Memories are facts about your business that the assistant keeps between conversations, so you are not re-explaining the same context every time you ask something.
Without them, every conversation starts from nothing. You would say "we do commercial HVAC, mostly service contracts, and our slow season is February" before every useful question.
What gets remembered
Durable facts rather than passing detail: what the business does, how it is structured, which customers or job types matter, standing preferences about how you want things reported.
Not the contents of every conversation. A memory is something worth carrying forward, and most of what is said in a chat is not.
Seeing and removing them
The Memories tab lists what is held. Each can be confirmed as still relevant, or deleted.
Both are worth using. A memory that was true a year ago and is not now is worse than no memory, because the assistant will reason from it confidently. If the business changed — a service line dropped, a location closed, a partner left — the memory about it should go.
Who can see what
Memories are filtered by who you are. Two things apply at once: what the memory is classified as, and who recorded it.
Someone without access to an area of the business does not get that area's context by asking the assistant about it, and a memory recorded in a sensitive context does not surface to everybody in the workspace. See roles and permissions.
The honest limit
A memory is a statement, not a verified fact. It was recorded because it was said or inferred, and the assistant treats it as background rather than as evidence.
Anything that matters to a number comes from your connected data instead — see how Omnyra reaches a number. If a memory and your data disagree, the data is right and the memory should be deleted.