Claude's memory is now one memory, shared between chat and Cowork

Memory that lives in one product and not the other is a feature you cannot rely on, because you never know which side of the wall you are standing on when you ask. Anthropic closed that gap on August 25: the memory used in chat is the same memory used in Claude Cowork, across web, desktop and mobile.

The part that matters is the settings screen

Saved memories are grouped by topic and each entry can be read, edited or deleted. Memory can be paused or reset at any time. That is what turns a memory feature from something that happens to you into something you operate, and it is the difference between a product you can put a client's material into and one you cannot.

Some categories are excluded by default rather than by promise. Health and beliefs are not stored unless the option to include sensitive topics is switched on, and a harder line is drawn around identification numbers, criminal history and immigration status, which are described as never stored.

Who has it on, and who has to ask for it

On Free, Pro and Max the memory is on by default, with sensitive topics still off. On Enterprise and Team it is the other way around: memory is off for the individual user until that user turns it on, which keeps the decision with the person whose work is being remembered rather than with the plan they were assigned.

Updates happen while the conversation runs rather than as a summary written afterwards, so what Claude keeps is closer to what was actually said than to a reconstruction of it.