Graph engineering academy · module 2 of 7 · 8 min

What is a graph?

What are nodes, edges and state, without any graph theory?

  • Read a graph: nodes do work, edges say what runs next, state is what they share.
  • Build a three-node workflow and run it.
  • Spot a cycle and the condition that ends it.

Forget the mathematics. In agent engineering a graph is a drawing you can run. Boxes do the work; arrows say what runs next; a shared notebook holds what the boxes need to tell each other.

Node
A box that does one piece of work: call a model, run a tool, wait for a person.
Edge
An arrow that says which box runs next.
Directed
Arrows point one way, so the order is clear.
State
A shared notebook every box can read and write.
Input and output
Where the run starts and where it ends.
Transition
Moving along an arrow from one box to the next.
Branch
A fork where a condition picks one arrow out of several.
Cycle
Arrows that lead back to an earlier box, so work can repeat.
Termination condition
The rule that ends a cycle: tests pass, three tries, a person says stop.

Build one

You have a request and a done node. Add an agent, select it, choose "Connect from here", then select Done. Connect Request to the agent the same way. Press Run. Then add a Check that loops back to the agent on "fail", and watch the step budget stop a cycle that has no way out.

Learning simulation
An empty graph with a request and a done node.RequestDone

2 steps ready0 model calls · 0 human · 0 failures · 0 recoveries
  • Not reachable from the input: Done.
  • Dead ends (no edge out): Request.

Press Run to execute the graph as drawn. Change a branch, a reviewer’s answer or a failure below and run again: the trace changes because the graph does.

Introduce a failure

Shared state fields

  • request
  • risk
  • plan
  • attempts
  • decision

Select a node to rename it, change its kind, move it, connect it or remove it. Add nodes with the buttons above.