Skip to main content

buttery

Buttery smooth, agent-friendly explainer animations in Python.

buttery.computer

❯ Use buttery to make an animated explainer of merging two sorted lists.

Merging two sorted lists, one comparison per beat

examples/merge_sorted.py: the merge is simulated in Python, then each step becomes keyframes.

Key features

  • The scene is a pure function of time. state(t) resolves every property at t. Nothing accumulates between frames, so every frame renders independently and in parallel, and a scrub to any t is exact.
  • Pydantic models are the schema. JSON is the product contract; the Python API is sugar over the same models. buttery schema prints the JSON schema an agent can build against.
  • Properties are expressions, not values. Tweens, keyframes, sin(6 * T), and references to other objects' properties (ring.x = dot.ref.x) form a dependency DAG, so edges follow nodes for free. Shorthand strings like "dot.r + 0.5" parse into the same tree, no eval.
  • Built for agents. validate / state / preview / render as a CLI, a Python API, and an MCP server. Every call returns {"ok": true, ...} or a structured error list, never a bare stack trace. A Claude Code skill is included.
  • Explainer primitives. circle rect line text group, and a code block that syntax-highlights Python and lets spans select by token, line, or character range.
  • Validation up front. Structural checks from pydantic (unknown fields, arity, colors) plus semantic checks: unique ids, references resolve, no dependency cycles.
  • Real motion blur. A skia-python renderer samples sub-frames across a configurable shutter, uses every core, and writes an .mp4 through ffmpeg or a PNG sequence without it.

More examples

❯ Use buttery to show a red-black tree left rotation one step at a time.

Red-black tree left rotation, one step at a time

examples/rb_rotation.py: nodes are groups, edges are lines whose endpoints reference the nodes, so the edges follow the rotation for free. The before/after trees are two tuples; the script diffs them.

❯ Use buttery to show every token color of the syntax highlighter, one kind per beat.

Every color of the syntax highlighter, side by side

examples/syntax_theme.py: a custom theme colors all six token kinds; the legend walks them one per beat and every token of that kind pulses in the code.

Architecture

Agent tool surface   validate / state / preview / render     (tools.py, cli.py, mcp_server.py)
Authoring            Python sugar  <->  JSON                 (objects.py, expr.py, parse.py)
Core                 Scene, primitives, expression DAG       (scene.py, evaluate.py)
Renderer             skia-python, motion blur via sub-frames  (render.py)

Install

pip install buttery    # or: uv add buttery
brew install ffmpeg    # for .mp4 output (PNG sequences work without it)

The renderer is skia-python, which ships large platform wheels; expect a heavier install than the code size suggests.

From a clone:

uv sync                # Python 3.11+, pydantic, skia-python, numpy, mcp
uv run pytest

Python

from buttery import *

dot = Circle("dot", r=0.3, fill="coral")
dot.x = tween(-3, 3, at=0, dur=1.5, ease="out_cubic")
dot.y = 0.2 * sin(6 * T)

ring = Circle("ring", fill=None, stroke="white")
ring.x = dot.ref.x
ring.r = dot.ref.r + 0.5

scene = Scene(duration=3).add(dot, ring, Text("label", content="Hello", y=-1.6))
scene.state(1.0)                       # plain data
scene.preview(1.0, path="p.png")       # quick low-res PNG
scene.render("out.mp4")                # motion blur, all cores, ffmpeg
scene.save("scene.json")               # the same thing as JSON

obj.prop reads the stored expression; obj.ref.prop makes a reference to it (the spec's dot.r + 0.5 became dot.ref.r + 0.5 because a plain attribute read cannot be both a value and a reference).

JSON

{
  "duration": 3,
  "objects": [
    {"id": "dot", "type": "circle", "r": 0.3, "fill": "coral",
     "x": {"op": "tween", "keys": [[0, -3], [1.5, 3]], "ease": "out_cubic"},
     "y": {"op": "mul", "args": [0.2, {"op": "sin", "args": [{"op": "mul", "args": [6, "t"]}]}]}},
    {"id": "ring", "type": "circle", "fill": null, "stroke": "white", "x": "dot.x", "r": "dot.r + 0.5"}
  ]
}

Shorthand strings like "dot.r + 0.5" parse (no eval) into the same op tree. uv run buttery schema prints the JSON schema from Scene.model_json_schema().

