Are there free AI chatbots?

Yes – as a free extension or open source. Free is then the licence, not the operation: database, language model usage and your working time for setup and maintenance remain. For TYPO3, dAi is such a free extension.

What does one conversation with the chatbot cost?

With knowledge-based bots using sparse context, usually well under one cent to a few cents for three questions and answers – depending on model, amount of passages and answer length. An answer cache lowers that further because recurring questions need no model call.

Why are some vendors so much more expensive?

Mostly because of enterprise contracts with minimum terms, many channels and systems, their own support team or per-conversation pricing at high traffic. For a website that is supposed to answer questions this is rarely necessary.

Is a chatbot worth it with few visitors?

If the same questions arrive regularly by e-mail or phone, yes – every answered question saves time in the team. With few visitors the usage costs are negligible anyway; what matters is inexpensive setup, for instance with the existing website content.

How do I calculate the costs in advance?

Estimate conversations per month, multiply by about three questions and three to four thousand tokens per question, and apply the token price of your model. An answer cache and prompt caching push the result down; a per-call log then shows the real numbers.

What does dAi Pro cost?

That depends on the scope: amount of content, desired features, operation on our server or yours. After the free demo you receive a fixed-price quote; ongoing AI costs stay transparent through the per-call log.

What does an AI chatbot for a website cost?

Between "free" and "six figures" lies everything with AI chatbots – and both can be true. The price depends less on the vendor than on what the bot is supposed to do, where its knowledge comes from and who maintains it. This page breaks the cost down into its components, explains the common pricing models, calculates what AI usage actually costs and names the items that are often missing from quotes. At the end comes the pricing logic of our own products dAi and dAi Pro – without small print.

The four cost blocks

Setup – connecting the knowledge sources, design adaptation, tuning the answer behaviour, legal texts, briefing. One-off; close to zero with builders, the largest item with integrations into specialist systems.
Licence or operation – the vendor's monthly flat rate or, when self-hosted, server, database and maintenance.
AI usage – the cost of language model calls, usually billed per token. It grows with the number of conversations and is often only estimated in the quote.
Maintenance – keeping content current, reviewing logs, closing gaps. Not an invoice item from the vendor, but real working time on your side – or automation that takes exactly this over.

Pricing models on the market

Monthly flat rate (SaaS) – usually tiered by pages, documents or conversations. Predictable, but with limits: whoever grows slides into the next tier.
Price per conversation or message – fair with little traffic, expensive with a lot; and an incentive for the vendor not to keep conversations short.
Enterprise contract – annual fee plus project, often with a minimum term. Sensible with many channels and systems, oversized for a website.
Project price plus usage – one-off setup at a fixed price, afterwards only the actual AI costs and optional maintenance. Transparent if usage costs are logged openly.
Open source or free extension – zero licence, but your own working time for operation and maintenance.

What AI usage really costs

Language models are billed per token – roughly one token per syllable – and prices range from fractions of a euro to a few euros per million tokens depending on the model. With a knowledge-based bot, one call consists of the question, the supplied text passages and the answer, together typically two to four thousand tokens. A conversation with three questions therefore usually costs well under one cent to a few cents – even a thousand conversations a month stay in the low double-digit euro range. It gets expensive through giant prompts, superfluous calls and premium models where a cheaper one would do. Three mechanisms keep costs small: pass only relevant passages instead of whole documents, serve recurring questions from an answer cache, and use prompt caching where the provider offers it – explained on RAG and CAG explained. How fast and at which token price four models answer the same questions is shown live by our RAG comparison.

Hidden costs missing from quotes

Content maintenance – a bot that answers from your website is only as good as the website. Outdated pages cost trust, not money – until the first customer demands a wrong promise be kept.
Legal review – privacy policy, data processing agreement, possibly a data protection impact assessment. A few hours with the data protection officer, but not zero; details in the guide Chatbot and GDPR.
Integration – ticket system, CRM, calendar: every connection costs coordination, even when the interface "exists".
Plan limits – the 500 pages, the 1,000 conversations, the second language: what happens when exceeded, and what does the next tier cost?
Switching vendors – without export and open interfaces you pay the setup a second time when switching.
Wrong answers – the most expensive cost type: wrong prices, invented conditions, annoyed customers. Grounding and source citations are therefore not a comfort feature but insurance.

Build or buy?

With open building blocks – vector database, embedding model, language model API – a chatbot can be assembled in a few days. The experience: the prototype is fast, operation is the project. Clean extraction of PDFs with tables and scans, detection of changed and deleted content, multilingualism, grounding quality, accessibility of the chat window, logs, cost control, security updates – each of these topics costs developer weeks. Self-building pays off when a development team is there anyway and the bot becomes a core product. For a website that is supposed to answer questions, a ready-made platform is cheaper in almost all cases – even after three years.

Three scenarios, three orders of magnitude

Small website with up to 50 pages, questions about services and opening hours: with a free TYPO3 extension or an inexpensive SaaS plan, running costs are in the double-digit euro range per month; AI usage is negligible with a few hundred conversations.
Mid-sized company with website, product catalogues and PDF data sheets, multilingual, with handover to sales: here setup counts – extracting documents cleanly, separating languages, configuring handover. A fixed project price in the low four-digit range plus ongoing AI costs is realistic; enterprise platforms would be oversized.
Portal with login and personal answers, connection to ticket system and customer data, on-premises operation: an integration project with several parties. AI costs stay small; integration determines the price.

Six ways to keep costs small

Start with the content that already exists – the website first, documents afterwards.
Test first, then buy – a demo with real content shows more in an hour than three presentations.
Have usage costs logged – per call, in euros; estimates do not replace numbers.
Choose the model to fit – for website questions a mid-sized European model is usually enough; premium models only where they make a measurable difference.
Automatic freshness instead of manual work – a bot that detects changes itself saves the most expensive resource: your time.
Keep switching open – export, open interfaces, replaceable model.

The pricing logic of dAi and dAi Pro

Our free TYPO3 extension dAi carries no licence fee: chatbot over your own TYPO3 pages, RAG and CAG, language model of your choice – you pay only the AI usage at your model provider and bring a PostgreSQL database. dAi Pro follows the "fixed project price plus usage" model: after a free demo with your real content you receive a fixed-price quote for your specific case – depending on the amount of content, desired features and operation on our server or yours. dAi Pro logs ongoing AI costs per call in euros, an answer cache serves recurring questions without a model call, and the language model remains freely selectable per collection. The entry point is the GDPR-compliant AI chatbot for your website – the demo is ready in 15 minutes.

Frequently asked questions about costs

: Yes – as a free extension or open source. Free is then the licence, not the operation: database, language model usage and your working time for setup and maintenance remain. For TYPO3, dAi is such a free extension.
: With knowledge-based bots using sparse context, usually well under one cent to a few cents for three questions and answers – depending on model, amount of passages and answer length. An answer cache lowers that further because recurring questions need no model call.
: Mostly because of enterprise contracts with minimum terms, many channels and systems, their own support team or per-conversation pricing at high traffic. For a website that is supposed to answer questions this is rarely necessary.
: If the same questions arrive regularly by e-mail or phone, yes – every answered question saves time in the team. With few visitors the usage costs are negligible anyway; what matters is inexpensive setup, for instance with the existing website content.
: Estimate conversations per month, multiply by about three questions and three to four thousand tokens per question, and apply the token price of your model. An answer cache and prompt caching push the result down; a per-call log then shows the real numbers.
: That depends on the scope: amount of content, desired features, operation on our server or yours. After the free demo you receive a fixed-price quote; ongoing AI costs stay transparent through the per-call log.