What is brain.db?
A single SQLite database that watches the YouTube channels, scrapes the RSS feeds, holds the transcripts, indexes the wikis, and answers questions in plain English. 528,000+ items and growing every night at 1 AM.
brain.db is the spine of 1n2.org. Everything that gets ingested — every Mad Bitcoins episode title, every WCN livestream description, every Bitcoin Group panel sign-off, every Wikipedia article on a Curio Card artist, every Google News result for "Bitcoin ATM scam" — lands as a row in one SQLite file on the server.
The numbers are big in the way personal databases get big when you let them run for years: 528,000+ items across the canonical sources, plus 1,666 episode transcripts, plus 78 community-built wiki articles, plus the daily news firehose. The shape of the data is deliberately boring — title, date, source, body, tags, embeddings — because the value isn't in the schema, it's in the coverage.
What gets ingested
The nightly cron at 1 AM pulls from a handful of feeds:
- YouTube — the World Crypto Network, Mad Bitcoins, and Bitcoin Group channels. Every video's title, description, view count, and comment thread.
- Transcripts — Whisper has been run over the entire back catalog. The transcripts are messy (more on that below) but searchable.
- Google News — daily fetches on a watchlist of topics: Bitcoin, sanctions, AI, mining, kiosks, Curio Cards.
- RSS feeds — a long tail of crypto and tech publications.
- Ethereum blockchain — Curio Card transfers and floor prices, indexed live.
- Social feeds — Twitter for the main accounts, scraped daily.
Once a thing is in the database, it stays. Episodes from 2013 sit next to episodes from last week.
How Gemma enriches
Raw ingestion is half the story. The other half is a local Gemma model that runs nightly across the new rows, doing the kind of work a human researcher would do if they had infinite time:
- Tagging — every transcript chunk gets topic tags so the search index actually returns something useful.
- Summaries — long episodes get a paragraph-level summary attached.
- Entity extraction — names, ticker symbols, locations, dollar amounts. These become the cross-references that turn a flat archive into a graph.
- Embeddings — semantic vectors so "stripper" and "exotic dancer" return the same row when somebody searches.
None of this is glamorous. It's the difference between a tape library and a research tool.
The /brain/ask endpoint
The user-facing surface of brain.db is /brain/ask — a Q&A box that takes a plain-English question and returns an answer cited to actual rows in the database. Ask it "when did Andreas Antonopoulos first appear on TBG?" and it pulls the panel sign-on dates from the Bitcoin Group archive, returns the date, and links to the episode.
The endpoint is intentionally narrow. It doesn't try to answer questions about the wider internet. It answers questions about what is in brain.db. That's both its limitation and its honesty.
The Whisper-mangle problem
Transcripts are not perfect. Whisper does its best on names it has never heard — "Curio Cards" sometimes becomes "Cure-EE-oh Cards"; "Andreas Antonopoulos" sometimes becomes "Andre Anono-poulous"; the word "Bitcoin" is occasionally rendered as "big coin" in noisy episodes. Anything that pipes brain.db results into a report has to know about this. The reports philosophy on 1n2.org has a whole section on it.
The galaxy view
Because brain.db has embeddings, you can plot it. Brain Galaxy is the visualization: 528K items projected into 2D, clustered by topic, with the dense knots showing where coverage is deep (Bitcoin Magazine, ATM scams, Curio Cards) and the sparse arms showing where it isn't (anywhere outside the show ecosystem). It's the database, but as a picture.
Why a personal database
The premise of 1n2.org is that one human plus one AI can keep a meaningful archive going if the plumbing is right. brain.db is the plumbing. The wikis, the reports, the dashboards, the magazines — they all sit on top of it. When a new Mad Bitcoins episode drops, the same row that goes into the database also surfaces on the homepage, gets quoted in the next report, and feeds the embeddings that let "/ask" answer the next question.
It's one database. It's the answer to nearly every question about why anything on 1n2.org exists.