AI CSV Analysis Without Uploading Your File

What happens to my file when I open it?

AskCSV analyzes CSV and TSV files inside your browser: you open a file, ask questions in plain English, and get answers with charts. "Privately" is worth spelling out precisely, so here is the whole path a file takes.

1. Open. You select a CSV or Excel file, or try a sample. The file is read in this tab with DuckDB compiled to WebAssembly; it is not sent to AskCSV.
2. Storage. It is kept in this site's storage on your device, separated by account (or by guest session if you have not signed in). It is not published, shared, or listed anywhere, and AskCSV has no server copy.
3. Question. When you ask, the browser sends AskCSV's server your question, the name and row count of each file or sheet in the chat, the names and types of the columns you chose to share, up to 10 sample rows from them, and the earlier questions, answers and SQL in the chat. The server passes that to the AI provider, which writes one read-only SQL query.
4. Query. The browser runs that SQL over the local file with DuckDB. Read-only means a query can select, filter, group, and aggregate, but cannot modify or delete your data, read other files, or reach the network.
5. Result and explanation. The totals, counts, shares, and trends in an answer come from that local query result and are computed deterministically in code. The result — up to its first 100 rows, from the columns you share — goes to the AI with those figures for the explanation, and the chart is built from the same result in the browser.
6. Follow-ups. You can keep asking in the same chat. Each question refines the analysis against the same local file, and the conversation stays in this browser.

What does the AI provider actually see?

AskCSV uses OpenAI as its AI provider (a sub-processor) and keeps what it sees to a minimum. To answer one question, OpenAI gets only:

• your question in plain English;
• the name and row count of each file or sheet in the chat;
• the names and types of the columns you chose to share;
• up to 10 sample rows from those columns, so the model can tell a date column from a currency column;
• the earlier questions, answers and SQL in that chat;
• the result of the query your browser ran — up to its first 100 rows (16 KB), from shared columns only — and short facts computed from the whole result: totals, counts, shares, and its highest and lowest rows.

Never sent:

• the file itself, or any row that is neither in the sample nor returned by a query in this chat;
• columns you turn off — they are left out of the sample and the results, and earlier answers that used them are left out of the chat history;
• error text from your browser — if a query fails, the AI is told only the kind of error.

A result can hold real rows: ask for the five largest orders and those five rows, in the columns you share, go to the AI. These limits are part of a fixed request format that both your browser and AskCSV's server check, and the server refuses a request that goes over them.

Before your first question you can review the columns and see the exact sample rows that will be sent. Columns that look sensitive — email, phone, identifiers, long text and secrets — are off by default. If you share no columns, only your questions, the file names and row counts, and earlier messages go.

Where does each part of my data go?

The privacy policy lists the sub-processors and describes each of these in more detail.

How long is my data kept?

Signing in does not upload the files already open in this browser, and it does not make them available on another device.

Does AskCSV connect to my database?

No. AskCSV works on files only. You open a CSV or Excel file; there is no direct database connection, no warehouse credential, and no live sync.

How do I check that the numbers are right?

Every chart shows the SQL that produced it, which makes an answer auditable instead of a black box:

• The generated SQL is visible, so you can compare what was calculated with what you meant to ask.
• Totals, shares, and trends are computed in code from the query result, so the arithmetic is not a language model's guess.
• The AI's job is to explain the result, not to produce the figures.

AI answers can still be wrong — most often because the generated query answers a slightly different question than the one you had in mind, for example summing gross revenue when you wanted net. Reading the SQL is the fastest way to catch that.

Frequently asked questions

Can I try it without an account?

Yes. As a guest you get $0.25 of credits (about 5 answers), and no sign-in is needed to open a file or try a sample. Credits are charged only for tokens the AI actually processes, so a request refused before the AI runs costs nothing.

What do I get with a free account?

Signing in with Google ties your credits and subscription to an account. A $5/month subscription includes $5 of usage every month (about 100 answers); extra credits start at $10 (about 200 answers). Signing in does not upload the files already open in this browser.

Which files can I use?

CSV, TSV, delimited TXT, and Excel .xlsx (every sheet is read). Only older .xls files need to be saved as .xlsx or CSV first. How large a file can be handled depends on the browser and device.

Does it work on my phone?

Yes. Files open and queries run in any modern browser, including phones and tablets. There is nothing to install, no model to download and no WebGPU requirement. An internet connection is needed for a new AI answer.

Technical references

• DuckDB is the SQL engine AskCSV runs read-only queries with, compiled to WebAssembly inside your browser.

Start with the sample or open a file you are authorized to analyze, and check the SQL behind the first chart yourself. For a hand-checkable example with known totals, see the six-row summary exercise.