MoreAI

MoreAI MetaAgent is an AI shopping agent for local Egyptian brands, built for this hackathon on Muse Spark, SAM 3.1 and Muse Image. The problem: local brands in Egypt are scattered across social media. To buy, you message page after page, ask for prices and wait, and you still can't put one outfit together across brands. For shoppers: Say what you want in your own words. It searches across the brands by meaning, and your budget is a hard limit in code. Muse Spark looks at the product photos and shows up to 3 pieces, each with a reason. Attach a photo of your outfit and ask what goes with it. It reads the colors and suggests what matches. Search with a photo: Muse Spark names each piece, SAM 3.1 finds where it is, and search by meaning finds similar pieces. It asks before it changes your cart or places an order. One cart across brands becomes one order per brand. It remembers your taste, and you can see and change it. For brands: One product photo is enough. Muse Spark reads it into a name, a description, a category and a color, and nothing goes live until the merchant approves it. How it's built: Our own agent loop on Meta's Responses API, with 9 tools and one backend request per model step. A live trace shows every model call and tool call with its real time. The model only proposes: budgets, stock and totals are checked in code. A Cloudflare Worker holds the API key and enforces a daily spending cap, and search by meaning runs on Cloudflare Vectorize. Demo data: the brands and products are made up, and all 200 product photos were made with Muse Image. Carts and orders stay in the visitor's browser. Shopper history, push campaigns, notifications, settings and ratings are labelled demo data. Try it, no login: https://moreai-metaagent.khaled-m2225.workers.dev