Skip to content

EXPLANATION

How Graph Builder works

Graph Builder makes an agent workflow visible: the steps it takes, what happens after each success, and how it handles a failure. Save a graph so you can run the workflow again.

Last updated

Read as Markdown

Agent graphs

An agent graph is the complete structure of an agent’s steps and connections. Every graph starts at Start and finishes at Done or Failed. When a step succeeds, the graph follows its green connection; when it fails, the red one.

This is the graph that opens when you choose New Graph in Graph Builder.

Steps

Each card on the canvas is a step. Add steps from the palette on the left, and change the selected step in the panel on the right.

  • Agent steps hand work to an agent. Choose a classification, such as researcher, coder, or reviewer, and write the instructions for that step.
  • Script steps run a program. Its exit code decides which connection the graph follows.
  • Your graphs are graphs you have saved. Add one to another graph as a single step.

Connections

To connect two steps, drag from a step’s port to the step that runs next. A green port leads to the next step after a success; a red port leads to what handles a failure, such as Failed or a step that recovers.

A connection back to an earlier step creates a loop: repeated behavior within the graph, which runs up to a set limit. Graph Builder warns you when a step has no failure route.

Guides

Start from a built-in graph to understand an existing flow, or create your own. Run the saved graph with a goal, follow its progress, and check its results.

Build your first AI agent graph walks through creating, connecting, saving, and running a graph. Publish to the Marketplace covers sharing a graph as an agent or submitting a plugin repository. Creating agents does not require a Panolayer subscription.