This Day Then

A first demo for a calm memory journal.

The prototype now includes a small built-in backend for private accounts and saved diary entries.

What works

• Audio-first warm-friend check-in UI
• Optional local Chatterbox TTS voice for more natural open-source speech
• Browser demo live voice mode with speech recognition/synthesis where supported
• Green meditative design with a breathing canvas background
• Drafts up to five memory lines from the real conversation transcript
• Creates accounts, logs users in with HttpOnly cookie sessions, and saves entries per user
• Sends a six-digit email code before creating password accounts
• Supports Google sign-in when OAuth credentials are configured
• Lets signed-out visitors use the full diary UI with temporary in-memory entries
• Stores diary data in a local JSON database
• Shows the selected date across five years
• Adds sample same-date memories for demoing comparison
• Light and soft dark themes

Try it

Run the local web service:

npm start

Then open http://127.0.0.1:3000.

To compare OpenAI Realtime voices directly, open http://127.0.0.1:3000/voice-lab.html. The lab lets you choose a Realtime voice, tune the personality notes, and start a live WebRTC voice test before changing the main diary flow. It also shows a local mic meter so phone testing can confirm that the browser is actually receiving microphone input.

By default the app writes its database to data/this-day-then-db.json. To use a different database path, set:

DIARY_DB_PATH=/absolute/path/diary-db.json npm start

Google sign-in

Create a Google OAuth client for a web application, then set these environment variables:

GOOGLE_CLIENT_ID=your-client-id
GOOGLE_CLIENT_SECRET=your-client-secret
GOOGLE_REDIRECT_URI=https://thisdaythen.com/api/auth/google/callback
CANONICAL_HOST=thisdaythen.com

Add the same redirect URI in Google Cloud Console. If GOOGLE_REDIRECT_URI is not set, the server uses the current request origin plus /api/auth/google/callback. CANONICAL_HOST keeps www.thisdaythen.com and thisdaythen.com on the same host so OAuth state cookies survive the Google redirect.

Email registration codes

Password registration now sends a six-digit verification code before the account is created. Production email delivery uses Resend:

RESEND_API_KEY=your-resend-api-key
EMAIL_FROM="This Day Then <hello@thisdaythen.com>"

EMAIL_FROM must be a sender address or domain verified in Resend. For local testing only, set EMAIL_DEV_MODE=1; the server will print and return the code instead of sending an email.

When OPENAI_API_KEY is configured, voice chat uses OpenAI Realtime WebRTC so one tap opens a live, hands-free conversation. The app also uses the Responses API with structured output to draft one to five lines from the user's transcript without adding unstated details or filler. The app keeps OpenAI speech generation with the fable voice as a fallback path when VOICE_MODE=openai-tts is set.

Useful OpenAI voice knobs:

• OPENAI_API_KEY - enables OpenAI voice generation
• VOICE_MODE=openai-realtime - prefer live voice chat, the default when a key is configured
• OPENAI_REALTIME_MODEL=gpt-realtime - model used for live voice chat
• OPENAI_REALTIME_VOICE=alloy - selected Realtime voice
• OPENAI_SUMMARY_MODEL=gpt-5.5 - primary model for memory-line summaries
• OPENAI_SUMMARY_FALLBACK_MODEL=gpt-4.1-mini - fallback if the primary summary model is unavailable
• VOICE_MODE=openai-tts - use Fable speech generation instead of live Realtime chat
• OPENAI_TTS_MODEL=gpt-4o-mini-tts - speech generation model
• OPENAI_TTS_VOICE=fable - selected voice

Realtime voices are alloy, ash, ballad, coral, echo, sage, shimmer, verse, marin, and cedar; marin and cedar are the best starting points for natural quality. fable is available in the Text-to-Speech fallback, not in Realtime live chat.

Natural local voice

For open-source TTS, run a local Chatterbox service in one terminal:

python3 -m venv .venv-tts
. .venv-tts/bin/activate
pip install chatterbox-tts
python local_tts/chatterbox_server.py

Then run the app with the local TTS bridge enabled:

npm run start:local-tts

The app will call POST /api/local-tts, which forwards text to http://127.0.0.1:7861/synthesize and plays the returned WAV audio. If the local model is unavailable, the browser voice remains as a fallback.

To clone or guide a specific voice, point Chatterbox at a short reference clip you have permission to use:

CHATTERBOX_REFERENCE_AUDIO=/absolute/path/reference.wav python local_tts/chatterbox_server.py

Useful knobs:

• LOCAL_TTS_URL - local synthesis endpoint, defaults in start:local-tts to Chatterbox
• VOICE_MODE=local-tts - force local TTS when Chatterbox is configured
• LOCAL_TTS_EXAGGERATION=0.45 - emotion intensity sent to Chatterbox
• CHATTERBOX_DEVICE=cuda|mps|cpu - force the model device