Agent Harness

An interactive walkthrough of a LangGraph run and the architecture around a production agent harness.

LangGraph run flow An animated walk through one agent run: client request, load state, agent node, router, human review, tools, reducer merge, the ReAct loop, writing memory, returning the result, time travel, and streaming. Client requestM6 · SDK Load stateM2 · M5 Agent nodeM1 · LLM RouterM1 · conditional edge Human reviewM3 · breakpoint Run toolsM1 · M4 · parallel Reducer mergeM2 · checkpoint Write memoryM5 · Store Return resultM2 · M6 Time travelM3 Streaming runs through every step (M3) ReAct loop tool call no tool call

Click any node to jump to its explanation, or use the controls to step through the run.

1 / 12
M6The client starts a run with the SDK