Every chat is a checkout.

Every competitor was building sellers a dashboard to log into and abandon. I bet the opposite: you should run an entire business from a chat.

Chidi is an AI assistant that runs a seller’s shop inside the chat their customers already use, answering, taking orders, handling payments, while the seller stays in control just by chatting back.

The hard part was never the replies. It was letting software act on its own and still be trusted with the money, so I designed and shipped it in code, where voice and mechanics can’t drift apart.

Across Lagos, Accra, and Nairobi, tens of millions of sellers run a business inside Telegram and WhatsApp DMs, often on patchy data.

It works until it doesn’t. A buyer messages “abeg, the red Adidas still dey?” at midnight and gets no reply. Orders live in someone’s head, and there’s no record of who paid. Fast at five orders, drowning at fifty.

Every tool answers by bolting on a CRM the seller never opens. My constraint was harsher: if a feature made the seller leave the chat, or broke when the network did, it didn’t ship.

I didn’t start with screens. I shadowed how sellers in Lagos, Accra, and Nairobi actually run a shop: a phone, a WhatsApp or Telegram thread, a notebook for orders, a bank app to confirm transfers. The chat already was the storefront, the catalogue, the checkout, and the customer record, all in one thread, on a connection that could drop mid-sale.

The same few behaviours showed up in every city. They became the rules the product wasn’t allowed to break.

Three decisions shaped everything after.

Chidi works both sides of the conversation, the same assistant on each.

Underneath is one brain and one source of truth, customers, catalogue, orders, payments, and a memory of every buyer’s history, so a returning customer is greeted, not re-interrogated.

Letting software answer customers is easy. Letting it touch money is the real design problem, and testing early builds with real sellers put the line exactly where it belongs: they loved the replies, but froze the moment Chidi might confirm a payment they hadn’t seen.

Chidi is autonomous, with the line drawn at the money. It replies, reserves stock, and drafts the next step on its own, but anything that moves money waits for the seller’s tap: a payment, a refund, a price change, a broadcast.

Every action leaves a quiet receipt with See and Undo, so the seller can always see what happened and pull it back.

The payment is the trust moment. When a transfer is matched, through Paystack or bank reconciliation, the buyer gets a confirmation formatted like proof, not a chat bubble, the thing they screenshot.

It is the pattern the rest of my work is moving toward: an agent that acts, with a human holding the keys to the money.

The second hard call was the network. These sellers work on patchy data, and a missed message is a missed sale.

So Chidi is offline-first. The inbox keeps working with no connection: every action, reply, fulfilment, broadcast, payment-confirm, drops into a local outbox, and everything syncs the instant the connection returns. No lost orders, no “message failed.”

Sellers adopted Chidi without changing how they work, the chat stayed the same, the business behind it got real. Active sellers retain at 71% by day 30, across three markets: Lagos, Accra, and Nairobi.

I can ask Chidi what someone bought, their size, what they asked about before. When I follow up, it feels personal, like I remembered everything myself.Amadi K., eyewear, Lagos

In Nairobi, a seller raised her best product 12% after Chidi flagged she was underpricing. Zero complaints, more orders.

Next: WhatsApp and Instagram on the same brain, and deeper cross-border reconciliation.

Chidi is where I first shipped an agent that handles a real business’s money, with the controls to make it safe, the exact problem I’m now building Tabs around.