Lab Cookbook

Create, evaluate, train and deploy your own agents with Prime Intellect Lab.

1. Setup
2. Environments and Evals
3. Building Your First Environment
4. Training with RL
5. Prompt Optimization
6. Warm Starts with SFT
7. On-Policy Distillation
8. Judges and Instruction Following
9. Tool Use and Search
10. Multimodal Environments
11. Coding Agents and Sandboxes
12. Synthetic Agent Environments
13. Custom Harnesses
14. Best Practices
15. Legacy Environments — reference only; not part of the main path

For platform reference (accounts, secrets, Hub workflows, hosted runs, inference deployments), see Lab Configuration. For quick definitions of Lab, Verifiers, evaluation, and RL terms, see the Glossary.

Repository Structure

This is a local map of the cookbook workspace.

• guides - ordered walkthroughs.
• configs - runnable Prime eval, GEPA, and training configs.
• environments - local environment packages used by guides.
• reference - thin pointers only; do not duplicate managed guidance here.