Articraft Data

A local-first dataset of articulated 3D asset records generated with Articraft.

Paper | Project Page | Articraft Harness

This repository is the plain data folder used by Articraft. It stores complete local records, category metadata, prompt/source material, and a compact records_manifest.jsonl for browsing. The Articraft code repo stays small; point it at this folder with --data-dir or ARTICRAFT_DATA_DIR.

Security Note: Articraft records include generated model.py source files. Only compile or execute model scripts from sources you trust.

---

Dataset Snapshot

• 10,788 local records
• 246 category folders
• records_manifest.jsonl with one row per complete record
• Source records, prompts, provenance, traces, ratings, labels, tags, and category metadata

Generated caches and local working state are intentionally ignored by this repo.

---

Use With Articraft

Clone the harness and this data folder side by side:

git clone https://github.com/mattzh72/articraft.git
git clone https://github.com/mattzh72/articraft-data.git
cd articraft

Set the data root once:

export ARTICRAFT_DATA_DIR=/path/to/articraft-data
uv run articraft status
uv run articraft library check --require-records

Or pass it explicitly:

uv run articraft viewer --data-dir /path/to/articraft-data
uv run articraft library list --data-dir /path/to/articraft-data
uv run articraft compile --data-dir /path/to/articraft-data <record_id>

To rebuild the manifest after editing records:

uv run articraft library rebuild-manifest --data-dir /path/to/articraft-data

---

Layout

• records/ contains one directory per Articraft record. Each record stores record.json plus revision-local prompts, source code, provenance, cost data, inputs, and traces.
• records_manifest.jsonl is the canonical browse manifest. Every row must correspond to records/<record_id>/record.json.
• categories/ contains category metadata used for filtering and labeling records.
• system_prompts/ stores hashed copies of generation system prompts referenced by provenance files.
• batch_specs/ and top-level guide files preserve historical prompt/category planning material.
• cache/, local/, generated URDFs, and generated asset folders are ignored local state.

---

Validate

From an Articraft checkout:

uv run articraft status --data-dir /path/to/articraft-data
uv run articraft library check --data-dir /path/to/articraft-data --require-records

Expected output for this snapshot:

records=10788 categories=246
Library data valid: categories=246 records=10788 manifest=10788

---

License

This dataset is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

When using, sharing, or adapting the dataset, attribute the Articraft project and cite the paper below. A suggested attribution:

Articraft Data, Matt Zhou et al., licensed under CC BY 4.0.
https://github.com/mattzh72/articraft-data

Citation

@article{zhou2026articraft,
  title     = {Articraft: An Agentic System for Scalable Articulated 3D Asset Generation},
  author    = {Zhou, Matt and Li, Ruining and Lyu, Xiaoyang and Song, Zhaomou and Huang, Zhening and Zheng, Chuanxia and Rupprecht, Christian and Vedaldi, Andrea and Wu, Shangzhe},
  journal   = {arXiv preprint arXiv:2605.15187},
  year      = {2026}
}