Changelog

A real certificate for your work — one that can prove it is genuine.

• Certificate of Verified Analysis — every saved workbook gets a proper certificate: the ATARS logo, a grade seal, every sheet with its quality, and a PDF ready to print.
• QR codes only ATARS can verify — each certificate carries signed QR codes holding every sheet's details. Give ATARS the PDF, or a photo or scan of it, and it says whether the certificate is genuine; add the .atars file and it rechecks every figure against the data itself.
• Pickle files open again, safely — a .pkl that holds a table opens; one that carries code is refused before it runs.
• Nested JSON opens as the table inside it — a file whose records sit inside it opens one row per record instead of stopping.
• Yes/no columns no longer stop the report — a True/False column is read as the category it is.
• Save as keeps every sheet — saving a workbook under a new name used to keep only the sheet on screen.

ATARS is live on the Microsoft Store (product 9NFRBPL4T2M8) — free, signed by Microsoft, and kept up to date by Windows. It is the same app as ATARS-Setup.exe, which stays available as a direct download.

• Your work saves itself — every change is written to your file as you make it.
• A save can never shrink your file — a workbook that held five sheets comes back holding five, and saving no longer erases work on sheets you had not opened.
• One file, not copies — opening the same file keeps one file, instead of leaving you with sales (2), (3), (4).
• Documents answer questions in your own words, and the document shelf shows what it actually read — saying so when a scanned PDF yields no text.
• AI survives a model being retired — ATARS finds a model your key can still use and tells you it switched.
• Cleaning never invents data, and the log names every column left untouched.
• Faster — a five-sheet workbook saves in about 0.6 s where it took 7.3 s, and clicking around a 10,000-row table no longer rebuilds downloads every time.

A free command-line edition on PyPI — pip install atars — running the same analysis engine on Windows, macOS and Linux.

• Three ways to run it — one-shot commands (clean, profile, chart, model, forecast, sql, report), an interactive shell, or a full-screen window with clickable tabs.
• 19 tabs in five groups — the same Understand → Prepare → Explore → Model → Report workflow as the desktop app.
• 17 chart types drawn in the terminal, a 22-op Prep Studio that prints the pandas it would run, and reports in Markdown, HTML and PDF.
• Fourteen automatic checks on every file opened — missing values, duplicates, skew, outliers, correlations, change points, drivers and more, each ranked and carrying its own evidence.
• Deliberately limited. No AI analyst, no Advanced Statistics, and no Word / PowerPoint / notebook exports — the desktop app remains the full product.

Your analysis becomes a file you own — plus numbers that prove themselves, and AI that reads what your spreadsheets could not.

• The .atars file is your analysis — data, cleaning, charts and notes in one file you can save, reopen and send to someone else.
• Sheets — keep several sheets in one file, and add, switch or remove them as you work.
• Metrics — define a business term once ("Revenue = sum of Sales, excluding returns") and every chart and report uses that same definition, so two people never present different numbers for the same word.
• Receipts — every figure carries its working: the column, the filter, how many rows were left out and why, and the arithmetic. Receipts are re-checked against your data when you export, and travel inside the report.
• Ask anything — one Insights conversation runs the analysis your question needs, writes and runs the code, and shows every step.
• Reads your text columns — comments, reviews and notes are grouped into themes, computed on your own computer over every row.
• Your documents — add policies, contracts or past reports; ATARS searches them, cites the passage it used, and remembers them between sessions.
• Local AI — install Ollama and download a model from inside ATARS, for AI that never leaves your machine.
• Sized to your computer — heavy work is planned around your machine's cores and free memory.

Reliability release — four fixes found in a full code audit before wider rollout.

• Filter fix — typing ( or [ in a "contains" filter now searches literally instead of erroring.
• Correct AutoML winner — a model scoring exactly 0.0 is no longer ranked below worse, negative-scoring models.
• Always-fresh AI answers — Aryabhata's whole-dataset statistics now refresh after any cleaning-setting or branch change.
• Big-table downloads — very large result tables (SQL output, segments) get a one-click "Prepare CSV" download.

The newest release — quality-of-life and intelligence upgrades across the app.

• Faster AutoML — a rebuilt engine with a Fast / Full-accuracy switch: Fast ranks models on a sample for a quick answer, Full trains on every row for maximum precision.
• Live AutoML progress — a real progress bar and per-model checklist show what's done, what's training, and what's left.
• Whole-dataset AI — Aryabhata now understands your entire dataset (per-column stats, correlations, first & last rows), so it answers questions about all your data, not just a sample.
• OpenRouter support — added as an AI provider, alongside OpenAI, Groq, Gemini, xAI, Mistral, Anthropic-compatible and local Ollama.
• Better filtering & charts — the global categorical filter now shows searchable, selectable values on large columns, and distribution charts have cleaner percentage labels.
• Self-scrolling AI chat — long conversations scroll inside the chat panel instead of stretching the page.

ATARS began in February 2026 and has grown into the complete, no-code analytics platform you use today. Everything below is available right now.

• Guided workflow — Understand → Prepare → Explore → Model → Report, from a raw file to a finished document.
• Any source — CSV, Excel, JSON, Parquet, Feather and TSV files, plus SQLite, MySQL, PostgreSQL, Redshift, Snowflake and BigQuery; combine multiple files by stacking or joining.
• SQL Lab — full SQL (joins, window functions, CTEs) on your data, powered by DuckDB.
• Cleaning & quality — automatic cleaning, a data-quality auditor and anomaly detection, all recorded in an audit trail.
• Charts & dashboards — 15 interactive chart types with slicers and pinnable dashboards, plus AI chart suggestions.
• Statistics & business analytics — t/z-tests, ANOVA, chi-square, correlation and non-parametric tests; cohort & retention, RFM segments and time-series decomposition.
• Machine learning — AutoML leaderboard, regression / classification / clustering, leakage-safe cross-validation, and single & batch prediction.
• Explainable AI — SHAP global & local explanations and driver analysis — see why, not just what.
• Forecasting & simulation — trend & seasonal forecasts with confidence bands, a what-if scenario simulator, and drift monitoring.
• Reports — one-click PDF, Word, PowerPoint, Excel and Markdown, plus a runnable Jupyter notebook, with optional AI narrative.