Decision Tree
Comparison
Path 1: Gymnasium Web UI
Best for: Researchers who want to explore environments, test agent behaviors, or demonstrate capabilities without infrastructure setup.
- Visual environment browser — See all available environments and their capabilities
- Built-in agent loop executor — Run agent models against tasks without writing code
- Real-time state inspection — Watch database and application state change
- Integrated verifier — Check task completion with one click
- Sample tasks — Pre-built prompts and verification criteria
Get Started
Web UI Guide
Complete your first task in 5 minutes using the Gymnasium interface
Path 2: Docker Images
Best for: Researchers running large-scale evaluations, integrating into training pipelines, or needing full programmatic control.
- Self-hosted environments — Run on your own infrastructure
- Full API access — Programmatic control over all environment features
- Parallel execution — Spin up multiple containers for scale
- CI/CD integration — Automate evaluation in your pipelines
- Custom agent loops — Use your own execution framework
What You’ll Need
- Docker installed (version 20.10+)
- Docker images from Scale (contact Scale to receive)
- Your own agent loop implementation
Get Started
Docker Quick Start
Load and run environment containers with your agent loop
Environment Availability
All three environment types are available through both paths:Verifier Access Differs:
- Website Environments expose
/verifierin Docker containers
Still Unsure?
Next Steps
Web UI Guide
Get started with the Gymnasium interface
Docker Quick Start
Set up self-hosted environments