Welcome to Uni-Agent’s documentation!

Uni-Agent is a framework for building, running, and training long-horizon agent workflows. It provides persistent sandboxes, tool-based interaction loops, and verl integration for scalable reinforcement learning.

The documentation is organized into three sections:

  • Quickstart — the main path from zero to training: install Uni-Agent, launch a sandbox, build a simple tool-using agent, scale interaction across many tasks, and train with fully asynchronous RL.

  • Guides — deeper walkthroughs and config references for going beyond the quickstart.

  • Blog — long-form posts: design notes, results, and write-ups of new features.

Uni-Agent overview