abax 0.1.4

A keyboard-first statistics and data-science workstation: a scriptable spreadsheet with built-in statistics, analysis, and visualization.

abax

A keyboard-first statistics and data-science workstation — an integrated environment for data work, built on a fast, scriptable spreadsheet. Load a dataset, explore it with 600+ formula functions (statistics and distributions, financial, engineering, database, and RF/amateur-radio), run built-in analyses (regression, t-tests, ANOVA, correlation), reshape it with pivot/group-by and recode, visualize with the grapher, hand a selection off to pandas, and script the whole thing with Python macros — across CSV, Excel, Parquet, SQLite, JSON, R, and more.

It also carries purpose-built tools: RF & antenna engineering (link budgets, Smith charts, a thin-wire Method-of-Moments solver with NEC .nec import/export), Jupyter integration (a lossless .ipynb round-trip, rich display, and abax-as-a-Jupyter-kernel), a dual-pane file manager with configurable command buttons and one-click archiving, and a budget wizard.

It runs as a Qt desktop GUI (the default), a vim-style terminal UI, or a headless CLI. The core is pure-stdlib Python; every heavier capability is an optional dependency with a graceful fallback. When a behaviour is ambiguous, abax follows gnumeric.

Install

pip install abax[gui]         # the Qt GUI (PySide6) — the usual choice
python -m abax                # launch it

abax installs full-fat by default: on first launch it auto-installs the remaining optional dependencies in the background (the data-science stack, Excel/Parquet I/O, the PTY terminal, Jupyter integration) so everything just works. It's best-effort and non-blocking — abax keeps using its pure-Python fallbacks meanwhile — and attempted only once per machine.

python -m abax deps           # install every optional dependency now (blocking)
python -m abax --deps         # show what's present + the auto-install status
ABAX_NO_AUTOINSTALL=1 …        # opt out (or set auto_install=false in settings)

pip install -e ".[dev]"        # a development checkout
# extras: .[gui] PySide6 · .[gui-pyqt] PyQt6 · .[tui] textual · .[excel] openpyxl
#         .[parquet] pyarrow · .[science] numpy/pandas/scipy/… · .[bayes] pymc (heavy)
#         .[jupyter] nbformat/ipykernel/anywidget · .[all] everything (default full-fat)
#         .[thin] GUI + light conveniences, no heavy science

The core is pure stdlib — the whole test suite passes with zero optional packages installed, and every optional dependency degrades gracefully.

Use

python -m abax view sales.csv            # print as a table
python -m abax get sales.csv D2          # compute one cell
python -m abax convert sales.csv out.xlsx
python -m abax tui budget.abax          # curses TUI (vim keys)
python -m abax gui budget.abax          # Qt GUI (PySide6/PyQt6)
python -m abax macro run totals book.csv --macros macros/
python -m abax deps                      # fetch all optional dependencies

Formats

Open and save by file extension — convert between any of them:

Format	Extensions	Notes
CSV / TSV	.csv .tsv .tab	comma and tab delimited
Excel	.xlsx .xlsm	via openpyxl; formulas preserved
XML Spreadsheet	.xml	Excel 2003 SpreadsheetML; formulas stored as R1C1
Markdown	.md .markdown	first-class GFM tables (alignment, escaping)
Jupyter	.ipynb	valid nbformat 4.5; round-trips the whole workbook losslessly (markdown-table fallback for foreign notebooks)
R	.R .RData	data.frame export; best-effort import
SQLite	.db .sqlite	tables / SELECT queries ↔ sheets (one sheet per table)
JSON Lines	.jsonl .ndjson	flat-file record DB; one object per line
Fixed-width	.fixed	whitespace-aligned columns
Native / JSON	.abax .json	lossless workbook; JSON is also the interchange format

python -m abax convert budget.csv budget.md      # → GitHub-flavored Markdown table
python -m abax convert budget.csv budget.xml     # → XML Spreadsheet (R1C1 formulas)
python -m abax convert budget.csv budget.ipynb   # → notebook (markdown + DataFrame)
python -m abax convert budget.csv budget.R       # → R data.frame

