Frequently asked questions

A real Windows desktop app. Get it from the Microsoft Store, where it is now live, or download ATARS-Setup.exe from the Download section, then launch ATARS from your Start menu — it opens in its own window like Excel or Power BI and runs fully offline. This website is only where you download it. There is also a free terminal edition (pip install atars) that runs on Windows, macOS and Linux, but it is a smaller, limited version — see the next question. Full Linux and macOS desktop builds are planned.

There are three ways to get ATARS, and two of them are the same product:

• Microsoft Store — the complete desktop app, now live and signed by Microsoft, so there is no SmartScreen warning and updates install themselves. The one to pick on most PCs.
• ATARS-Setup.exe (direct download) — the same app as a standalone installer. Per-user install, no admin rights — for PCs where the Store is blocked or unavailable.
• Terminal edition — pip install atars — a deliberately limited free edition for the command line. It shares the same analysis engine, but it is not as powerful as the app: no AI analyst, no Advanced Statistics, 17 chart types instead of 19, and no Word, PowerPoint or notebook exports. Some charts (treemap, sunburst, violin, bubble) cannot be drawn honestly in a terminal at all. It exists for servers, SSH sessions and scripted work — and it is currently the only way to run ATARS on macOS and Linux. Full terminal-edition guide →

If you are on Windows and just want to analyse data, use the desktop app.

Yes — this is Microsoft SmartScreen’s standard notice for any new app it hasn’t seen many downloads of yet, not a virus detection. ATARS is a free student-built app and isn’t (yet) code-signed with a paid certificate, so Windows plays it safe. To install: click More info, then Run anyway. Only ever do this for the installer downloaded from this site or our official GitHub releases page. If you would rather not see the warning at all, install the same app from the Microsoft Store — Store packages are signed by Microsoft.

No. ATARS is a guided, no-code platform. Cleaning, charts, statistics, machine learning, forecasting and reporting all run through clicks and menus. SQL Lab is there if you want to write queries, but you never have to.

Any structured tabular dataset from any domain — CSV, Excel, JSON, TSV, Parquet and Feather files, or databases & warehouses: SQLite, MySQL, PostgreSQL, Amazon Redshift, Snowflake and Google BigQuery. You can also drop several files and combine them by stacking rows or joining on a key.

Yes — fully. Auto-detection, all chart types, SQL, cleaning, statistics, AutoML, forecasting and exports run 100% offline. AI is optional. Install a local model and the AI analyst (Aryabhata) works with no key and nothing leaving your computer; or add a provider key to use a cloud model and get AI narratives on your reports.

OpenAI, Groq, Google Gemini, xAI/Grok, OpenRouter, Together AI, Cerebras, SambaNova, Mistral and Anthropic-compatible endpoints — plus any OpenAI-compatible endpoint, and local models through Ollama. Many offer a free tier with no card: Groq, Gemini, Cerebras, SambaNova, OpenRouter and Mistral, and a local Ollama model needs no key at all. ATARS shows the free options, their limits and their catches inside the app.

Three ways. ATARS is checked against the standard libraries: a correctness benchmark compares its results with pandas, NumPy, scikit-learn and SciPy on data with a known answer, and 14 of 14 checks pass on the current code. More than 900 automated tests guard the engine, including checks that fail if a core calculation silently changes. And your own figures carry receipts, re-derived from your data when you export. The checks cover core calculations rather than every possible result, and are run by the developer, not an independent auditor. Every check and its result is listed in the Methodology.

Aryabhata follows 18 short notes on statistical method — when you may claim a cause, how to treat missing data, how small a sample is too small, how to spot a model that looks too good — written for ATARS and drawing on 49 published works by researchers such as Tukey, Gelman, Pearl, Rubin, Breiman and Hyndman. For each question it looks up the notes that apply, on your own computer, and the numbers in its answer come only from what ATARS computed. The authors are not affiliated with ATARS, and the notes guide the AI rather than make it infallible. The full list with sources is in the Methodology.

Yes. Install Ollama and download a model from inside ATARS, and Aryabhata runs entirely on your machine — no key, no account, no internet needed once the model is downloaded. Answers are shorter than a large cloud model's, and it needs a few GB of disk and around 8 GB of RAM. Worth knowing about the free cloud tiers: some providers may use free-tier prompts to improve their models, so keep confidential data on a local model.

ATARS runs entirely on your own computer — your files are processed locally and never uploaded, sold or shared. Out of the box nothing leaves your machine. The only features that use the internet are ones you switch on: a cloud AI provider you choose, web search when you allow it, and an optional update check. See the Privacy Policy and Data Usage Policy.

They are written context your answers can quote — a refund policy, an SLA, a contract, last quarter's report. Open AI → Insights, expand 📚 Documents, add files (.txt, .md, .pdf, .docx, .csv) or paste text, and press Index them. Then ask an ordinary question — "what is our refund window?" — and the passage that answers it is retrieved and cited under The working. Documents never supply a number: figures are always computed from your data. Everything stays on your computer and is remembered between sessions.

Almost always because the PDF is a scan — a picture of words rather than words — so there is no text to read out of it. The Documents shelf now tells you: it lists each file with how many words came out of it and the opening line it read, and warns you outright when a file yielded nothing. If that happens, export the document as text (or paste the part you need) and index that instead.

It finds the passage that answers your question even when it shares no words with it — "can I get my money back?" reaches a paragraph that only ever says "refund". It is off until you switch it on, in the Documents section. The first time, a small open model (about 90 MB) is downloaded to your computer; after that it runs offline, and neither your documents nor your questions are ever sent anywhere. Ordinary keyword search keeps running either way, so turning it on can only add results, never lose them.

ATARS loads full tables without silently capping rows; in-memory work is limited only by your computer's RAM. For truly huge databases, SQL Lab can push aggregation to the database itself — so even billion-row tables summarise fine.

Yes — SQL Lab lets you run full SQL (joins, window functions, CTEs and aggregations) on your dataset, powered by DuckDB. No database server to set up, and you can push a query result back into your working set.

Yes. Built-in cohort & retention analysis, RFM customer segmentation, and time-series decomposition (trend / seasonality / residual), alongside AutoML model ranking and what-if scenario simulation.

Yes. The Explainable AI module uses SHAP to show global feature importance and local, per-prediction contribution breakdowns, so you can see why a result was produced.

Polished PDF, Word (DOCX), PowerPoint (PPTX), Excel and Markdown reports, a runnable Jupyter notebook (.ipynb) that reproduces your analysis, plus CSV, JSON and chart images. Reports end with a "Where these numbers came from" section, where each tracked figure is re-checked against your data at export time. With AI enabled, every export also gets an executive narrative and plain-English interpretation.

It is your analysis, saved as one file: the data, the cleaning, charts, notes, metric definitions and receipts, with several sheets inside if you use them. Save it anywhere, reopen it later, or send it to someone with ATARS and they get the same analysis. Because it contains your data, treat it like the dataset itself when you share it.

Yes — ATARS is free to download and every feature is included. No payment, no account, no trial timer. AI is optional — run a local model free with no key, or use a free key from providers such as Groq, Gemini, Cerebras or OpenRouter. Grab it from the Download section.

Students, researchers, analysts, educators, startups, businesses, NGOs and institutions — anyone who needs to turn data into decisions without a data-science team.