How to read a workflow's graph

When you need this

Use this whenever you want to see what’s actually in a flow before you dig into one step — how many nodes it has, how they depend on each other, where the loops and branches are, or what a specific node’s prompt or schema looks like. Studio has two screens for this, reading the same underlying node data: the canvas, a visual graph you navigate by clicking nodes, and the node list, a flat list you scan top to bottom. Pick whichever matches how you’re already looking at the problem.

Steps

• Open a flow and you’re on its canvas (the screen Studio lands you on — see your first workflow). Every node is a card. An ordinary step is a small card with a colored bar for its kind, the node’s kind tag and name, a subtitle (the agent and inference for an llm node, the function name for a code or tool node), a dot if it has an unresolved problem, and small counts of its inputs and outputs. A loop, switch, parallel, or map node is drawn as a larger container card holding its member nodes, tagged with how many members it has.
• Read the edges: a solid arrow is an ordinary dependency between two nodes; a dashed line looping back is a loop node’s repeat edge. A floating legend (top right) and a zoom control (bottom left) sit over the graph without covering it. Cards aren’t draggable and nothing here is editable — clicking is the only interaction, and it selects a node rather than moving it.
• Click any card to open the node inspector, a panel that slides in from the right without covering the canvas. It has up to six tabs: definition, input, prompt, output, config (labeled agent instead when the node has agent settings), problems — though prompt only appears if the node has one, and problems only if it has any.
• Flip the formatted/raw switch near the top of the inspector to change every section on the current tab from a human-readable view to the same data as raw JSON, and back. It’s the same node data either way, just two ways of reading it — useful when the readable view hides a field you need to see exactly as written.
• Switch to the node list, at the same flow, when you’d rather scan every node than click through the graph one dependency at a time. It’s a flat list on the left, indented to show which container a node belongs to, with a detail panel on the right for whichever node you click.
• The node list’s detail panel isn’t the canvas inspector reused — it’s a separate part of Studio, built for scanning rather than navigating. It shows the same node, but with a different tab set: definition, prompt, schemas, problems, and no separate input, output, or config tabs. Its definition tab folds in what the canvas splits across input and output — the node’s bindings, its declared outputs, and its upstream and downstream links, all in one place. schemas covers the node’s input, output, and form JSON schemas together, one section each.

Example

Terminal

aqven new my_project --template showcase

In the support_case flow this creates, the triage node is an llm step with one upstream node (prepare) and about a dozen downstream nodes that read its output. On the canvas, its card is tagged llm, subtitled gemini · triage — its agent and inference id. Click it and the inspector opens on definition, showing its description and that single upstream link to prepare; switch to prompt to read the actual prompt text, formatted or as raw JSON. Open the same flow’s node list and click triage there instead: the tabs read definition, prompt, schemas, problems, but the prompt tab shows the identical text, because both screens read it from the same file.

See also

• How to call a model, How to write a step in Python, How to reuse a flow as a step, How to give an agent a tool, How to pause for a person, How to route by a value, How to repeat a step with a limit, How to run a step over a collection, How to branch into parallel steps, and How to narrow a dynamic value to a type — each teaches the fields the inspector’s tabs are showing you.
• Node specifications — the full generated field reference behind every tab.