The problem
How it works
The OpenHands repository now ships Agent Canvas, a web and CLI front end for starting agent conversations and automations. It runs the open-source OpenHands agent by default and can switch to third-party agents that speak the Agent-Client Protocol. The README offers three local setups: an npm install with no sandbox, one Docker container with a mounted projects folder, or a mode where each conversation gets its own container. Automations can run on a schedule or on webhook events and connect to services such as GitHub, Slack and Linear; the README gives splitting GitHub issues into tasks as an example.
Results and limitations
The README describes setup and features, not a measured outcome. It warns plainly that the no-sandbox install gives the agent full access to your filesystem, and that the per-conversation Docker mode isolates conversations but not the Canvas or automation service itself.
Before you try it
Prefer one of the Docker options, mount only the project folders the agent needs, and give it a repository token scoped to those repositories. Keep the server bound to localhost; the README says exposing it on a LAN or the internet needs a strong API key and the separate self-hosting hardening guide. You bring your own model key, so usage costs depend on the provider and how long agents run. Review every pull request an agent opens before merging. Results depend on your setup and were not independently reproduced by Top 200 Repos.
Tools and repositories
- Tools
- OpenHands Agent Canvas (npm or Docker) · Docker for sandboxed conversations · An LLM provider key
- Prerequisites
- Docker (recommended) or Node.js 24 and uv for the npm install, plus an API key for the model you choose.
Try it yourself: Turn one issue into a reviewed pull request
Fill the fields to adapt the prompt. Empty fields stay as visible [placeholders].
You are working in the repository [Repository] inside a sandbox. Read issue [Issue number or link] and restate the problem and the expected behaviour in two or three sentences. List the files you plan to change and the tests you will add or update, then stop and wait for my confirmation. After I confirm, make the change on a new branch named [Branch name], run [Test command] and report the results honestly, including any failures. Do not push, open a pull request, change CI settings or touch secrets until I approve the diff. Keep the change limited to what the issue asks for.Run it in the Docker sandbox mode and read the diff yourself before approving the push.
Related videos
About OpenHands/OpenHands
One AI Agent, Multiple LLMs: Automatic Model Routing in OpenHands (opens in a new tab)
Published by the project’s official channel.
Watch on YouTube (opens in a new tab)About OpenHands/OpenHands
OpenHands, Open source, and open models (opens in a new tab)
Published by the project’s official channel.
Watch on YouTube (opens in a new tab)Independent creators. Listing a video implies no endorsement, sponsorship or partnership.
Sources
- OpenHands/OpenHands README (Agent Canvas) (opens in a new tab)
Project documentation · github.com · Undated · Source checked 12 Nov 2026, 03:30 UTC
What this source supports
- Agent Canvas runs the open-source OpenHands agent and third-party agents such as Claude Code, Codex, Gemini CLI, Pi and OpenCode, or any ACP agent
- Local setups without a sandbox, with one Docker sandbox, or with a Docker container per conversation
- Warns that the no-sandbox install gives the agent full access to the filesystem
- Automations run on a schedule or on webhook events and connect to Slack, GitHub, Linear and other services
- Works with any LLM through LLM profiles; can also run on OpenHands Cloud or Enterprise; MIT licence