Paste a table from Excel or use the sample, and write your question. The model only decides what to compute; code computes every figure over every row, and the steps are written beside the answer.
| ماه | منطقه | گروه کالا | تعداد سفارش | مبلغ فروش (ریال) | تخفیف (ریال) |
|---|---|---|---|---|---|
| فروردین | تهران و البرز | لبنیات | ۲۱۰ | ۳۱۵٬۴۰۰٬۰۰۰ | ۱۲٬۷۶۰٬۰۰۰ |
| فروردین | تهران و البرز | روغن | ۱۹۵ | ۵۰۹٬۵۴۰٬۰۰۰ | ۱۲٬۵۶۰٬۰۰۰ |
| فروردین | تهران و البرز | برنج | ۱۱۴ | ۶۷۵٬۰۴۰٬۰۰۰ | ۲۴٬۰۵۰٬۰۰۰ |
| فروردین | تهران و البرز | حبوبات | ۱۰۸ | ۲۰۲٬۶۱۰٬۰۰۰ | ۳٬۸۷۰٬۰۰۰ |
| فروردین | تهران و البرز | نوشیدنی | ۲۲۴ | ۲۲۱٬۲۴۰٬۰۰۰ | ۲٬۶۸۰٬۰۰۰ |
| فروردین | شمال‌غرب | لبنیات | ۱۰۹ | ۱۵۱٬۵۳۰٬۰۰۰ | ۵٬۸۰۰٬۰۰۰ |
Total sales by region over the six months
| منطقه | Total sales (rial) |
|---|---|
| تهران و البرز | 15,593,130,000 |
| شمال‌غرب | 6,878,040,000 |
| شرق | 5,938,630,000 |
| جنوب | 5,870,700,000 |
| غرب | 4,781,010,000 |
These figures were computed by code over all 150 rows, not by the model; the model saw the column names and text values, never the numbers. Example question; its plan was written beforehand.
Contact us for dashboards and questions over your own company’s data
Copy a table from Excel or Google Sheets and paste it, or give it a CSV file. The table stays in your browser.
The model sees only the column names and the text columns’ values, and writes a query plan: which rows, grouped by what, which sum, average or count.
Code runs the plan over every row and builds the table and the chart. The steps are written beside the answer, so you can repeat them in Excel.
A language model makes mistakes adding up hundreds of rows, and can state a number it made up with confidence. Here the model only says what to compute, and code computes every figure from the real rows; if it misread the question, the written steps show it.
Questions answered by a sum, average, count, lowest or highest, with conditions on columns, grouped by one or two columns, sorted, and “the top few”. Forecasts, custom formulas and joining tables are not in this trial.
In data analytics projects we build the same approach over an organisation’s accounting, sales and inventory data: read-only access, one definition per metric, and beside every answer the query that ran and the source of each figure.
Yes. The example questions are unlimited, and each internet address can ask up to 10 questions of its own an hour and 30 a day.
The rows and numbers stay in your browser. For each question only the column names, each column’s kind and the values of text columns go to an AI service outside Iran: all of them when a column has fewer than 30 distinct values, so conditions are written correctly, otherwise a few examples. No value from a number column is sent.
The computation steps sit beside every answer; if a condition or grouping is wrong, ask again using the exact column names. If a needed column is missing, it says these columns cannot answer it.
Not yet. Select and copy the rows in Excel and paste them here, or save the file as CSV. Up to 20,000 rows and 40 columns.
Yes. Persian and Latin digits, «٬» or “,” between thousands and «٫» or “.” as the decimal point are all read as numbers.
We build the same approach over your database, with read-only access, on your own infrastructure. A discovery call settles which decisions and metrics it should answer.