Sandflow

Sandflow is an open source dataset library providing 100,000 3D objects written as code using Sand3D.

Sandflow provides a 3D-as-code dataset library containing 100,000 3D objects written in Sand3D, an LLM-friendly language. The collection includes CAD parts, mechanical parts, and articulated mechanisms, covering items like furniture, lighting, vehicles, and toys. Each 3D object is represented by a Python file that defines its parts, shapes, holes, positions, and materials. The dataset is built from five open collections, including Objaverse, CAD parts, Articraft, NVIDIA SimReady, and AutoMate. The programs generate meshes that work with Blender, three.js, and glTF engines, and the full library is available on Hugging Face and GitHub.

Sandflow at a glance

Best for: Software engineer, Designer
Runs on: Web, open source
Pricing: Free open source
Released: November 2026
Category: AI research tool
Website: sandflow.ai

Key features

Contains 100,000 3D objects written as code
Uses Sand3D, an LLM-friendly language
Represents each object as a Python program
Includes 10,000 plus articulated objects and 36,000 joints
Provides meshes compatible with Blender and three.js
Available on Hugging Face and GitHub as open source

Use it for

Read 3D object programs with large language models
Modify 3D object dimensions by editing code values
Export meshes to Blender and glTF engines
Build on open source 3D code libraries

Alternatives to Sandflow

SolidCraft — It runs on the desktop, in the browser, via CLI, and through an MCP server.
SOMA — Design 3D models instantly using natural language and AI coding agents.
Lobole — Drives the GUI, Python scripts, and AI agents through a single unified command layer.
DeerFlow — Executes long tasks using sandboxes, memories, tools, and subagents.
Sandcastle — Orchestrates AI coding agents in isolated Docker, Podman, or Vercel sandboxes.
CADAM — Integrates directly into SolidWorks, Onshape, and Fusion.
Mesh LLM — Run large local models across multiple machines through one OpenAI-compatible endpoint.
Apache Hamilton — Creates modular, testable, and self-documenting data pipelines using Python code.
MLflow — Complete observability and evaluation for production AI apps.
LanceDB — Provides one table for the whole AI data loop and high-throughput data access.
Metaflow — Scales projects easily from laptops to the cloud with infrastructure integration.
DataFlow — Automates and scales data preparation through reusable, operator-based pipelines.

All Sandflow alternatives