# Introduction (https://www.agentenvframework.com/docs)

> What AgentEnv Framework is and the concepts it is built from

**AgentEnv Framework** is Scale AI's open-source framework for building RL environments, and for
running and grading agents in them. Its repository is
[`scaleapi/agentenv-framework`](https://github.com/scaleapi/agentenv-framework), and its CLI is
`agent-env`. Everything runs locally by default, so you need no cloud account to try it.

Start with [Getting Started](https://www.agentenvframework.com/docs/getting-started.md), which installs the framework and runs a first
task in two commands.

## Core concepts

* **[Environments](https://www.agentenvframework.com/docs/environments.md)**: the world an agent acts in, usually an MCP server and the
  state its tools change.
* **[Artifacts](https://www.agentenvframework.com/docs/artifacts.md)**: versioned data, such as a set of emails or a Docker image, that
  any environment can load.
* **[Agents](https://www.agentenvframework.com/docs/agents.md)**: anything that speaks A2A, from your own agent to a wrapped harness
  such as [Claude Code, Codex, Gemini, Hermes Agent, OpenCode, pi, OpenHands, Letta Code, Pydantic
  AI or Browser Use](https://www.agentenvframework.com/docs/agents/supported-agents.md), or a person.
* **[Tasks](https://www.agentenvframework.com/docs/tasks.md)**: a repeatable run, a DAG of steps that deploy envs and agents, prompt
  the agents and grade their work.
* **[Registry](https://www.agentenvframework.com/docs/registry.md)**: the four stores that keep what you register and what your runs
  record.
* **[Plugins](https://www.agentenvframework.com/docs/plugins.md)**: your own env types, task steps, stores and more.
* **[Sandboxes](https://www.agentenvframework.com/docs/sandboxes.md)**: where envs and agents run, from your own Docker to Modal,
  CoreWeave, E2B, Sail, Tensorlake, Tenki, Vercel, smol machines, boxd, Daytona, boat, Agent37
  or, for macOS computers, Lume.
* **[Training](https://www.agentenvframework.com/docs/training.md)**: a bundle's tasks as RL environments, trained on with TRL, where
  every rollout is an episode and every reward a verifier's score.

[Core Concepts](https://www.agentenvframework.com/docs/core-concepts.md) shows each one at work and why it exists.