Day 3 -- OpenHands: The Autonomous Agent Platform

OpenHands is the heavyweight of terminal agents: 81.5K GitHub stars, 7,067 commits from 519 contributors, and a philosophy that puts every agent session inside its own Docker container. It is not a CLI you install and forget. It is a platform -- a browser-based control center where agents run in isolated sandboxes, share skills across teams, and operate on schedules you define.

This day covers the full OpenHands workflow: pulling the Docker image, running your first agent, navigating the Agent Canvas web UI, hardening sandbox boundaries, building reusable skills, and deploying to the cloud or enterprise Kubernetes cluster.

What you will have by the end of today

• OpenHands running as a Docker container with Agent Canvas on localhost:3000
• An agent that completes real tasks inside an isolated sandbox
• A .agents/skills/ directory with at least one custom skill your agent uses
• Working knowledge of sandbox mount points, network isolation, and security boundaries
• A Cloud account and an understanding of when to reach for Enterprise

Lessons

1. Docker Setup & First Run -- pulling the OpenHands image, mounting workspaces, and running your first agent task in under ten minutes
2. Web UI & Agent Canvas -- the browser-based control center for conversations, files, terminals, and automations
3. Sandbox Isolation & Security -- how Docker containers contain agent actions, plus mount points, network policies, and security boundaries
4. Skills & Customization -- writing skills that give your agent domain knowledge, keyword triggers, and path-based rules
5. Cloud & Enterprise -- signing up for OpenHands Cloud, exploring managed sandboxes, and the enterprise deployment options

Next: Day 4 -- Goose: The MCP-Native Agent Harness