Skip to main content
An agent uses LazyCloud the way you do. On your machine it writes the decorated function, runs lazycloud run and lazycloud deploy, and reads the result. No command needs a browser or prompt, and --json puts the outcome on stdout as one document with a nonzero exit on failure. On a devbox the agent itself runs in the cloud, on a container whose disk keeps its files, with the same CLI.

​
Point an agent at the docs

  • https://docs.lazycloud.dev/llms.txt lists every page with a one-line summary; https://docs.lazycloud.dev/llms-full.txt holds the full text.
  • Any page URL with .md appended returns that page as Markdown.
  • https://docs.lazycloud.dev/mcp is an MCP server that searches the docs. Each page’s MCP option, next to Copy page, gives the command that adds it to an agent.
An agent with the full text in context can write a workload definition without guessing at option names.

​
Install and sign in

Approving the code login prints signs the CLI in. The sign-in lives under ~/.lazycloud, so an agent that runs the CLI afterwards is signed in too. CI and headless agents use an access token; see Authentication. The agent’s loop is the quickstart loop: edit, uv run pytest, uv run lazycloud run, uv run lazycloud deploy.

​
Run the agent in the cloud

A devbox installs Codex, Claude Code, OpenCode and Pi when its image builds, through agent_harnesses. lazycloud devbox <name> login <agent> signs one in from your laptop’s browser. See Coding agents.