desmosdotpy 0.1.0

Build interactive Desmos graphs from Python — geometry, sliders, buttons, and a Python-to-Desmos-LaTeX compiler.

desmos.py

Build interactive Desmos graphs from Python.

Write your math in idiomatic Python with tuples, attribute access, list comprehensions, ternaries, recursive iteration. desmos.py compiles it to a Desmos state JSON and embeds it in a self-contained HTML page that runs entirely client-side.

You can find a demo with a Mandelbrot explorer (examples/mandelbrot.py) built in ~40 lines of Python.

---

Installation

pip install desmos-py

For development:

git clone https://github.com/chris-zy-liu/desmos.py
cd desmos.py
pip install -e ".[dev]"
pytest

---

Quickstart

from desmos import Graph

g = Graph(bounds=(-10, 10, -10, 10))

a = g.slider("a", -5, 5, value=1, step=0.1)   # interactive slider
g.expr(r"y = a x^{2}", color="#c74440")        # raw Desmos LaTeX

g.to_html("parabola.html")

Open parabola.html in a browser. Drag the slider — the curve updates.

---

Concepts

Graph

The top-level builder. Holds the viewport, an ordered list of expressions, and the ticker (if any).

g = Graph(
    bounds=(xmin, xmax, ymin, ymax),
    title="My Graph",          # HTML <title>
)

Expressions

Every visible or hidden element in a Desmos graph is an "expression". You add them through helper methods on the Graph object. The primary ones:

Method	What it does
g.expr(latex, **style)	Add a raw-LaTeX expression. Escape hatch.
g.var(name, value)	Hidden variable: name = value
g.slider(name, min, max, value, step)	Variable with a slider UI
g.circle, g.rectangle, g.point, g.line, g.arrow, g.text	Geometric primitives
g.button(text, at=(x,y), action=...)	Clickable point that runs an action
g.ticker(action, rate_ms=30)	Action that runs continuously
@g.func	Compile a Python function to a Desmos function
g.iterate(step, x0, name=..., extra_params=[...])	Recursive iteration
g.heatmap(value_fn, x_range=..., y_range=..., max_value=...)	Coloured point grid
g.field(Ex_fn, Ey_fn, x_range=..., y_range=...)	Vector field arrows (live, vectorised)

Output

g.to_html("out.html")     # write self-contained HTML
g.to_state_json()         # return the Desmos state as a JSON string
g.to_state()              # return the state as a dict (for inspection / tests)

---

Variables and sliders

N    = g.slider("N", 10, 100, value=40, step=1)
zoom = g.var("zm", 1.5)

Both return a Var object that:

1. Knows its Desmos name (e.g. z_{oom} — see Identifier naming).
2. Supports arithmetic — zoom * 0.5, N + 1, -zoom — by building LaTeX expression trees lazily. This is not a general expression DSL (for that see @g.func); it's just enough to write button actions naturally.

zoom * 0.5    # builds LaTeX: z_{m} \cdot 0.5
N + 1         # builds LaTeX: N + 1

---

Actions and buttons

A Desmos action is an assignment that runs on demand:

zoom.set(zoom * 0.5)   # action: zm -> zm * 0.5

Bind it to a button:

g.button("Zoom in", at=(5, 5), action=zoom.set(zoom * 0.5))

Pass a list to sequence multiple assignments atomically:

g.button("Reset", at=(5, 4), action=[
    cx.set(0),
    cy.set(0),
    zoom.set(1),
])

A ticker runs an action continuously at a fixed rate:

t = g.var("t", 0.0)
g.ticker(t.set(t + 0.016), rate_ms=16)   # 60 Hz time counter

---

Geometric primitives

Every primitive returns an Expression whose style attributes you can mutate after the fact.

# Circle
g.circle((0, 0), radius=1, color="#2d70b3")
g.circle((cx, cy), radius=r, line_style="DASHED")   # parameterised on Vars

# Rectangle (filled)
g.rectangle((-2, -1), width=4, height=2,
            color="#c74440", fill_opacity=0.3)

# Draggable point
p = g.point((1, 1), draggable="XY", color="#fa7e19", label="drag me")

# Line segment
g.line((0, 0), (3, 4), line_width=3)

# Arrow with filled triangle head (static endpoints — head geometry is
# precomputed in Python; for dynamic vector fields use vectorised polygons)
g.arrow((0, 0), (1, 1), color="#000000", head_size=0.15)

# Text label
g.text("Origin", (0, 0))

The position / radius / size arguments can be plain numbers or Var / LatexExpr instances — in which case the shape updates whenever the underlying variable changes.

---

@g.func — Python compiled to Desmos

The decorator reads the function's source with inspect.getsource, parses it into a Python AST, and emits Desmos LaTeX:

@g.func
def step(z, c):
    return (z.x ** 2 - z.y ** 2 + c.x,  2 * z.x * z.y + c.y)

becomes (roughly):

step(z, c) = ((z.x)^2 - (z.y)^2 + c.x,  2·z.x·z.y + c.y)

Note that a Python 2-tuple translates to a Desmos point, with .x / .y accessors.

