Alive

A web app that records your voice, transcribes it, and turns your days into a web of the words and labels you keep returning to, so you can see the patterns in your life and live with more purpose.

Built for the Galaxy S24 Ultra viewport (412×915, 19.5:9). One file: index.html, plus sw.js for quiet reminders.

Run it

Serve the folder over HTTP (browsers block storage and the microphone on file:// pages):

python -m http.server 5173

then open http://localhost:5173. On a phone you need HTTPS for the microphone (a tunnel such as ngrok works).

What it does

• Records audio and transcribes it (OpenAI Whisper with word timings, or the browser's speech recognition).
• Each recording becomes a card tied to its audio and transcript: tap to read the transcript, which lights up word by word as it plays.
• A web of keywords and labels, with a ring of your last 60 days.
• Time capsules, quiet-moment reminders, key points and tags per entry.
• Shared sync: one Firebase database for everyone (see below), or an optional encrypted sync to your own GitHub repo.

The Alive server (works on every device, nothing to configure)

The folder api/ holds Vercel serverless functions (free Hobby plan) between the site and OpenAI / GitHub. They keep the two OpenAI keys and the GitHub token as environment variables, so they are never in this repo or in the page, and they only accept requests from this site. They also fix the model (gpt-4o-mini) and cap the reply size, so nobody can run up the bill with expensive calls. The site itself stays on GitHub Pages; Vercel only runs api/.

1. Create a free account at vercel.com (Continue with GitHub is the easy way).
2. Add New… → Project → import alive-app → Framework Preset: Other → leave the rest as is → Deploy.
3. Settings → Environment Variables → add these (Production), then Redeploy:

   Name	Value
   OPENAI_TRANSCRIBE_KEY	your Whisper key
   OPENAI_TEXT_KEY	your text key
   GITHUB_TOKEN	fine-grained token: only alive-app, Contents: Read and write
   GITHUB_REPO	detorotodorov-alexander/alive-app
   ALLOWED_ORIGIN	https://detorotodorov-alexander.github.io

4. Copy your project's address plus /api (for example https://alive-app.vercel.app/api) into apiURL in config.js and empty openaiTranscribeKey and openaiTextKey there. Commit.
5. Because keys were once in this repo's history, rotate them in OpenAI (create new keys, update the two Vercel variables, delete the old).

Limits: a recording up to about 20 minutes (the request body limit is 4.5 MB at the app's 24 kbps). Reading entries and audio doesn't use the server at all; the app reads the public repo directly.

Demo setup (GitHub is the database)

• History: entries/ holds the past entries (a Spanish demo history, Aug–Oct 2026). Any device that opens the site loads them from this public repo; no login.
• Recording live: on the computer used for the demo, press and hold the title to open Settings and paste a GitHub token (fine-grained, this repo only, Contents: read and write). It is saved in that browser only. From then on every recording is uploaded here: the audio to audio/<id>.webm, the entry to entries/<id>.json, and the entry again once it has its key points and its "echoes".
• Echoes from your past: after each new recording the AI reads the whole history and writes how today connects to it, with links to the past entries it draws on (open a card to see it).

Shared keys and database (config.js)

The app has no accounts and no per-user setup. Everyone uses the same configuration, which lives in config.js:

Setting	What it is
openaiTranscribeKey	OpenAI key used only for speech-to-text (Whisper).
openaiTextKey	A second OpenAI key for everything text: key points, tags, summaries, insights, reminder messages (model: textModel, default gpt-4o-mini).
databaseURL	One shared Firebase Realtime Database for all entries. Empty = entries stay on each device.

Edit config.js on GitHub (pencil icon), commit, and the live site updates in about 30 seconds.

Create the shared database (2 minutes): in the Firebase console, create a project, then Build → Realtime Database → Create database. Open the Rules tab and set { "rules": { ".read": true, ".write": true } } (explicit rules don't expire like "test mode" does). Copy the database URL into databaseURL.

This repo is public, so anything in config.js is public. Use keys made only for this project, restricted to the endpoints they need (Audio for the Whisper key, Chat for the text key), with a hard monthly spend limit. OpenAI may auto-disable keys it finds in public repos. The open database rules also let anyone read and write every entry, which is fine for a demo and not for real journals. A real release should put the keys and the database behind a small server.

Everything is online in the shared database: entries and their audio (24 kbps voice, stored in ~900 KB chunks, about 240 KB per minute). Any device can play or transcribe any recording; each device keeps a local copy as a cache. On the free Firebase plan (1 GB) that is roughly 70 hours of recordings.

Project notes

• PIPELINE.md explains how text is processed so nothing is invented (gate, extraction with verbatim quotes, verification, links).
• REMOVED.md lists what has been hidden or removed from the UI and how to bring it back.
• archive/v1-voice-journal/ is a snapshot of the app before the latest redesign.
• Without config.js keys, the app falls back to the browser's speech recognition and local text analysis; keys typed in Settings still work for local testing.