OpenHands

Open-source autonomous coding agent you can run on your own machine

AI Coding & DevelopmentFree planOverseas★★★★☆ 4.0

Visit OpenHands official site ↗ View similar tools Compare alternatives →

What is OpenHands?

OpenHands, previously known as OpenDevin, is an open-source platform for AI agents that write code. The agent runs in a sandboxed environment with a shell, an editor and a browser, so it can modify files, run commands and check its own work rather than only suggesting text. It ships as a web interface, a command-line tool and a headless mode for scripting, and it accepts most model providers as well as local models. Because it is open source and self-hosted, teams can keep repositories, credentials and prompts inside their own infrastructure, which is the main reason it gets chosen over managed alternatives.

Last updated: 2026-09-20. This site only provides an index; for exact features, pricing, and licensing, see the official website.

Key features

Sandboxed runtime with shell, editor and browser for each agent
Web interface, CLI and headless mode for scripted automation
Model-agnostic setup supporting hosted APIs and local models
Repository-aware tasks such as bug fixing and writing tests
Docker-based deployment with configurable agent permissions
Open source under a permissive licence with active community work

Pros & cons

Strengths

Self-hosting keeps source code and credentials in-house
No per-seat licence; you pay only for model usage
The sandbox limits what the agent can reach on your machine

Watch out for

Needs real setup effort and some Docker familiarity
Success rate on complex tasks varies with the model you attach
Fewer guardrails and admin controls than commercial agent products

Best for & use cases

developers, self-hosted agent workflows, coding agent research and cost-sensitive teams

If you're comparing similar products, check the alternatives below, or browse all tools in the AI Coding & Development category.

FAQ

How is it different from a chat assistant?

It acts rather than advises. The agent edits files, runs commands and inspects the results inside a sandbox, then iterates, which is closer to delegating a task than asking for a snippet.

What do I need to run it?

A machine with Docker and enough memory for the sandbox, plus access to a model through an API key or a local runtime. The setup is documented, but it is not a one-click install.

Is it safe to point at a repository?

The sandbox is the main protection and permissions are configurable. Even so, work on a branch you can discard, and review the diffs before merging anything the agent produces.