Supported subset (v1, strict)

Python	Desmos
a + b, a - b, a * b, a % b	a + b, etc. (% → \operatorname{mod})
a / b	\frac{a}{b}
a ** b	a^{b}
-a, +a	unary -a, +a
a < b, <=, >, >=, ==, !=	corresponding LaTeX
a < b < c	piecewise condition {a<b, b<c}
a if cond else b	piecewise {cond: a, b}
f(x, y)	f(x, y) (user funcs) or builtin
(x, y)	Desmos point (x, y)
(x, y, z), [x, y, z]	Desmos list [x, y, z]
p.x, p.y	point component access
a[i]	a[i + 1] (Python 0-based → Desmos 1-based, auto-converted)
[expr for k in iter]	[expr for k = iter]
range(n)	[1...n]
range(a, b)	[a...b]

Built-in functions mapped to Desmos equivalents:

abs, sqrt, sin, cos, tan, exp, log / ln, floor, ceil, round, min, max, sum (→ \operatorname{total}), len (→ \operatorname{length}).

Explicitly rejected (raises DesmosTranslationError)

for / while loops, if statements (use ternary), mid-function assignments, lambda, generator expressions, list comprehensions with if filters or multiple generators, *args / **kwargs, keyword arguments, f-strings. The error always reports the offending line number.

Why so strict?

Desmos's evaluation model is purely functional. It does not allow mutation. The translator stays within constructs that map cleanly. If you hit a limitation, you can always drop down to g.expr(r"...") and write raw LaTeX.

---

g.iterate — recursive iteration

Desmos doesn't allow recursion inside arrow expressions, but it does allow recursive function definitions via piecewise base cases. g.iterate emits one:

orbit = g.iterate(step, x0=(0, 0), name="orbit", extra_params=["c"])

becomes:

orbit(k, c) = { k = 0 : (0, 0),  step(orbit(k-1, c), c) }

Read: "orbit at iteration k is (0, 0) if k=0, otherwise one step applied to the previous orbit value, threading c through unchanged."

• step is a LatexExpr returned by an @g.func-decorated function.
• The first argument of step is the iterated state.
• extra_params lists names passed through unchanged on each recursion.
• Returns a LatexExpr for the new function so you can reference it in other @g.func bodies.

---

g.heatmap — colored pixel grid

g.heatmap(
    value_fn,                                  # @g.func returning a number
    x_range=(cx - zm, cx + zm, RES),           # math region (may be Vars)
    y_range=(cy - zm, cy + zm, RES),
    max_value=N,                               # value_fn's normaliser
    palette="fire",                            # "fire" or "grayscale"
    point_size=12,
    display_bounds=(-2, 1, -1.5, 1.5),         # optional: where pixels live
)

Two grids are emitted internally:

• Math grid — at x_range × y_range. Follows pan/zoom variables. Sampled by value_fn.
• Display grid — at display_bounds (defaults to the viewport). Static. This is where pixels are drawn.

The separation is what makes zoom work properly: shrink the math region and the pixel positions stay put, so the picture magnifies instead of compressing into a smaller patch.

The color for each pixel is hsv(240·(1 − value/max_value), 1, 1) for the fire palette — blue at low values, red at high.

---

g.field — live vector field

g.field(
    Ex_fn, Ey_fn,                # @g.func returning x and y components
    x_range=(xmin, xmax, n),     # n = count (int) or stride (float)
    y_range=(ymin, ymax, n),
    length=0.25,                 # shaft length (every arrow is direction-only)
    head_size=0.08,              # length of each V-head wing
    color="#000000",
    line_width=1.2,
    name="f",                    # identifier prefix; change if you want two fields
)

The arrow geometry is built entirely as vectorised Desmos LaTeX, so the field updates live whenever its inputs change — drag a charge, move a slider, etc. For single static arrows (Python-precomputed head) use g.arrow instead.

---

Worked example: Mandelbrot explorer

See examples/mandelbrot.py for the full file. The shape:

from desmos import Graph

g = Graph(bounds=(-2, 2.5, -1.5, 1.5))
RES = 50

N  = g.slider("N",  10, 100, value=40, step=1)
cx = g.var("cx", -0.5);  cy = g.var("cy", 0.0);  zm = g.var("zm", 1.5)

@g.func
def step(z, c):
    return (z.x**2 - z.y**2 + c.x,  2*z.x*z.y + c.y)

@g.func
def magsq(z):
    return z.x**2 + z.y**2

orbit = g.iterate(step, (0, 0), name="orbit", extra_params=["c"])

