How your knowledge base builds itself
Two different layers that are easy to mix up
Section titled “Two different layers that are easy to mix up”Wiki is text: articles about topics and entities that the agent writes and keeps up to date from your own data. The Brain isn’t text — it’s a 3D map: the same data, but as a network of links, where wiki articles are just two node types among many others (notes, mail, tasks, meetings, contacts, documents…). You read the wiki; in the Brain, you look at what’s connected to what.
It fills itself in — that’s confirmed, not marketing
Section titled “It fills itself in — that’s confirmed, not marketing”If you just keep notes, get mail and make recordings the ordinary way, doing nothing special for a “knowledge base” — it genuinely fills itself in. After a note is saved or a document uploaded, a background queue, with no button involved, extracts durable facts, people, companies and topics, and adds graph edges. The memory-status screen says exactly that: if it’s empty, “it will fill in on its own as you work.”
Notes also have a manual path for links — [[bracketed]] wiki-links with
autocomplete, and a “Find links in text” button for an already-saved note
— but those add to the automation, they don’t replace it.
Three spots worth checking rather than trusting blindly
Section titled “Three spots worth checking rather than trusting blindly”Money and meetings need your confirmation. Money facts from email and tasks extracted from a meeting summary only become structured objects once you explicitly agree — unlike ordinary concepts and links, that isn’t silent automation.
Guessed links can be wrong. An automatic link carries a confidence level, not a guarantee — uncertain ones are marked as a suggestion, and a node’s panel in the Brain always shows “Extracted from” so you can check. Glancing at the Clusters view now and then isn’t decorative: it shows what’s sitting on the periphery with weak links — things that didn’t get picked up on their own.
Where to search for what’s accumulated
Section titled “Where to search for what’s accumulated”Search inside the 3D map itself is literal, by node name — it’s just a filter on the scene, not a search by meaning. Real semantic search lives at the module and agent level: ask the agent in your own words and it searches by content, even if the exact words of your query aren’t in the text.