Sumora

Sumora gives a small business a working team of AI agents from one conversation, and every agent runs on Meta's Muse models. Tens of millions of one-to-five-person businesses in India and the US run on WhatsApp, a notebook and the owner's evenings. They don't need another dashboard. They need staff. The owner describes the business in one sentence, drops in a rate card or menu, and answers a short interview. Sumora then: - designs an agent team for that business (front desk, bookings, outreach, quotes, website, daily brief); - builds a private website; - runs the day-to-day work over WhatsApp and the web. Anything that spends money, refunds, messages a new contact or publishes waits for the owner's approval, which the owner can give from WhatsApp or Command. Agents with eyes: for businesses whose work is visual, the Inspect tool has SAM 3.1 locate regions in a customer's photo (holes, stains and tears in fabric, for example) and Muse Wonder judge each one. The result is an annotated photo, a verdict and a reply for the customer. It runs on a production stack: Temporal for durable agent runs and approvals, Postgres with tenant isolation, Cloudflare Workers and Containers, and the WhatsApp Business Platform as our sender. Sumora's outbound product is already live with 4 client businesses: 1,077 people messaged, a 22% average reply rate and an 18-hour median to first reply (Aug 5 - Nov 2, 2026). This agent platform runs on staging and launches next. TRY IT (judges): https://staging-app.trysumora.com Sign in: judges@trysumora.com / Muse-B43hyV_DqTQE-2026 1) Open Drishti Vision, a sample AI startup that inspects fabric for textile mills in Surat. Sumora built its team and brain from one sentence and its rate card. 2) Command: see what each agent did and which Meta model did it. Approve or decline what's waiting. 3) Inspect: upload a fabric photo (or use examples/fabric.jpg from our repo). SAM 3.1 locates the defects and Muse Wonder judges them. 4) New business: describe any small business in one sentence and watch Sumora staff it and build it. Code for the Meta layer, with live-run output: https://github.com/devanshranjan10/sumora-meta-hackathon