Practical overview
When it fits
Use it when an evaluation requires an agent to act in a defined environment and the team needs to run and score tasks there. The built-in hello task offers a simple CLI starting point; local environments require a running Docker daemon.
Who it is for
- Researchers, engineers and domain experts creating realistic reinforcement-learning environments.
- Teams building reproducible environments and tasks to evaluate AI agents.
Documented uses
- Build containerized environments with defined artifacts, tools and dynamics.
- Run agent tasks inside environments and grade agent behavior.
- Extend the framework with environment types, providers, task steps and stores through plugins.
Requirements & limits
The framework is open source under the Apache-2.0 license. The documented installation options include uv and pip. Local environment runs require Docker; the built-in hello task does not.
- Python 3.11.4 or newer is required.
- A plain installation uses local stores and needs no cloud SDK.
- Explorer, Tensorlake, AWS and GCP capabilities require their corresponding optional extras.
- The explorer requires an .agentenv/config.toml file; the Tensorlake backend requires an API key in its configuration.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Install the framework and run hello
Install the agent-env CLI with uv and run its built-in hello task.
Before you start
- Python 3.11.4 or newer
- uv
Steps
Install the CLI with the command below.
uv tool install agentenv-frameworkRun
agent-env run hello.
Expected outcome
Runs the built-in hello task; it needs no Docker, model or configuration.
Guide limitations
- The hello task does not demonstrate running a custom environment.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 13 October 2026
- Original website title
- scaleapi/agentenv-framework · Official homepage title · retained in its source language