OpenHands
Developers juggling several AI coding helpers lose track of them. OpenHands gives them one control center to start agents and automate chores like turning issues into tasks.
How you use it: A developer installs it, ideally in a protected Docker setup, and opens the control center in a browser.
In plain English
The README presents Agent Canvas as the user-facing part of OpenHands. It is a developer control center for starting conversations and automating tasks. The examples it gives are generating reports that publish to Slack, and breaking GitHub issues into tasks.
It runs locally by default and can also connect to agent backends in Docker, on virtual machines, or on company infrastructure. You can optionally use OpenHands Cloud or OpenHands Enterprise. The open-source OpenHands agent is included, and the README says you can also use Claude Code, Codex, Gemini CLI, Pi, OpenCode, or any agent that speaks the Agent Client Protocol.
The README is frank about safety. The no-sandbox install runs the agent server on the machine you install it on, and the agent then has full access to your filesystem. Docker options isolate conversation work more carefully. Local listeners bind to 127.0.0.1 so a session key is not exposed on your network, unless you choose a wider bind.
Practical ways to use it
Example 1
Keep coding-agent chats in one screen
The README says you can switch between local, remote, and cloud backends from the same Agent Canvas.
Example 2
Run an agent while the laptop is shut
The quick start says a server in the cloud lets agents keep running, and makes it easier to trigger them from Slack, GitHub, or Datadog. The self-hosting guide covers the security work.
Example 3
Automate a repeating developer chore
The README describes automations that integrate with Slack, GitHub, and Linear, on a schedule or from a webhook.
Example 4
Use a coding agent you already pay for
The feature table says Canvas can run OpenHands itself or third-party agents such as Claude Code and Codex, as long as they speak the Agent Client Protocol.
Who it is for
Developer. You are expected to install Node.js or Docker and to understand that an agent may read and write files.
Difficulty: Developer setup. The shortest path is a global npm install. The safer everyday path in the README is a Docker sandbox, which has more steps.
Getting started
- 1
Choose a sandbox before you start
Option 1 in the README runs on the machine directly and warns that the agent gets full filesystem access. Option 2 uses Docker and only mounts a projects folder you choose.
- 2
For a local trial, install the CLI
The README requires Node.js 24 or later and uv. Then install @openhands/agent-canvas globally and run agent-canvas. Open http://localhost:8000.
- 3
For Docker, prepare a projects folder
The README sets a projects path and runs the agent-canvas image, publishing port 8000 only on 127.0.0.1. Open http://localhost:8000/canvas. The image tag in the README at review time was 1.40.0.
- 4
Keep the first install off the public internet
The README tells you to follow the self-hosting notes before you expose a server, and not to inject the session key if the port is reachable beyond your own computer.
What you need first
- Accounts
- A model provider key, unless you bring a local model. OpenHands Cloud and Enterprise are optional and are not required for the local install.
- Hardware
- Node.js 24 or later and uv for the npm path. Docker for the sandbox path.
- Cost
- The repository license on GitHub is MIT. Model use is billed by your provider. The README mentions OpenHands Cloud and Enterprise as optional paid infrastructure and does not print those prices in the quick start.
Cautions
- Without a sandbox, the agent has full access to the filesystem of the machine you installed it on. The README prints that warning in the direct-install options.
- Do not publish port 8000 on a public address with the session key injected. The README explains when the key is injected and when you must type it.
- On Linux Docker Engine, the permission tips include making a projects folder writable by every local user. The README says to keep that folder away from secrets.
Questions people ask
What does OpenHands do?
Agent Canvas is a self-hosted screen for running coding agents and automations. It can use the OpenHands agent or other agents that speak the Agent Client Protocol.
Who is OpenHands for?
Developers who want a control center for coding agents on their own machine or on a server they operate.
What do I need first?
Node.js 24 and uv for the direct install, or Docker for the sandboxed install, plus a language model you are allowed to call.
What should I watch out for?
The no-sandbox install gives the agent your whole filesystem. Prefer the Docker path, keep the port on localhost, and read the self-hosting notes before anyone else can reach it.