anyplotlib

anyplotlib is a fast, interactive plotting library for Jupyter, built on anywidget and a pure-JavaScript canvas renderer. It follows matplotlib's object-oriented API — create a Figure, call methods on Axes — so switching is often a one-line change:

import anyplotlib as apl

fig, ax = apl.subplots(1, 1)        # same shape as plt.subplots(1, 1)
ax.imshow(data)                     # pan, zoom, and inspect — live
fig                                 # display in a Jupyter cell

If you have used matplotlib's OO interface, you already know most of anyplotlib. What you gain is interactivity that stays fast on large data — without a kernel round-trip per frame.

Why another plotting library?

Matplotlib is a superb tool for publication-quality static figures, but its interactive notebook story (ipympl) re-renders the whole figure on the Python side for every frame. anyplotlib makes the opposite trade-off:

• All rendering happens in the browser. Python serialises compact state (raw image bytes, base64-encoded float arrays) once; pan/zoom/drag never touch the kernel.
• Each image, line collection, or marker group is a single canvas object, so blitting works and drag interactions run at full frame rate.
• The scope is deliberately limited. The OO API only (no plt.plot() global state), a curated set of plot types and marker styles, and raster canvas output rather than vector graphics. For print-quality SVG/PDF figures, matplotlib remains the right tool.

Features

• Plot types — plot (1-D lines with markers, linestyles, legends, log y), imshow (2-D images with colormaps, colorbars, scale bars, overlay masks), pcolormesh (non-uniform 2-D meshes), bar (grouped, horizontal, log, value labels), and 3-D plot_surface / scatter3d / plot3d.
• Layouts — subplots, matplotlib-compatible GridSpec indexing (slices, spans, negative indices), width_ratios/height_ratios, sharex/sharey linked pan-zoom, and floating inset axes with minimize/maximize.
• Markers — static overlays (points, circles, ellipses, rectangles, polygons, arrows, line segments, text, h/v lines) with matplotlib-style kwargs and live .set() updates.
• Widgets — draggable overlays (RectangleWidget, CircleWidget, AnnularWidget, CrosshairWidget, PolygonWidget, VLineWidget, HLineWidget, RangeWidget, …) that report positions back to Python.
• Events — a two-tier callback system: pointer_move fires every drag frame for cheap updates; pointer_settled / pointer_up fire once for expensive recomputation. Plus key_down, wheel, double_click, and per-line scoped handlers.
• Interactive docs — the bundled anyplotlib.sphinx_anywidget extension makes any anywidget figure live in Sphinx Gallery pages via Pyodide — no kernel or server needed.
• Embeddable anywhere — figures don't require Jupyter. Export self-contained HTML (fig.save_html("plot.html")), mount the renderer directly in an Electron app or web page via the JS mount() API, or run a live Python backend over any transport with anyplotlib.embed.FigureBridge (full callback support). See the embedding guide in the docs.

import numpy as np
import anyplotlib as apl

fig, (ax_img, ax_spec) = apl.subplots(1, 2, figsize=(900, 400))
img  = ax_img.imshow(stack.mean(axis=2), cmap="viridis")
spec = ax_spec.plot(stack[64, 64], units="eV")

cross = img.add_widget("crosshair", cx=64, cy=64)

@cross.add_event_handler("pointer_move")   # every drag frame — keep it cheap
def update(event):
    spec.set_data(stack[int(cross.cy), int(cross.cx)])

Installation

pip install anyplotlib

Works anywhere anywidget does: JupyterLab, Jupyter Notebook, VS Code, PyCharm, Google Colab, and marimo. Dependencies are intentionally light: anywidget, numpy, traitlets, and colorcet (no matplotlib required).

Documentation

Full docs, a live example gallery (interactive in the browser — no install), and the event-system guide are at cssfrancis.github.io/anyplotlib.

Development

git clone https://github.com/CSSFrancis/anyplotlib
cd anyplotlib
uv sync                              # install with dev dependencies
uv run playwright install chromium   # browsers for rendering tests
uv run pytest                        # full suite (unit + Playwright + visual)
make html                            # build the docs locally

The architecture is a single anywidget.AnyWidget (Figure) that owns all traitlets; plot objects are plain Python classes that serialise their state dicts to per-panel traits, and figure_esm.js renders them. See AGENTS.md for the codebase guide and anyplotlib/FIGURE_ESM.md for a map of the JS renderer.

License

MIT — see LICENSE.