OpenHands

★ 79.1k · GitHub →

Self-hosted control center for coding agents and automations.

What is OpenHands?

OpenHands (formerly OpenDevin) started as an open-source "AI software engineer" and has evolved into a self-hosted developer control center for coding agents. Rather than being one agent, it is a platform: an agent server with a REST API, a web GUI, a CLI, and an automation server for schedules and webhooks — all of which can run on your laptop, a spare machine, or a cloud VM.

The architecture is its differentiator. Agents run in sandboxed environments (Docker is the recommended isolation), and the platform is deliberately agnostic on both models and agents: it works with essentially any LLM, and via the Agent-Client Protocol it can even orchestrate other agents like Claude Code, Codex, or Gemini as backends. Integrations with GitHub, Slack, Linear, and Notion let agents pick up work from where your team already tracks it.

The core is MIT-licensed (an enterprise/ directory carries a separate license), and the company behind it sells OpenHands Cloud for hosted execution. The honest trade-off: you get maximum control and ownership, but there are more moving parts than a single-binary CLI agent — expect to spend time on setup, sandbox configuration, and choosing models before you get to the actual coding.

Key capabilities

  • ▸ Sandboxed agent execution — Docker isolation keeps agents away from your host filesystem
  • ▸ Runs anywhere: local GUI/CLI, headless server, or a cloud VM that keeps working while you sleep
  • ▸ Model-agnostic (any LLM) and agent-agnostic via the Agent-Client Protocol
  • ▸ Automation server: schedule agents or trigger them from webhooks, GitHub, Slack, Linear, Notion
  • ▸ MIT-licensed core with an optional commercial cloud (OpenHands Cloud)

When to use it

Pick OpenHands when you want agents running on infrastructure you control — asynchronous background work, scheduled automations, webhook-driven fixes — with any model and hard sandbox isolation. It suits teams with compliance requirements or anyone building an always-on agent workstation. Don't pick it for quick interactive pairing on your laptop: a single CLI agent like Claude Code, Gemini CLI, or Aider gets you coding in one command with none of the platform overhead.

Alternatives

Claude Code ★ 135.4k Cline ★ 64.2k Aider ★ 46.9k Gemini CLI ★ 105.7k

Frequently asked questions

Is OpenHands free and open source?

Mostly yes. The core is MIT-licensed and free to self-host; only the enterprise/ directory carries a separate commercial license. You pay for the LLM you connect (bring your own API key or local model), and optionally for OpenHands Cloud if you want hosted execution instead of running your own server.

What is the difference between OpenHands and Claude Code?

Claude Code is a single terminal agent tied to Claude models; OpenHands is a self-hosted platform that runs agents in sandboxes, works with any LLM, and adds a GUI, scheduling, and webhook automations. OpenHands can even drive Claude Code as one of its agent backends via the Agent-Client Protocol.

Does OpenHands require Docker?

No, but it is strongly recommended. You can install via npm and run agents directly, though then they have full access to your filesystem. Docker mode sandboxes the agent with volume mounts, which is the safer default for autonomous work — especially for scheduled or webhook-triggered runs you are not watching.

Star count updated 2026-07-02 · refreshed weekly from the GitHub API · browse all tools →