Skip to main content
Agents are somewhat hard to build - you have to define the tools it can call, maybe find a code execution environment, figure out how to give it access to files, and the whole thing is hard to test. Trelent Agents makes it easy to build and run agents using Docker sandboxes. Your sandbox defines the tools the agent can use - just normal programs and CLI tools - and then you can create runs to execute prompts against it using your choice of harness (Claude Code, Codex, or Gemini) paired with a model. Testing is easy too, just exec into your docker container locally and install whichever coding agent you are most familiar with, and give it the prompt you want to test. It will have the exact same tools as your live Trelent Agent.

Quickstart

Build a sandbox and run your first agent.
Check out the agents-demo repo for a complete working example.