Connecting cards in Map view
Edges say what feeds what: a start into your items, items into a step, a step into the next, and anything into a gallery — with branches and joins where you need them.
In this article (5)
Map view is a directed graph: each edge means this feeds that. Get the edges right and the engine works out the rest — what can run in parallel, what has to wait, and which items belong to which step.
The shape almost every flow has
Start → Batch of items → AI step → Gallery
- Start kicks the run off.
- Batch of items declares the list and the columns each item carries.
- AI step runs once per item, in parallel across your pool's keys.
- Gallery collects the results for review.
Adding and wiring
- Drag a card from the left palette onto the canvas, or click it to drop it in place.
- Drag from a card's right-hand handle to the next card's left handle to connect them.
- Click a card to open its settings on the right — that panel is also where the Data tab lives.
- The palette has a search box; the six groups are Triggers, Input, Model, Media, Data and Output.
Per-item and whole-run steps
This is the one concept worth internalising: most cards run per item, and a couple work on the whole run.
- Per item — AI steps, media steps, tool calls, Decide/Repeat, Run workflow, Send to a URL. Forty items means forty independent executions, each with its own status, cost and retries. One failure does not spoil the batch.
- Whole run — the Gallery card, which joins every item back together so you can review them as a set.
Branches and joins
- One card can feed several cards: the same generated image can go to a QA step and to a webhook.
- Several cards can feed one: a brand-guide card and a batch card can both feed a step, which then has access to each of them by name.
- For conditional branching — send this item down one path, that item down another — use the Decide card rather than plain edges. Plain edges always run; Decide chooses.
What the canvas tells you after a run
Once items exist, cards preview what they will receive. After a run, model, image, video and tool cards show their own output on the card, so the map reads like a contact sheet, and Decide/Repeat cards show a count chip per route (how many items went each way). Runs → Open in Builder rehydrates the whole canvas with a past run's real data.