyone |
Your data is yours, and we treat it that way
Your uploaded file is parsed and aggregated in memory, then discarded. We do not keep the original. The aggregated data behind your dashboard is stored in our database against your account so you can open the dashboard again.
Your data is never used to train a model, never used to improve prompts without your knowledge, and never sold or shared with anyone other than the model provider that picks chart types and writes the takeaways for your dashboard.
Exporting and deleting
You can export the aggregated data behind any dashboard as a CSV from the dashboard page on Pro and Team plans. You can delete any dashboard from its page. You can delete your whole account from Settings, which removes every dashboard and the account record.
Both are immediate. No support ticket needed.
For the technical detail on how the model routing works, see How the AI works. For what happens to your file and who can read it, see Security. For data you want to export or delete, see Your data.