
Teya is an open-source AI family agent that transforms any cheap Android phone into a wall-mounted household assistant. Unlike traditional voice assistants that forget everything between sessions, Teya combines cloud-based AI reasoning (via Mistral) with on-device memory, voice recognition, and direct control of phone functions—allowing it to remember family details, recognize who is speaking, manage calendars and shopping lists, and place approved phone calls. It keeps all household data and sensitive operations local by design, with no server backend required.
Summaries like this, in your inbox every morning.
Sign up free →What happened
A developer released Teya, an open-source AI family agent that runs on any Android phone (8.0 or newer) mounted on a wall. The agent listens, remembers household details, and performs tasks like managing shopping lists, calendars, timers, expenses, and making approved phone calls. It uses Mistral's cloud API for reasoning and speech processing, while keeping household memory, voice recognition, and device control local on the phone itself.
Why it matters
Unlike voice assistants designed in the early 2010s (which forget everything and can't act across apps), Teya combines modern AI intelligence with actual household context—recognizing who's speaking, remembering family details across sessions, and taking action on calendar invites, reminders, and calls. For the person carrying the household's mental load (appointments, meals, logistics), it offers a second brain that runs quietly in the background without requiring expensive hardware or a complex server backend.
What to watch
The app is sideloaded directly as an APK (not through app stores) and requires a household Mistral API key; users pay for their own API usage. Smart-home control via BLE/Matter is on the roadmap but not yet built. The codebase is open on GitHub; the developer invites contributions for new tools, visualizations, and additional AI providers.
Teya began as a solo project and is now open-source. The developer built it to solve a concrete family problem: modern AI is intelligent but disconnected, while existing home voice assistants are dumb but present. The solution was to take an inexpensive Android phone, add modern AI reasoning via Mistral, and layer on household context—memory, voice recognition, and local integrations—that earlier-generation assistants lack.
The agent currently handles shopping lists ("we're out of milk" is added automatically without repeat requests), voice-controlled calendar events with email invites to household members, timers and alarms that announce themselves in Teya's own voice, reminders that become either timers or calendar entries depending on context, expense logging with exact arithmetic ("12 euros for fruit" is categorized immediately), and phone calls to approved family contacts only—a five-year-old can say "Call Grandma" and place a normal hands-free cellular call without access to a dialer. The agent recognizes each household member's voice through on-device voice enrollment, greets them by name, and maintains separate memories for each person.
Memory is the core differentiator. At the end of each conversation, the Mistral model distills what is actually worth keeping into one short note, or decides there is nothing worth keeping at all. Every night, those loose notes consolidate into durable facts about the household, and anything not reinforced fades on a forgetting curve—aging the way human memory does. The household can ask Teya to remember something ("remember I'm allergic to peanuts") and later ask it to forget, and all saved memories can be reviewed and edited by hand.
Technically, Teya is a native Kotlin app that connects an LLM, text-to-speech, speech-to-text, wake-word detection, and voice-activity-detection models to Android's native SDK, turning model decisions into tool calls—placing a call, adding a shopping-list item, saving a memory. Reasoning, speech-to-text, and text-to-speech run through Mistral's cloud API; the wake word (a self-trained "hey_teya" model), barge-in detection, voice ID, shopping list, expense log, family memory, household roster, and contacts allowlist stay on the device. There is no backend server; family data and device control are local by default because telephony and microphone are Android-local APIs a cloud server could not reach anyway. The app is installed by sideloading the APK directly, not through the Play Store, and requires the household to enter its own Mistral API key during onboning. There is no auto-updater; new versions are downloaded from Releases and tapped to install over the existing app, preserving household data and settings.
The developer notes that the microphone and speaker built into a cheap Android phone were tuned for close, arm's-reach use, and picking up a wake word or clear speech across a room asks more of them than that. Early testing is encouraging but limited; a bigger room may need an external microphone array or speaker. iOS is not supported because Teya needs deep, unsandboxed access to Android's APIs (telephony, notifications, always-on foreground service) that iOS does not allow. Smart-home control via BLE/Matter is on the roadmap but not yet built. The codebase is open; the developer invites contributions for new tools, visualizations, better wake-word and voice-activity-detection tuning, and additional LLM providers via GitHub issues or pull requests.
Teya addresses a long-standing gap in consumer AI: while ChatGPT and Claude offer genuine intelligence, they sit behind a chat window disconnected from actual household infrastructure and daily life. Traditional voice assistants like Alexa, designed around simple if-then commands from the early 2010s, lack context and memory—they can set a timer or play music but cannot track who is picking up the kids from football or understand how that connects to standing calendars and household logistics.
The design choice to run on a cheap Android phone is foundational. A modern Android device already includes a display, 4G/5G, microphone, speaker, GPS, calendar, timer, and access to millions of apps—all the hardware a family agent needs. By wall-mounting a single visible node and keeping the rest of the home's infrastructure (smart bulbs, locks, calendars) invisible, the builder creates a unified interface the whole household sees, while offloading heavy reasoning to Mistral's cloud API. This avoids the current industry pattern of buying dedicated local hardware (such as Mac Minis for coding agents) just to host orchestration frameworks.
Privacy and safety are built into the architecture itself. Household roster, per-person memory, and contact allowlists live on the device; raw transcripts never touch disk; only consolidated memory notes and data needed for each turn reach the cloud. Voice enrollment and speaker identification run on-device, and phone calls are placed as normal cellular calls through the phone's SIM with no VoIP or audio capture—making the system safe enough to leave within reach of children and guests.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
No discussion yet for this article
Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.
Get Started FreeFree · takes 30 seconds · unsubscribe anytime
1 minute a day. The AI essentials.
200+ sources · Email / LINE / Slack