An agent should do the work, not just answer questions.
The conviction
Most AI tools stop at the chat window: formulate an answer, leave the rest to the human. An employee works differently: they open the system, check the documents, make the routine decision themselves, and only flag the exception. We build agentctl so an agent does the second thing, not just the first. That's the difference between a tool you operate and a colleague who takes work off your plate.
Self-hosted by principle, not by constraint
We could have built agentctl as simpler, hosted SaaS. We deliberately didn't. Swiss companies (banks, insurers, public sector, but just as much the SME down the street) decide for themselves where their data lives. Infrastructure sovereignty isn't an enterprise add-on for us, it's the default case. That makes the product harder to build. It's still the only version we can stand behind.
How we work
An agent never gets a new, separate role with us. It inherits the rights of the real employee identity it acts on behalf of. Every action is logged the same way a human's would be. Any language model can be used, no lock-in to a single provider. These three decisions run through the entire product; they aren't features bolted on afterwards.
Who we build for
We build first for Swiss companies with their own systems, their own data, and their own regulatory requirements: from SME accounts receivable to a bank's regulatory reporting review. Not just the large, regulated industries: the same logic (an agent that acts like an employee and can be reviewed just as one would) holds up in any company past a few dozen employees.
Where we stand today
agentctl is a young Swiss company, founded by Daniel Dutli and Mohammed Ajil. We're deliberately small and build alongside our first customers rather than planning everything upfront. What's running today is running in production, not a paper pilot; see the numbers on the homepage. What's still missing, we'll tell you just as plainly if you ask.