Asih

Problem statement: Build for Your World. Asih is a calm AI helper for older adults, and its name is the Indonesian word for caring: warm, compassionate care for others. It is not an open-ended chatbot. It offers a few fixed, guided flows that end with an action the person chooses, because the goal is independence with control, not a companion that decides for someone. Every flow starts with honesty. The opening says Asih is an AI helper, not a person or a doctor, and can make mistakes. A red-flag pre-screen sends urgent words to a Get help screen before the agent runs; the app never calls emergency services itself and gives no medical advice. Before the first photo, voice note or message, a read-aloud notice explains what is sent, that the app asks Meta not to store its answers with store:false, and what Meta may still retain. The person must acknowledge it. Appointment help reads a letter photo or labeled example. It extracts date, time, place, clinic, person, phone and what to bring, marks unreadable text, and has the person confirm one item at a time. Then they can save it, download a calendar file, add their own questions or see transport links. Find a group takes an interest and optional ZIP or city, or pasted listing text or a screenshot, and returns search terms and public calendar types. After a listing is summarized, the person confirms date, time and place, sees a checklist and transit help, and can edit an invitation. The invite opens a WhatsApp click-to-chat link that the person sends themselves. In live testing the group step returned empty event and plan fields, so it is demonstrated with labeled example data, not claimed as working live end to end. Check an ad, for health ads on Facebook or Instagram, accepts ad text, a screenshot or an example. Code quotes warning signs in the ad's own words, then always shows fixed wording saying it cannot tell whether the ad is real or the product works and asking the person to check with someone they trust. It is a second opinion, never a safe or scam verdict, and not medical advice. It also walks through Why am I seeing this ad, Accounts Center ad preferences, reporting an ad and a Meta Ad Library search. The app is in English and Spanish, with formal usted; Spanish still needs a fluent-speaker review. A native Wear OS companion records a short voice request, sends it to the same server, and routes it to the right flow; anything needing a photo stays on the phone. The design draws on a review of 15 peer-reviewed sources, which produced twelve guidelines, and five simulated older-adult persona reviews. Those reviews were simulated, not real participants.