What for: an open-source developer agent (ex-OpenDevin) that writes code, fixes bugs, runs tests in a sandbox and opens pull requests.

What it is and who it's for

OpenHands (formerly OpenDevin) is a powerful open-source developer agent that writes code on its own, fixes bugs, runs tests in a sandbox and opens pull requests. For teams and engineers who want an autonomous "AI developer" with a strong SWE-bench score and the option to self-host. What sets it apart: high scores on real-task benchmarks and isolated sandbox execution.

Key features

Autonomous task solving: code, bugs, refactoring, PRs.
Isolated execution in a sandbox (Docker) — safe by design.
The agent gets a browser and a terminal to check and run things.
Any model (best results with top-tier ones).
GitHub integration (issues → PR) and CI scenarios.
Self-host and a cloud version.

Get started in 5 minutes

Run it with Docker (you need Docker Desktop):

docker run -it --rm -p 3000:3000 \
  -v /var/run/docker.sock:/var/run/docker.sock \
  docker.all-hands.dev/all-hands-ai/openhands:latest

Open http://localhost:3000, set your model and key.
Give it a task (fix a bug, add a feature) — the agent runs it and shows you the result or the PR.

When to take it, and when not

✅ Take it if: you want an autonomous self-hosted AI developer with strong results on real tasks.
✅ Take it if: you want an issues → PR loop and isolated, safe execution.
❌ Skip it → "Cline"/"Kilo Code" are better if you prefer interactive work in VS Code with control over each step.
❌ Skip it → "Aider" is better if you want a lightweight terminal helper with no Docker infrastructure.

The honest price

Open-source and free to self-host; you pay for model tokens (noticeably so with top models). There is a cloud version with paid plans. Exact amounts are on the provider's side and change over time.

Gotchas

An autonomous agent burns a lot of tokens on hard tasks — keep an eye on the budget.
You need Docker and a bit of DevOps for a comfortable self-host setup.
Results depend heavily on the model; with a weak one, quality drops.
Autonomous changes still need a PR review before merging.

🤖 Prompt accelerator

"I'm deploying OpenHands via Docker to work autonomously on the repo <LINK/STACK>. Help me pick a model that balances quality and cost, set up the GitHub integration (issues → PR), and phrase the task so the agent does it in the sandbox and opens a clean PR. Give me acceptance criteria for reviewing the result."