Postgres as the brain of a personal scraper fleet

Fetching a page is the easy part of a personal scraper project. The harder part is explaining what the page supports, whether it changed, and whether the collector is still working. I built ai_resets in Python with a PostgreSQL database named ai. It tracks AI usage limit resets, plan changes, promotions, free credits, and student offers, from official pages, RSS, Reddit, X watchlists, Bluesky, YouTube, Hacker News, and offer aggregators. This article describes how Postgres gives the project a shared memory that I can inspect with SQL. Each kind of knowledge has a home, including sources, collected items, page snapshots, structured offers, verifications, predictions, and alerts. Repeated collection is ordinary, and conflict handling refreshes an existing row. The pipeline records the attempt alongside the result, with statuses such as running, ok, partial, error, and skipped. A reachable page does not establish every statement associated with its URL. Observation time is kept separate from event time. The database earns its place when a displayed claim can be followed back through verification, captured evidence, and the collection attempt.