One run, start to finish.
A workflow runs on its own schedule. Here's one, and the only time it bills.
A step is active while it's working — including while it waits on a model or an app to answer. It stops billing the moment the run is parked on a person, a schedule or a Wait step. A step that doesn't use a model counts at least a second, and a call to a connected app has a small minimum of its own — the step that starts a run, like this one, has neither.
What you're never charged for.
The rate follows the work.
Each minute is priced by the kind of step doing it. You don't pick a plan — the step you build decides.
Every rate is on your Billing screen before you start, and every run shows what it cost when it finishes.
Agent steps run on Claude models — the long-running agent work is built on Claude's managed agent infrastructure. Model steps can mix OpenAI, Anthropic and Google.
No plans. Everything's in.
There's one way to pay, and it unlocks everything.
A ceiling you set. Spend that stops there.
If work stops, nothing you built is locked away — it stays open to read, and tables, files and skills export any time.
neonloops is in open beta: the app is live, the rate is published, there is no trial, and the first workspaces are onboarded one call at a time.
neonloops is in open beta: the app is live, the rate is published, there is no trial, and the first workspaces are onboarded one call at a time.