Foreign JSON. Opening a .json file auto-detects what it is: a native abax workbook, the spec's interchange envelope ({app, schema_version, data}), a qrpn calculator save ({stack, registers}), a list of records, or a dict of columns.

python -m abax view qrpn-save.json --sheet stack   # read a qrpn calculator save

Formulas

600+ functions across aggregate, conditional, math, lookup, logical, text, date, statistics, financial (incl. bonds), engineering, RF/amateur-radio & antenna, info, and the LET/LAMBDA functional families:

=SUM(A1:A10)                         =VLOOKUP("banana", A1:B9, 2, FALSE)
=IF(B2>0, "ok", "bad")               =INDEX(B1:B9, MATCH("kiwi", A1:A9, 0))
=SUMIF(A1:A9, ">100", B1:B9)         =IFS(s>=90,"A", s>=80,"B", TRUE,"F")
=ROUND(AVERAGE(C1:C5), 2)            =TEXT(0.1234, "0.00%")
=DATEDIF("2026-01-01", TODAY(), "M") =SWITCH(grade, 1,"one", 2,"two", "other")
=PERCENTILE(A1:A99, 0.9)             =CORREL(A1:A50, B1:B50)

Operators + - * / ^ % & and comparisons; ranges (A1:C3); absolute refs ($A$1); bare TRUE/FALSE. Errors are values (#DIV/0!, #NAME?, #N/A, #REF!, #SPILL!, #CALC!, #CIRC!, …) and propagate. Control-flow functions (IF, IFERROR, IFS, SWITCH, CHOOSE) are lazily evaluated. Circular references surface as #CIRC!, never a crash.

Dynamic arrays. A formula that yields an array spills across neighbouring cells (Excel-style): =SORT(A1:A9), =UNIQUE(B:B), =SEQUENCE(3,3), =FILTER(A1:A9, B1:B9>0), and the reshaping family (TRANSPOSE, VSTACK, TAKE, MMULT, …). Operators broadcast over ranges (=A1:A3*2), array constants (={1,2;3,4}) and array IF work, A1# references a spill range, and @ takes a single value. See docs/formula-reference.md.

Editing: copy / paste / fill / sort

Gnumeric-style grid editing. Fill series autodetects progressions — 1, 2, …, Mon, Tue, …, Jan, Feb, …, ISO dates, and Item 1, Item 2, …. Pasting shifts relative references ($-anchored refs stay fixed).

• GUI: Ctrl+C / Ctrl+V (relative paste; values go to the system clipboard as TSV for other apps), Ctrl+D / Ctrl+R fill down/right, Del to clear; Data menu for Sort and Fill series; Tools → Copy selection as Markdown.
• TUI: y/p yank/paste, :copy A1:B3, :paste C1, :fill down A1:A20, :fill series A1:A12, :sort A1:C10 B desc.

The GUI

A complete menu bar with logical shortcuts, plus a command palette reachable two ways: Ctrl+Shift+P or pressing : on the grid (vim/gnumeric feel). The palette lists every action — including any loaded macros.

Menu	Highlights
File	New Ctrl+N, Open Ctrl+O, Save Ctrl+S, Save As Ctrl+Shift+S, Quit Ctrl+Q
Edit	Copy Ctrl+C, Paste Ctrl+V, Clear Del, Fill Ctrl+D/Ctrl+R, Find/Replace Ctrl+F, Palette Ctrl+Shift+P
Insert	Sheet Shift+F11, Function Shift+F3
Format	Theme Ctrl+T, OpenDyslexic font, conditional formatting, vim mode
Data	Sort, Fill series, Recalculate F9
Sheet	Next Ctrl+PgDn, Previous Ctrl+PgUp, Rename (or use the tabs)
Tools	Calculator Ctrl+K, Python console Ctrl+Shift+Y, Clipboard Ctrl+Shift+V, File manager Ctrl+Shift+F, Budget wizard, Scientific (Matrix / Signal / ODE / ML / RF toolkit / Smith chart / Antenna pattern), Install optional features, Macros, recording
Help	Keyboard shortcuts F1, About

Vim navigation works in the grid too (j/k/h/l, g/G, /). Sheet tabs sit at the bottom (click to switch, double-click to rename). The formula bar autocompletes function names and shows a live argument-hint tooltip. Eight themes ship — Obsidian, Dark One, Nord, Solarized, CRT green/amber, Light, and High-contrast (≥7:1) — matching the star/qv palettes; an OpenDyslexic font can be fetched on demand. Widgets are screen-reader labelled.

