On August 25, Anthropic announced that it has merged the memory used by Claude chat with the memory used by Claude Cowork, its work-execution mode. What Claude learns in one place is now available in the other, so you no longer have to restate the background every time you hand over a task. The stored material can be reviewed, edited, or deleted from settings, and the feature is on by default for a wide range of users, including those on the free plan.
No more re-briefing
Until now, conversations in Claude chat and the actual work handled by Cowork lived in effectively separate places. You could spend several exchanges shaping the direction of a project, then have to write out the assumptions and the names of everyone involved again once it was time to produce the document. The AI you talked things through with and the AI that did the work were, in practice, two different colleagues inside one service.
This update removes that boundary. Cowork can draw on the context built up in chat, and anything Cowork picks up while working flows back into chat. Anthropic points to examples such as writing a progress update for a manager or putting together a conference agenda. Details like attendee counts, the host city, and the confirmed speakers carry over from earlier conversations, so the request itself can be a single short sentence.
Memory grows while the conversation is still going
The way memories are created has changed as well. Claude previously summarized a conversation once it ended and stored that; now important topics are written to memory as the conversation unfolds.
It looks like a small change, but it matters a lot if you move back and forth between chat and Cowork. A condition you settled on moments ago is reflected right away, even in an ongoing session. You no longer have to close out a conversation before handing off the work.
You can open up what Claude remembers
Anthropic paired the change with a commitment to show users what Claude is holding on to. Saved material appears in the memory section of settings as a list of short files organized by topic, and each one can be read, rewritten, or deleted individually. Edits carry into later conversations. There are also controls to pause memory generation or reset it entirely.
By default, sensitive information is not added to memory. That covers health conditions, race and ethnicity, religious beliefs, political views, gender identity, and similar categories. If you do want those remembered, enabling the include sensitive topics in memory setting allows it, and the app notifies you when such a topic is saved.
Some information is never stored, regardless of your settings. Government-issued ID and Social Security numbers, criminal history, immigration status, and other data that falls outside the acceptable use policy all sit in that category. The scope of memory widens for convenience, while the line around what gets handled stays fixed by the company.
Plans covered and how to get started
Shared memory is on by default for the free plan, Pro, and Max, and works on the web as well as in the desktop and mobile apps. On iOS and Android, the updated memory feature only appears once the app is on its latest version. If nothing looks different on your end, checking for an app update is the quickest place to start.
When you delegate work to an agent, more time tends to go into sharing background than into the instruction itself. Unifying memory is not a flashy new capability, but it cuts directly into that overhead, which makes it a meaningful change for everyday use.
Summary
On August 25, Anthropic unified memory across Claude chat and Claude Cowork, its cloud work mode. Context gathered in conversation carries straight into the work, and memory now updates while a conversation is still in progress. Saved items can be reviewed, edited, or deleted one by one in settings, and sensitive categories such as health and beliefs are excluded by default. The feature is already active across plans, including the free tier, and mobile users need to update their app.
