Autoplot

Autoplot parses headers and merges multiple files through a graphical import flow. You set the delimiter, header row, skipped rows, and encoding, then check a preview table before anything becomes data. Batches of files collapse into one structured project, so a folder of per-run CSVs becomes a single analysable dataset. When the parse is wrong, you fix it in the preview instead of debugging a script, which saves the kind of hour that usually disappears into a traceback you didn't ask for.

Parsed columns become named, durable variables that every plot, fit, and workspace can reference. You can derive, filter, and split them, and batch-create across many files at once. This is where the app saves the most time over a notebook: define a column once, and every later figure reuses the same vocabulary. No re-deriving the same column in a new cell. Define it once.

X&Y plots cover scatters, lines, step, bars, stem, lollipop, area, and ribbon styles, with dual axes and in-card running averages or smoothing. Histograms show density-normalized distributions with Gaussian-mixture or custom fits. For 3D data, the same x, y, z values render either as a flat heat map with contours or in a true 3D scene. Every figure overlays as many series and analysis outputs as you need. Change one card and the whole board follows. For 3D work, newer builds support up to 500,000 scatter or animated path points with safer editing, which is enough for the particle and field datasets that would choke a lighter plotting app.

Change a card and the figure follows, with no re-running a script and no third-party tools in the loop. Export produces vector PDF at journal DPI with embedded fonts, plus PNG and JPG. Newer builds also export any figure as an editable Matplotlib Python script with its data attached, so you can keep refining outside the app. That path matters when a journal or collaborator wants the raw plotting code.

The analysis surfaces go past basic stats into methods that specific fields depend on. Free covers PDF and log-binned distributions with power-law analysis. Higher tiers add CDF/CCDF, x_min diagnostics, finite-size-scaling collapse, and E–S scaling. If your paper leans on heavy-tailed data, the tools you'd normally hand-code are built in as named operations.

The assistant reads your intent and dataset schema, never your raw values, and applies an open-weight model hosted on DeepInfra's U.S. infrastructure. It writes Python you review line by line before anything runs, and file operations stay sandboxed to a folder you choose. You can chain a request into a full task: import a file, derive a variable, build a plot, annotate it, and export, all from one plain-language instruction. Disabling the assistant leaves the non-AI toolkit untouched.

Autoplot is a real SwiftUI app for Apple Silicon, not a web page in a window. Documents launch fast, scrolling is native, and large CSVs open straight from disk with no browser size ceiling or upload step. The scientific Python stack runs inside the app, so imports, fits, plots, Compose, and export all work with no connection. Only the hosted assistant, account, and billing paths need the network.