+the `mgrep` for your entire life — a semantic search engine over all your personal data. you type a query like "what was that restaurant in tokyo" or "when did i talk to sarah about the internship" and it searches across notes, emails, messages, calendar events, photos, documents, and browser history to surface the actual memory. the gap it addresses: you know you experienced or captured something, but locating it requires remembering which app it lives in and which search syntax to use. that friction is high enough that most people just give up.

+

+the technical approach would be a unified ingestion pipeline that chunks and embeds data from all sources into a single vector store, then exposes a natural-language query interface. the indexing layer is the hard part — connecting to all data sources (Apple Messages, Gmail, Notes, Obsidian, browser history, photos via OCR/CLIP) and keeping embeddings fresh. the search layer itself is relatively straightforward once data is in a common format. a local-first design would be important for trust: this is maximally sensitive personal data and running it through a cloud service would be a non-starter for most people.

+

+this is essentially [[axon|axon]] applied to historical data rather than real-time context. the [[universal-data-capturer|universal data capturer]] is the forward-looking complement — capturing everything going forward so there is something to search. [[me-model|me model]] is the more ambitious version: rather than search, train a personal LLM on all this data so it can answer questions about you. [[always-on-ai-assistant|always-on AI assistant]] generates the continuous transcript that would make life search most powerful. the closest existing product is Rewind.ai, which is screen-recording-based and Mac-only — life search is broader (all data sources) and query-based rather than timeline-based.