Three decisions shape everything here, and all three are unfashionable.
The market is full of products promising one assistant that does everything. In practice a general agent is mediocre at each individual job, and mediocre is unusable when the output feeds a business process. An agent that classifies tickets correctly 94% of the time is worth deploying. One that does everything at 70% is worth nothing, because you have to check all of it.
So we ship eight agents that each do one job, with a measured accuracy number attached, and we tell you what each one is bad at. Choosing the wrong agent is the most expensive mistake a customer can make here, and it costs us nothing to prevent it.
Seat pricing exists because it is predictable revenue, not because it reflects value. You buy eleven seats and use four. The vendor is fine with that.
We charge for work that happened. A quiet month costs almost nothing. A busy month costs proportionally more and you can see exactly why, run by run, in your history. There is no platform fee underneath waiting to bill you anyway, and there is no annual commitment designed to lock in revenue before you know whether the product works.
The uncomfortable consequence is that our revenue drops when a customer's usage drops. We think that is the correct incentive.
We are not building a workspace you live in. There is no chat interface, no document editor, no knowledge base, no seat for every employee. Agents take input and return structured output, and your systems do the rest.
This means we are the wrong choice if you want software that replaces a department's tooling. We are the right choice if you have a process that works and want a specific expensive step in it automated.