Adoomi crawls your public website pages and takes PDF, Word and plain-text uploads, then shows you the whole source list before any of it answers a customer. Retraining runs on a schedule and again when a source changes, so a corrected page reaches the assistant without anyone rebuilding anything.
Adoomi reads your public website — services, policies, FAQs, opening hours, contact routes — and builds the assistant's knowledge out of what is actually published there. There is no separate knowledge base to write from scratch.
Price lists, policy documents, service sheets and the internal FAQ your team works from go in as PDF, Word or plain-text uploads. Facts you know but have never published become answerable without publishing them on the site first.
You see which sources the assistant is drawing on and how much each one contributes, and you approve what it may use. Nothing answers a customer that you have not looked at, and a source that is misleading it can be taken back out.
Retraining runs on daily and weekly cycles, and again when your sources change. Fix a wrong sentence on your website and the correction flows through to the assistant on its own — nothing is rebuilt and nobody re-teaches it by hand.
Give Adoomi your web address and it crawls the public pages. For most businesses that is the bulk of the knowledge done in one step, and it is the part of setup you do not have to prepare anything for.
Anything customers ask about that you have never published — a price list, a returns policy, a service sheet — goes in as a PDF, Word or plain-text file. Watch your plan's allowance for source content while you do it: Free 50 KB, Local 10 MB, Growth 20 MB, Scale 30 MB. On a large site, pick the pages customers actually ask about — services, prices, policies — and leave the rest out.
Before go-live you see every source and how much each one is contributing, and you read the answers the assistant proposes from them. Remove a source that is pulling it the wrong way, and change any answer you would not want a customer to read.
Paste the single line Adoomi gives you into your site and the chat window is live. Connecting the site, reviewing and adding that line takes ~10 min in total. After that the sync looks after itself; you come back only to add or replace a source.
Reads the public pages of your site — services, policies, FAQs, contact and opening details — and lists what it took, so you can see the coverage rather than assume it. Limit: public pages only; member areas, customer portals and anything behind a login cannot be crawled and have to be uploaded as a document instead.
Takes PDF, Word and plain-text files for the facts that never made it onto a page, such as a price list, a policy document or a service sheet. Limit: uploads are static — revise the document and the assistant keeps answering from the version you gave it until you upload the new one.
Shows the conversations the assistant actually had, so a poor answer can be read in context and traced back to the source worth fixing. Limit: reviewing is a manual read — nothing scores your content for you or edits a source on your behalf.
Collects the questions the assistant could not answer, which is the most honest list of what to write or upload next. Limit: a gap records that the question was asked, not the answer to it — someone still has to publish or upload the missing fact.
Your teamPricing page edited — next sync re-reads it.
CustomerHow much is a standard service now? I have seen an older figure quoted somewhere else.
AdoomiThe current price is the one on our pricing page as it reads today — it was updated this week, so the figure you saw last month is out of date.
CustomerAnd do you do weekend slots? Someone told me you had started.
AdoomiWeekend slots are not covered by anything we have published, so I will not guess at it. Leave me your name and email and I will pass the question to the team, who can confirm properly.
Your teamEmail alert: enquiry about weekend availability, not covered by any current source. Publish it or answer this one by hand — either way it shows up as a knowledge gap.
The price answer changed because the page changed. Nobody retrained the assistant: the scheduled sync re-read the pricing page and the new figure came with it. The weekend question is the other half of the same rule — no approved source, no answer — so it went to a person and landed on the list of content worth adding.
The crawl reaches what any visitor can reach. Member portals, customer accounts, booking back-ends and anything behind a login stay invisible to it. Where a gated fact matters to customers, it has to arrive as an uploaded document instead.
Stock levels, prices and availability are as current as the last sync, not as current as your till. A business with fast-moving inventory is better served by an assistant that points at the product page than one that promises a number it cannot see. Shop-style questions of that kind sit with the product and order answers feature page.
Source content is capped per plan: Free 50 KB, Local 10 MB, Growth 20 MB, Scale 30 MB. A large site, a deep blog archive or a full catalogue can exceed that, and the honest answer is that you choose what matters rather than crawling everything.
A document is a snapshot of the day you uploaded it. Nothing watches the file for edits, so a revised price list or policy has to be uploaded again. The website crawl re-reads your pages on its own; your files do not re-read themselves.
Syncing fills nothing in. Where an answer is not in an approved source, the assistant hands the question to a person and captures the enquiry rather than guessing, and it stays unanswered until someone publishes the page or uploads the file.
Everything synced is EU-hosted with GDPR export and deletion built in — remove a source and it stops informing answers. The chat window sets no tracking cookies.
Two sources feed the assistant: a crawl of your public website pages, and documents you upload as PDF, Word or plain-text files. Nothing else goes in, and you see the full source list, plus how much each source contributes, before any of it answers a customer.
Retraining runs on a schedule, in daily and weekly cycles, and again when a source changes. Edit a page on your site and the correction flows through on the next sync. Nothing is rebuilt, no content is re-imported, and nobody has to re-teach the assistant by hand.
Allowances run by plan: Free gives 50 KB of source content at £0, Local 10 MB at £8, Growth 20 MB at £20 and Scale 30 MB at £75 a month. A credit is one AI answer — see the pricing page for the plan ladder.
Login-protected pages stay out of reach. The crawl covers public pages only, so member areas, customer portals and account pages are invisible to the assistant. Where those facts matter to customers, export the content and upload it as a document instead.
Prices, stock and availability are as current as the last sync and never live, because nothing reads your till or stock system. Fast-moving inventory is better handled by pointing customers at the product page than by promising a number the assistant cannot see.
Unanswered questions go to a person rather than to a guess. The assistant hands over and captures the enquiry with contact details and context, and the same questions collect as knowledge gaps: your list of the pages and documents worth adding next.
Adoomi starts with your public business knowledge and routes the moments that need a person to your team.