Find, format, calculate, script

• Find / Replace (Ctrl+F) — regex with backreferences, scoped, in-formulas or values; TUI :find/n/N, :s/pat/repl/.
• Conditional formatting — value→color rules (comparisons, between, contains, color scale) rendered in the GUI and the TUI; persists with the workbook.
• Function browser (Shift+F3) — searchable list of every function (incl. UDFs) with signatures; TUI :func.
• RPN calculator (Ctrl+K) — an HP-style scientific RPN keypad with an X/Y/Z/T stack, rolled in from the qv project. ← Cell pulls the active cell onto the stack; → Cell writes X back. TUI :rpn is a REPL (< pull / > store). Its save format is qv-compatible ({stack, registers}).
• Python console (Ctrl+Shift+Y) — a REPL wired to the live workbook (doc, wb, cell(ref), put(ref, val), rpn, refresh()). TUI :py <code>. Console/scripts/macros run at a selectable code-isolation level — in-process, an out-of-process resource-limited worker (default), or a strict OS sandbox (no network, writes to a scratch dir only).
• Clipboard manager (Ctrl+Shift+V) — copy history with pinning; TUI :clips/:clip.
• Graphing (Data → Graph) — plot a function of x or a selected column; QPainter in the GUI, HP-48-style braille in the TUI (:plot sin(x) -6 6).
• Equation editor (Insert → Equation) — type LaTeX, get a live Unicode preview and MathML (via pandoc when present, pure-Python fallback otherwise); insert into a cell. TUI :eq \frac{a}{b}.
• Number formats (Format → Number) — General/Integer/Currency/Percent/ Scientific/Thousands per cell, persisted with the workbook; TUI :fmt percent A1:A9.
• A toolbar with the common actions, for mouse-first usability.

RF, antenna & signal engineering

For amateur-radio operators and RF engineers (the Radio menu), plus 60+ RF formula functions:

• RF math — DBM2W, W2DBM, VSWR, FSPL, FRIIS, EIRP, WAVELENGTH, XL/XC, RESFREQ, Z0COAX, SKINDEPTH, the Maidenhead grid locator (GRIDSQUARE, GRIDDIST, GRIDBEARING), the US band plan (HAMBAND) and CTCSS tones — with an RF toolkit dialog (link budget / coax / antenna dimensions / L-network matching) and a Smith chart.
• Antenna modeling — analytic dipole/array patterns with a polar viewer; dipole input impedance (DIPOLER/DIPOLEX/RESONANTLEN); and a real thin-wire Method of Moments solver generalised to arbitrary 3-D wire structures (Yagis, bent wires) with NEC .nec deck import/export. The Antenna pattern viewer exports SVG and NEC decks.
• Signal/DSP — no-numpy FFT/STFT/spectrogram, Welch PSD (real + complex I/Q), interpolation, Butterworth/FIR filters, and ODE solvers via the Signal / data and ODE solver tools. See docs/rf-toolkit.md.

File manager & budgeting

• Dual-pane file manager (Tools → File manager, Ctrl+Shift+F) — a Worker / Directory Opus-style browser: two panes where operations act on the active pane's selection with the other pane as target. Copy / move / delete / rename, one-click .zip and .tar.gz creation and safe extraction, recursive find by name and file contents, and configurable command buttons whose shell commands expand {dir}/{path}/{sel}/{dest} placeholders (Python, not Lua — persisted per user).
• Budget wizard (Tools → Budget wizard) — set up income and 50/30/20 categories and drop in a live budget sheet where Spent is a SUMIF over an expenses log; logging an expense updates the budget through the formula engine.

Jupyter integration

.ipynb export is valid nbformat 4.5 and round-trips the whole workbook losslessly. A Sheet renders as an HTML/Markdown table via the IPython rich-display protocol. abax can run as a Jupyter kernel (python -m abax.kernel, after pip install abax[jupyter]) and expose an editable sheet widget (anywidget). See docs/jupyter.md.

