Give an AI agent a goal and watch it work the graph

“Agentic AI” usually means a black box that goes away, does something, and hands back a result you cannot inspect. Grafy runs the agent on the canvas: it plans the steps, executes each one as a node, and leaves a trail you can read, correct and rerun.

A goal in, a graph out

Tell the agent the outcome you want. It drafts a short plan, then works through it — generating, editing, extracting — dropping each result onto the graph as it goes.

Every step is inspectable

Because the work is nodes, not a hidden loop, you can see exactly what the agent did and why. Disagree with a step? Branch from the one before it and take over by hand.

Rerun and adjust

The plan and its steps are reproducible. Swap a model, change a prompt, or re-run from any point without starting over — the agent and you share the same canvas.

Give an agent a goal and watch it work.