Give Stelt the knowledge behind your business
Stelt answers your visitors from information you supply — your website, documents and notes. Train it on the things people actually ask about, and its answers stay grounded in your business rather than generic AI.
Your business already has the answers
The information visitors want usually already exists — on your website pages, in your service and pricing details, your FAQs, your policies and process documents, and in things your team simply knows. The problem isn’t that the answers don’t exist; it’s that visitors can’t always find them.
Stelt turns the material you choose to provide into a conversational way to get those answers. (It only knows what you add — it doesn’t reach into private systems on its own.)
Bring knowledge in from several sources
You build Stelt’s knowledge from a few kinds of source, and they all feed one place — the knowledge Willow answers from.
One page, or more of your site
Point Stelt at a single page to train from just that page — useful when one page holds what you need. Or turn on a site crawl, and it follows your internal links to gather content from across your site, so you don’t have to add every page by hand.
It’s a focused crawl of your own domain — built to cover a full business site, not a one-page snippet, and not a search of the wider web.
One thing worth knowing: some websites restrict automated crawlers. If Willow can’t access a site, the owner or developer may need to allow crawler access before Stelt can train from those pages — a quick setup step on some sites, not something every site needs.
Add what isn’t on your website
Plenty of useful information never makes it onto a public page. Stelt lets you add that directly, as PDF or Word documents or as typed notes:
A service guide
The detail behind what you offer.
A policy document
Terms, returns, how you handle things.
A process explanation
How something actually gets done.
An FAQ document
The questions you already answer a lot.
Notes to remember
Details that live in someone’s head.
Illustrative examples of material you might supply — not promises about specific document handling.
From source material to a grounded answer
“Grounded” is the important word. In plain English: Stelt is instructed to answer using the business information it has been trained on, rather than inventing business-specific details. That’s what keeps answers on-brand and about your business.
You add a source
A page, a document, or a note.
Stelt learns it
The material becomes part of its knowledge.
A visitor asks
In their own words, by voice or text.
A grounded answer
Drawn from your material, not guessed.
Being honest: no AI is perfect, and grounding reduces made-up answers rather than guaranteeing none. Stelt is built to say when something is outside what it knows and hand off, rather than bluff.
Keeping website knowledge fresher
Businesses change — prices, hours, services. A website source can be set to re-read your pages on a schedule, so as those pages change you don’t have to keep manually re-adding the same source. It re-reads periodically to stay reasonably current; it isn’t a real-time sync, and changes you make elsewhere (in a private system, say) aren’t picked up unless they’re in a source Stelt reads.
One knowledge base, many kinds of conversation
The same trained material powers everything Stelt does. Once your knowledge is in, it answers general website questions, handles customer-support questions, works in voice or text, helps show the right content, and supports moving toward an inquiry where configured — all from one knowledge base. It’s the foundation under the whole website assistant.
What Stelt knows — and what it doesn’t
Understanding the edges makes the whole thing trustworthy:
| Stelt can know | Don’t assume Stelt knows |
|---|---|
| Information from sources you’ve added and trained | Changes you haven’t published or added yet |
| Content found in a configured website crawl | Private systems it hasn’t been connected to |
| Details from your uploaded documents | Live inventory, stock or real-time availability |
| Notes you’ve supplied manually | Anything you’ve never supplied — from the wider web |
None of this makes it weaker — it’s what makes the answers dependable. The richer and more current your sources, the more it can help.
A simple training workflow
Setting it up is manageable — not a data project:
- Add your sources — a website page or crawl, documents, or notes.
- Let Stelt learn the material so it becomes part of its knowledge.
- Test the questions your visitors ask and see how it answers.
- Update or refresh as your business changes — add a source, or let a website source re-read.
It’s a strong fit for information-heavy businesses — services with detailed processes, sites with FAQs or policy documents, and anyone answering the same questions over and over. It works for any business, trained on your own content — see it by industry.
Frequently asked questions
How does Stelt learn about my business?
You give it your information — a website page or a crawl of your site, uploaded PDF and Word documents, and notes you type in. Stelt uses that supplied material to answer visitors, so its answers are specific to your business rather than generic.
Can Stelt learn from my website?
Yes. You can train it from a single page, or turn on a site crawl so it follows your internal links and gathers content from across your site. It only reads your own public pages — it isn’t a web search engine.
How many website pages can Stelt crawl?
A crawl follows your site’s internal links across your own domain and is built to cover a full business site — well beyond a single page. It focuses on your reachable pages; it doesn’t crawl the wider web.
Can I train Stelt from a PDF?
Yes — you can upload PDF material and Stelt will read its text as a source.
Can I use Word documents?
Yes — DOCX documents are supported the same way as PDFs.
Can I add information manually?
Yes. You can type in notes or text directly, which is handy for details that live in someone’s head rather than on a page.
What happens when my website changes?
A website source can be set to re-read your pages on a schedule, so you don’t have to manually re-add unchanged sources as they change. It re-reads periodically rather than syncing in real time.
Does Stelt search the whole internet for answers?
No. It answers from the material you’ve trained it on — your pages, documents and notes — not the open web.
Will Stelt know information I haven’t provided?
It shouldn’t. Stelt is instructed to answer from your trained material rather than invent business-specific details, and to hand off when something is outside what it knows. If an answer is missing, that’s a signal to add the source.
Does training Stelt change my website?
No. Training reads your sources to build Stelt’s knowledge; it doesn’t edit your website or your documents.
How does this knowledge get used in a conversation?
When a visitor asks something, Stelt draws on the trained material to answer — out loud or in text — and can pull up a relevant page or document to go with it.
Train an assistant on your business
Start a free trial and point Willow at your website, documents and notes.
Start free trial