Agent Canvas Overview

Run OpenHands through the Agent Canvas UI and connect it to local or remote backends.

The Agent Canvas is the Beta UI for running OpenHands from your own machine or against backends you manage. It gives you one place to start conversations, switch backends, configure models, manage MCP servers, and work with automations.

Agent Canvas is in Beta. Expect the install path, packaging, and some UI details to keep evolving.

What Agent Canvas Includes

When you run agent-canvas, you get a complete local stack:

• The Agent Canvas web UI
• An OpenHands agent server
• The automation backend used for scheduled and event-driven workflows
• A local ingress endpoint that serves the UI and routes API traffic

Key Concepts

Concept	What It Means
Agent Canvas UI	The browser interface for conversations, customization, settings, and automations
Active Backend	The backend Agent Canvas is currently using for conversations and synced settings
Local Backend	A backend running on your own machine or another machine you control
Cloud Backend	A backend backed by OpenHands Cloud services when available
Customize	The top-level area for Skills and MCP Servers
Settings	Backend-synced configuration for Agent, LLM, Condenser, Verification, Application, and Secrets

How It Fits with Other OpenHands Products

• Agent Canvas is the recommended local and self-hosted UI path.
• Local GUI (Legacy) remains documented for older Docker-based workflows.
• OpenHands Cloud is the hosted product.
• OpenHands SDK is the Python framework behind the agent system.

Where to Start

• Setup — Use the published npm package to run Agent Canvas from your terminal.
• Connect and Manage Backends — Switch between local and remote backends.
• Customize and Settings — Configure skills, MCP servers, and backend-synced settings.
• Automations — Work with the bundled automation backend.
• Self-Hosting Agent Canvas — Run Agent Canvas on a VM or other dedicated machine.