WhatsApp chatbot development for AI agents that answer in Arabic, English and Urdu

We build on the WhatsApp Business Platform through Meta's Cloud API, the official route for businesses that need automation at volume. Your number is registered to a verified Meta business account, incoming messages reach our service through a webhook, and the agent replies through the same API. The WhatsApp Business app on a phone can't run this kind of automation, and unofficial tools that script WhatsApp Web put the number at risk of a ban.

Two platform rules shape every design. When a customer messages you, a 24-hour customer service window opens, and inside it the agent can reply freely. Outside that window the business can only start a conversation with a pre-approved message template, such as an appointment reminder or a delivery update, and only with contacts who have opted in. Meta charges for template messages by category and the recipient's country, so we decide early which messages need templates.

We have shipped this message path in public code. WhatsApp-Me, Safock's open-source plugin under the MIT licence, runs two-way messaging between an AI coding agent and a phone over the WhatsApp Cloud API, with verified webhooks carrying replies back to the agent.

Most WhatsApp bot problems trace back to decisions made before the build. These are the questions we would put to any vendor, including us.

• Does the agent run on Meta's official WhatsApp Business Platform, or on a tool that scripts WhatsApp Web?
• Are the Meta business account, the phone number and the message templates registered to your company?
• Which messages need pre-approved templates, and what will Meta charge for them in your customers' countries?
• When a chat moves to a person, does your team see the full history and a summary?
• Was the agent tested on real messages in the languages your customers use, including Roman Urdu?
• Where are transcripts stored, who can read them, and when are they deleted?

Customers in Saudi Arabia, the UAE and Pakistan rarely stick to one language. We set the agent to reply in the language the customer writes in, covering Arabic, English, Urdu script and Roman Urdu, and we test it on real message samples from your business before launch. For IJT Pakistan, the country's largest student network, our team built JamiatGPT, an AI assistant that answers in Pakistani languages.

Hand-off is part of the design from day one. The agent passes the chat to a person when the customer asks for one, when the message is a complaint or a payment dispute, or when it isn't confident in its answer. Your staff pick up the conversation in a shared inbox with the full history and a short summary, and the agent stays out of the chat until they hand it back.

WhatsApp suits jobs where customers expect a quick reply and the answer lives in one of your systems. These are the patterns we plan around.

Every conversation is logged to your CRM or database with the contact, the outcome and a summary, so sales and service teams follow up from one record instead of scrolling through a phone.

• Clinics and hospitals: book, move or cancel appointments, send reminders, and answer questions about timings and departments
• Retail and e-commerce: order status, delivery updates, return requests and product questions answered from your catalog
• Real estate and showrooms: qualify leads, share brochures, book viewings and log each enquiry in the CRM
• Schools and training providers: admissions questions, fee reminders and class schedules
• B2B suppliers: order confirmations, dispatch updates and payment reminders to trade customers

WhatsApp messages carry personal data, so data protection belongs in the plan from the first week. In Saudi Arabia that means the Personal Data Protection Law (PDPL) and questions about where data is stored and processed. The UAE has its own federal data protection law, and regulated sectors such as healthcare and banking add rules of their own in every market.

We treat these as design decisions: which data the agent may read, how long transcripts are kept, which region hosts the database, and which model provider processes the text. We build to the requirements your legal or compliance team sets. We don't give legal advice or certify compliance.