Skip to content

Home

llmcad

A minimal, LLM-friendly CAD library in Python.

PyPI version Python versions License

Documentation: https://llmcad.org

Source Code: https://github.com/llmcad/llmcad

llmcad is a Python CAD library designed from the ground up to be LLM-friendly, by abstracting 3d spatial reasoning (e.g. absolute coordinate positioning), which LLMs typically struggle with. It wraps OpenCASCADE (via OCP) with a minimal, explicit API that makes 3D modeling predictable and debuggable for AI agents.

The key features are:

  • Minimal API: ~28 core concepts. Shapes, sketches, operations, booleans, assemblies — nothing more.
  • Named faces and edges: Every face and edge has a semantic name (top, front, left_edge, ...), so LLMs can reference geometry unambiguously.
  • Face-local coordinate systems: Each face carries its own 2D frame. Positioning on faces uses intuitive local offsets — no global coordinate math.
  • Visual debugging: Built-in multi-view snapshots, face coloring, and edge visualization for LLM feedback loops.
  • Immutable operations: Boolean ops return new bodies. No mutation, no surprises.

Installation

pip install llmcad

Example

from llmcad import Box, Cylinder, Rect, extrude, fillet, snapshot

# Create a base plate
plate = Box(100, 60, 10, name="plate")

# Add a boss on the top face
boss = extrude(
    Rect(30, 30).place_on(plate.top),
    amount=20,
    name="boss",
)
plate = plate + boss

# Cut a hole through the boss
hole = extrude(
    Rect(10, 10).place_on(plate.faces["boss_end"]),
    through=True,
    name="hole",
)
plate = plate - hole

# Fillet the top edges of the boss
plate = fillet(plate.faces["boss_end"].edges, radius=3)

# Render a multi-view snapshot
snapshot(plate, "plate")

License

This project is licensed under the terms of the MIT license.