Memory and the Brain
What Suvi remembers, how to inspect it, and how to correct it.
Suvi's memory is the reason it improves week over week rather than resetting. The Brain page in the dashboard is where you can see it.
Three parts and a habit
The notebook. Every conversation and document, searchable by words and by meaning. Ask about "that Dubai buyer" and Suvi finds the right person even if nobody used those words.
The filing cabinet. Suvi's own notes about you: preferences, running projects, things it has learned. It re-reads these before every conversation.
The map. People, companies, deals and commitments as connected entities. Who bought what, who owes whom, what was promised and when.
The habit is reflection: at intervals during quiet hours, Suvi reviews what happened, decides what mattered, and files it.
Two kinds of time
Every fact carries when it was true in the world, and separately when Suvi believed it. Those are different questions, and keeping them apart is what lets you ask "what did we think the terms were in March" as distinct from "what were the terms in March".
When a fact changes, the old version is not overwritten. It is marked superseded and a new version is written. History stays queryable underneath the current answer.
Correcting it
Tell Suvi directly, in chat. "Ravi moved to Gulf Traders in June" or "that is the wrong Priya" both work. Corrections write to memory and change future drafts.
You can also edit entities on the Brain page: merge two contacts that are the same person, split one that is two, or mark a fact as no longer true.
Correcting early is worth more than correcting later. A wrong entity that sits in the graph for a month has shaped a month of drafts.
What it will not do
Memory is per instance. Nothing is pooled across customers, and your content is not used to train models for anyone else.