Agent as a Service
A chatbot replies. An agent books the appointment, opens the ticket, updates the record. We build agents that work inside the systems you already run, and we keep them running after launch.
Can I move my appointment to Thursday?
Done. Thursday at 10:00, the confirmation is in your inbox.
Where the line is
One customer, one request: move an appointment from Tuesday to Thursday. Scroll to see how each kind of system handles it.
A decision tree someone wrote in advance. The moment a request falls outside it, the customer gets a menu that does not fit.
Reads your help center and answers correctly. Polite, accurate, and the customer still has to log in and do it themselves.
While it is still talking, it is already in your calendar, your customer record and your email. When the call ends, the job is done.
Be honest about this
General agents got very good in 2026. Before you pay anyone to build one, it is worth knowing where they stop. The limit is not how smart they are. It is what they are allowed to touch.
We are not going to pretend the free option does not exist. If your case fits one of the first two rows, take it. We would rather tell you that on the first call than sell you a project you did not need.
What “as a service” means
An agent works on day one. Then models update, your tools change their APIs, and customers ask things nobody planned for. As a service means we handle all of that, every month, so the agent you launched keeps doing its job.
Every conversation and every action is logged. We get the alert when something fails, before a customer has to tell you.
When providers release new models, we test them on your real past conversations first. Nothing changes behaviour silently.
Your CRM, calendar and ticketing tools change over time. We fix the connections before they break, not after.
A monthly spending cap and a usage report. No surprise invoices because the agent had a busy week.
We review the conversations handed to your staff and teach the agent what it missed. It handles more every month.
When something goes wrong, you talk to the people who built it, within an agreed response time.
Guardrails
Agents get things wrong sometimes, like people do. What matters is what happens next. Every agent we build works inside limits you set: what it does alone, what needs a person to approve, how much it can spend, and how to undo anything it did.
Wants to refund €340 to Maria Rossi for order #4821, damaged on delivery. This is above your €100 limit for refunds without approval.
You choose which actions need a person. The agent asks, waits for the answer, and only then acts.
Caps per action and per month. The agent cannot go over them, however hard a customer insists.
Every action is logged: who asked, what changed and when. Ready to export for your auditors.
Anything the agent changed can be reversed by your team, straight from the activity log.
Data & AI Act
Since 2 August 2026, an AI that talks to people in the EU has to make clear it is not a person. However human it looks or sounds, our agents say it in the first line, keep personal data where it belongs, and hand over to a human whenever someone asks.
Article 50 of the AI Act requires it. We build it into the first sentence of every conversation, voice or chat.
A data processing agreement with you, EU hosting where the setup allows it, and your data is not used to train shared models.
“Can I talk to someone?” always works. The agent passes the conversation to your staff with the full context.
Purpose, data sources, risks and logs are written down as we build. When stricter rules arrive, you are not starting from zero.
Designed with a Data Protection Officer on the team, not reviewed by one at the end.
Already running AI?
Many chatbots and agents went live in a hurry. Since then models have changed, prompts have drifted and the AI Act now applies. We review what is already running and tell you, in plain words, what is fine and what needs fixing first.
AI Act Article 50, GDPR and your agreements with model providers, checked against how the agent actually behaves.
Prompt injection, data leakage and what the agent could do with the permissions it has, tested with real attack patterns.
We read a sample of real conversations: wrong answers, dead ends, and handoffs that never reach a person.
A prioritised list with the effort for each fix. Your team can do it, or we can.
Works on any agent or chatbot, whoever built it. If everything is fine, the report says so.
Pricing
An agent costs something to build, something to keep running, and a little to use. Each part is priced on its own, so you always know what you are paying for and to whom. The ranges below are where most projects land.
From the first call to an agent working inside your systems.
One-time. Depends on channels and integrations.
Everything that keeps the agent working, and getting better, after launch.
Depends on how many agents and channels you run.
Every conversation costs a little: the AI model charges per message, and voice agents use phone minutes.
At typical volumes. Paid to the providers, not to us.
Every project starts with a technical assessment. You get the build price and the monthly figure in writing before anything starts, and if a simpler tool would do the job, we will tell you.
Request a technical assessment
FAQ
What buyers ask us most often about AI agents, costs, data and what happens after launch.
An AI agent built for your business and run by us after launch. We build it to work inside your systems, such as your calendar, CRM, ticketing, email and phone, and then monitor it, update the models, maintain the integrations and improve it every month. You pay a one-time build fee and a monthly service fee.
A chatbot answers questions, usually from a script or a help center. An agent finishes the task: it checks the calendar, moves the booking, updates the customer record and sends the confirmation, inside your systems and within limits you set.
For personal tasks, you should. General agents work with the open web and your personal accounts. They cannot work inside your company CRM, ERP or ticketing with your permissions, approval rules and audit log. When the agent touches customers, money or records, that is what you need.
Anything with an API: Google and Microsoft calendars, CRMs such as HubSpot or Salesforce, ticketing tools, ERPs, email, WhatsApp and web chat, and phone lines for voice agents. If a system has no API, the technical assessment tells you what is possible and what is not.
Usually 3 to 8 weeks from the technical assessment to a live agent, depending on channels and integrations. You get the timeline in writing together with the price, before any work starts.
Three parts: a one-time build, typically €1,500 – 6,000; a monthly service fee of €150 – 450; and AI and phone usage paid directly to the providers, usually €30 – 150 a month at typical volumes, with a cap you set. Exact figures come in writing after the technical assessment.
You do. The AI model and phone accounts are in your company's name, and the code and configuration are yours. If you stop working with us, you keep everything and can hand it over to another team.
It works inside limits you set: what it does alone, what needs a person to approve and how much it can spend. Every action is logged, and your team can reverse a change straight from the activity log. Each mistake becomes part of the monthly review, and we fix the cause.
Our agents say they are an AI in the first sentence, as Article 50 of the AI Act has required since 2 August 2026. We sign a data processing agreement with you, use EU hosting where the setup allows it, and document purpose, data sources and risks from day one, with a Data Protection Officer on the team.
That is part of the monthly service. We test new models on your real past conversations before switching, so the agent's behaviour never changes without anyone noticing.
Always. “Can I talk to someone?” works at any point in the conversation. The agent passes it to your staff with the full context, so the customer does not have to repeat anything.
Yes. A voice agent answers the call, understands the request, acts in your systems while still on the line and confirms by email or SMS. It introduces itself as an AI and can transfer the call to a person.
Yes, whoever built it. Our AI audit checks compliance, security, answer quality and permissions, and ends with a prioritised fix plan. It costs €600 – 1,500 one-time, with the report in 5 – 10 working days.
Still have a question? Ask it on a 30-minute call. If an agent is not the right answer for you, we will say so.
Request a technical assessmentAn agent is often one piece of the job. These are the other pieces we build.
Want AI inside your own product, not a separate agent?
AI features built into your software, with prompt injection defence, output validation and data isolation from the start.
Learn more →Worried about what your AI could be tricked into?
Security assessments and penetration testing, including attacks aimed at AI agents and the systems they can reach.
Learn more →Your systems need fixing before an agent can use them?
We clean up the code the agent will depend on, so it connects to something stable instead of something fragile.
Learn more →We'll get back to you within 1 business day.