NoteRecall: semantic note search

Find meetings and client conversations by meaning, not exact words

Problem

After months of notes, keyword search only works if you remember the exact wording. Users recall fragments like 'a client raised a deadline concern' but not which meeting or phrase. Existing note apps either do plain full-text matching or ship the entire vault to a cloud model — slow and token-hungry.

Solution

Build a local SQLite FTS index plus a vector sidecar over notes, retrieve candidate snippets locally, and send only the hits to an LLM for summarization and citation. Users describe what they remember in natural language and get back the most likely notes with highlighted passages.

Users

Consultants, freelancers, PMs and researchers who keep long-term notes

MVP

Import Markdown/plain-text notes; local full-text + vector dual index; natural-language query returns cited snippets; filter by time/person; optional local small model

Revenue

One-time desktop license ($20–40) or subscription for sync and multi-device; free local-model tier as funnel

Why now

Local embedding models and SQLite vector extensions are now lightweight enough for offline desktop apps, and big note apps won't build local-first semantic search for niche users.