Web tools, mobile tools, model-backed workflows and the logic that ties them together. The builder talks to any warehouse, database or API you already run, so a team can put a working piece of internal software in front of the people who need it — without standing up a service first.

Both draw on the same data layer. Web suits dense review and admin screens; mobile suits a short task done away from a desk, such as a stock count or a site check.

Makers create and edit; Viewers open what has been shared with them and act on it. Anyone who has never opened the editor in a quarter is almost always a Viewer.

Yes — run it on our managed cloud, in your own cluster, or in an air-gapped network. The same build works in all three, and moving between them is a configuration change.

Queries run against your own stores; results are not copied into ours. Access is by role, every action is logged, and we hold SOC 2 Type II with a report available under NDA.

There is. Verified education and non-profit workspaces get the team plan at 40% off, and coursework accounts are free for the length of a term.

A long-running task that reads your data, decides on a next step and carries it out, checking in with a person at the points you mark as sensitive.

Because the cost of a run tracks how long it thinks, not how many messages it sends. Hours are the honest unit, and they make a forecast easy to sanity-check.

Heavier models cost more per second of reasoning. We pass that through at cost rather than blending it into one rate that quietly overcharges lighter work.

Repetitive work with a clear finish line: triaging a queue, reconciling two systems overnight, or drafting a first pass a person then signs off.

Only active reasoning and tool calls. Time spent waiting on a person to approve a step is not billed, and neither is a run that fails before its first step.

Start on the smallest bundle. The usage page projects a month from your first fortnight, and you can move up a tier mid-cycle without losing what you have paid for.

Yes. Add your own keys per workspace and the model spend goes straight to that provider; we then bill runtime only.

Not on the bundled rate — inference is already inside the hourly price. It is billed separately only when you supply your own keys.

It is an add-on to any paid plan. Seats stay as they are, and the hours appear as their own line on the same invoice.

Runs pause rather than roll on quietly. You will get a note at 80%, and topping up or lifting the cap restarts everything where it stopped.