A multilingual AI agent that understands the way your customers actually write.

In North Africa and across the MENA region, nobody types a DM in textbook Arabic. They write in Derja, often in Latin letters with numbers standing in for sounds (“3aslema, 9adech el prix?”). They switch to French mid-sentence, drop an English word for the product, and when typing gets tiring, they just send a voice note.

A generic multilingual chatbot is built for clean, single-language input. It runs the message through “Arabic” or “French” detection, gets neither, and replies with a menu, a fallback message or an answer in the wrong language. An Arabic chatbot trained on Modern Standard Arabic is no better: Derja is not MSA, and Arabizi is not Arabic script.

So the conversation falls back on a human who has to read, decode and reply to every single one. Which is exactly the work you bought the bot to remove.

Language is not a nice-to-have on top of automation. For a merchant selling in Tunisia, Morocco, Algeria or across the MENA region, it decides whether automation works at all.

Nothing about the way your customers write has to change. The agent adapts to them, message by message.

Unifunl’s AI agent replies in English, French, Arabic, Tunisian Derja, Moroccan, Algerian, Emirati, Omani and more, in Latin or Arabic script. It adapts to the language the customer uses, including when several are mixed in one message, which is how most people in the region actually write. Unifunl is one of very few platforms built to handle Tunisian Derja natively.

The languages you configure set the agent’s defaults, not its limits. If your agent is set up for Derja and a customer writes in English, French or Emirati Arabic, the agent answers in the language the customer used, so nobody gets turned away for writing the “wrong” language.

It understands inbound voice notes and images, not just typed text. A customer can send a voice message or a screenshot of a product and get a real answer, without anyone on your team having to listen, transcribe or translate first.

Language is only half of it. The agent answers from your synced store catalog, prices, variants and stock, so a correct answer in Derja is also a correct answer about your products. It runs on WhatsApp, Instagram, Facebook Messenger, TikTok and your website widget from one inbox.

These Unifunl customers run their front line in Tunisian Derja, French and Arabic, the way their customers actually write.

Most multilingual chatbots work by detecting a language and matching the message against scripted replies. That holds up for clean English or French. It breaks the moment a message is Derja in Latin letters, or three languages in one sentence, because there is no script for that.

Unifunl’s agent does not route by language. It reads the whole message, whatever mix it is written in, reasons over your catalog and policies, and answers in the customer’s own language. The same agent then takes the order, follows up and hands off, so multilingual is not a separate module bolted on top.