AI agents for businesses: they read, sort and draft. You approve.

An email arrives with an invoice, someone opens it, works out what it is, records it. A quote request comes in, someone reads it, looks up the price list, replies. A supplier changes its terms on its website and nobody notices for three weeks. These are ten-minute jobs, a hundred times a month. A chatbot doesn't do them: it has no access to your systems and cannot take responsibility for an action. An agent can, on one condition: every action goes through a person before it runs.

1. 1

   1. It watches

   It reads the sources you assign: mailboxes, document folders, web pages, CRM or management software records. On a schedule or on an event.

2. 2

   2. It sorts

   It works out what came in and puts it in the right class, with a confidence level. Below the threshold, it asks.

3. 3

   3. It drafts

   It fills in the entry, writes the draft reply or the alert, with data taken from your systems. It doesn't send, it doesn't save.

4. 4

   4. It waits for your yes

   Everything ends up in an approval queue. A person approves, corrects or discards. Every choice stays in the log and teaches the system.

• Agent in production on your cloud or ours, connected to the agreed sources

• Approval queue with adjustable confidence thresholds

• A log of every action: who, what, when, with which data

• Acceptance test on real cases, with the threshold written in the contract

• Optional operations plan: monitoring, source updates, threshold reviews

• 90-day warranty on the delivered system

Tell us about the process. If it's something we do well, you get a proposal with scope, acceptance threshold and price in 5 working days. If it isn't, we tell you who to call instead.

• How much does an AI agent for my business cost?

  A fixed-price project for the first process, with the acceptance test and threshold written down first, then an optional operations plan. The price depends on how many sources to connect and on the integrations with your management software. We put the number in writing after a thirty-minute call.

• How is this different from n8n, Make or Copilot Studio?

  We use them when they're enough, and we tell you so. They connect simple flows. An agent on a process with messy documents, decisions with consequences and management software without an API is something else: it needs code, checks and a log.

• What happens when it makes a mistake?

  Nothing runs without confirmation, so a mistake is a wrong draft in the queue, not an action taken. The approver corrects it, the correction stays in the log and moves the thresholds. After acceptance, for 90 days we fix systematic errors ourselves.

• Can it act on its own, without approval?

  Technically yes. We don't deliver it that way on actions with consequences (payments, messages to customers, accounting entries). On reversible actions, like an internal alert, the step can be removed after an observation period, with the numbers in hand.

• Does it connect to my management software even without an API?

  Usually yes: file imports, database writes, or whatever route your vendor supports. It's one of the first things we check, before promising you anything.

• Is my data used to train models?

  No. The system runs on your cloud or on our tenant, the data stays where we agree and does not train third-party models.

• Does it integrate with the systems we already have?

  That is the starting point. The agent reads and writes where you work today: management software, CRM, mailboxes, Microsoft 365, shared folders. If a system has no API, we go through file imports or the database. We don't ask you to change tools.