the find
− 0.6 mA knowledge graph of an entire digital life. Parsers walk a phone backup, mail and chat exports, music history and a photo archive, and distill them into a graph: 249 nodes, 1,178 edges, 79 detected communities, spanning 2006–2026. On top sits a self-hosted web app — full-text search across every message thread, semantic queries, views for trips, festivals and music. The README states the covenant plainly: data never leaves your hardware.
This is the Amazon-scale personalization discipline turned inward — a recommendation platform whose entire user base is one person. It is also the memory that the AI companions read from.
cross-section
− 1.3 mFastAPI over SQLite, self-hosted at home, behind a gate that fails closed. Deployment is push-to-deploy — a git push to a bare repo, a hook, a service restart. The graph build itself is fully extractive: 100% of edges traced to source material, zero inferred.
inscription
− 2.0 mField note, from the graph report: The structure is autobiography. The channel mix timestamps when each person entered the corpus — a Gmail era (2006–2018), an iMessage era (2019–2024), a WhatsApp era (2025–2026) — so the platform-migration arc reads as a life story in protocol headers. The superlatives, kept as unnamed aggregates: a single thread of 162,797 messages running since 2006; one friendship of ~54,500 messages across eight different channels; two small friend pods with cohesion 1.00 — circles where everyone knows everyone, with the math to prove it. ⟲ And 70 distant lights — isolated nodes, tied to nothing else in the graph, filed honestly as knowledge gaps.