@g.func
def inside(k, c):
    return 1 if magsq(orbit(k, c)) < 4 else 0

@g.func
def escape(c):
    return sum([inside(k, c) for k in range(N)])

g.heatmap(escape,
          x_range=(cx - zm, cx + zm, RES),
          y_range=(cy - zm, cy + zm, RES),
          max_value=N,
          display_bounds=(-2, 1, -1.5, 1.5))

g.button("Zoom in", at=( 1.7, 1.2), action=zm.set(zm * 0.5))
g.button("Zoom out", at=( 1.7, 1.0), action=zm.set(zm * 2.0))
g.button("Pan left", at=( 1.7, 0.6), action=cx.set(cx - zm * 0.3))
g.button("Pan right", at=( 1.7, 0.4), action=cx.set(cx + zm * 0.3))
g.button("Pan up", at=( 1.7, 0.2), action=cy.set(cy + zm * 0.3))
g.button("Pan down", at=( 1.7, 0.0), action=cy.set(cy - zm * 0.3))
g.button("Reset", at=( 1.7, -0.3), action=[
	cx.set(-0.5),
	cy.set(0.0), 
	zm.set(1.5)
])

g.to_html("mandelbrot.html")

That's it — a fully interactive Mandelbrot explorer compiled to a single HTML file. The grid resolution × iteration depth is the perf knob (50×50 × 40 iters ≈ 100K evaluations per redraw).

---

Worked example: electric field of two charges

See examples/electric_field.py for the full file. Two draggable point charges (+1 red, -1 blue) and a 13×9 grid of arrows showing the field direction. Drag either charge to watch the field redraw.

Most of the work goes into the field equations; the arrows themselves are one call to g.field:

# Two draggable charges, as raw expressions.
g.expr(r"P_{1}=(-1, 0)", drag_mode="XY", point_size="20", color="#c74440")
g.expr(r"P_{2}=( 1, 0)", drag_mode="XY", point_size="20", color="#2d70b3")

@g.func
def Exfrom(p, q, src):
    return q * (p.x - src.x) / ((p.x - src.x)**2 + (p.y - src.y)**2 + 0.02)**1.5
# Eyfrom defined symmetrically

@g.func
def Ex(p): return Exfrom(p, 1, P1) + Exfrom(p, -1, P2)
@g.func
def Ey(p): return Eyfrom(p, 1, P1) + Eyfrom(p, -1, P2)

g.field(Ex, Ey,
        x_range=(-3, 3, 13),  # (min, max, count)
        y_range=(-2, 2, 9),
        length=0.25, head_size=0.08)

P1 and P2 are Desmos identifiers — they're undefined Python names, but the translator only needs them to compile to LaTeX (auto-subscripted to P_{1}, P_{2}). A +0.02 softening term in the denominator avoids the singularity at the charges themselves.

Under the hood, g.field emits four helper functions (mag, tip, wingL, wingR), three parallel point lists (T, L, R), and one vectorised polygon call polygon(G, T, L, T, R). That last call is the key trick: Desmos's polygon(L1, ..., Lk) with k equal-length point lists produces N k-vertex polylines — here 5 vertices per arrow tracing tail → tip → wingL → tip → wingR, drawing the shaft and V-head as one self-intersecting polyline per grid point. The whole field redraws in one pass when you drag a charge.

---

Identifier naming

Desmos parses bare multi-character names as products: zoom would mean z · o · o · m. To avoid this, desmos.py automatically subscripts:

• g.var("zoom", ...) → Desmos variable z_{oom}
• g.var("cx", ...) → c_{x}
• @g.func def escape(...) → Desmos function \operatorname{escape}
• Single-character names like g.var("a", ...) pass through unchanged.

You can opt out by including a _ or \ in the name, in which case the name is passed through verbatim: g.var("a_1", 0) → a_1.

---

Output

g.to_html("path.html")     # writes a self-contained HTML page
g.to_state_json()          # the underlying Desmos state, as JSON
g.to_state()               # same, as a Python dict

The HTML loads Desmos via their public Calculator API on a demo API key. That key works on any domain for non-commercial use. For production deployment, request your own key from desmos.com/api and edit desmos/template.html.

---

Examples

python examples/quickstart.py        # parabola with a slider
python examples/geometry.py          # circles, rectangles, points, lines, arrows, text
python examples/electric_field.py    # two charges + live vector field
python examples/mandelbrot.py        # the showcase

Each writes an .html file in the current directory. Open it in any browser.

---

Tests

pytest

37 golden-file tests for the AST translator currently. All pass on Python 3.10–3.14.

---

Roadmap

Not yet implemented but designed-for:

• Live server mode — g.serve() with a websocket bridge so Python callbacks can fire on button clicks / slider changes.
• Operator-overloaded expression DSL — symbol objects so simple math doesn't require @g.func.
• Translator extensions — for-loop unrolling, simple assignment inlining, enumerate, filtered comprehensions.
• Jupyter / Colab integration — _repr_html_ so graphs render inline.
• Animation primitives — g.animate(var, keyframes) sugar over tickers.

---

License

MIT. See LICENSE.