LogoHow To CSV

Data Chat

Upload a dataset to begin

Supported formats: .csv,.xlsx,.xls,.xlsm,.xlsb,.tsv

Free users: Up to 1MB per file.
Sign in to upload files up to 10MB.
Upgrade for even larger files.

AboutData Chat

Ask questions about your dataset in plain English (or Italian, French, Spanish...) and get trends, statistics, charts and outliers back instantly — matched locally, no cloud AI, no API key.Category: Analysis & Visualization
Asking "what is the trend of sales over time" without writing a formulaFinding outliers by describing them in plain EnglishGetting a percentage breakdown by category on demand
The Data Chat tool is compatible with:Any CSV or Excel datasetNon-technical usersQuick exploratory analysis. This tool alone can help you manage and analyze your data effectively. However, a composed workflow using QUICK STATS, CORRELATION, LIVE CHAT may provide even more powerful data processing capabilities. Consider exploring these related tools for a comprehensive data solution.
Most "chat with your data" tools ship your rows to a cloud LLM to answer a question that's really just a statistics query in disguise — "what's the trend", "top 10 by revenue", "any outliers". Data Chat handles that entire class of question with a local intent classifier that matches your question to one of about 25 known analysis operations, then runs real statistics (Pearson correlation, IQR outlier detection, linear regression, percentile-based binning) directly in your browser. Your question is matched against trained phrasings for things like trend detection, group-by aggregation, correlation, distributions, top/bottom-N ranking, and free-text filters like "where region is Europe" or "excluding the Fog category" — all resolved against your actual column names and values, not just keyword matching. Answers come back as text plus, where it makes sense, an inline chart or table generated on the spot: a line chart for a trend, a scatter plot for a correlation, a histogram for a distribution. Because there's no LLM in the loop, responses are instant and there's nothing to configure — no API key, no per-query cost, and no risk of a hallucinated number, since every answer is computed directly from your data rather than generated by a language model. The one exception: your typed question text (not your data) is sent to Google Translate so non-English queries still match correctly.

Frequently Asked Questions:

Does this use ChatGPT or another LLM?

No — it matches your question to a known analysis intent using a local classifier, then computes the real answer with statistics functions running in your browser. No language model is involved and no data ever leaves your machine.

What languages does it understand?

English natively, plus Italian, French, Spanish, German, Portuguese and more — non-English questions are translated to English before matching, so the underlying data itself is never sent anywhere.

What happens if it doesn’t recognize my question?

It tells you plainly rather than guessing — ask it "help" for a list of supported question types, like trend, top N, correlation, distribution, or group by.

Can I filter the data as part of a question?

Yes — phrases like "where region is Europe", "excluding the Fog category", or "in [A, B, C]" are parsed and matched against your actual column values before running the analysis.


Developed by HowToCSV for our valued CSV Haters worldwide.