Skip to main content
OpenHands Agent Canvas: A Control Center for Coding Agents and Automations
AI WorkflowProjectSource project · deployment required

OpenHands Agent Canvas: A Control Center for Coding Agents and Automations

OpenHands Agent Canvas is a self-hosted developer control center for starting coding-agent conversations, switching between local and remote environments, and configuring scheduled or event-driven automations. It supports OpenHands, Claude Code, Codex, and ACP-compatible agents.

Delivery

Upstream repository and source ZIP; deployment required

Project language

TypeScript

This is an upstream source snapshot at a fixed commit, separate from any ENHE package.

No verified installer is provided. Check the upstream repository for setup and deployment requirements.

Project details

This workflow comes from a GitHub repository. Review the delivery and verification status below before using it.

Application area

Programming & development

Project type

Agent project

License

MIT

Delivery

Upstream repository and source ZIP; deployment required

GitHub stars

92,444 · 11/2/2026

Package verification

No ENHE package available

Link-only reason: This item is available as an upstream repository link only.

Project Overview

Project Overview

Review core features, use cases, and delivery details.

Project overview

The OpenHands/OpenHands repository provides Agent Canvas, a self-hosted developer control center for coding agents and automations. It runs locally by default and can connect to agent backends deployed in Docker, on virtual machines, or in the cloud. The project supports the OpenHands agent out of the box. It can also use Claude Code, Codex, Gemini CLI, Pi, OpenCode, and agents compatible with the Agent-Client Protocol (ACP). The repository's primary language is TypeScript, and its license is listed as MIT.

Use cases

- Individual development: Start agent conversations locally to work on projects. - Team collaboration: Connect to a shared Agent Server for tasks such as code review and dependency updates while retaining a personal local environment. - Workflow automation: Configure scheduled or event-driven tasks, such as publishing reports to Slack or breaking GitHub issues into tasks. - Continuous operation: Deploy services on a cloud server so tasks can continue when a personal computer is turned off. Deployment details and security configuration require the self-hosting documentation.

What it does and problems it solves

Agent Canvas provides a shared interface for starting conversations, choosing backends, and managing automations, reducing the work involved in switching between agent environments. One frontend can connect to multiple Agent Servers and switch between local, remote, and cloud backends. The project can work with an automation service to trigger tasks through schedules or webhook events and integrate with services including Slack, GitHub, Linear, and Notion. The README also describes a multiple-Docker-sandbox mode that runs new conversations in separate containers, supporting concurrent agent execution. This repository primarily owns the frontend control center, backend selection, and local service orchestration. Agent services and automation services are implemented in separate repositories.

How to use

1. Choose an installation method from the next section and start Agent Canvas. Before first use, understand which files the selected method allows agents to access. 2. Open the running interface in a browser. The supplied material does not retain the complete access address for npm or source launches. Obtain the appropriate entry point from Quickstart in the original repository below this page; consult the same section for the Docker entry point. 3. Start a new agent conversation in the interface. OpenHands is available by default. For other agents or model configuration, consult the ACP Agents and model settings documentation referenced by the original repository. The supplied material does not include complete configuration steps. 4. To connect another execution environment, add a backend directly from the interface and switch to it. Consult the corresponding backend documentation for its configuration details. 5. For scheduled or event-driven tasks, consult the Automations documentation to configure automations and third-party service connections. The supplied material describes these capabilities but does not provide complete operating instructions. 6. When using multiple Docker sandboxes, choose separate project directories or Git worktrees for concurrent tasks to avoid conflicting file edits. Existing local conversations are not automatically converted to container conversations. This mode isolates conversation execution; it does not sandbox the entire Canvas or automation service.

Installation

The README provides the following alternative installation paths. Choose the one that fits your needs; you do not need to complete all of them. 1. Install directly on the host: Prepare Node.js 24 or later and uv, run `npm install -g @openhands/agent-canvas`, then run `agent-canvas`. This starts the full local stack by default. Warning: this method runs agent-server directly on the host, giving the agent full filesystem access. The supplied material does not include installation instructions for Node.js or uv. 2. Use a single Docker sandbox: Install Docker first. Use Docker Desktop on macOS or Windows, or Docker Engine or Docker Desktop on Linux. Create a host directory containing your project folders and assign it to `PROJECTS_PATH`. Then follow the Docker launch command in the original repository's Quickstart to configure ports and directory mounts. The agent can access every project under that directory. Equivalent Windows commands are in `README.windows.md`. 3. Use multiple Docker sandboxes: On macOS or Linux, prepare Node.js 24 or later, uv, and a running Docker installation. Ensure the current user can execute `docker` commands. Run `npm install -g @openhands/agent-canvas`, followed by `OH_CONVERSATION_RUNTIME=docker agent-canvas`. New conversations run in separate containers. For Windows WSL 2 instructions, consult `README.windows.md`. 4. Run from source: Prepare Node.js 24 or later, the repository-pinned npm 11.12.1, and uv. Run `npm --version` to check the version, then obtain the clone address from the original repository below this page and clone the project. From the parent of the cloned directory, run `cd OpenHands`, followed by `npm ci` and `npm run dev`. This method also gives the agent direct access to the host filesystem. 5. Before exposing the service to a local network or the internet, consult `docs/SELF_HOSTING.md` and the README's API-key guidance and complete the relevant security configuration. Do not use the local quickstart configuration directly for an internet-facing deployment.

Source and usage notes

- Original repository: https://github.com/OpenHands/OpenHands - Source revision: 172d7f180096cfe06f1a875d2fcbbcb1fae2b6bb This listing links to the upstream repository. It does not provide a verified download package, and the complete workflow has not been independently validated. This description was organized and translated with AI from public upstream material. Refer to the original repository for capabilities, setup requirements, costs, and license terms.

Changelog

0 records

Important version notes will be added when availability, delivery details, or usage guidance changes. Review the current page information before purchase or use.

Share This Page

Scan the QR code to view this page, or copy its link to share.

User comments

Log in to comment.

FAQ

How can I access this project?

Choose View Original Project to open the upstream GitHub repository. Follow its setup instructions to deploy the source.

What should I check before using it?

Check the upstream setup instructions, runtime requirements, version, and the verification status shown on this page. A repository link does not mean ENHE has verified the complete workflow.

Does access include third-party services and permissions?

Check the upstream license and each service provider's terms. Model APIs, accounts, hosting, and other dependencies may have separate requirements or charges; follow the stated delivery scope.