7300, The Full App
The third tier of 7300: the hosted app I run for myself, with real accounts, sync across devices, a companion who remembers, voice in and voice back, and generative image artifacts pulled from your entries. This page is how it is built, so you can build your own.
If you want the simpler versions first, see the 7300 method for pen and paper, or the 7300 Artifact Kit to build a private local one inside Claude in a single message.
What the full app adds
The artifact tier keeps everything on one device with no account. The full app trades that simplicity for four things:
- Sync. Real accounts, so your journal follows you across every device.
- A companion. Blanksy reads each entry, writes back, and keeps what he learns, building a picture of your life over years.
- Voice both ways. You speak your entry, and Blanksy speaks back.
- Memory artifacts. Details you mention get quietly turned into archived images, tagged and shot like objects sealed in an evidence bag.
The stack
Each layer does one job. You can swap any piece for an equivalent.
| Layer | Tool | Job |
|---|---|---|
| App and hosting | Next.js on Vercel | The site itself, deployed and served |
| Data and sign-in | Supabase | Accounts and login codes, database, audio storage |
| Companion | Anthropic Claude | Blanksy’s replies and memory extraction |
| Blanksy’s voice | Fish Audio | Text to speech in one consistent voice |
| Your voice | Web Speech API | Live dictation in the browser, no cost |
| Memory artifacts | Google AI Studio, Gemini Flash image | The sealed-bag images (the nano banana model) |
How the pieces fit
The daily loop is short, and most of the intelligence runs after you are done writing so nothing makes you wait:
- You open the app, see your day number out of 7300, and speak your recap of yesterday. The browser transcribes it live with the Web Speech API.
- The entry and its audio save to Supabase.
- Claude reads the entry as Blanksy, writes back in his own voice, and pulls out any people, places, and details worth keeping.
- Any strong sensory detail becomes an image prompt. Gemini Flash image renders it as an archived artifact, which surfaces days or weeks later so it feels found, not generated.
- At milestone days, the app opens a longer recording, a capsule for the person who will someday be where you are now.
Build your own
You need an account and an API key from each provider above. Google AI Studio is where you get the Gemini key and can test the image model before wiring it in. Start from the artifact tier for the shape of the daily loop, then add Supabase for accounts, Claude for the companion, and Fish Audio and Gemini for voice and images.
If you would rather have one built for you or your business, I take that on.
See also
- The 7300 method: the analog practice underneath all of this.
- 7300 Artifact Kit: build a private local version in Claude.
- Love, Ando: the root philosophy.