Graphwright

Graphwright is a sheet for designing the graphs behind AI agent systems: goal decompositions, workflows, agent capabilities, team structures, and communication patterns. Place boxes, connect them with typed arrows, and give every step the context it needs: what it does, its inputs and outputs, its constraints. When the design is ready, export one text bundle that any language model can read, with a Mermaid diagram, a JSON record set, and a card for every node. The vocabulary follows the research paper "Graph Engineering in the Era of LLM Agents". Each graph belongs to one layer and offers that layer's node and edge kinds, so the structure means the same thing to you and to the model. When a step fits none of them, a Custom box takes your own name for it. A worked Getting started example shows the shape every workflow needs before an agent runs: intake, context gathering, planning, a routing decision, agents, merging, verification, memory, delivery. • Five graph layers with a starter template each • Plain white sheet with grid, rulers, pinch zoom • Move, connect, and Apple Pencil ink modes • A details panel for every box and arrow • Custom boxes with your own type name and color • Export to Mermaid, JSON, and node cards in one bundle • Graphs sync between your iPad and Mac through your own iCloud • No accounts, no analytics, no servers of ours