Anthropic unifies Claude's memory across chat and Cowork in single integrated system

Anthropic rolled out a unified memory system this week that merges how Claude tracks context across its chat and Cowork services. Previously, the two platforms operated in isolation—Cowork stored project-specific details while chat maintained separate user knowledge. Now, what Claude learns in one place carries seamlessly into the other.

How the integration works#

The new architecture means Claude arrives at each task already familiar with your preferences, priorities, and ongoing projects. An example from Anthropic illustrates the payoff: discuss conference logistics in chat, then ask Cowork to build a budget—the system already knows the attendee count, location, and speaker roster without prompting for a recap.

Another key change is real-time memory updates. Previously, Claude would refresh its understanding only after a chat session ended. Since Cowork often runs for hours or days, the system now ingests and stores new context continuously as you work, making multi-session projects far more coherent.

What Claude remembers and what it won't#

By default, Claude sidesteps storing what it identifies as sensitive content: health records, religious beliefs, race, ethnicity, and gender identity. However, users can toggle a setting to override this safeguard if they prefer the system to retain such information. The mechanism is binary—there's no granular control to include some sensitive categories while excluding others.

Certain data never enters memory regardless of settings: government identification numbers, criminal history, immigration status, or anything that violates Anthropic's Acceptable Use Policy.

The memories themselves appear as small topic-organized files that users cannot edit directly but can modify through conversation with Claude in the settings menu.

Rollout and competitive positioning#

The unified memory launches on by default across Free, Pro, and Max tiers on web, desktop, and mobile. Enterprise and Teams users have the feature disabled by default. Anthropic also emphasized that memory data receives no ad-targeting treatment—a distinction it made pointedly in reference to OpenAI's recently introduced Computer History feature, which tracks user activity on Mac devices.