Smart Journal
Hypothesis: privacy-first architecture as a product differentiator — what happens when data never leaves the device.
A private reflection system where AI runs entirely on-device using Gemma. Every journaling app sends your thoughts to a server. This one doesn't — and the entire product architecture follows from that single decision.
The problem
Every AI journaling app processes your data on a server. The AI model lives in the cloud, your entries travel over a network, and your most private thoughts are stored somewhere you don't control. Privacy is an afterthought, not an architecture decision.
The decision
On-device AI as the non-negotiable constraint — and everything the product became because of it. If the AI can't run on your device, it doesn't run. That single decision eliminated a class of product features, forced a different UX architecture, and made privacy the core value proposition rather than a checkbox.
What that forced
- ·No cloud sync — local-first storage with optional encrypted export
- ·Model size constraints — Gemma 2B fits on device; GPT-4 does not
- ·Offline-first UX — the AI must work without a connection
- ·No analytics, no telemetry — what you write stays on your device
- ·Battery and thermal management as UX concerns, not just engineering ones
The UX
On-device AI changes what reflection looks like. Without a server, there's no history sync, no cross-device continuity, no “your AI learned about you over time.” The product had to be valuable in a single session, on a single device. That constraint produced a cleaner product — a daily reflection tool, not a life-logging system.
The AI surfaces patterns within your own entries, suggests prompts based on what you've written before, and offers structured reflection frameworks — all without ever reading your journal on a server.
Stack
Flutter · Gemma 2B · Android + iOS
Status
In development
Core constraint
On-device only — no server AI
Category
Privacy-first · Personal tools