Reference
This is a curated, step-through model of one agent loop, not a recording of a real system. Diagrams show roles and order, not real models, prompts or timings. The plant, the lantern and the tools are stand-ins: real agents read screens and files, call software tools, and may run for hours. Challenge answers are practice, not a measure of mastery.
Progress is remembered only in this browser, on this device, using local storage. Nothing about your visit is sent or synced anywhere; Start again clears it.
The loop shown is one simple shape: observe, think, act, check, repeat. Real agents take detours: they chain several tools in one lap, retry failed actions, ask for help mid-task, and run many loops at once. What stays true is the order of ideas: goal, look, choose, do, check.