The TUI

python -m abax tui — a vim-first curses interface that degrades to ASCII + 8-color over SSH. Normal/insert/command modes, the eight themes (live :theme), Tab autocomplete with argument hints, an RPN REPL (:rpn), a Python one-liner (:py), and the same editing/find/macro commands as the GUI via :.

Macros

Extend abax with plain Python files — command macros that drive a workbook, and user-defined functions callable inside formulas:

# macros/sample.py
@macro("totals")
def totals(ctx):
    n_rows, n_cols = ctx.sheet.used_bounds()
    for c in range(n_cols):
        col = chr(ord("A") + c)
        ctx.set(f"{col}{n_rows+1}", f"=SUM({col}1:{col}{n_rows})")
    ctx.recalc()

@register_function("TAXED")          # now usable as =TAXED(A1, 0.08)
def taxed(args):
    return numbers([args[0]])[0] * (1 + (as_number(args[1]) if len(args) > 1 else 0.08))

abax --macros macros/ macro list           # discover macros + UDFs
abax --macros macros/ macro run totals book.csv
abax --macros macros/ view sheet.csv        # UDFs available everywhere

Reachable from the TUI (:macro <name>), the GUI Tools → Macros menu, and the command palette. Macros run your Python code — by default isolated (an out-of-process, resource-limited worker), optionally in a strict OS sandbox, or in-process; only load files you trust. Auto-discovered from CONFIG_DIR/macros/*.py.

Recording

Record your edits and abax writes the macro for you:

• GUI: Tools menu → Start recording / Start relative recording / Save recorded macro… / Replay recording. A ● REC / ● REL indicator shows in the title bar.
• TUI: :rec start/stop, :rec rel for a relative recording, :rec save mymacro.py, :rec replay.

A recording is a list of actions, so it round-trips through JSON and emits clean @macro Python you can hand-edit. Relative recording generalizes a pattern: record =A1*2 at B1, replay at B4, and it writes =A4*2 — cells and relative refs shift by the same offset.

abax --macros macros/ macro run rowcalc data.csv --at B4   # replay relative to C-col B4

Autocomplete & argument hints

While typing a formula, abax completes function names — built-ins and your UDFs. In the TUI, Tab completes (single → NAME(, multiple → common prefix); once inside a call it shows the signature with the current argument marked — VLOOKUP(lookup, table, »col_index«, [approximate]). In the GUI, the formula bar shows a completion popup and a floating signature tooltip that tracks your cursor.

Build

just install    # dev setup
just test       # tests (pass with zero optional deps)
just pyz        # abax.pyz — one-file, optimize=2, compressed, stripped
just wheel      # wheel (docstrings kept)
just check      # lint + test + pyz + smoke

Layout

abax/
  core/      stdlib-only engine at the root: references, tokenizer, parser,
             evaluator, functions, sheet, workbook; fill/series; translate
             (ref-shift), r1c1, completion. Pluggable libraries in subpackages:
    core/io/        csv/markdown/notebook/r/xml/exchange/sqlite/flatfile I/O
    core/calc/      RPN (12C/15C/16C), algebraic, TI calculator engines
    core/science/   linear algebra, calculus/ODE, signal, stats, ML, finance
    core/format/    number formats, cell styles, conditional formatting, palettes
  engine/    adapters: excel_io/ods_io/parquet_io, document façade (dispatch)
  gui/       Qt front-end: MainWindow + mixins (view/palette/calc/console/
             macros/tools), menus, dual-surface theming, _qtcompat. Widgets in
             gui/grid/, gui/dialogs/, gui/calc/, gui/console/
  tui/       curses TUI (vim-first, SSH-safe): capabilities/themes/commands/
             editor/keys/render/app
  macros.py  macro engine + UDF registration
  recorder.py  macro recording (absolute & relative)
  app.py     CLI entry (lazy imports, --help/--version/--deps fast paths)
tests/       pytest suite

See docs/architecture.md for the architecture and invariants, and docs/formula-reference.md for the complete function list.