Saitabau-Dennis/Mpesa-AI-Lift

# M-Pesa AI Lift **Turn a Lipa na M-Pesa till into an AI commerce layer — no website, no developer.** M-Pesa AI Lift is a working prototype of an intelligent assistant embedded in the M-Pesa / Safaricom ecosystem (Zuri / WhatsApp-style). In one chat, a customer can browse, order, and pay with M-Pesa STK Push; a merchant can digitize stock, take sales, read till insights, and explore working-capital eligibility from their own transaction history. This repo is the demo you can run locally in a browser. The product identity is **M-Pesa AI Lift**; the session is linked to a sample shop (**Baraka General Store**) so tools have real inventory and till data to act on. --- ## Why it exists Millions of Kenyan SMEs already accept M-Pesa. What they usually *don’t* have is: - a digital storefront or catalog - guided shopping and payment in the channel customers already use - live stock awareness after sales - plain-language business insights from till history - a path from that history to credit / working capital M-Pesa AI Lift sits on top of that till relationship and closes the loop in conversation. --- ## What you can do in the demo ### For customers (shopping) 1. Ask for a product (“Do you have milk?”). 2. Choose quantity. 3. Give your **real** M-Pesa number when asked (the agent must not invent one). 4. Confirm the order summary. 5. Receive an STK Push, enter PIN, and get a receipt when payment succeeds. ### For merchants (business) - Stand up a shareable digital catalog from current stock - Adjust inventory and record POS-style sales - Pull till transaction summaries (“How’s business?”) - Run a **credit score + suggested working-capital limit** from synthetic till history There is **no merchant/customer toggle**. One Zuri-style thread routes intent from natural language. --- ## Quick start **Requirements:** Python 3.10+, a virtualenv, and an OpenRouter API key. ```bash git clone cd mpesa-lift-prototype python -m venv .venv source .venv/bin/activate …