Eval Protocol

Eval Protocol is an open-source framework that enables reinforcement learning fine-tuning for AI agents running in production environments. It abstracts away infrastructure complexity by providing a standardized HTTP interface that works across Python, TypeScript, Docker, and other languages and frameworks.

The system handles agent orchestration, metadata management, and trace storage automatically. Once an agent exposes the EP standard, it can be trained with any supported trainer including Fireworks RFT, TRL, or Unsloth without environment rewrites. It's designed for applied AI teams, research engineers, and MLOps teams working with complex multi-turn or tool-using agents.