The Engine For RL Environments

Built on game engine technology and engineered for shipping velocity, WorldQL is the collaborative engine for reinforcement learning environments

Game Engine Editor

The Collaborative Environment Engine

Build and ship RL environments faster

Real-time Collaboration

Multiple people and agents can edit the same environment at once, like Google Docs

Maximum Compatibility and Flexibility

Built on web technologies, Dreamlab is easy to deploy and works for all types of applications

Fast Iteration

Build, test, and deploy environments in hours instead of weeks

Defeat Reward Hacking

Analyze session replays to discover issues early in training runs

Physics / Game Environments

Robotic Arm Manipulation Task

Robotic arm environment showing random movement behavior
AI Analysis

“It looks like the model is making high speed random movements in an attempt to achieve the goal by chance. To address this, I'll add a cost per movement to the reward function…”

Web Applications

Simulated CRM / Salesforce Clone

AI Analysis

“Viewing this session replay, it appears the model is clicking the submit button rapidly to count the same customer update action multiple times. Fixing this race condition in the UI…”

Version control UI showing commit history and file changes

Seriously Good Version Control

Track every change, branch experiments, and collaborate easily. Version control built for iterating on complex environments.

Visual History

See how changes in your environment impacted results

Easy To Use

Start collaborating instantly. No software downloads or Git setup required

Merge Conflict Resolution

Resolve merge conflicts visually without leaving your browser

“One of the best version control UIs I've seen in a browser”

  • pytorch logo
    OpenEnv
  • Docker
  • Standalone

Build anything, run anywhere

WorldQL is built on web technologies and supports anything that can run in a browser. In addition, the engine includes the Rapier physics engine, a key-value database, and multiplayer support for multi-agent environments

Environments can be exported in one click to OpenEnv-compatible Docker containers.

We specialize in environments for training domain specific computer use models. We also can provide RL environments for generic coding models focused on game development, one of the most accuracy sensitive domains of software.

Get in Touch

We help labs build custom environments and datasets at scale. We work with a network of over 10,000 registered developers to deliver great results fast.

Ready to get started?

Send us an email and we'll get back to you within a few hours to discuss your project. Weekends and holidays included :)

Send us an email

You can also book a call.