Dukon

Empty shelves cost retail $1.7 trillion a year, and most of the world's groceries are still sold by corner shops with one owner and one phone. Dukon turns that phone into a shelf-to-order agent. The owner snaps the shelf. Muse Spark looks first and names what is there; SAM 3.1 finds every pack and every price tag; Muse Spark reads each one: name, count, price. Dukon compares walks over time and turns each empty spot into money lost per day — one empty box of Snickers in a Bishkek supermarket: 567 som (≈ $6.49) a day, almost $200 a month from one box. Then a Muse Spark agent with tool calling drafts orders per supplier and negotiates with real suppliers in WhatsApp. The model proposes, the code decides: the owner's rules (budget, max stock, minimum discount, banned suppliers) are enforced by code, and a blocked step stays in the log with its reason. Suppliers answer like people ("we can only deliver on Thursday"); Dukon understands the sentence, changes the order, reconfirms, and the owner only approves. The owner can ask by voice in Kyrgyz — Muse Spark 1.2 listens to the audio itself. Seven languages, PDF export, branches on a map, no sign-up. Live: https://dukon-woad.vercel.app Video: https://youtu.be/avagyzTkrYQ Code: https://github.com/AdilkhanEEAIR/dukon