Sado

Sado — one voice, every language. A voice-first AI companion that hears and speaks Tajik, Uzbek, Russian and English, and gets things done on your phone and computer. THE PROBLEM Mainstream AI assistants speak about 25 languages. Tajik and Uzbek, spoken by more than 40 million people, are not among them, and Meta's hosted speech API does not cover them either. For millions of people, including migrant workers who receive papers in Russian they cannot read, "talk to your phone" still does not work. WHAT IT DOES - Hears and speaks four languages (Tajik, Uzbek, Russian, English) and works out which one you used. - Remembers and reminds: facts you share, reminders and shopping lists, all visible and deletable in the app. - Explains any paper: photograph a Russian utility notice or share a contract PDF; Sado explains it in your language. - Acts on your phone: calls and texts (only after your "yes"), music, alarms and timers, calendar, volume, battery, maps, apps. - Acts on your computer: call it by saying "Sado"; it reads the screen, reads what a coding agent in the terminal answered and types it the next task, finds and opens files, controls media, and runs commands after your yes. - One Sado across devices: sign in with Google and the owner's phone drives the computer while the computer reaches the phone; everyone else gets their own private Sado. - Runs on your own computer: speech recognition, voices, the wake word and memories stay local; only the reasoning goes to Muse Spark. - A living 3D sphere drawn on the GPU ripples with your voice and with Sado's. CHALLENGES - No hosted speech model covers Tajik or Uzbek, so we paired Muse Spark with open Meta speech models and wrote number, date and money normalisation for each language. - Making an agent that acts safe: every action that reaches another person or changes the computer is a single-use, expiring request that only the user's yes can run; the model can never confirm it. - Giving judges a working demo without exposing the owner's computer or memories: per-account spaces and an owner-only computer link. WHAT'S NEXT More languages from Omnilingual's 1,600 through pluggable language packs, an always-on wake word on the phone, smart-home control in Home mode, and a hosted GPU backend so anyone can install the app and talk.