AI Business Chatbots From Nilex: What They Do, What We Connect Them To and the Limits We Set

An AI business chatbot from Nilex is an assistant that knows your business: it answers customers from your own content, recommends from your real products, and is connected to your data instead of a generic script. Depending on the project, it can also look things up inside your system and, where you agree to it, act there, such as creating or updating records, checking orders or summarising what is happening.

That description fits many products on the market, so this article is specific. It shows the four assistants we have built and launched, what each one is connected to, and the limits we set on every one of them. If you are deciding whether a chatbot is worth it for your company, the useful question is not "does it use AI?" but "what does it know, what can it touch, and what stops it from going wrong?"

Key takeaways

• Our assistants answer from the business's own content and recommend from its live catalogue; the coffee and appliance assistants talk with customers in Egyptian Arabic.
• We have built four: a legal book assistant for Dar El Ahram, a coffee expert for Anas Saleh Coffee, an appliance advisor for Bashayer El Khair and the sales assistant on our own site.
• The strongest ones have your experts' rules written in, such as AC sizing or blending rules, so they ask the right question before recommending.
• In the coffee and appliance assistants, prices in replies are checked against the catalogue or calculated by the server, not typed freely by the model.
• Every assistant we built has limits around it, such as filtering or screening, usage caps, protection of personal data or a hand-off to people.

What does an AI business chatbot from Nilex do?

The AI business chatbot service covers a range, from an assistant that answers questions from your content in a website widget, in Arabic and English, to one connected to your system's data. In practice the jobs fall into four groups:

1. Answering: questions about products, ordering, shipping, terms or services, from content you control.
2. Advising: asking about the customer's situation and recommending the product that fits, the way a good salesperson would.
3. Calculating: working out a size, a blend or a price from rules and data the server checks.
4. Handing off: passing the customer to your team when a person is needed.

Assistants inside a business system are part of our custom work as well: the custom ERP service lists an AI assistant inside the system that you can ask questions, and the e-commerce service includes AI product recommendations. Which actions an assistant may take inside your system should be decided per project, because every action is something to secure.

Which assistants has Nilex built?

Legal book assistant: Dar El Ahram

A lawyer rarely knows the exact title of the book he needs; he knows his case. The assistant on Dar El Ahram's store reads the catalogue of 800+ legal titles, including the table-of-contents PDFs of the books, and answers a reader's legal question with the books that fit and why. It gives the price and whether a book is in stock, answers questions about ordering and shipping, and streams its answers so the reader sees them as they are written. The full build is in our legal book assistant case study.

Coffee expert: Anas Saleh Coffee

The coffee expert on Anas Saleh Coffee's store talks in Egyptian Arabic, Franco or English. It explains any coffee sold in Egypt, recommends products from the live catalogue and designs a custom blend from the customer's own words: a request like «تقيلة ووشها كتير ومش مُرّة، بطعم شوكولاتة» becomes a Turkish blend with exact percentages, the roast chosen for each bean and the cardamom level. The server checks every blend before the customer sees it, including robusta and spice limits and the price per weight from the shop's own prices. The customer can adjust the blend in an editor with the price updating live and add it to the cart, and the roastery receives the full recipe with the order. See the coffee blend assistant case study.

Appliance advisor: Bashayer El Khair

On Bashayer El Khair's store, customers talk to the assistant the way they would talk to a salesman. It works out the right size before suggesting anything: a 20 m² bedroom with afternoon sun gets the air-conditioner horsepower Egyptian installers would choose, and a family of five gets a suitable fridge capacity. It then shows matching products with prices and an add-to-cart button, compares models side by side and looks up the manufacturer's specs when a listing is thin. Large open spaces and installation questions go to the team, and the owner can read the conversations from the dashboard. Details are in the appliance advisor case study.

Our own sales assistant

The assistant on our website runs on Claude and behaves like a consultant: it asks one short question at a time, recommends the matching system or service with its real plan and answers security, payment, warranty and data questions only from facts we have written. It hands off through WhatsApp, a call-back form that creates a CRM lead, or the contact page, and it does not book meetings, see customer data or take actions. We describe it in how we built our own AI sales assistant.

What do we connect a chatbot to?

The quality of an assistant comes mostly from what it is connected to. Across our projects, these are the sources that made the difference:

• Your live catalogue, so recommendations are products you actually sell, at today's price and stock.
• Your documents, such as the table-of-contents PDFs that let Dar El Ahram's assistant match a legal question to a book.
• Your experts' rules, written as data and code: blending rules and Egyptian spice mixes for coffee, AC and fridge sizing for appliances.
• Your written policies, such as the knowledge sections our admins edit for security, how we work and sales notes.
• Your cart and your team: an add-to-cart button for the customer, and WhatsApp, a call-back form or a CRM lead for your staff.

The principle behind these builds: where your own facts should decide, such as your prices or your terms, the assistant should answer from them, not from the model's general knowledge.

What limits do we set?

Every assistant accepts text from anyone, so limits belong in the build, not added after a problem. The risks are explained in AI chatbot security and prompt injection; here is how they appear in our builds:

The price and hand-off guardrails are covered in depth in keeping a store chatbot honest, and the cost side in keeping chatbot costs predictable.

When is a chatbot worth it for your business?

From the four projects, a chatbot is most useful when at least one of these is true:

• Customers ask the same advisory questions again and again, like "which AC for my room?" or "which book for my case?"
• Your catalogue is too large to browse, or your customers do not know the product names.
• The product is configured to the customer, like a coffee blend, and the configuration follows rules you can write down.
• Questions arrive outside working hours and go unanswered until morning.

If your customers mostly need a phone number and an address, a well-built website with a clear contact page will do more for you. A chatbot is not a substitute for a good catalogue and correct prices; all four of our assistants were added on top of stores and pages that already worked.

How Nilex helps

Start from the AI business chatbot page, where the packages are listed, and read the client stories on /clients. The Assistant package comes as a widget for your website; if you are starting a store, the e-commerce store service includes AI product recommendations. Then contact us with the questions your customers ask most often; that list is a good starting point for an assistant.

Frequently asked questions

Does the chatbot understand Egyptian Arabic?

Yes. The Anas Saleh Coffee and Bashayer El Khair assistants talk with customers in Egyptian Arabic, and the coffee expert also understands Franco and English. Our own sales assistant works in Arabic and English. For more on dialect, see our guide to chatbots that understand Egyptian Arabic.

Can the chatbot invent a price or a discount?

Not in the Bashayer El Khair and Anas Saleh Coffee assistants: prices in replies are checked against the catalogue, and custom blends are priced by the server from the shop's own prices.

Can it take actions inside my system?

It can be connected to your system's data, and actions such as creating or updating records can be part of a project. Which actions are allowed is part of the agreed project scope. Our own website assistant deliberately takes no actions at all.

Which AI models do you use?

Our client store assistants use the OpenAI API, and our own website assistant runs on Claude from Anthropic.

Can I read what customers asked?

In the Bashayer El Khair project, the owner reads the assistant's conversations from the dashboard. What is stored, and for how long, should be decided when the assistant is set up; on our own website, conversations are kept for 90 days.