llms.txt
Compare with other sites' llms.txt files →# AgentEnv Framework
> The open-source framework Scale AI uses to build RL environments: composable environments behind an MCP gateway, any agent in any sandbox, tasks as DAGs.
These docs are also an MCP server, at https://www.agentenvframework.com/mcp.
## Start here
- [Introduction](https://www.agentenvframework.com/docs.md): What AgentEnv Framework is and the concepts it is built from
- [Getting Started](https://www.agentenvframework.com/docs/getting-started.md): Install AgentEnv Framework with uv, uvx or pip and run a first task, with no Docker, model or cloud account
- [Core Concepts](https://www.agentenvframework.com/docs/core-concepts.md): The concepts AgentEnv Framework is built from and why each one exists, starting with environments
- [Sandboxes](https://www.agentenvframework.com/docs/sandboxes.md): Sandbox = The isolated machine or container that your environment or agent resides in
- [Model Providers](https://www.agentenvframework.com/docs/model-providers.md): Model provider = The endpoint your agents and judges send their model calls to
- [OpenEnv](https://www.agentenvframework.com/docs/openenv.md): Serve a bundle's tasks as OpenEnv environments, and import a bundle with openenv import
## Environments
- [Environments](https://www.agentenvframework.com/docs/environments.md): Teaching you how to create & deploy a realistic RL environment
- [Creating your own environment](https://www.agentenvframework.com/docs/environments/creating.md): Write an environment as an AgentEnvEnvironment, with the tools an agent calls and the data plane the framework uses
- [Composing environments](https://www.agentenvframework.com/docs/environments/composing.md): Combine small environments into a multi env that an agent sees as one MCP server
- [Deploying your environment](https://www.agentenvframework.com/docs/environments/deploying.md): Package, register and deploy an environment, use it through its card, and tear its instances down
- [Choose the environment's topology](https://www.agentenvframework.com/docs/environments/topology.md): The shapes an env instance can take, from the default gateway to a server on its own or a topology of your own
- [Environment gateway topology](https://www.agentenvframework.com/docs/environments/gateway-topology.md): The default topology: one MCP endpoint for all of an env's servers, plus a virtual clock, triggers, per-role tool access and a consistency mode
- [The Environment Card](https://www.agentenvframework.com/docs/environments/environment-card.md): An Open Protocol for Environments
- [Virtual Clock](https://www.agentenvframework.com/docs/environments/virtual-clock.md): The virtual clock for ensuring environments are timeless
- [RBAC](https://www.agentenvframework.com/docs/environments/rbac.md): Decide which agents access which tools
- [Triggers](https://www.agentenvframework.com/docs/environments/triggers.md): Make an environment react during a run. When a tool call, a state or a virtual time matches, the gateway calls a tool, changes a role's tools or gives an agent an instruction.
- [More Useful Env Extensions](https://www.agentenvframework.com/docs/environments/useful-env-extensions.md): Some extensions we have found useful
## Artifacts
- [Artifacts](https://www.agentenvframework.com/docs/artifacts.md): What an artifact is, and the pages that create artifacts, use them with environments and agents, and collect new ones from a run
- [Creating your artifacts](https://www.agentenvframework.com/docs/artifacts/creating.md): Put files as file artifacts and environment artifacts, bundle them into a universe, and put new versions
- [Using Artifacts with Environments](https://www.agentenvframework.com/docs/artifacts/with-environments.md): Manage the state of your env using artifacts, capture the state of your env into artifacts
- [Using Artifacts with Agents](https://www.agentenvframework.com/docs/artifacts/with-agents.md): Give your agent files and skills using artifacts
- [Collecting Artifacts from Deployed Agents and Environments](https://www.agentenvframework.com/docs/artifacts/collecting.md): Capture the state of deployed agents and envs as artifacts, for a judge to grade or for a later run to resume from
## Agents
- [Agents](https://www.agentenvframework.com/docs/agents.md): What an agent is, and the pages that take one from a Python class to a deployed agent with extensions and skills, and human agents
- [Supported Agents](https://www.agentenvframework.com/docs/agents/supported-agents.md): Supported agent = An agent AgentEnv Framework ships, ready to deploy into any environment
- [Creating your own agent](https://www.agentenvframework.com/docs/agents/creating.md): Write an agent as an AgentEnvAgent that answers A2A tasks with Claude and the tools of the envs it is given, calling itself on sub-tasks
- [Deploying your agent](https://www.agentenvframework.com/docs/agents/deploying.md): Deploy your agent on any sandbox and connect with any env
- [Useful A2A extensions](https://www.agentenvframework.com/docs/agents/extensions.md): The urn:agentenv extensions the framework relies on to provide full functionality
- [Skills](https://www.agentenvframework.com/docs/agents/skills.md): Write a skill once, use on any agent
- [Human agents](https://www.agentenvframework.com/docs/agents/human-agents.md): A person is an A2A agent too
