WHAT SHIPPED
● One memory across chat and Cowork, two-way. Announced 25 August.
● The caveat most coverage buries: only when Cowork runs in the cloud. Local Cowork sessions do not use memory at all.
● Memories now form during a chat, not by reviewing it at the end.
● No option to separate the two. Anthropic confirmed there is no way to split chat and Cowork memory.
What changed
Until Tuesday, Cowork's memory was bound to individual projects and had nothing to do with what Claude knew about you from chat. Those were two separate systems that happened to sit in the same product.
They are now one. What Claude remembers from your conversations is available the moment you hand Cowork a task, and anything that comes up during a Cowork task carries back to chat. Anthropic's framing: what Claude remembers from your chats is there when Cowork runs a task in the cloud, and vice versa.
The practical version: ask Cowork to draft an update for your manager and it already knows who that is and how they like updates written, without you re-explaining.
The limitation nobody is leading with
CLOUD ONLY
Shared memory works when Cowork runs in the cloud. Cowork sessions running locally on your own computer do not use memory at all.
If you chose local execution for privacy or because your files live on your machine, this update does not reach you. That is in Anthropic's own documentation and largely absent from the coverage.
Memories now form while you talk
The other behavioural change is easy to miss. Previously both systems created memories by reviewing a conversation once it ended. Memory is now generated on the fly during the chat.
That means something said early in a long session can shape what happens later in the same session, rather than only affecting future ones. More useful most of the time. Also worth knowing if you tend to think out loud before arriving at what you actually want.
What you can control
| Control |
Where and what it does |
| See everything stored |
Settings > Memory, organised under Topics. Read, edit or delete any item |
| Turn it off |
Toggle Generate memory from chats. Pause keeps existing memory without adding more |
| Sensitive topics |
Off by default. Health, ethnicity, religious beliefs, politics and gender identity are excluded unless you opt in |
| Separate chat from Cowork |
Not possible. Anthropic confirmed there is no option — they share one system |
| Projects |
Chats inside projects are excluded from the cross-chat synthesis |
Defaults differ by plan, and they are inverted
| Plan |
Memory default |
| Free, Pro, Max |
On. Web, desktop and mobile. Cowork is not on Free |
| Team, Enterprise |
Off. Enterprise members can only enable it individually once an owner enables it for the organisation |
Worth noting the inversion. Consumers get it on by default; organisations get it off. That is the right way round, and it means an enterprise admin has a decision to make rather than a setting to discover after the fact.
Why Anthropic is doing this
It fits a pattern. Cowork became available inside the Claude for Chrome extension recently, and conversations in the Chrome side panel now form part of your usage history. Chat and Cowork have been converging for months, and a single memory is the logical endpoint of that.
The practical benefit is the gap between research and action. You think a problem through in chat, then hand execution to Cowork. Previously that handoff meant re-explaining everything. Now it does not.
What to actually do
| If you are... |
Do this |
| On Pro or Max |
Open Settings > Memory and read what is actually in there. It is on by default and you may not have looked |
| Running Cowork locally |
This does not apply to you. Local sessions do not use memory |
| Using Claude for personal and work things |
There is no partition. If that matters, use separate accounts or keep work in projects |
| An enterprise admin |
Off by default. Decide deliberately rather than discovering it later |
| Tempted by sensitive topics |
It is opt-in for a reason. Turning it on means Cowork sees it too, since there is one store |
The wider point
Shared memory is the thing that makes an agent handoff work. Every measurement this month has pointed at the same constraint: agents perform when the context and the target are specified, and struggle when they are not. A memory that carries across surfaces is context specification you do not have to repeat.
It also means the memory store becomes a more consequential surface. One system now feeds a chatbot and a computer-use agent, and the thing you told it casually in March is available to something taking actions on your behalf in August. Reading what is in Settings > Memory is a ten-minute job worth doing once.
FAQ
What changed with Claude memory?
As of 25 August 2026, Claude chat and Claude Cowork share a single memory rather than maintaining separate ones. Context flows both ways. Memories are also now created during a conversation rather than by reviewing it afterwards.
Does it work with local Cowork sessions?
No. Shared memory applies only when Cowork runs in the cloud. Sessions running locally on your computer do not use memory.
Can I keep chat and Cowork memories separate?
No. Anthropic has confirmed there is no option to split them, because they now use one shared system.
Is memory on by default?
On for Free, Pro and Max across web, desktop and mobile. Off by default for Team and Enterprise organisations, where an owner must enable it before individual members can.
What about sensitive topics?
Health, ethnicity, religious beliefs, politics and gender identity are excluded by default. You can opt in through Include sensitive topics in memory — which also makes them available to Cowork.
How do I see or delete what Claude remembers?
Settings > Memory, where everything is organised under Topics. You can read, edit or delete any individual item, or pause memory generation while keeping what exists.
Does memory cover chats inside projects?
Chats within projects are excluded from the cross-chat synthesis Claude builds from your history.