TARS

Millions of people can’t use the devices in their own homes, because the buttons, menus and letters are in a language they can’t read. TARS is a phone web app for them: older people, people who can’t read, and people who only speak their own language. They point the camera at whatever they’re stuck on (a thermostat, a laptop, an English letter, a medicine box) and ask in any of 25 languages. TARS answers out loud, one short step at a time. It highlights the exact button when asked “which one?”, checks through the camera that each step worked, and warns about scams before they reply, pay or call. A code orchestrator runs six agents, each with its own backup model, all on Meta models: Talker (Muse Spark 1.2): speaks, and starts thinking before the person finishes talking Checker (Muse Spark 1.3): confirms each step through the camera Planner (Muse Spark Open on Baseten): thinks the task through in the background Scam watcher (Spark 1.2, then 1.3): two looks must agree before it warns Pointer (Muse Spark 1.3 + SAM 3.1): finds where to press Translator: puts the start screen in the phone’s language Muse Voice Transcribe turns speech into text in real time.