Clawblox is an open-source engine for building simulated worlds for embodied LLM agents.
- engine core written in Rust with Rapier for 3D rigid-body physics
- worlds scripted in Luau, a lightweight language developed by Roblox, and natively support multi-agent
- simple HTTP protocol to embody coding agents in simulated worlds
- runs headless from the terminal for long-running experiments
- customizable Three.js graphics rendering running in the browser
Paper: https://machineplaying.org
macOS / Linux:
curl -fsSL https://clawblox.com/install.sh | shWindows (CMD):
curl -fsSL https://clawblox.com/install.cmd -o install.cmd && install.cmd && del install.cmdWindows (PowerShell):
irm https://clawblox.com/install.ps1 | iexFrom source:
git clone https://github.com/nacloos/clawblox.git
cd clawblox
cargo install --path .clawblox run worlds/mesa-island --agent codex
# open http://localhost:8080This starts the world server and a managed coding agent connected to it.
clawblox init my-game
cd my-game
clawblox run
# open http://localhost:8080This creates a world with world.toml (config), main.lua (world logic), and
API.md (agent instructions).
| Command | Description |
|---|---|
clawblox init [name] |
Scaffold a new world |
clawblox run [path] |
Run locally (Ctrl+C to stop) |
clawblox run [path] --resume <snapshot.json> |
Resume a world from a saved snapshot |
clawblox start [path] |
Start in background |
clawblox stop |
Stop background server |
clawblox login |
Register/login |
clawblox docs |
View engine docs |
- Agents: run coding agents in worlds
- Worlds: worlds and the world protocol
- Scripting: Luau scripting guide
- Custom renderers: Three.js renderer guide
- Running worlds: local runs, recording, and resume
Clawblox isn't tied to the built-in engine. Any server that implements the world
protocol (join, observe, and input over HTTP) can be a Clawblox world, and agents
connect to it the same way. worlds/mesa-island is a complete reference
implementation you can study or copy from.
See Worlds for the protocol and World interface for the implementer contract, including the launch environment and snapshot/resume requirements.
Please cite as:
@misc{cloos2026machineplaying,
title={Is this machine playing?},
author={Nathan Cloos and Antonio Norelli and Daniel Durbin and Jacob Andreas and Daniela Rus and Phillip Isola},
year={2026},
eprint={2610.07130},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2610.07130},
}
This research was sponsored by the Department of the Air Force Artificial Intelligence Accelerator and was accomplished under Cooperative Agreement Number FA8750-19-2-1000. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of the Department of the Air Force or the U.S. Government. The U.S. Government is authorized to reproduce and distribute reprints for Government purposes notwithstanding any copyright notation herein.