Q&A pairs
The question in your customer's words, the answer in exactly the wording you want. The fastest way to lock down a sensitive reply like a cancellation policy.
Upload your documents — your agent answers from them
An agent that doesn't know your business gives generic answers. The knowledge base is where you put what a well-trained employee knows: prices, policies, opening hours, and the questions customers keep asking — built from your own documents.
How does the agent answer from my business documents?
You upload your files — PDF or Word, scans included — and the platform reads, splits and indexes them so search works by meaning rather than by literal keyword. When your customer asks on your text chat channels, the agent answers from the passage that fits; with no clear answer it escalates or logs a callback instead of inventing one. Before any of that, you can test any question and see which file the answer came from.
What can go into the knowledge base?
Five content types, mixed however you like — and you decide what each agent may read.
Files
PDF, Word (.docx), plain text (.txt / .md). Price lists, service guides, handbooks — uploaded as they are, with no rewriting.
Links
A page on your own site added as a source instead of copied across — re-ingest it whenever the page changes.
Snippets
A short fact that doesn't deserve a document: parking is free for customers, card and cash accepted.
Working hours
Days, shifts and holidays. The agent builds answers on them and logs a callback instead of promising a slot while you're closed.
My documents are scans — can it read those?
It can. Most OCR engines return Arabic as disconnected letters or with the word order reversed; ours treats Arabic as connected script, so letters join correctly and tables aren't flattened into noise. This is about reading joined letters, not diacritics — your documents don't need to be vocalised. Processing runs in the background and you watch each file's status in the dashboard.
How does the agent find the right answer in all that content?
Documents are split into passages in a way that respects Arabic, and each passage is indexed by its meaning — so retrieval works by meaning, not wording. A customer asks how long a whitening session takes; your file says “teeth whitening session duration: 45 minutes”; the system knows it's the same subject. Every answer carries its source, so you fix a wrong fact at the document rather than arguing with the agent.
Does it answer from my documents in my customers' conversations?
Yes — on the text chat channels. The agent searches your knowledge base and phrases the reply from your content rather than from general information. With no clear answer it doesn't improvise a number: it escalates to your employee or logs a callback.
How do I check the answers before customers hear them?
A retrieval test panel lets you type the question the way your customer would and see which passages came back and from which file. Above it sits the test chat, where you talk to the agent itself and read the full reply before publishing. Any change can be uploaded and re-tested in the same sitting.
So you train an AI model on my documents?
No. Your files are stored for you and indexed for search; at question time the relevant passage is retrieved and handed to the agent to phrase its reply from — so the answer is grounded in text you can go and read. Two consequences: edits and deletions take effect immediately with no retraining, and your organisation's data is isolated at the database level.
How often should I update it, and who's allowed to?
Update it whenever something changes that an employee would say to a customer — a price, your hours, a new service. Permissions are set per employee, so an operations lead can edit pricing while a receptionist has read-only access, and sensitive changes sit alongside the append-only audit log. One suggestion: turn the ten questions you get most on the phone into Q&A pairs.
Where are my documents processed?
Permanent storage of your data is in the Gulf region (Doha), while OCR processing of knowledge-base documents runs through a European processing endpoint. We do not claim everything sits inside Saudi Arabia, because that isn't accurate today — and disclosing cross-border transfer is itself the correct posture under the PDPL.
The full detail is on the security and data page.
Knowledge base questions
Which file formats do you accept?
PDF, Word (.docx) and plain text (.txt / .md), alongside Q&A pairs, links, snippets and working hours. Scans are read with Arabic OCR that returns correctly joined letters.
Does the agent answer from my documents in chat?
Yes, on the text chat channels — from your own content, and it escalates or logs a callback when there's no clear answer. You can try any question in the test chat before publishing.
Do I have to rewrite my documents in a particular format?
No, upload them as they are. The only thing worth writing by hand is a sensitive reply you want worded exactly — and that belongs in a Q&A pair.
Can each branch read only its own material?
Yes. Knowledge-base scope is set per agent, so agents never cross into each other's files.
Every customer channel in one platform?
Yes — phone calls, WhatsApp, Telegram, SMS, email and your website chat all land in one team inbox. The same agent answers them all from your business's own information, and your staff can step into any conversation.
Phone calls
Answers in your customer's dialect and books inside the call.
WhatsApp Business
Replies and follows up in the thread the customer started.
Telegram
The same agent, the same knowledge, on Telegram.
SMS
Reminders and confirmations that reach every phone.
Email enquiries land in the same inbox.
Website chat
Your site's visitors talk to the same brain — try it below.
One inbox, one agent, full context at handoff.
Ready to see it answer your customers?
A 30-minute intro demo — we show you the platform live and answer your questions.