§ Overview
A personal archive you can search in everyday language. Save a photo or a few words about your day, and years later find it again by asking vaguely, the way you actually remember it. A query like "the dance I saw in Bali" finds the right story even when neither word was ever written down. A native iOS app backed by a Python API, with an AI chat that answers only from your own stories and cites every one it uses.
§ Key features
- 01Capture a story in seconds: a photo, a few words, or both
- 02Semantic search in everyday language, so a story turns up without remembering its exact words
- 03A vision model writes a neutral caption, reads visible text, and tags every photo, so a photo with no note at all can still be found
- 04Ask your stories directly: an AI chat that answers only from your own archive and cites the stories behind each answer
- 05Related stories connect automatically
- 06A journal view of what you saved, plus a Plus subscription through RevenueCat
§ Challenges & solutions
Success was defined as recall, not AI: the moment someone types a vague question about something they forgot they saved, and it comes back. Every technical choice was measured against that. Search, embedding and keyword matching run concurrently in a single query against Postgres with pgvector, so relational data, vectors and full text live in one database instead of a separate vector store. In the chat, a model citing a story is a request, not a guarantee, so every citation is validated against what was actually retrieved and an answer with no surviving citation is dropped. Before building the app at all, the product had to pass a 30-day gate as a Telegram bot in daily personal use.