Run AutoGPT with explicit approval steps instead of --continuous unless you accept unbounded spending and actions. Save and resume agent state between runs rather than letting a loop spin indefinitely. When exposing the serve UI, set AP_SERVER_PORT deliberately and never leave the default port reachable where you do not want it.

Context: AutoGPT's official classic usage docs document the run modes. The run subcommand executes a single agent and saves its state on termination so agents can be resumed later; the CLI defaults to run when no subcommand is given. Continuous mode runs the AI with no user authorization, 100 percent automated, and the docs explicitly say it is NOT recommended: it is potentially dangerous and may cause the AI to run forever or carry out actions you would not authorize. The serve subcommand exposes an Agent Protocol API with a UI on port 8000 by default, configurable through the AP_SERVER_PORT environment variable.