Agent Orchestration Explained
Agents as graphs, steps and checkpoints.
An interactive guide to agents as graphs: steps, checkpoints, multi-agent patterns and durable runs, on an engine checked against LangGraph.
How this is checkedEach of 39 sessions also runs in real LangGraph 1.2.14, and after every one of 71 operations its whole checkpoint history equals the engine's. How it is tested →
What you'll learn
A multi-step agent is a discrete-event system: nodes run in steps, write to a shared state, wait at barriers, save checkpoints, pause for people, crash and resume. This site takes LangGraph's execution model apart, super-step by super-step, and then durable execution: replaying a workflow's history after a crash without paying for its side effects twice. Then it scales up: multi-agent patterns compared on one task, fan-out against a rate limit, reliability with retries and verifiers, hundreds of workflows sharing one endpoint, and a pattern chosen from the simulation rather than from opinion.
The animations are driven by Agent_Loop_Sim's orchestration module, identically in Python and in your browser. Its graph runtime is checked against real LangGraph 1.2.14 in CI: 39 sessions, 71 operations (runs, crashes, resumes, interrupts, edits, time travel, Send fan-outs) and 232 checkpoints, every one equal. No language model runs anywhere.
Part of a family of sixteen companion sites in four tracks, all in the site menu above: Foundations (Decoder, Training), Hardware (Kernels, Systolic, Accelerators, Racks), Inference (Serving, Architectures, Numerics, Trade-offs) and Agents (Harnesses, Protocols, Context, Orchestration, Evals, Security). It is the fourth of the six agent sites. How this one was built, and how to check it: about.