Skip to Main Content

EveryMuse

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

EveryMuse — Demo video

EveryMuse — Meta's models on every ear. When your hands are full, even a small question can mean stopping, finding your phone, unlocking it and typing. EveryMuse is a personal AI agent for the Bluetooth audio glasses or earbuds you already own, with an iPhone app and a browser client. The phone connects the headset to the agent; no custom glasses firmware is required. Ask for weather, remember where you parked, create a reminder or request background research. Its distinctive feature is Heads-up: when enabled, it notices useful moments in nearby speech, such as a friend's dinner plan, and offers help. It asks before acting. Rewind answers questions such as "What was that address?" from the last three minutes of nearby speech. Find it answers questions about a supplied photo and outlines the relevant object. The prototype combines real-time speech recognition, tool calling, web search, memory, photo understanding and asynchronous research. It supports language selection, reconnects after interruptions and shows receipts for tool actions. Heads-up is opt-in; Rewind holds nearby transcripts temporarily in server memory. The application does not retain raw audio, video or photos. Requested notes, action receipts and limited session-recovery data are stored separately. The repository includes the iPhone client, web client, backend, automated tests and reproducible evaluation scripts with saved results. The demo labels archived recordings, synthesized microphone input, recreated reply audio and a recorded Rewind evaluation. It does not demonstrate a physical headset or a locked iPhone. Try it: https://earshot-yuansshi.fly.dev Availability note: Meta API credit was exhausted during final testing, interrupting fresh speech requests. Typed fallback and the recorded demonstrations provide alternative ways to review the prototype.

Team