Not a chat widget bolted on, not an API key and a job done.
The part that matters: the model can run on your own server, and your data does not leave.
Most people who say "I want AI" actually mean level one. Here are all three so you can see where you land.
Producing site content faster
Your site answers customer questions itself
$530launch price
Data must not leave your server
From around $1,360
This table decides where your data goes, what you pay, and how much control you keep.
Swipe the table sideways to see all columns →
| Cloud API (the usual AI chat) |
On-premise (what I build) |
|
|---|---|---|
| Where data goes | Your product material and customer questions land on someone else's server | Everything stays on your own server |
| Compliance | Fails audits where data residency matters | Meets data-does-not-leave requirements |
| Cost shape | Priced per call — the more you use it, the more you pay | One-off deployment plus hardware; usage is free |
| Upfront | Low, works once connected | Needs hardware; assessed before quoting |
| Control | If they change pricing, the API or retire a model, you follow | You decide the model, version and update cycle |
| Offline | Requires an internet connection | Runs on your internal network |
| Maintenance | Handled by the provider | I deploy and document it; managed care available |
| Best for | Non-sensitive material, tight budget, wanting a trial | Factories, healthcare, legal, government, or long-run cost sensitivity |
A buyer asks "what is your MOQ" or "do you have CE" at three in the morning. Time zones being what they are, you cannot watch the site all night.
AI answers from your knowledge base first, escalates to your email when unsure, and you pick it up in the morning.
You have 200 products and a buyer has to read twenty pages of PDF to find a pressure rating. Most give up after two.
Feed in the spec sheets, and "working pressure of DN50" returns an answer in three seconds.
Translating product pages one language at a time is where most companies stop, which is why so many sites stay English-only.
AI drafts, a human finalises. Roughly a third of the cost and ten times the speed.
Better to say this up front than to take your money and build something useless.
Legal opinions, medical diagnoses, financial audits, tax filings — anywhere a wrong answer causes real damage.
AI can draft and organise. A person must decide. I label this explicitly.
Five FAQs and three products. A plain FAQ page is faster and more accurate — this is overkill.
Wait until you have twenty-plus items. I will tell you that straight.
"How much stock is left" and "where is my order" — AI does not know. That means integrating your ERP and shipping systems.
Possible, but it needs system integration, which roughly doubles the work and the price. I will say so first.
Built alongside the website
The site answers for you
$530launch price
Data cannot leave your server
Hardware assessment first
Not sure which level? Send me what material you have and I will tell you which one makes sense — and if none does, I will say that.
Send the scenario. I will tell you which level, roughly what it costs, and what you need to prepare.
Scan and say "AI" so I know what it is about. Replies within 4 working hours.
On your phone: press the QR code to save it to your album, then open WeChat and use Scan, then Album.