Claude and Cowork Now Share One Memory

Anthropic has removed the memory barrier between Claude chat and its Cowork agent. With the August 2026 update, Claude can carry relevant context from a chat into a cloud-based Cowork task, while information learned during Cowork can flow back into later chats. 17181926

The practical effect is less rebriefing when moving from planning in chat to having Cowork perform work. The important limitation is that locally run Cowork sessions do not use this shared memory. 2126

What the shared memory changes

Previously, context built in Claude chat did not automatically follow a user into Cowork. The update makes the relationship two-way:

• A conversation can establish relevant project details, priorities, people, or preferences that Cowork can use later.
• A Cowork task can add information that becomes available in subsequent Claude chats.
• Anthropic does not provide a setting to keep separate memories for chat and Cowork; the two surfaces share one memory system. 1921

This is designed for workflows that move between discussion and execution. Someone might first use Claude to think through a project and then hand Cowork a task related to it. Cowork can start with the context already saved instead of requiring the user to repeat the background. 171832

Memories are created while work is happening

The update also changes when Claude builds memory. Rather than waiting until a conversation has ended and then summarizing it, Claude can create and update memory topics during an ongoing chat or Cowork session. 212728

That makes the shared system more useful for extended work: new decisions or preferences can become part of the context while the interaction continues, and later work can draw on the updated information. The available evidence describes this as real-time or live updating, but it does not specify how quickly every individual change appears across surfaces. 2127

Users can manage memory topic by topic

Saved information is exposed under Settings → Memory as individual Topics. Users can review what Claude has stored, edit a topic, or delete it. 17222526

This gives users more control than a single all-or-nothing memory store. It also makes it possible to correct outdated information—for example, a project priority that has changed—without removing every other saved topic.

There is one structural trade-off: because chat and Cowork use the same memory, users cannot create a separate memory boundary between those two products. Deleting or changing a shared topic therefore affects how both surfaces can use that information. The sources confirm that there is no option to split the memories, but do not describe every consequence of editing or deleting a topic. 1921

Sensitive information is opt-in

Memory excludes sensitive topics by default. The documented examples include health, race, religion or beliefs, political views, and gender identity. Claude stores memories in these categories only if the user explicitly enables the relevant sensitive-topics setting, and users receive a notification when Claude does so. 2223

The provided evidence does not identify the complete list of information that remains permanently ineligible for memory. It is therefore not possible to state that full list reliably here.

Which plans and platforms get the feature?

For individual accounts, memory is enabled by default on Free, Pro, and Max plans. Reports describe availability across Claude’s web, desktop, and mobile experiences, although the chat-to-Cowork sharing itself depends on Cowork running in the cloud. 172225

For Team and Enterprise plans, memory is off by default. An organization owner can enable it, making availability conditional on the organization’s setting rather than automatically active for every business account. 222526

The available sources do not establish additional seat, regional, or deployment-specific eligibility rules for Team and Enterprise users. Organizations should therefore verify the feature in their own administrative settings before assuming that every account has access.

Cloud Cowork is the key limitation

The update does not mean that every Cowork session can access the shared memory. Anthropic’s support documentation specifies that the memory connection works when Cowork runs in the cloud; Cowork sessions running locally on a user’s computer do not use it. 26

That distinction matters for anyone switching between desktop workflows. A task may be launched from the desktop Cowork experience, but the relevant question is whether the task is being executed in the cloud or locally—not simply which app window is open.

Claude Code remains outside the shared system

Anthropic’s change covers Claude chat and Cowork, not Claude Code. Claude Code remains separate from the shared Chat/Cowork memory arrangement. 19

In short, the update creates one memory across Claude’s conversational and cloud work surfaces, while preserving a boundary around Claude Code and locally run Cowork sessions. For users who regularly plan in chat and execute tasks in Cowork, the main benefit is continuity; for teams and privacy-conscious users, the important details are the administrator-controlled default, the cloud-only limitation, and the topic-level controls.