CubieVis

Qt GUI for cubie: define a system, pick an integrator, enter a batch, solve, view the traces.

Install and run

uv venv; uv pip install -e .[test]
.venv/Scripts/python -m cubievis

cubie[cuda12]/cubie[cuda13] adds a CUDA toolkit when the machine has none.

Worktrees share the main checkout's .venv: python tools/worktree_setup.py (run by orca through orca.yaml) layers a .venv that imports the worktree's src.

Using it

1. Add system…: equations (dx = ...), a Python f(t, y, p), or a .txt/.py/.cellml file; Parse fills the category tables (CellML shows its parsed equations); Edit system… reopens the current one; Make constant/Make parameter move selected rows, here and on the main window's tables.
2. Enter y0 and parameters as numbers, lists or numpy expressions; pick grid type, variables to save/summarise (All/None), their cadences, and summary metrics (peaks takes n in its second column).
3. Pick an integrator (type to search) by Go; step sizes and tolerances sit beside it, All settings… holds the rest; a row passes only when its checkbox is set.
4. Set timing, directory and results folder; the label by Go estimates runs and disk size; the log shows phase, elapsed time, then the result path. Set y0 from run copies a run's final state into y0.
5. The viewer opens the batch; summaries are a second dataset in it. X, Y, Z and Sweep each take time, a swept input, or a variable or its first or second time derivative, with log/lin (log by default for log-spaced inputs) and limits. Y (holding time or an input) cycles axis/collapse/animate; Sweep cycles collapse/animate. Lines/Surface switches the drawing; inputs on no axis get sliders; zooming time in 2D reloads that span at finer resolution; hovering names the run. Older dataset.json batches are converted to a store when opened.
6. File → Save/Load session… keeps systems, inputs and settings in a JSON file.

Layout

Tests

NUMBA_ENABLE_CUDASIM=1 .venv/Scripts/python -m pytest --ignore=tests/gui/test_main_window.py --ignore=tests/gui/test_viewer_pane.py
.venv/Scripts/python -m pytest tests/gui/test_main_window.py tests/gui/test_viewer_pane.py tests/test_run.py   # GPU, needs a display
.venv/Scripts/python bench/speed_check.py store|render <batch>     # store build and raw render timings
.venv/Scripts/python bench/viewer_check.py <batch> <store>         # each viewer mode

AI slop warning

Any AI input into Cubie is reviewed line-by-line, so it is largely slop-free and the poor design choices are all human-made. This GUI did not get the same degree of human care - if it steers cubie, is usable, and looks ok, then it is clean enough. The code is therefore an AI wasteland - it will get cleaned up when it starts to fight development, not before, so read it at your peril.