AI agent · chatbot · automation
We don't “introduce AI”. We hand one concrete job to an agent: customer-service questions, pre-screening quote requests, content production. What you see in the bottom-right corner of this page is one of them — our own agent, in production.
A chatbot answers. An agent completes a task: it searches, decides, writes, forwards — and knows when to call a human.
A 2026 AI agent is a language model (Gemini, OpenAI or Claude) plus three things: your own knowledge (documents, price list, FAQ, CRM — connected through a RAG system), tools (calendar, e-mail, forms, shop API, whatever the model may use), and rules (what it may do alone, what a human must approve). That is what turns it into a colleague rather than a toy.
We build three jobs most often, because their return is the clearest:
A custom AI system is priced after an assessment, because the order of magnitude depends on the number of integrations and the state of the data. What every project includes:
Two weeks, at the end of which you know what the system will do, what it costs and what it returns.
In sprints, with a working version every two weeks.
We don't hand over and disappear.
30 minutes: which process hurts, what the volume is, what data exists. If an agent makes no sense, we say so.
After a two-week assessment you get an architecture, a timeline and a fixed price for the pilot.
One process, on real data, with measured results. If the success criteria are not met, you can stop here.
Further processes build on the working pilot — sprint by sprint, at a predictable cost.
For Hungarian and EU clients this is the first question, rightly. Your data does not train a public model: we use enterprise APIs where the provider contractually commits not to learn from the traffic. EU data centres where the model allows it (Gemini and OpenAI EU regions, Claude via AWS Bedrock in Frankfurt). Data minimisation: the agent only sees what the task needs, and personal data is redacted from logs too. The privacy notice and the data-processing agreement are part of the project, not an afterthought.
Frequently asked questions
We give a fixed price after the assessment, because the order of magnitude depends on the integrations and the state of the data — a website customer-service agent built on your FAQ and a system wired into a CRM, a shop and a calendar differ several times over. At the end of the two-week assessment you get an exact number for the pilot.
Yes, naturally and grammatically — the Gemini, OpenAI and Claude models are all good in Hungarian. Quality comes from your own material and from testing: in the pilot we tune it on real Hungarian conversations.
It depends. If you need a website FAQ bot, an existing platform is enough, and we will tell you so. Custom development is needed when the system has to connect to your processes and data — CRM, shop, internal documents, approvals.
In EU data centres, over enterprise APIs where the provider does not learn from the traffic. Infrastructure on AWS with a serverless architecture, in your account or ours — as compliance requires.
It hands over to a person, with the full history and a summary. We set the rules together: what it may handle alone, what needs approval.
Assessment two weeks, a pilot typically 4–8 weeks, then expansion sprint by sprint. The 3 months of support count from go-live.
Tell us which process takes the most time right now — within 48 hours we tell you whether an agent is worth building for it, and what the pilot would be.
Itt vagyok!
Kérdezz bátran a SoundCam szolgáltatásairól, vagy beszéljük át a projektedet.
Beszélj velünk az ajánlatról.
