Legacy
A memory for later
A private memory vault that captures photos, voice, and stories, and uses them to build an AI version of you that can keep speaking to your family, your loved ones, and your children long after you pass away.
Why I'm building this
I'm a mom. One day I started thinking about what happens when I'm no longer here, and whether my kids would really remember the small stuff. The stories behind the photos. The way I'd tell them. The sound of my voice.
Most of what survives a person is the things they wrote down, the photos they took, and whatever other people happen to remember. That is a lot of gaps. Photos show you what something looked like, but not what it felt like, or what led up to it, or what happened right after. Videos help, but most of us don't film our ordinary days, and the ordinary days are the ones our kids will want back.
I wanted to build something that captures the story alongside the photo. A voice recording of me telling it. A written version when I can't record. The date, the people, the context. So that decades from now my kids can ask "Mom, what was I like as a baby?" and actually get an answer, in my voice, with the details I'd give if I were still here to tell them.
It started as a personal thing. Then I realized that if I want this, other people probably want it too.
So I'm building it for myself first, but with the architecture in place from day one to become something other people can use. A quiet, careful product for anyone who wants to leave their voice behind for the people they love.
The memories you save here aren't just an archive. Every photo, every voice recording, every story you tell becomes training material for a private AI trained to be you. It remembers your stories. It knows your personality. It knows how you talk, what you cared about, how you'd answer.
When you're no longer here, your kids, your grandchildren, your partner, the people who love you, they can open the app and ask you something. About a day. About a person. About who they were as a baby. About what you thought of them. And they get an answer back, in your voice, from the version of you who actually lived it.
This is not a scraped chatbot cobbled together from social media posts after you're gone. It's something you build deliberately, over years, while you're alive. A careful, chosen version of yourself, saved for the people who'll still want to hear from you after you're no longer around to tell them yourself.
What this is meant to do
Technical architecture
Legacy is built in three stages. Stage 1 is live today. The architecture is multi-tenant-ready even while only one user (me) is live on it. Every data record carries a userId, every storage path is user-scoped, and every API call is designed as if there were already a hundred users. When it's time to open it up, switching from personal to public is an additive change, not a rewrite.
Installable on iPhone from Safari, using Add to Home Screen. Works fully offline. No App Store tax, no review cycle, instant deploys.
- Vanilla HTML / CSS / JS, no build step
- IndexedDB for structured records and Blob storage
- Service worker for offline caching
- MediaRecorder API for voice capture
- GitHub Pages and Actions for CI/CD
Server layer for sync, cross-device access, and feeding the AI. Designed around the same schema as the local app, so the migration is mostly plumbing.
- FastAPI with async PostgreSQL
- Cloudflare R2 for media (no egress fees)
- Presigned URL uploads, client to storage direct
- Whisper API for voice transcription
- Fly.io for hosting and managed Postgres
Chat with an AI that answers as you, grounded in the memories you've saved. Not a generic chatbot. A specific person, speaking in your voice, about your life.
- Anthropic Claude or OpenAI GPT-4 via API
- RAG over memories, semantic and filter hybrid
- Persona core always in system prompt
- Voice samples as few-shot for tone
- Prompt caching to keep per-turn costs near zero
The whole thing is designed to survive even if the business doesn't. No vendor lock-in, no proprietary format, no hostage data.
- One-tap export, JSON with base64 embedded media
- Documented, versioned schema
- Encryption at rest (backend phase)
- No training on user data, ever
- Designated heir access for estate planning
Where this is going
The business case
I'm building this primarily because I want it to exist. But the market for it is real. People increasingly understand that AI is going to reshape what it means to leave something behind. There's a growing space around "digital legacy," but most of it is either passwords for heirs or generic chatbots trained on scraped Facebook posts. Neither is what I'd want for my kids.
Legacy takes the opposite approach. A careful, private capture experience first. Then, only after years of real memories, the AI layer on top. Earned, not extracted.
Direct to consumer SaaS, billed via Stripe. 97% revenue retention because it's a PWA, no Apple 30% tax. At scale, infrastructure costs are roughly $1 to $10 per user per month depending on usage, giving healthy margins.
Long-form plans (annual or multi-year prepaid) make sense here given the emotional permanence of the product.
Primary segment: parents of young children who want to leave something substantive behind. Secondary: grandparents, people processing a diagnosis, adult children capturing stories from aging parents before it's too late.
The unifying thread isn't demographic. It's the desire to leave your voice behind for someone specific.
Legacy is not trying to be a social network, a photo cloud, or a generic AI companion. It's a quiet, private, emotionally serious tool designed for one use case and executed carefully.
The product aesthetic (warm paper tones, serif typography, zero feeds or followers) is part of the moat. It tells the user what this is and what it isn't.
Classic subscription-product dynamic amplified by emotional weight. Once someone has captured five years of memories here, they're not leaving. And they shouldn't have to. Portability guarantees (full JSON export) turn that into a trust asset instead of a lock-in risk.
The AI layer compounds the effect. The longer you've used Legacy, the richer the AI gets. That's a flywheel that only this kind of slow-captured dataset can produce.
Stage 1 is live. Open it in Safari on an iPhone and tap Share, then Add to Home Screen, to install it as a real app. The repo is public. Everything is there, including the architecture docs.