How it works
A shared core, then everything at once
Running agents in parallel is easy. Making four independent outputs look like one brand is the part that needs a design — this is that design.
Your brief becomes a run
Submitting freezes a graph — which agents run, in what order, with which versions — and that snapshot is what executes. Editing a template later never rewrites what an earlier run produced, so a kit you downloaded stays reproducible.
brand-core resolves first
One agent derives the palette, typographic scale and voice from your brief. Nothing else starts until this exists, because everything else is downstream of it — that ordering is the entire reason the deliverables cohere.
Specialists run concurrently
Letterhead, business card and social templates start together the moment the core lands. Each reads the core rather than re-deriving its own interpretation, and each produces its artifact independently.
Deliverables arrive as they finish
There's no final assembly step to wait on. A slow or retrying agent delays its own artifact and nothing else, and you can watch each one land in the run view.
What that buys you
Two properties that follow from the design
You pay for what arrives
Credits are held when a run is submitted and settled when it completes. Work that fails is refunded rather than billed — the net cost of a run is the cost of the deliverables you actually received.
Agents are versioned
Each agent is registered and versioned independently. Improving the business-card agent can't silently change what the letterhead agent does, and a run records which versions produced it.
The fastest way to understand it is to run one
The Free plan includes enough credits for a complete kit, and doesn't ask for a card.