Repolode
Submit
scaleapi

scaleapi/agentenv-framework

Python SDK and CLI for building, deploying and running containerized agentic RL environments and the tasks that grade agents inside them.

View on GitHubWebsite
Stars
417
Forks
57
Open issues
39
Last commit
2 hours ago

What is it?

AgentEnv Framework is a Python SDK and CLI for creating environments that AI agents can act in, along with the tasks that score what the agent did. Each environment is a containerized server that speaks the open agentenv-framework-protocol. The framework builds environments into versioned images, deploys them behind a gateway, connects an agent to them, and runs the verifier that grades the result.

The main commands are agent-env run hello, which runs a built-in task without Docker or model configuration, and agent-env up, which opens an explorer for local environments. Environments and agents run as containers on the local Docker daemon by default, and other sandbox providers are documented. Local stores work out of the box, while AWS and GCP storage backends are optional extras. Tasks can be exposed as OpenEnv environments, and plugins can add env types, providers, task steps and stores.

Key features

  • 01Builds environments into versioned container images and deploys them behind a gateway.
  • 02Connects agents to environments and scores their actions with task verifiers.
  • 03Installs with pip or uv and runs a built-in hello task with no extra setup.
  • 04Supports local stores by default, with optional AWS and GCP backend extras.
  • 05Lets installed plugins add env types, sandbox providers, task steps and stores.
  • 06Exposes tasks as OpenEnv environments that any OpenEnv client can reset and step.

Use cases

  • Building reinforcement learning environments where agents use tools and are graded by a verifier.
  • Evaluating LLM agents on multi-step tasks inside isolated containers.
  • Adding custom environment types or sandbox providers through a plugin package.
  • Running agent tasks in a specific time zone with local-hour scheduling.

Quick start

  1. 1

    Install the agent-env command as a uv tool

    uv tool install agentenv-framework
  2. 2

    Run the built-in hello task (needs no Docker, model or configuration)

    agent-env run hello

Things to know

  • Running environments locally requires a running Docker daemon.
  • Requires Python 3.11.4 or newer.
  • The project is at version 0.9.x and has no written compatibility or deprecation policy, so breaking changes are possible.

In collections

Open-source AI agent frameworks & appsMCP serversReinforcement learning frameworks & training environmentsTesting frameworks & test automation tools

Topics

#reinforcement-learning#agent-evaluation#llm-agents#sandbox#docker#mcp#a2a#agentic-ai#ai-agents#llm-evaluation

FAQ

What is agentenv-framework?
Python SDK and CLI for building, deploying and running containerized agentic RL environments and the tasks that grade agents inside them. AgentEnv Framework is a Python SDK and CLI for creating environments that AI agents can act in, along with the tasks that score what the agent did. Each environment is a containerized server that speaks the open agentenv-framework-protocol. The framework builds environments into versioned images, deploys them behind a gateway, connects an agent to them, and runs the verifier that grades the result.
Is agentenv-framework free to use? What license does it have?
agentenv-framework is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is agentenv-framework still maintained?
Actively maintained. The last commit to agentenv-framework was on Nov 7, 2026.
What can I use agentenv-framework for?
Building reinforcement learning environments where agents use tools and are graded by a verifier. Evaluating LLM agents on multi-step tasks inside isolated containers. Adding custom environment types or sandbox providers through a plugin package. Running agent tasks in a specific time zone with local-hour scheduling.
How do I install agentenv-framework?
Install the agent-env command as a uv tool: uv tool install agentenv-framework Run the built-in hello task (needs no Docker, model or configuration): agent-env run hello
Where can I get agentenv-framework?
The source code and installation instructions are on GitHub: https://github.com/scaleapi/agentenv-framework

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects