Help centre generation

An AI knowledge base generator that sounds like you wrote it

Give it your website. Get a help centre back as drafts — in your own voice, with your own policies, ready to read before anything is public.

The short answer

An AI knowledge base generator turns content you already have into help-centre articles. inrelay reads your website — policies, FAQs, documentation — works out your brand voice from it, and writes the articles and collections as drafts. You review them; nothing publishes on its own.

Looping animation: replies you have already sent are read, matched to your voice, and become a drafted help-centre article on refunds and returns.

What it prevents is the empty help centre. Everybody agrees a knowledge base would deflect most of the repeat questions, and almost nobody writes one, because the first article is forty minutes and there are thirty of them. The blank page is the whole reason the project never starts.

01 · One URL

How it works in inrelay

Open the knowledge base, hit Generate articles, and type the address of your store, your site or your docs. Everything after that is watching a progress line.

The generate-articles panel
  • It reads the pages that answer things. Policies, FAQs, shipping and returns, documentation — not your product listings, which make terrible help articles. For an ecommerce store that is usually the returns and delivery pages doing most of the work.
  • Tell it what kind of business you are. A commerce or software setting shapes what it looks for and how the collections come out.
  • You watch it work. The run reports crawling, then generating, then how many pages it read and how many articles and collections it produced.
  • It writes in your voice. The same voice profile the drafting agent uses — derived from your own site on day one, and refined from your real replies as you send them.
  • It marks what it does not know. Where a fact is genuinely missing from your site, the draft says so rather than inventing a policy you never had.
  • A website is not the only input. Knowledge sources also take a single URL, a crawl, an uploaded document or text you paste — and generation can run from one of those instead.
  • Everything it produces is a draft. There is no auto-publish mode, and there is not going to be one.
Generated drafts in the knowledge base manager
02 · It keeps writing

The second run writes itself

A generated help centre is a starting point, and the useful question a month later is which article you are still missing. inrelay answers it from the inbox: closed conversations are grouped into topics, and each topic is scored for whether your knowledge base actually covers it — covered, thin, or a flat gap.

A gap comes with the obvious next move attached. One action sends that topic’s real questions back through the same generator, in the same voice, and files the result as a draft in a review queue inside the knowledge base manager. The article you write next is the one your customers have been asking for, phrased the way they asked it.

It also notices when an article has gone out of date: the source page it was written from is re-read, drift is detected against what was there before, and a proposed rewrite arrives in the same queue for you to accept or ignore. See topic analytics for the reporting side of it.

The article suggestion queue
03 · Forty minutes an article

Why it matters

Self-service is the cheapest support there is, and the reason small teams do not have it is never disagreement about the value. It is that a help centre is thirty pieces of writing standing between you and the benefit, and the benefit only arrives at the end.

Generation moves the work from writing to editing, which is a different activity with a different failure mode. Editing thirty drafts that already know your returns window and your shipping times takes an afternoon, and an afternoon is a thing a founder can actually find. The articles then do three jobs at once — they answer customers directly, they ground the chatbot, and they make every draft the agent writes agree with your published policy.

04 · Drafts, always

How inrelay does it differently

The tempting version of this feature publishes. It would demo beautifully: paste a URL, watch a finished help centre appear, share the link. We will not build it, and the reason is the same one behind every other decision here — the moment a machine can publish a policy under your name, you have stopped being able to promise your customers anything.

So the output is always drafts, the review queue is always a person, and the generator marks its own uncertainty instead of smoothing over it. It is the same shape as approve before send, applied to writing rather than replying: the machine does the typing, and a human decides what is true.

The practical version of that promise: your first run is free on every plan, including the free one, so you can see what it makes of your site before deciding whether any of this is for you.

05 · Set-up

Setting it up

Open the knowledge base

It is in the desktop app, beside the inbox. No separate product to provision.

Type your address and pick your kind

Your store, site or docs URL, and whether you are commerce or software. Then start the run and carry on with something else.

Read the drafts

Fix the things only you know, delete what does not apply, and publish the ones that are right. Anything the generator was unsure of is marked for you.

Let the queue do the rest

From then on, gaps and stale articles arrive as suggestions rather than as a project. The help docs cover every step.

Generation is on every plan — one free run per workspace to start, then a monthly allowance on the paid plans. The knowledge base itself is included everywhere, in the widget or as a public help centre.

06 · Close neighbours

Works with the rest of the inbox.

Generation is one piece of the inrelay feature family — these are its closest neighbours.

KNOWLEDGE BASE

Where the drafts land

Collections, custom domains, SEO settings and search — the help centre the generator fills in.

See the knowledge base → AI CHATBOT

What reads them back

Every article you publish is another question the bot can answer instead of escalating.

See the chatbot → TOPICS

What is still missing

Topic analytics score your coverage and turn the gaps into the next drafts.

See topic analytics →
Start today

Your help centre, by this afternoon.

Free to start with your own inbox — every feature, no card, never per seat.

Start freeSee pricing

Questions, answered.

Does it publish articles automatically?
No, and it is not going to. Every article and collection the generator produces is a draft. You read them, edit them and publish the ones that are right — the same rule the drafting agent follows for replies.
Will the articles sound like a robot wrote them?
They are written against your brand voice, worked out from your own site to begin with and refined from the replies you actually send. They will still need your edits — that is what the review step is for — but they start from your phrasing rather than from generic support English.
What if my site doesn’t say something the article needs?
The draft marks the missing fact instead of inventing one. That is deliberate: a made-up returns window is worse than a blank you fill in yourself, because you would never have known to check.
How many times can I run it?
Your first run is free on every plan, including the free tier, and the paid plans carry a monthly allowance shown in the panel before you start. Drafts filed from a coverage gap or a stale article don’t count against it — they aren’t crawls.

Start running your DMs.

Give your customers faster answers and your team their evenings back.

Download formacOS Download on theApp Store Get it onGoogle Play

Desktop and mobile. Free for solo.