OpenHands runs agents that plan and execute multi-step engineering work: reading a repository, editing files, running tests, and iterating until a task is done. Each agent works inside a sandboxed runtime rather than on the host machine.
It is the closest open equivalent to the hosted autonomous coding products, and because the runtime is yours, the blast radius of a confused agent is a container rather than your laptop.
Decisive facts
- Sandboxed execution runtime per agent session
- Repository-wide planning, editing, and test running
- Browser access for agents that need documentation
- Any model provider, including self-hosted endpoints
- Headless mode for CI and batch evaluation