Introducing AgentEnv: An Open-Source Framework for Building RL Environments
Scale AI, Monday, October 5th, 2026
Scale open-sources AgentEnv, the framework behind its reinforcement learning environments for AI agents.
Scale is open-sourcing AgentEnv, the framework that runs all of its reinforcement learning environments, designed to be realistic, accurate and scalable while letting engineers, researchers and domain experts work in parallel.
Environments are composed of versioned blocks, including apps such as Slack, Gmail, a browser or Blender, seed-data Artifacts, and Rules such as clocks, triggers and roles.
Each environment publishes an Environment Card under the open AgentEnv Environment Protocol so any agent, sandbox or cloud can connect.