Custom AI agents

Custom AI agents for small businesses, built around one task in your process

Custom AI agents for small businesses are software that takes on one defined task in your process, such as checking damage photos or filing notes after a sales meeting. Each agent works with your data and follows your rules. When a case needs judgment, it hands that case to a person on your team.

We build agents for European companies with 10 to 500 staff, then host and maintain them. If a standard tool already does the job, we tell you before we build anything.

Last updated: 28 October 2026

Agent or chatbot

What a custom agent does that a chatbot doesn't

Off-the-shelf chatbots and assistants answer questions, and a person does the work that follows. A custom agent does a piece of that work inside your process and leaves a result your team can use.

Off-the-shelf chatbot or assistant

  • Waits for someone to type a question, then answers it.
  • Knows what it was trained on and whatever you paste in.
  • Leaves the follow-up to the person who asked.
  • Works the same way for every company that uses it.

Custom AI agent

  • Starts when something happens in your process, such as a new report, email or voice note.
  • Reads from and writes to your systems, with the access you approve.
  • Finishes the task, for example a complete case file or a scheduled follow-up.
  • Follows your rules and passes unclear cases to a named person.

An agent is often one part of a wider automation project. Our AI automation page explains how we decide which step to start with.

AI automation for small businesses
How we build one

How we build a custom AI agent

  1. 01

    We map the process first

    We sit with the people who do the task today and write down how it really runs, exceptions included. The agent is designed from that map.

  2. 02

    We define what the agent may and may not do

    Which systems it reads, what it is allowed to change, and the point at which a person takes over. You approve those limits before we build.

  3. 03

    We test it on real cases

    Before launch we run it on your past cases and in your languages, and compare its output with what your team decided. It goes live when the results hold up.

  4. 04

    We run it and maintain it

    We host it on EU servers at Hetzner in Germany, watch how it performs and adjust it when your process changes.

These steps follow the method behind all our builds: measure the process first, change one thing, then check the result against the starting point.

How we work
What it costs

What a custom AI agent costs

You pay a one-time setup fee and a monthly fee. The setup fee covers mapping the process, building the agent and testing it on your cases. The monthly fee covers hosting, maintenance and updates.

We scope both before we start and don't bill by the hour. If the build takes longer than planned, your price stays the same.

What moves the price

  • How many decisions the agent makes and how many exceptions it has to recognise.
  • Which of your systems it connects to, such as your CRM, ERP, inbox or phone line.
  • How many cases, documents or messages it handles each month.
  • How many languages and channels it has to cover.

New features and larger changes are scoped and priced separately, before work on them starts. We give you a figure once we have seen your process.

Limits

When you don't need an agent

An agent costs money to build and to run. You are usually better off without one when:

  • The task follows the same fixed steps every time. A simple automation or a script does it for less.
  • You want a website to answer common customer questions. A standard chatbot product is cheaper and good enough.
  • The task comes up a few times a month. A person handles it for less than an agent costs to build.
  • The data the agent would need is missing or spread across notebooks and inboxes. Sorting that out comes first.

If your case looks like one of these, we tell you in the first call and point you to what fits better.

FAQ

Questions about custom AI agents

What is the difference between an AI agent and an AI assistant?

An assistant answers when someone asks it something, and a person does the work that follows. An agent is given a task in your process and carries it out, for example checking a damage report or filing a meeting note. It works within limits you set and hands unclear cases to a person.

How long does it take to build a custom AI agent?

It depends on the task and the systems involved, and we agree the scope with you before we start. We don't bill by the hour, so a longer build doesn't cost you more.

Where is our data stored?

Our agents run on servers in the EU, at Hetzner in Germany. Hosting is part of the monthly fee.

Who owns the agent?

You do. Your data stays yours, and so does the system we build for your process. If a build turns out to fit other businesses too, it only becomes a product of its own with your agreement, the way TraxAgent did.

What happens when the agent gets something wrong?

Before launch we set the point at which the agent stops and passes a case to a person, for example when it is unsure about a photo or a document. Those cases go to your team for a decision. When we see the same mistake more than once, we fix it as part of maintenance.

Can the agent work with the software we already use?

We connect it to the tools your team already works in, such as your CRM, ERP, inbox or phone line. Which systems it reads from and writes to is part of the scope we agree before the build.

Who maintains the agent after launch?

We do. The monthly fee covers hosting, maintenance and updates. New features or bigger changes are scoped and priced separately, before we start on them.

Do we need a technical team?

No. We handle the build, the hosting and the maintenance. What we need is time with the people who do the task today, mostly at the start.

Which task would you hand to an agent first?

Tell us how it runs today. We will tell you honestly whether an agent fits and what it would take to build.

Pick a time that works for you. 15 minutes. No pitch.