AIBOSSPRO

Outside the US, small service businesses (barbers, studios, clinics, detailers) get booked on WhatsApp, and most of their owners can't answer it while they work. I know this firsthand: AIBOSSPRO builds WhatsApp agents for exactly these businesses on monthly retainer, one bespoke build at a time. AIBOSSPRO Brain turns that bespoke work into one product an owner can set up alone. The owner describes the business once, in a voice note on the onboarding page; the Brain answers customers in their language, holds bookings, and prepares branded posts from the owner's own photos. The owner approves every booking and every post. Nothing is booked without a tap, and nothing is published automatically: approved posts are exported for the owner to post. On Nov 3, two business owners in Israel tried the production build hands-on. Yoel, who runs a services business: "This is genius. I'm sure I'll use it." A barbershop owner, after it generated posts from photos of his own shop: "You built this? I want it for myself." What makes it hold up: - Every model-generated reply passes a deterministic verifier before it is sent. It checks prices, hours, availability and discount conditions against the owner's approved rules, in English, Hebrew, Spanish and French. A blocked reply is corrected once; a second block hands the conversation to the owner. - Owner control is structural. Approvals are idempotent. HUMAN mode makes zero model calls. Every model-generated reply is labelled as AI. A health-category business is kept administrative: no medical advice. - Costs are bounded. A durable spend ledger reserves an estimated cost before every model call and settles after. Measured in the recorded golden run: $0.0017-$0.0093 per customer turn (n=4). - Judges don't need WhatsApp. npm run demo runs the same engine through a local terminal simulator. With no key it replays the recorded golden run and checks seven invariants; with your own Model API key, npm run demo -- --golden runs it live and checks the six that apply to a live run. provenance/GOLDEN-RUN.md lists every model call in the recorded golden run with model, tokens, cost and provider request id. Live on a WhatsApp Cloud API test number (one business per number); four demo businesses in the web app and simulator. 225 tests. All code in the repository was built during the hackathon window (first commit Oct 27); the git history is the record. Known limits, bundled third-party fonts and dependencies are listed in the README. Bonus categories: Best Agentic Use-Case (Muse Spark plans with native tool calls, bounded by the verifier and owner approval) and Best Creative Build (SAM 3.1 and Muse Image produce the post and story assets; Muse Wonder checks them).