FastBuild123

Turn any build123d model into a live, interactive parametric web UI. Annotate your part function's parameters with a FastAPI-style @app.part() decorator and get auto-generated sliders in the browser with instant 3D preview — the type-annotated signature is the UI definition.

Quick start

Run a bundled example without installing anything:

uvx fastbuild123 demo enclosure

Or install it into your own project:

uv add fastbuild123      # or: pip install fastbuild123
fastbuild view my_model.py

A browser page opens with parameter controls on the left and a live 3D view on the right. Drag a slider or save the file — the model updates in place, keeping your camera pose, tree state, and clipping planes. A model that raises keeps the last-good geometry on screen with an error banner; fix the file and it recovers.

Defining parts

from typing import Annotated, Literal
from build123d import Part, Box
from fastbuild import App, Group, Range

app = App()

@app.part()
def enclosure(
    width:  Annotated[float, Range(20, 200, step=1, unit="mm")] = 80,
    wall:   Annotated[float, Range(0.8, 5, step=0.1, unit="mm")] = 2,
    ribs:   Annotated[int, Range(0, 8), Group("Ribs")] = 4,
    vented: bool = False,
    style:  Literal["rounded", "chamfered"] = "rounded",
) -> Part:
    ...

Widget mapping: float/int + Range → slider, bare numbers → number input, bool → checkbox, Literal/Enum → dropdown. Every parameter needs a default — defaults are the initial build. Multiple @app.part() functions in one file get a part selector. Add Group("Label") to a parameter's annotation to place it under a named collapsible section in the panel; ungrouped parameters render first.

Headless export

fastbuild export my_model.py enclosure --param width=120 -o enclosure.step

Values are validated against each parameter's Range; output format follows the extension (.step/.stp/.stl).

AI coding agents

FastBuild123 ships its own agent skill doc — API rules plus caching best practices in the Agent Skills format. Install it into your project so Claude Code, Codex, Cursor, and friends pick it up automatically:

fastbuild agent-docs --install   # writes .claude/skills/fastbuild123/ and .agents/skills/fastbuild123/

Re-run after upgrading the package; plain fastbuild agent-docs prints the doc to stdout instead. The library is also indexed on Context7 (configured via context7.json) for agents using the Context7 MCP server.

Requirements

Python 3.10–3.13 on macOS (Apple Silicon or Intel), Linux (glibc 2.31+, x86-64 or aarch64), or 64-bit Windows — wherever cadquery-ocp publishes wheels. Expect a ~500 MB install; the OpenCASCADE kernel is most of it. Alpine/musl is not supported.

Development

uv sync
uv run pytest
uv run fastbuild demo enclosure

Examples live in fastbuild/examples/ so they ship with the package and stay runnable via fastbuild demo. Specs live in specs/. The 3D component is a vendored, pinned build of three-cad-viewer (fastbuild/static/vendor/, see VERSION there) fed by ocp-tessellate output over a versioned websocket protocol.

License

MIT — see LICENSE. Bundles three-cad-viewer (MIT), whose license is included at fastbuild/static/vendor/LICENSE-three-cad-viewer.