White Paper
The AI for everyone. Settled 2026-11-16.
Talkmy™ is a voice-first AI app for seniors. One screen, one bright button, a chat thread. Press and hold, speak in any language, get an answer read aloud in that language. No sign-in, no account, no language picker, no typing required. The north star: a universal translator communicator — press button and get answers.
A note from the founder
The founder, Tony, states where Talkmy came from in two sentences:
“This app was made with love, the founders love for his parents.”
“I carved it in the only shape I knew, strong intelligent immigrant parents that were gated from the best of what their country had to offer.”
Contents
1. The problem
Seniors are locked out of AI by the door, not by the room
The AI itself is not the barrier. A senior who can hold a conversation can use a conversational AI. The barrier is everything around it: an account to create, a password to keep, a prompt box that assumes typing, a language setting buried in a menu. The mainstream tools have no senior mode. The teaching lane — free classes and paid courses — proves the demand is real, then ends every lesson by handing the senior a stock account at the mainstream tool, sign-up wall and prompt box intact.
The adult child is the unpaid front line
Every family has one person who takes the questions. Is this junk mail? Is this real? What does this letter say? The questions are small. They arrive anyway, all day, at the worst times — at work, at dinner, at the end of a long day when the person answering is hungry and tired. The founder's statement of the problem is personal, because it is his:
“It's mentally exhausting to the point that sometimes I dread going over there and eating with them because they'll ask me questions I'm too overwhelmed to deal with. I just want to eat.”
The friction is not a lack of love. It is a routing problem. Every small question in two households routes to one person, because there is nowhere else for it to go.
The bother tax, and the scam door
The parent pays a tax the child never sees: hesitation. A parent who does not want to bother a busy son simply does not ask. The letter sits unopened — and an unanswered question is exactly how a scam gets in. The fake bank letter, the gift-card bail story, the urgent wire request all depend on the senior having no one to ask in the moment. Asking has to cost nothing, or people will not ask, and the door stays open.
2. The insight
The customer is the child. The user is the parent.
Talkmy has two people in it. The parent uses it. The adult child buys it, installs it, and gives the one sentence of briefing. Payer, installer, and briefer are the same person. The promise to them:
“Talk about the things they've done, instead of doing the things that need to be done.”
Talkmy absorbs the chores so the calls home become conversation about their lives again. What still reaches the child is what deserves a child's brain — in the founder's success criterion, reducing my help to just judgement calls.
Solve for me, build for others
Talkmy was built to solve the founder's own nightly problem first. That is not a limitation; it is its redundancy. The worst case is bounded in advance: user number one is served every night, so the work cannot go to zero. It is the founder's root decision principle — don't make decisions until you can visualize the escapes — applied to building.
Nobody ships this shape
A landscape scan (2026-11-16) found the field split into three crowded lanes and one empty one: AI companions sold by the month (ElliQ, Sage, Tiina, Meela); simplified hardware (GrandPad, a locked tablet the family administers, inert without its plan); and teaching services that end at the stock ChatGPT login. The empty lane: no product found whose job is to be the front door to general AI for a senior — open the app, speak in plain words in your own language, get real general-purpose help, with no account ceremony, no prompt craft, no companion character, and no family dashboard watching. Custody in the field is inverted: conversations live on the maker's servers and die when payment stops. Talkmy inverts the doors back. (Competitor prices are as published in November 2026; snapshots, not commitments.)
3. The product
The chain
- Open the app, you are in it. No landing page, no gate, nothing to click first.
- Glass hellos. Floating, quiet glass bubbles drift in the chat area — hello in the top twenty languages, in groups of about seven, dissolving and replaced. Nothing bright: brightness is a budget, and the whole budget is spent on the Talk button. Each bubble carries its instruction inside it, in its own language: the greeting, plus “Hold talk to speak.”
- Tap, or speak. Tap her language's bubble and every other hello disappears while the whole screen fades out and reappears in her language. Or simply hold the button and speak — the same fade happens into the language she spoke.
- The manual, on time. Once her language is known, the chat box quietly holds starter text in that language: what she can say, what she can do. “I'm giving them the manual only when I know they can understand it.”
- Hold, speak, hear the answer. The button dents under her finger; a live meter shows her voice being heard; the answer comes back read aloud first, in her language, one thought at a time. A photo can ride along with the question — show it a pill bottle or a letter and ask what is this, without typing a word.
The redundancy stack
No single layer carries the teaching. Each assumes the previous one might fail: the installer's one sentence (“You just use it like we chat, like a person, and you can send photos. I installed it for you. It looks like this.”); the floating hellos; the tap — Apple's setup screens taught the world that the greeting screen is where you tap your language, and Talkmy honors that convention instead of fighting it; speech detection itself; and a teaching pulse on the Talk button that runs across as many launches as it takes, until she clicks and completes a talk successfully — demonstrated understanding, not a timer — then goes still forever. The deepest layer is not in the app at all: a motivated adult child who would gladly spend one kitchen-table hour to buy back years of small stuff.
The hellos, precisely
Top twenty languages, divided by three, ordered by population weighted by likely users, with the main seven drawn from the region the phone already reports — no account, no asking. A phone in California floats English, Spanish, Chinese, Vietnamese, Tagalog, and Korean first; a phone in Bavaria floats German, Turkish, Polish, Russian, and Arabic. Grouped, nobody is ever more than about two dissolves — roughly fifteen seconds — from their language. On a tap or a first message, the crowd bows out and only her language remains.
