South Africa Local AI Pilot English transcript 01 / The opportunity Imagine opening an AI assistant on your phone that can help with a school lesson, a government form, or an everyday question. Now imagine the assistant running from a hub in your own community. This is a proposal to test shared local AI in South Africa, funded by a yearly contribution of twenty-five US cents per mobile subscription. That's about four rand and nineteen cents. Here's how it could work. 02 / How local AI works A school, clinic, business, or community centre hosts a server. You join its Wi-Fi and open the assistant in your browser. The server runs the model and sends the answer back to your phone. Once the model and reference materials are installed, people within Wi-Fi range can use the assistant without internet access or mobile data. The hub still needs electricity and enough computing capacity for the people it serves. 03 / Ready-to-use equipment The equipment would arrive with the model, apps, and approved materials already installed. The installer would set up secure Wi-Fi, accounts, permissions, and a trusted local connection. Before opening the hub, staff would test it and learn how to use it. Regular updates and on-site support would keep it running. 04 / Power for remote communities Where grid power is unreliable or unavailable, a solar and battery system could keep a hub running. It could power the router and computers and charge phones and power banks. A larger AI server would need more capacity. Each site would size its system for the equipment it uses, the hours it opens, and the support it can maintain. 05 / South Africa pilot economics The proposed contribution is twenty-five US cents a year, or about four rand and nineteen cents at the thirteenth of November exchange rate. South Africa's communications regulator, ICASA, reported about a hundred and seventeen point three million mobile cellular subscriptions at the end of September twenty twenty-five. If every subscription in that baseline paid, the gross annual total would be about twenty-nine point three three million US dollars. In South African rand, the same total is about four hundred and ninety-one point eight five million. 06 / Approval and collection Before any charge, the funding plan would need lawful approvals and agreements with participating mobile operators. One option is for operators to collect the annual contribution and pass it to a dedicated fund. The plan would need to explain who pays, how exemptions work, and how the fund reports its spending to the public. 07 / From revenue to a service At ninety percent collection, the illustrative gross total is about twenty-six point four million US dollars. In South African rand, that is four hundred and forty-two point six six million. At seventy-five percent, it's about twenty-two million US dollars. In South African rand, that is three hundred and sixty-eight point eight nine million. Servers, electricity, training, and support would come out of those totals. The regulator's count includes machine-to-machine subscriptions. People can hold more than one mobile line, and eligibility could narrow the billing base further. 08 / Government and residents A local council could help residents find permit forms and see which documents to submit. Staff could search approved policy documents and draft replies. The assistant would show its sources and direct official decisions to staff. Sites could test answers in languages used locally, including English and the languages spoken in their communities. 09 / Schools and learning A learner could ask for another explanation of fractions or help with a challenging problem. Teachers could prepare activities from approved South African curriculum materials. The hub at the school could offer those tools during an internet outage. Teachers would check the answers, protect learners' information, and test how well the assistant works in the languages their classes use. 10 / Healthcare support Clinic staff could use a hub to find reviewed guidance, explain care instructions in plain language, or prepare routine paperwork. Trained professionals would verify outputs and make clinical decisions. Sites would need to protect patient information and account for South Africa's Protection of Personal Information Act. 11 / Businesses, farmers, and families A small business could draft an invoice, check stock, or prepare answers to common customer questions. A farmer could look up reviewed guidance before speaking with an extension officer. Families could practise an interview or learn a digital skill. For information such as today's prices or weather, the assistant would need a current source. 12 / Each organisation's app Each organisation could have its own portal. A person signs in and searches the information their account can access. The assistant could also prepare a form or check an appointment through an approved app connection. Before making an important change, the service would show what it plans to do and ask the user to confirm it. 13 / Installable apps Progressive web apps could bring approved lessons, forms, and business tools to compatible phones. Where installation is supported, users can add the app to their home screen and download what they'll need later. A learner could carry an exercise home, or a shop owner could draft an invoice on the phone while offline. 14 / Farmers and agriculture apps A farmer could take photos of crops, record field observations, and track feeding or health checks for livestock. The app could keep downloaded guides and saved records available offline. At a hub, or on a phone that can run its own AI model, the farmer could use the assistant to prepare questions for an extension officer or veterinarian. It takes qualified advice to diagnose serious problems. 15 / Save now, sync when connected While offline, the app could save supported forms and queue changes on the phone. Once the user reconnects to an authorised server, it can send the queued changes and fetch fresh content. The app should show which work is saved, which is waiting, and which the server has received. It also needs to deal with interrupted transfers, duplicate entries, and conflicting edits. 16 / Three ways to use it At the hub, local Wi-Fi connects a phone to the shared AI and apps without mobile data. Away from the hub and fully offline, a phone can still use downloaded content and save supported forms. With internet access, people farther away can sign in to the app through a secured gateway. Each route serves a different need. 17 / AI on your own device Some smartphones, tablets, and computers can run a smaller AI model themselves. Once the model and its materials are installed, a compatible device can answer new questions even when it's disconnected from the hub. Performance depends on the device and model. Organisational records still require the right permissions and connection. 18 / Internet access via ngrok For people using the internet, a service such as ngrok can give an authorised local app an HTTPS address. Both the hub and the remote user need internet access. The app would require authentication and use access limits and activity logs. Its users would sign in through the gateway before reaching the services their accounts allow. 19 / A practical South Africa pilot Start with a small set of schools, clinics, municipal offices, and businesses. Give each hub reliable power and local support. Test priority languages, privacy, answer quality, and how many people can use it at once. Work with operators and communities to evaluate a transparent funding plan. Measure cost and usefulness before deciding whether to expand. 20 / The vision The goal is to make useful AI easy to reach from a phone. Local hubs serve nearby users. Downloaded apps keep selected tools available offline. A securely connected app can reach people farther away. South Africa can test these approaches with communities, learn what works, and share those results across Africa.