AI agents that do real work, with a human in the loop
We build AI agents and assistants that handle support, sales enquiries, documents and back-office tasks for businesses in Croatia and the EU, within limits you set.
Overview
An AI agent is software that uses a language model to understand a request, decide which steps to take and carry them out in your systems - look up an order, draft a reply, extract data from an invoice, open a support ticket. Unlike a plain chatbot, it acts. We build agents for narrow, well-defined jobs, connect them only to the tools and data they need, and log every action they take.
The model is the easy part; the guardrails are the real work. We define what the agent may do on its own and what needs a person's approval, test it against real examples before launch, and protect it against prompt injection and data leaks. We use models from providers such as Anthropic, OpenAI or Google, or open-source models hosted in the EU, depending on the task and your data-privacy requirements.
When it makes sense
- Your support team answers the same questions about orders, delivery and returns all day.
- Sales enquiries arrive by e-mail and wait hours for someone to qualify them and reply.
- Staff copy data from invoices, contracts or delivery notes into the system by hand.
- You want to use AI, but not by sending customer data somewhere you cannot control.
What we build
06- 01
Support assistants
Answer questions from your documentation and order data, hand over to a person when unsure, and draft replies for review.
- 02
Sales and enquiry agents
Read incoming enquiries, ask for missing details, qualify them, and prepare quotes or CRM entries for your team.
- 03
Document processing
Extract data from invoices, contracts, delivery notes and forms, validate it and enter it into your ERP or accounting software.
- 04
Internal knowledge assistants
Let staff ask questions about your manuals, procedures and records, with answers that cite the source document.
- 05
Guardrails and approvals
Defined permissions, human approval for sensitive actions, prompt-injection defences and strict limits on what data the agent sees.
- 06
Evaluation and monitoring
Test sets built from real cases before launch, then logs, quality checks and cost tracking for every conversation and action.
How we work
-
Research
A short, fixed-scope phase: we pick one task, collect real examples and check feasibility, then deliver a written brief and a fixed price.
-
Design
We define the agent's tools, permissions, approval points and fallbacks, and choose the model and where the data is processed.
-
Build
We build the agent, test it against real cases, and first run it in draft mode, where a person approves every action.
-
Operate
We monitor quality and costs, review errors, update prompts and models, and widen the agent's autonomy only when results justify it.
No price list. A fixed price before we start.
Every project is different, so the price depends on what you need and how much of it. The first call is free of charge, and you always get a fixed price in writing before any work starts - no surprises.
- 01The first call is free and commits you to nothing
- 02You get a fixed price in writing before work starts
- 03You own the code and data once the work is paid for
Questions
-
01 What is agentic AI?
Agentic AI describes AI systems that do not just answer questions but pursue a goal: they plan steps, use tools such as your CRM, ERP or e-mail, check the results and continue until the task is done or a person needs to decide. An AI agent is one such system built for a specific job, for example processing incoming orders or handling first-line support.
-
02 How much does an AI agent cost?
The price depends on the project and its scope - the task, the systems the agent connects to, and how much testing and approval logic it needs. We do not publish a price list. The first call is free of charge, and we always agree a fixed price for the build in writing before any work starts. Model usage is billed by the AI provider, and we estimate it upfront.
-
03 What is the difference between an AI agent and a chatbot?
A chatbot answers questions in a conversation. An AI agent can also act: it reads data from your systems, fills in forms, creates records, sends messages or prepares documents, following rules and permissions you define. A support assistant that checks the real status of an order and starts a return is an agent; one that only quotes the returns policy is a chatbot.
-
04 Is our data safe with an AI agent, and is it GDPR-compliant?
It can be, if it is designed that way. We use AI providers' business APIs, which do not use your data to train models, choose EU data processing where available, and give the agent access only to the data its task needs. Personal data is minimised and logged under GDPR rules. For sensitive data we can use open-source models hosted in the EU or on your own infrastructure.
-
05 How do you protect AI agents from prompt injection?
Prompt injection is when text in an e-mail, document or web page tries to give the agent new instructions. We treat all incoming content as untrusted data, give the agent only the permissions its task requires, require human approval for actions like payments, deletions or external messages, and test the agent against known attack patterns before launch and after every change.
-
06 Which tasks are a good fit for an AI agent?
Tasks that are frequent, mostly text-based and have a clear correct outcome: sorting and answering enquiries, extracting data from documents, preparing quotes or summaries, searching internal knowledge. Many AI projects fail because they start too broad or without clear limits, so we begin with one task, measure the results against real examples and expand only once the agent proves reliable.
Related services
- 03 Business process automation Orders, invoices, reports and approvals that move between your systems on their own, with people only where judgement is needed.
- 04 System integrations: API, ERP and CRM API, ERP, CRM and accounting integrations - Minimax, Pantheon, Luceed, Synesis and more - so data is entered only once.
Which task would you give an AI agent?
Describe the task and the systems involved. The first call is free, and the price is agreed in writing before work starts.
Completely free. No commitments.
Message sent
Thanks - we'll get back to you within one business day.