Tool
OpenHands
An open-source platform for AI software-engineering agents — Agent Canvas as a self-hosted control center, plus agent servers and SDKs — so agents can edit code, use a terminal, and run in sandboxes you choose rather than only living inside someone else’s IDE.
What it is
OpenHands (formerly OpenDevin; org All-Hands-AI / OpenHands) is a multi-repo stack: Agent Canvas (frontend / control center), an Agent Server / SDK for running agents, and related automation pieces. You can run agents locally, in Docker, on VMs, or point Canvas at remote backends — including optional OpenHands Cloud. Docs live at docs.openhands.dev; the main Canvas tree is All-Hands-AI/OpenHands.
What it replaces
It replaces “hope the IDE agent finishes the ticket” with a more explicit agent runtime you can self-host — closer to Devin-style repo agents than to autocomplete. It can sit beside editor agents (Cline, Cursor, Copilot) rather than only replacing them. It does not replace your code review process, and it does not make a weak model suddenly safe with production credentials.
License
The OpenHands/OpenHands repository LICENSE is the MIT License. Related OpenHands repositories (SDK, automation, clients) publish their own notices — check each tree. OpenHands Cloud and paid enterprise / commercial offerings are separate products; MIT on the open repos does not automatically cover every hosted or enterprise feature you might buy.
Where it falls over
- Without a sandbox, an agent with shell access can trash or exfiltrate a machine — the project’s own docs warn about unsandboxed installs.
- Model quality and cost dominate; the open runtime still burns API tokens or local GPU time.
- Multi-service ops (Canvas, agent server, Docker, secrets) is more moving parts than installing a VS Code extension.
- Repo layout and product naming have moved quickly (OpenDevin → OpenHands → Agent Canvas); pin docs to the version you deploy.
- Cloud / enterprise add-ons are not “all MIT forever” just because the core LICENSE says MIT.
Who it's for
Teams that want auditable, self-hosted coding agents with model choice and are willing to operate sandboxes. Researchers and builders comparing open agent stacks. Wrong tool if you only wanted inline autocomplete with zero ops — use Copilot/Cursor — or if you cannot isolate agent filesystem access.
Related in this cluster
- Aider — lighter Apache-2.0 terminal pair-programmer
- Cline — open coding agent inside the editor
- Cursor — proprietary AI editor / agents
- Tabby — self-hosted coding assistant with an ee/ split
Verdict
Use OpenHands when you want an open, self-hosted agent control plane and you will own sandboxes, secrets, and model bills. Prefer Docker/sandbox paths over bare-metal agent servers on your laptop. Read Cloud/enterprise terms separately from the MIT LICENSE on the open repo. Stars are not a substitute for a dry-run on a throwaway clone.
← Tools · Home · openhands.dev · GitHub