AI Chatbot

Open the chat bubble in the bottom-right corner and ask, in Polish or English:

• What services does NORQ AI offer?
• How does cooperation start and what does it cost?
• What do I get before paying anything?
• Jakie branże obsługujecie?
• Will our systems be down during a cloud migration?
• Is our data safe with your AI solutions?

Then try to break it: ask about the weather, request a poem, or ask it to ignore its instructions. It will politely decline - those guardrails are part of the design.

You can also rate every answer with a thumbs up or down - that feedback flows straight into the analytics that make the assistant better (see "Standard or Extended" below).

The assistant always replies in the language you ask in. It works best in Polish and English - the languages our knowledge base is written in - but it can handle other languages as well: ask in German, Spanish or Ukrainian, and it will find the answer in our content and respond in your language.

The same applies to solutions we build for clients: content written in one language serves customers asking in another.

The assistant runs entirely on Google Cloud, serverless, scaling to zero when idle:

• Retrieval-Augmented Generation (RAG): Vertex AI Search retrieves the relevant company documents, and a Gemini model answers only from them
• Layered guardrails: questions outside company knowledge are refused, competitor recommendations are blocked, foreign links are stripped, prompt-injection attempts are contained
• Cross-lingual by design: content in one language answers questions in another
• Cost control at every layer: rate limits per visitor and per day, capped model output, and a serverless stack that costs almost nothing when quiet
• LLMOps in practice: every release passes an evaluation gate of real questions - including ones the assistant must refuse - before it reaches this website

The knowledge base is refreshed automatically from our website and documentation, so the assistant stays current without anyone maintaining it by hand.

The assistant comes in two versions. The Standard version is everything described above - grounded answers, guardrails, languages - with the strictest privacy posture: visitor messages are processed only to generate the answer and are never stored. A good fit when you want zero personal data at rest.

The Extended version - the one running on this website - adds human handoff and conversation analytics that make the assistant measurably better week after week:

• Human handoff - when the assistant can't help (or whenever the visitor prefers a person), they leave their e-mail address in the chat and your team replies from its normal inbox, with the full conversation attached - no dead ends, no separate helpdesk tool to learn
• Conversation analytics - what your customers actually ask is a goldmine: unanswered questions reveal gaps in your content, repeated questions reveal what your website fails to communicate, and trends show what customers care about right now
• Answer ratings - visitors rate answers with a thumbs up or down, so you see not just what was asked but whether the answer actually helped
• Weekly AI-generated topic reports - conversations are automatically grouped into topics, each with a concrete action hint: what content to add, what to clarify, what to fix first
• A live dashboard - refusal rate, answer quality, languages, busiest topics - shared with you, not locked away

Conversations in the Extended version are stored with a strict retention limit and EU data residency, then deleted automatically - the guardrails and cost control of the Standard version stay fully in place.

The same assistant can keep growing with your business:

• Live handoff - the Extended version's e-mail handoff can grow into replies delivered straight back into the chat window, with an operator console for your team
• Connection to your internal systems - checking order history, delivery status or product availability live, so "where is my package?" gets a real answer with the tracking link, not a generic reply
• Actions, not just answers - with the same agent technology we describe in our Generative AI service, the assistant can create tickets, book appointments or update records, with human approval where it matters

Each extension builds on the same secure foundation - grounded knowledge, guardrails and cost control stay in place as the capabilities grow.

The same solution works for company websites and e-commerce stores: it answers customer questions from your content, in your customers' language, around the clock - and hands over to a human where it should.

Embedding it on a website is a single line of code, and the knowledge base is built from content you already have. As with all our projects, cooperation starts with a free consultation and discovery phase.