Not a search box over old transcripts. A living model of how one mind actually thinks — and how it answers questions it was never asked.
Most "AI knowledge bases" store what was said and search it back to you. 2ndself does something almost no one is doing: it builds a small, deduplicated model of how a person reasons — their frameworks, standards, decisions, and the why behind each — so it can think in their voice on questions that were never in the transcript.
The whole design is one inversion: stop optimizing to remember everything, start optimizing to reason like the person.
If you read nothing else, read these. The docs explain each in depth, tied to the live code.
Every entry must pass one test — would this help answer a question you were never asked? That's what lets it speak on new things.
A meeting yields a handful, deduplicated — not a transcript. When you repeat yourself, one entry gets stronger instead of ten piling up.
Confidence (how sure) and durability (how lasting) are judged separately — so core principles never fade while today's takes are allowed to.
When you run the same move the opposite way somewhere, that's stored as a variant with its context — not flagged as you contradicting yourself.
A new opposing position supersedes the old one, which is kept in history — so the model tracks how your thinking actually evolves.
The corpus self-heals, fails closed on anything unverified, and stays a small, deduplicated model — not an append-only log that rots.
One writer turns conversation into a model, once a day. One read-only API answers questions, instantly. They never touch the same data at the same time — the single-writer rule.
Every field, every threshold, every design decision — pulled from the live code, not from memory.
Open the documentation →