Stilops · AI agents for businessesWatch the work happen, before it happens.

Stilops lets your organization hire AI agents for a specific business function — finance, support, legal, logistics and more. Every agent is built on Claude and always works the same way: it proposes its work first, rehearses it as a test run with no side effects, and only executes once you have approved it.

In three steps. The agent writes a concrete proposal, rehearses it as a test run so you see exactly what would happen, and then waits for you. Only your explicit approval actually executes the action — decline, and nothing happens.

A rehearsal with no consequences — of this task, now, in your own environment. Not a simulation up front in a test environment to see whether the idea holds, but the same steps with the same outcome, only without anything leaving the building. The agent prepares everything — drafts, amounts, recipients — and shows you the result, but sends nothing and changes no record. So you're judging an outcome, not an intention.

No — only what can change something. An agent that can only read and look things up (without connections) produces a report right away; there is nothing to execute, so nothing to approve. Whether a task is like that, Stilops decides from the tools the agent gets, never from the wording of the task.

No. Decline the proposal and nothing happened, so nothing counts. Approve it and let the action run, and the test run and the execution count together as one task against your quota, not two. There is one limit: test runs that don't become a task are free up to three times your included tasks per month — with 50 included tasks, that's 150 — and beyond that, each further test run counts as a task. You can see where you stand on your usage page; ordinary use never gets near it. Why this way? A system that charges you for the moment you pay attention teaches you, over time, to pay less attention.

From € 99 to € 499 a month excluding VAT, depending on the role — the price per role is on this page and in the catalogue. Every role includes a number of tasks per month: 100 for Customer Service, 50 for the Secretary, 8 for the heaviest build roles. Go beyond that and you pay per extra task (from € 2.50), and only after you've agreed to it yourself; without that agreement the agent simply stops until next month.

No. There's no canvas of nodes for you to set up and maintain. You hire a role from a catalogue of 31 functions — Secretary, Customer Service, Bookkeeping Support, Legal Support, Website Builder and so on — and that role knows what its job is. All you do is judge what it puts in front of you.

The ones you connect yourself. In the portal you add a REST API with your own key (bearer token, API key or basic auth) or an MCP server; from then on it's available to every agent in your organization. Reading is always allowed, also in a test run; writing only if you enable it on the connection, and even then only in a live run you approved yourself. For Moneybird, Mollie, Sportlink, KVK Handelsregister, bol (Retailer API), Shopify, OpenRouteService, Slack, Zendesk, TopDesk and Figma the portal prefills the form and it works directly with a key from the service. Exact Online, Google Workspace, Microsoft 365 and Teams only issue OAuth tokens: you first have to register an app yourself, and a guide walks you through it. If your system has no API, there's an Integration Specialist role that writes the connecting code.

No, and that's a choice. Other vendors let routine approvals pass automatically over time, based on a confidence score. At Stilops the moment stays exactly where it is, even after years. On the good days you won't notice the difference; you'll notice it on the one day something goes wrong, and by then it's too late to ask for it back.

The mechanism is the same for every field. Organizations put Stilops agents to work in legal, logistics, real estate, construction, financial advice, retail, software and consultancy, among others — always for a well-defined function with a fixed scope.

You do, always. No consequential action runs without a human approving it. And every proposal, test run and decision is kept, so you can always review what an agent did, what it wanted to do and who approved it.

Security is the starting point: an agent cannot do anything irreversible on its own — only your approval executes anything. On top of that come the usual safeguards: two-factor authentication (MFA) you enable per account, strongly hashed passwords, http-only session cookies over https, an append-only audit trail, and only functionally necessary cookies. Payments run through Stripe, so credit-card details never reach our servers.

Create an account in the portal and hire your first agent. Shortly after, you will see the first proposal come in — and there too, nothing happens until you approve it.