All-Hands-AI/OpenHands

Claude Code and OpenAI's Codex get the headlines. OpenHands does most of the same job, except it is fully open-source, self-hostable, and free. That combination is why it sits at 75,700 stars and climbing. OpenHands is an autonomous AI coding agent. You give it a task in plain language — fix this bug, add this feature, refactor this module — and it does what a developer does: reads the code, writes and edits files, runs commands in a real terminal, browses the web for docs, and iterates until the task is done. It is not autocomplete. It is an agent that takes a goal and works the problem end to end. It ships in three shapes. A CLI for terminal-first developers who want the agent where they already live. An SDK for wiring the agent into your own apps and pipelines. And a GUI for a visual, chat-style workspace. Under the hood it is model-agnostic — point it at Claude, GPT, or a local model — so you are not locked to one vendor's pricing or roadmap. The codebase is roughly 63% Python and 35% TypeScript, and it runs every action inside a sandboxed Docker environment so the agent can execute code without touching your host. The catch is the same catch every autonomous coding agent has, and it is the one worth saying out loud. An agent that can out-code you can also out-bug you. The data from 22,000 developers shows AI output rising far faster than quality, and a self-hosted agent given free rein is the sharpest version of that trade-off. Run it, but review what it ships like you would a junior engineer's PR — and pair it with a dedicated reviewer like Qodo to catch what it gets wrong. New to building with agents? Our roundup of the best AI coding tools shows where OpenHands fits against the commercial options, and our explainer on what MCP is covers how agents like this plug into external tools.