## Tasks
- [Tasks](https://www.agentenvframework.com/docs/tasks.md): What a task is, and the pages that take one from its steps to a stored, versioned task and its runs
- [What is a task step](https://www.agentenvframework.com/docs/tasks/task-steps.md): A task step is a node in a graph of computation
- [Creating a task](https://www.agentenvframework.com/docs/tasks/creating.md): Define your task as a DAG of task steps
- [Running a task](https://www.agentenvframework.com/docs/tasks/running.md): A task, once designed, can be run many times
- [Important Task Steps](https://www.agentenvframework.com/docs/tasks/important-steps.md): The Task Steps we have found most valuable
- [More task steps](https://www.agentenvframework.com/docs/tasks/more-steps.md): Every built-in step besides the important six, grouped by family with its key fields, how to list the registry, and how your own step joins it
## Registry
- [Registry](https://www.agentenvframework.com/docs/registry.md): The four stores that keep what you register and what your runs record
- [Document Store](https://www.agentenvframework.com/docs/registry/document-store.md): Where every versioned record, and every record of a deploy or a run, is kept
- [Object Store](https://www.agentenvframework.com/docs/registry/object-store.md): Where the bytes that registry records point at live, written once under versioned keys
- [Image Store](https://www.agentenvframework.com/docs/registry/image-store.md): Where the container images that envs and agents run are pushed, tagged by version
- [Secret Store](https://www.agentenvframework.com/docs/registry/secret-store.md): Where the values behind secret references come from, so no secret sits in your config file
## Cookbooks
- [Cookbooks](https://www.agentenvframework.com/docs/cookbooks.md): Complete recipes that take you from a fresh install to a graded run
- [Single-Turn Prompt in an RL Env Graded by an Agent Judge](https://www.agentenvframework.com/docs/cookbooks/single-turn-agent-judge.md): An accounts-payable agent works a week's invoice queue against policy in one prompt, and a judge agent grades every decision it made, from an empty folder to a score
- [Multi-Turn Conversation Between Two Agents in an RL Env Graded by an Agent Judge](https://www.agentenvframework.com/docs/cookbooks/multi-turn-two-agents.md): A support agent works an enterprise escalation across a conversation with a second agent that plays the customer, and a judge agent grades the whole conversation
## Plugins
- [Plugins](https://www.agentenvframework.com/docs/plugins.md): Add env types, providers, task steps and stores to the framework from a package of your own
- [Environment Plugins](https://www.agentenvframework.com/docs/plugins/environment-plugins.md): Add an env type from an installed package, using OpenCiv3's client on a GPU VM as the example
- [Environment Provider Plugins](https://www.agentenvframework.com/docs/plugins/environment-provider-plugins.md): Write a topology of your own that deploys the openciv3_client env onto a GPU VM and returns its record
- [Sandbox Provider Plugins](https://www.agentenvframework.com/docs/plugins/sandbox-provider-plugins.md): Run envs and agents on compute of your own, here GPU VMs with a display for the OpenCiv3 client
- [Task Step Plugins](https://www.agentenvframework.com/docs/plugins/task-step-plugins.md): Write task steps in a package, register them with an entry point and run them in a task, with OpenCiv3's match and recording steps as the example
- [Registry Plugins](https://www.agentenvframework.com/docs/plugins/registry-plugins.md): Ship a store in a plugin package, with a content-addressed object store for OpenCiv3's recordings as the example
## Training
- [Training](https://www.agentenvframework.com/docs/training.md): Train a model with AgentEnv, using the training frameworks you are already familiar with
- [TRL](https://www.agentenvframework.com/docs/training/trl.md): Train using Hugging Face's TRL
- [Unsloth](https://www.agentenvframework.com/docs/training/unsloth.md): Train using Unsloth
- [NeMo RL](https://www.agentenvframework.com/docs/training/nemo-rl.md): Train using NVIDIA's NeMo RL
- [prime-rl](https://www.agentenvframework.com/docs/training/prime-rl.md): Train using Prime Intellect's prime-rl
- [verl](https://www.agentenvframework.com/docs/training/verl.md): Train using verl
- [SkyRL](https://www.agentenvframework.com/docs/training/skyrl.md): Train using SkyRL
- [Miles](https://www.agentenvframework.com/docs/training/miles.md): Train using Miles
- [slime](https://www.agentenvframework.com/docs/training/slime.md): Train using slime
## Reference
- [CLI reference](https://www.agentenvframework.com/docs/reference/cli.md): Every agent-env command, subcommand, and flag
- [Python API](https://www.agentenvframework.com/docs/reference/python-api.md): Use AgentEnv Framework as a Python library: models, stores, and QueryBuilder
- [Docs MCP server](https://www.agentenvframework.com/docs/reference/mcp.md): Search and read these docs from Claude Code, Cursor, VS Code or any other MCP client
- [Troubleshooting](https://www.agentenvframework.com/docs/reference/troubleshooting.md): Common failures and what they actually mean
## Optional
- [Full documentation](https://www.agentenvframework.com/llms-full.txt): every page above in one file