The acceptance test
“If they can't understand what to do when they open my app, we've failed.”
The parity check that runs it: the founder installs Talkmy on his father's phone, gives the one sentence, then watches silently. His parents are the first beta testers.
4. The doctrines
No unexpecteds
“Unexpecteds are what causes a break.” For a senior mid-thought, a surprise is the thread snapping: where did that come from, what did I press, did I break it. Predictability is the accessibility feature — everything happens where it always happens.
All speech lives in the chat box
“Any speech that needs to come up, goes in the chat box.” Answers, prompts, the starter manual, deferral lines — all of it arrives as text in the thread. No pop-ups, no overlays.
Know, or defer
The scam guard is built in and deliberately not advertised — an advertised protection is a promise, and a promise creates a duty; unadvertised, it is a kindness. Its law: either it knows, or defers. Know — grounded in what was actually shown, naming only the concrete signs genuinely present (urgency, gift-card payment, a sender that does not match, a callback number), never inventing details. Ask — one small gentle request at a time. Defer to people — family, or the bank via the number on the back of her own card, never a number inside the suspicious message. The canonical line: “Let's check together.” Confidence never gets a vote.
Privacy by structure, not by policy
Talkmy only listens while the button is held. No ambient capture, no background recording. Her capsule — her language, her people, how she likes answers given — is built locally on her phone and never uploads. Anyone can copy a button. The moat is the capsule, in the user's own custody.
The asking is never metered
No prompt caps on her screen, ever. Asking again costs nothing: tap any answer and it re-reads itself aloud, as many times as she likes.
5. Architecture
- Ears and voice stay on the phone. Recognition and synthesis run on-device where the platform offers them. Only transcript text travels to the model; audio never leaves the phone, and the capsule file never uploads.
- The brain: Gemini Flash via Firebase AI Logic, with App Check attestation — no API key inside the app — and invisible anonymous authentication, so the no-sign-in doctrine holds. Distilled capsule facts shape each request as context; raw history is not shipped around.
- The offline option: Apple Foundation Models, fully on-device, for users who want nothing to leave the phone. It needs a newer iPhone; on a phone that cannot run it, one calm line in the chat box says so, in her language, and Talkmy keeps working the usual way.
- Messaging is the platform's own door. Talkmy resolves “my son” from the capsule, drafts the message, and presents the phone's own send screen. She taps send. Nothing can be sent behind her back.
- The help hand-off. When she needs a human, the app interviews her gently, assembles a clean summary, and forwards it to the dedicated contact she named while calm — a phone number and an email, that's all it stores. “Even if and when it needs a human, it's still mostly signal.”
- Platform order: iOS first, carved to finished; Android mirrors once. Never both in parallel.
6. Business model
- The core is never sold by the question. Asking is uncapped from the parent's side of the glass. She never sees a price tag on asking for help.
- The children pay, and send it to their parents. Premium is bought by the adult child and gifted onto the parent's phone. The invoice and the gift wrap are the same box.
- Premium carries the business: advanced features, and at the top a staffed service layer — live operators a senior can reach through the app, bought by family as peace of mind. The care features are never the advertised product.
- Filtered advertisers, if advertising ever comes — ratified in advance: “Filtered advertisers. If we have to take less money because we're honest, then so be it. Won't let our seniors be prey.”
- Pricing posture: honest production cost, not rent extraction. A great idea shared freely; a better evolution sold fairly.
7. Roadmap
- Help hand-off — the gentle interview, the summary, the dedicated contact.
- Fall detection, emergency contacts, suggested prompts — prompts only ever as soft invitations.
- The child companion pipe — records and logs for the adult child, opt-in, for those who want always connected, for those with more needs. The records are the parent's; the child sees what she granted, and the grant is revocable.
- Memory care — the far horizon, under a fixed boundary: “Digital care, not as the backbone, but as an additional tool.”
“Talkmy is the tool that allows for the load-bearing parts to be frictionless.”
8. State and honest limitations
State, 2026-11-16: name, icon, tagline, and domain locked; website live; iOS rounds 1 and 2 built and running on a physical test iPhone with the full first-run chain; App Store Connect record created; submission package prepared. The model round — Gemini plus the offline option — is in flight, wired and verified before anything is sent to the App Store. Replies in the current build are stubs.
Limitations, unsoftened:
- The parents test has not run. Motion, real speech, and spoken audio cannot be proven from a simulator. Until the handover happens, every usability claim in this paper is a design claim, not a measured one.
- Starter texts are best-effort translations, flagged for native review before any public launch.
- The scam guard is a discipline, not a guarantee. Know-or-defer bounds its errors, but deferral depends on a human being reachable.
- Email depends on the phone's Mail being configured, which on many seniors' phones it is not. Texting is the reliable lane.
- The offline option excludes older phones by hardware — and says so in the app rather than failing silently.
- Distribution is a sentence, not a channel. The model assumes sons and daughters carry it phone to phone. That is a belief about how families work, held by a founder who lives it nightly — unproven at any scale beyond one family.
Talkmy™ — The AI for everyone.
← Back to Talkmy Privacy Policy