How to use OpenHands as an AI software engineer

Unlike an assistant that completes code in your editor, OpenHands takes a whole task ("add the tests for module X") and works on its own: it reads files, writes code, runs commands and fixes the errors.

1. Install

Pick OpenHands from the app catalog.

2. Access it through an SSH tunnel

OpenHands has no authentication of its own, and an agent that executes code, left open on the internet, means full access to your server. That is why it is bound to localhost only. You reach it through an SSH tunnel from your computer:

ssh -L 8000:127.0.0.1:8000 root@SERVER_IP

Then open http://localhost:8000 in your browser. See also SSH key authentication.

3. Add a model key

From the interface, you add an API key for a model (Anthropic or OpenAI). Without it the agent does not work. Complex tasks consume many tokens, so keep an eye on costs.

4. Give it a task

You describe what you want in plain language. The agent explores the project, proposes a plan and starts working, showing you every command it runs and every file it changes.

Best practices

• Work on a copy: give it access to a clone of the project, not to production.
• Review everything: read the diff before accepting changes, exactly like a pull request.
• Dedicated server: the agent runs arbitrary commands, so keep it on a server separate from your production ones.

A lighter alternative

If you only want code suggestions in your editor rather than an autonomous agent, see the AI coding assistant.