Quickstart

This page gets you from a CAD file to structured data in a few lines. It assumes you have installed the toolkit.

Convert a CAD file (CLI)

cadling ships a command-line entry point:

# Convert a CAD file to JSON or Markdown
cadling convert model.step --format json -o model.json

# Chunk a CAD file for RAG
cadling chunk model.step --max-tokens 512 --overlap 50 -o chunks.jsonl

# Show file information
cadling info model.step

Convert a CAD file (Python)

from cadling import DocumentConverter, ConversionStatus

converter = DocumentConverter()
result = converter.convert("model.step")

if result.status == ConversionStatus.SUCCESS:
    doc = result.document
    print(f"Parsed {len(doc.items)} items")

    json_data = doc.export_to_json()
    markdown = doc.export_to_markdown()

Tokenize geometry

from geotoken import GeoTokenizer

tokenizer = GeoTokenizer()
tokens = tokenizer.tokenize(vertices, faces)

Encode a STEP file with a neural model

import torch
from stepnet import STEPEncoder, STEPTokenizer, STEPFeatureExtractor, STEPTopologyBuilder

tokenizer, extractor, builder, encoder = (
    STEPTokenizer(), STEPFeatureExtractor(), STEPTopologyBuilder(), STEPEncoder()
)

token_ids = torch.tensor([tokenizer.encode(step_text)])
topology = builder.build_complete_topology(
    extractor.extract_features_from_chunk(step_text)
)
embedding = encoder(token_ids, topology_data=topology)  # [1, 1024]

Which models are trained

Several models now ship trained with reproducible, honest metrics: ll_brepnet (B-Rep segmentation, test mIoU 0.828), ll_stepnet (face-count classifier, val acc 0.976), ll_ocadr (geometry-grounded, 0.919 vs 0.313 shuffled), and ll_gen’s program-based generators (valid CAD: AR 0.914 / latent diffusion 0.934). The neural models train and run natively in MLX on Apple Silicon as well as PyTorch. Remaining task heads (e.g. ll_stepnet property prediction/QA) ship as architectures — train them on your data before relying on their outputs.

Train / run natively on Apple Silicon (MLX)

Each neural package has an mlx/ trainer that runs on Apple Silicon. The ones with existing PyTorch checkpoints convert the real weights and prove parity:

python ll_stepnet/mlx/train_classification_mlx.py --mode parity   # acc 0.976, argmax 1.0 vs PyTorch
python ll_brepnet/mlx/train_brepnet_mlx.py        --mode parity   # mIoU parity vs PyTorch
python ll_gen/mlx/ar_generator_mlx.py             --mode train    # valid CAD generation 0.914
python ll_gen/mlx/latent_diffusion_mlx.py         --mode train    # latent-diffusion generation 0.934
python ll_ocadr/mlx/train_ocadr_mlx.py            --mode train    # geometry-grounded multimodal

Where to go next

• cadling — full CAD parsing, chunking, and SDG.
• geotoken — adaptive geometric tokenization.
• ll_stepnet — neural STEP/B-Rep models.
• ll_ocadr — geometry-aware LLM input.
• ll_gen — generative CAD orchestration.
• ll_clouds — point-cloud processing.