Coordinates: world units, origin at center, y up. The frame is view_width (default 8) units wide. Primitives: circle rect line text code group. text wraps at max_width and honors newlines. code is a monospace block on a fixed character grid; its spans select characters by line, by Python token, or by chars range and give them their own fill, background and opacity; theme: "default" (or a kind -> color mapping) syntax-highlights Python tokens. Ops: add sub mul div neg sin cos abs min max clamp smoothstep noise. Eases: linear in_quad out_quad in_out_quad out_cubic in_out_cubic spring. Colors: hex or CSS names; fill/stroke can be tweened between colors.

Agent surface

uv run buttery validate scene.json
uv run buttery state scene.json --t 1.25
uv run buttery preview scene.json --t 1.25 --out p.png --scale 0.25
uv run buttery render scene.json out.mp4 [--no-motion-blur --samples 8 --shutter 0.5 --workers N]
uv run buttery mcp            # MCP server on stdio: same four tools + scene://schema resource

Every call returns {"ok": true, ...} or {"ok": false, "errors": [{"path", "message", "object", "property", "t"}]}. Never a bare stack trace.

Claude Code skill: skill/ (symlink or copy it into ~/.claude/skills/buttery). Register the MCP server with claude mcp add buttery -- uv run --directory /path/to/this/repo buttery mcp. Or from the published packages, no checkout needed: claude mcp add buttery -- npx -y buttery-mcp (the buttery-mcp npm shim runs uvx buttery mcp; see npm/).

Validation and evaluation rules

  • Structural validation is pydantic (unknown fields rejected, arity checked, colors checked).
  • Semantic validation: unique ids (global across groups), references resolve to numeric animatable properties, no dependency cycles, tween key kinds match the property kind.
  • Evaluation compiles each expression to a closure once, walks the DAG in dependency order, and never mutates the scene.
  • Motion blur: samples sub-frames spread across shutter × frame interval, averaged (0.5 = 180° shutter).

Layout

src/buttery/
  expr.py        AST models (Op, Tween, Ref), operator overloading, sugar (T, sin, tween, keyframes, ...)
  parse.py       shorthand parser -> AST, constant folding
  objects.py     Circle, Rect, Line, Text, Code (+ Span), Group
  code.py        Python tokenizer, line / token / chars selection -> character ranges, highlight themes
  scene.py       Scene, semantic checks, state(t)
  evaluate.py    compile + topological evaluation
  render.py      skia rasterizer, motion blur, parallel render, ffmpeg
  tools.py       validate / state / preview / render (JSON in, JSON out)
  cli.py         `buttery` command
  mcp_server.py  MCP server
examples/        bounce.py, squash_bounce.py, merge_sorted.py, rb_rotation.py, code_walk.py, syntax_theme.py (Python) and their .json, launch_demo.json
skill/           Claude Code skill
tests/

Not in v1

Equations/LaTeX, 3D or cameras, GUI, audio, plugins, manim parity.

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

buttery-0.3.2.tar.gz (4.1 MB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

buttery-0.3.2-py3-none-any.whl (39.2 kB view details)

Uploaded Python 3

File details

Details for the file buttery-0.3.2.tar.gz.

File metadata

  • Download URL: buttery-0.3.2.tar.gz
  • Upload date:
  • Size: 4.1 MB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for buttery-0.3.2.tar.gz
Algorithm Hash digest
SHA256 85f175a46369c4ae95a787aee3d15cf1b46b085b36bc575c8cebe7767260bd7c
MD5 3d911140938cbbab6ae1909e9d7970bc
BLAKE2b-256 1869bbe0d5f634e9e6df8a486652cc4bf4c89aaf3945b527b15ce40b3a6b2da8

See more details on using hashes here.

Provenance

The following attestation bundles were made for buttery-0.3.2.tar.gz:

Publisher: publish.yml on fletchgraham/buttery

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file buttery-0.3.2-py3-none-any.whl.

File metadata

  • Download URL: buttery-0.3.2-py3-none-any.whl
  • Upload date:
  • Size: 39.2 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for buttery-0.3.2-py3-none-any.whl
Algorithm Hash digest
SHA256 9c97b7cdf0f9879b9ac96c4cdc0140d6d84b19e5ff546809fa22ff951d56df23
MD5 fdb685a459592f8b36afeb76683ef43d
BLAKE2b-256 eec0e5691ac0a36a8fa64dac7a3336e6fa8acd2c8dd7020e6918f22ddd2be241

See more details on using hashes here.

Provenance

The following attestation bundles were made for buttery-0.3.2-py3-none-any.whl:

Publisher: publish.yml on fletchgraham/buttery

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Release history Release notifications | RSS feed

This release

0.3.2 This release

2 files

0.3.1

2 files

0.3.0

2 files

0.2.0

2 files

0.1.0

2 files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page