Website assistant
without a language model
Answers questions about your services, searches your blog and guides visitors through a price calculator to an enquiry. The logic runs entirely in the browser — no cost per conversation, no user text sent to a model provider, no invented answers.
Rules · blog search · calculatorFirst we checked whether an LLM was needed at all.
Pay once or keep paying.
Next to “from €99 a month” a fixed price looks expensive at first. What matters is the number at the end of the term — and there the one-off investment beats any subscription, because the assistant costs nothing to run.
| Option | One-off | Running / month | Total after 3 years |
|---|---|---|---|
| Assistant Basis (this offer) | €2,900 | €0 | €2,900 |
| Assistant Plus (this offer) | €4,900 | €0 | €4,900 |
| Live-chat software with an AI moduleLime Connect Pro from €105 + AI Agent from €99 | — | from €204 | from €7,344 |
| Agency FAQ chatbot (industry range) | €3,000 – 8,000 | €200 – 400 | €10,200 – 22,400 |
| AI chatbot as a SaaS planmoin.ai Business, annual billing | — | €750 | €27,000 |
Put differently: against live-chat software with an AI module, Assistant Basis pays for itself in about 14 months; against a SaaS plan, in just under 4. All amounts exclude VAT.
To be fair: these are not like-for-like comparisons. The subscription products include features this assistant does not have — live-chat hand-off to staff, an agent inbox, ticketing, and freely generated AI answers to arbitrary questions. If you need those, you are better served there, and I will tell you so in the conversation. This comparison holds for the case at hand: a consulting assistant on your own website that answers recurring questions and leads to an enquiry.
What dropping the language model changes.
Both variants answer visitor questions. They differ in running cost, data flow and liability risk.
What the assistant does.
Six building blocks — all proven in production.
Answers about your services
Curated topics with keyword lists instead of a prompt. Every answer is written, reviewed and links to the right page.
Blog search in the chat
Search across an index generated at build time, results weighted by word matches — with no search server.
Guided price calculator
Project type and add-ons produce a price and duration range, leading straight into the enquiry form.
No user text leaves the page
All logic runs in the browser. There is no model provider that inputs could go to.
Loads only on demand
Loaded on first hover, focus or click — additionally pre-warmed during an idle phase. Never in the critical render path.
Bilingual DE / EN
Every answer exists in both languages; the language follows the page being viewed.
The measured numbers.
Values from the production build of pragma-code.de, not from a datasheet.
~14 KB transferred
The assistant module weighs about 14 KB gzipped (42 KB uncompressed), the stylesheet about 2.5 KB.
~1,150 lines of JavaScript
Plain JavaScript with no framework dependency, plus one component and one stylesheet.
Sixteen topics
Intent detection via curated keyword lists; short keywords up to four characters require an exact word match, so “AI” does not fire inside “campaign”.
Static search index
The blog index is generated as JSON at build time and shipped — no search service, no running cost.
Dynamic import
The heavy module is pulled in via dynamic import and does not weigh on page load.
Runs on any hosting
Static files are enough — no Node server, no backend, no API key in operation. WordPress, Shopify, JTL, Shopware, TYPO3 or static: all possible.
Two scopes, both at a fixed price.
The fixed price is agreed in writing after the first conversation. All amounts exclude VAT. Afterwards there is no cost per conversation — only optional upkeep of the answers. Payable in three instalments at no extra cost if you prefer (Basis €1,000 / 1,000 / 900, Plus €1,700 / 1,700 / 1,500 — on order, at the check-in and on acceptance). If you would rather hand over the upkeep, it attaches to the existing automation plan (€590 / month, up to 6 h) — optional, not a precondition.
Consulting assistant with your topics and your blog search — single language, without calculator.
Adds a guided price calculator wired into your enquiry form, plus a second language.
From the real questions to the assistant.
Content decides, not technology. That is why collecting the real questions comes first.
Collect the questions
From emails, call notes and search queries: what do prospects actually ask?
Write the answers
Each answer is drafted, approved by you and linked to the matching page.
Integration & calculator
Integrate into your site, wire up blog search, optionally the calculator through to the enquiry form.
Live & refine
After two weeks we review which questions fell through — and add exactly those topics.
Frequently asked.
Limits included — so you know when this is the wrong fit.
Isn’t this a step back from an AI chatbot?
What does it cost to run per month?
What happens with a question that isn’t covered?
Do I need a cookie banner or a DPA for this?
Does this run on WordPress, Shopify or JTL?
Can I see it first?
More on the topic.
Chatbots, automation and the question of when AI actually pays off.
AI Chatbots & Voice Agents
When a language model really pays off in support — and when rules are enough.
Read more
Agentic AI Workflows 2026
Where autonomous AI agents make the difference — and where they are overkill.
Read more
Marketing Automation 2026
How automated funnels fill the sales pipeline.
Read moreWhere the assistant can be installed.
The assistant is plain JavaScript, a stylesheet and one JSON file as its search index. So the answer comes down to two questions: are you allowed to add your own JavaScript, and how is the search index produced? It needs no particular CMS.
| Platform | Integration | Search index | Status |
|---|---|---|---|
| WordPress / WooCommerce | Script in the child theme or via a snippet plugin | Index from the WP REST API, rewritten on publish | Straightforward |
| Shopify | Added to the theme (theme.liquid) or as a theme app extension | Index generated from blog and product data via a Liquid template | Straightforward |
| JTL-Shop | Own child template (NOVA) or a small plugin | Index via export or a dedicated route in the template | Straightforward |
| Shopware 6 | Twig extension in your own theme or an app script | Index from category and blog data at theme build time | Straightforward |
| TYPO3 | Included via Fluid template or TypoScript | Index as a dedicated page type serving JSON | Straightforward |
| Astro, Next.js, Nuxt, static sites | Drop in the component, on-demand loading included | Index is created at build time — no extra service needed | Straightforward |
| Webflow, Squarespace | Custom code in head/footer — on plans that allow custom code | Index is generated and uploaded on publish cycles, not automatically per post | With limitations |
| Wix | Only via HTML embed or custom element — the assistant then runs inside a sandboxed frame | Search index has to be maintained manually | With limitations |
| Site builders without custom code | No integration possible while the plan disallows your own JavaScript | — | Not possible |
Reference status: in production the assistant currently runs on Astro (pragma-code.de and alexander-ohl.com). For the other platforms the table describes the standard integration path — which one fits your case is settled before the fixed-price quote, not after.
What do your visitors actually ask?
We look at your most frequent enquiries and say openly whether a rule-based assistant is enough — or whether your case needs a language model.