ATARS in your terminal

1 · Install

One command. No compiler, no account, nothing to configure:

Python 3.11 or newer. It pulls pandas, NumPy, scikit-learn, statsmodels, DuckDB, PyArrow, Matplotlib, ReportLab and Textual — all wheels, so there is nothing to build. Upgrade later with pip install --upgrade atars.

Prefer an isolated install? pipx install atars works too, and keeps the command on your PATH without touching your system Python.

2 · Three ways to run it

The same engine, reached three ways, depending on what you are doing.

• A full-screen window — type atars with no arguments and you get a terminal UI with clickable tabs, charts and buttons.
• A line-based shell — atars shell opens a prompt that keeps one file loaded across commands, with tab completion for commands and column names, and history on the arrow keys.
• One-shot commands — atars quality sales.csv and friends, for scripts, pipes and CI.

atars -f sales.csv starts with that file already loaded, skipping the picker. Because the shell opens a file once, the engine's result cache stays warm for the whole session — the second command on a dataset is a lookup, not a recompute.

3 · The terminal window

19 tabs in five groups, in the same left-to-right order the desktop app uses — so if you know the app, you know the window:

• Understand — Overview · Data · Quality · Metrics
• Prepare — Clean · Prep · Combine · Features · SQL
• Explore — Charts · Stats · Pivot · Trends · Business
• Model — Models · Explain · Drift · Forecast · Discover
• Report — Report

Opening a file gives you three routes, because a terminal cannot receive a dropped file: your operating system's own file dialog, a folder tree filtered to readable files, or your recent files. A workbook with more than one sheet asks which sheet, rather than silently analysing whichever tab happened to be first.

Prep Studio builds a transform recipe step by step from 22 operations — add steps, undo, apply, save the result, or print the pandas that reproduces it without ATARS installed. Combine joins a second file, and Use as data makes the joined table the one every other tab works on.

Every result table has a Save table button directly underneath it, writing a CSV next to the file you opened. Click a column heading to sort. A table stops at 500 rows and says so on its last line, rather than letting you read a cut as the whole file. Every pane scrolls, so a small terminal clips nothing. Analysis runs on a worker thread, so the window keeps responding while AutoML works.

Single-key controls: o open · e export CSV · f5 refresh · q quit · 1–5 jump to a group. A focused text box takes its own typing first, so q in the SQL box is a q.

4 · Charts

17 chart types drawn in the terminal, grouped by what they are for:

• Distribution — histogram, ECDF, box, strip
• Relationship — line, area, scatter, density heatmap, correlation
• Comparison — bar, horizontal bar, grouped, stacked, share, funnel, waterfall, pivot heatmap, count

Each chart declares what it needs, so the pickers it does not use are hidden rather than silently ignored. Every chart that groups by a category also takes an aggregation (sum, mean, count, median, min, max), a top-N cut, a sort order and optional value labels — and it says how many categories the cut left out instead of dropping them quietly.

Treemap, sunburst, violin, bubble and density contour are not here. A terminal cell has no sub-cell density and no size channel, so those cannot be drawn honestly; the desktop app draws them properly.

5 · What it tells you about a file

Opening a file runs fourteen checks over the table and ranks what comes back, most consequential first: missing values, duplicates, constant and identifier columns, numbers stored as text, skew, outliers, correlations, concentration, gaps between groups, trend over time, change points, what drives the main measure, how predictable it is, and relationships that reverse inside groups.

Every finding carries the number it is about and the evidence it came from, and no single check can crowd out the rest. Nothing calls a model. Every sentence is arithmetic over the table, which is exactly why each claim can be checked.

6 · Reports

Reports export as Markdown, HTML and PDF, carrying the data's own findings plus whatever you chose to keep. What reaches the document goes through the same two doors the desktop app uses:

• A fixed answer for this data files itself — cleaning, dataset understanding, features, moving averages, decomposition, change points, cohorts, growth, concentration, segments, drift. There is nothing to choose between, so nothing is asked.
• A value you went hunting for waits for Add to report — statistical tests, models, explanations, pivots, SQL and forecasts. You run those ten times to find the one worth stating; the other nine are not findings.

Everything arrives as a table, not as a sentence about a table. The document also prints the column profile, descriptive statistics, the strongest correlations, the category breakdown, missing values and outliers — so every figure it states can be checked inside the document itself, without the app. A chart cannot be drawn in Markdown, so each kept chart carries the numbers it was drawn from, with a line or two saying what they show. Where a long table is cut, the report says how many rows there really were.

7 · Command reference

Every command works on its own, and every one takes --json.

Exit codes mean something — 0 fine, 2 you asked for something impossible, 3 that feature is in the desktop app — so these compose in a shell script:

atars --help returns in about 9 ms: no command imports pandas until it actually runs. Thread limits are applied before NumPy loads, and n_jobs comes from the real core count, never -1.

8 · Recipes & reproducible pipelines

A recipe is a JSON list of transform steps that runs the same way every time.

recipe code prints runnable pandas with no ATARS import at all, so a pipeline you build here keeps working for a colleague who does not have this installed.

9 · From Python

The command line is a printer over a small Python API, so anything one can do, the other can do too.

10 · What it does not have

Stated plainly, because a limited edition that hides its limits wastes your afternoon:

• No AI analyst. Aryabhata, natural-language questions and narrated reports are desktop-app features. The wheel ships a stub where the AI package would be, and a test fails the build if a provider import or a credential read ever reaches the published artifact.
• No Advanced Statistics module.
• 17 chart types, not 19 — treemap, sunburst, violin, bubble and density contour need sub-cell rendering a terminal does not have.
• No Word, PowerPoint, Excel or Jupyter-notebook export. Markdown, HTML, PDF and CSV only.
• No lagged correlation or seasonality strength, no Benford tests, no group-distribution tests, no row-level anomaly explanations, no batch folders.
• No database or warehouse connections. Files only — CSV, Excel, JSON, Parquet, Feather, TSV. The desktop app connects to SQLite, MySQL, PostgreSQL, Redshift, Snowflake and BigQuery.

Everything that is here has an equivalent in the desktop app, on purpose: the free edition should never be ahead of the paid product at finding something.

11 · Privacy & licence

It runs entirely on your machine. There is no account, no telemetry and no upload — with no AI package in the wheel, there is no credential to read and nowhere for a request to go. See the Privacy Policy and the Licence.

View atars on PyPI ⬇ Get the full desktop app 🛍️ Microsoft Store