Whenever
Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

Whenever is IFTTT for the physical world. You give any camera a rule in plain English, and the camera becomes the trigger for any action. A café owner says: "Whenever a guest raises a hand and no waiter reaches the table within a minute, alert the manager." A family says: "Whenever a car blocks our gate, send me the picture." An operator says: "Every Monday at 9, send me last week's busy hours." Whenever turns each sentence into a live, tested program in about a minute, for about $0.15 to $0.35. How it works. Muse Wonder writes each rule as real Python: its sensors, its code and its tests, in a sandbox. A second Wonder review reads the code against the person's words before the rule goes live. Muse Spark and SAM 3.1 watch the frames. A free detector on the server (YOLOX, body points, faces) counts people, tables and cars, so the paid models look only when something may matter. Every alert carries its proof: the frame, the outline and each fact that passed. Ask answers questions about the past, such as "When did this parcel arrive?", from a journal and 12 hours of footage. It searches cheaply first with Meta's Perception Encoder, then confirms with Spark at full size. It runs on real cameras today: the author's five EZVIZ and Hikvision cameras on an Azure server, plus phones, browser cameras and any RTSP or ONVIF camera. Alerts go to browser push, Telegram, email and ntfy. A new action is any web service that the person describes in plain words. Who it is for: small shops, restaurants and homes that own cameras but have nobody to watch them, and CCTV operators who watch many cameras for clients. A human operator costs about $18 an hour. Whenever lets one operator cover many more cameras, and each alert proves itself. Everything in the repo was built during the hackathon, from 30 October 2026. 599 offline tests run on every push. Rule code runs in a Landlock and seccomp sandbox with no keys and no network. Try the live demo: https://demo.4-240-49-230.sslip.io. Demo access code: amber-timber-5946 Its cameras are recorded clips in a loop: a gate and a driveway by day and by night, and a cafe.