Answers from your data — not from the internet.
An assistant that answers from your own data, uses your systems when a number is needed, and is built to say so when the answer is not there. For visitors on your site, for your team inside the CMS or CRM — or both.
Teams answering the same questions again and again. Teams whose answers live in many different documents. Teams that tried a generic bot and could not trust its answers.
Not a fit if you only need a contact form — an assistant is the wrong tool for that.
Which components a project needs is agreed in the proposal; none is included by default.
Knowledge base
Your documents, pages and price lists prepared and indexed so the assistant finds the right passage. How and how often it is updated is agreed.
Assistant for visitors
On your site: answers about products, help with choosing and with preparing a request. The request can go into a CRM if one is connected.
Assistant for your team
Inside the CMS or CRM: finds materials and data, drafts, and carries out the actions you have agreed it may take.
Answer checks
Answers tied to retrieved sources, dialogue logs, review and correction, usage limits and cost control.
An AI assistant is planned during the build of a new site, or added later to a site IndexDock built — without buying the site again. It is not included in the site's starting price and not listed on the pricing page: its data sources, actions and integrations are agreed per project. Integrations shared with a CRM or a calculator are estimated once. Model usage and other AI services are external costs.
What does the cost of an AI assistant depend on?
On what it has to know and do: how many documents and data sources it answers from, whether it only answers or also acts through your systems, whether it serves visitors, your team or both, and which integrations it needs. Model usage is a running cost paid to the AI provider, like hosting. Assistants are not listed on the pricing page; scope and cost are agreed per project.
Can we train a chatbot with your own data?
Usually it does not need training. The common approach is retrieval: your documents are indexed, the relevant passage is found at the moment of the question, and the model answers from that text. When a document changes, the index is updated and the assistant answers from the new version — no retraining.
Is there a way to stop AI from hallucinating?
Not completely. What reduces it: answering only from retrieved passages of your own documents, showing the source, and giving the assistant a clear way to say it does not know. Instructions alone — "do not make things up" — are not enough. Logs and regular review catch the answers that still go wrong.
Why do RAG systems fail?
Often in retrieval rather than in the model. Documents split in the wrong places, so the answer sits across two chunks and neither is found whole. No refusal path, so when nothing relevant is found the model answers anyway. And an index that is not kept up to date, so the assistant repeats outdated information.
Is an AI agent the same as a chatbot?
A chatbot answers. An agent acts — it checks stock, calculates a price, books a slot or files a request by calling your systems. From outside the interface can look the same. The difference is whether it can only tell you something or actually do it, and that difference defines most of the work.