Claude Cowork Can Now Remember What You Discussed in Chat

AI assistants often lose their usefulness at the exact moment a conversation needs to become action. A user may spend days discussing a project with a chatbot, only to brief an agent from scratch when it is time to draft documents, organize files, or complete another task. Anthropic’s latest Claude update is designed to remove that handoff.

What is changing

Anthropic is combining the memory systems used by Claude’s regular chat experience and Claude Cowork. Information developed in one environment can therefore be available in the other. Project context, recurring preferences, and useful topics can follow the user from research and planning into execution, making Claude feel more like one continuous assistant than two products sharing a brand.

Anthropic’s example is straightforward: a user might previously have discussed a conference with Claude, including its location, expected headcount, and speakers. Later, the user could ask Cowork to prepare a manager update or create an agenda without re-entering those details. The feature is not presented as a complete transcript archive. Instead, Claude turns relevant themes into reusable memory that can support future work.

The timing of memory updates is also changing. Claude can now add topics while a conversation is still underway, rather than waiting until the chat ends to summarize it. That should make information available sooner when a user switches between chat and Cowork during an active project.

More visibility into retained context

Anthropic is also exposing Claude’s stored memory to users. They can read what has been retained, make corrections, or delete individual information. This matters because shared memory can be helpful only when it remains understandable and controllable. A mistaken preference or outdated project detail could otherwise influence later outputs repeatedly.

Privacy remains a central boundary. By default, Claude will not store personal or sensitive categories such as health information, race or ethnicity, religious beliefs, political views, or gender identity. Users may choose to enable “include sensitive topics in memory,” and the app will notify them when a sensitive topic is saved.

Some information is excluded regardless of that setting. Anthropic says Claude will not store government-issued identification, Social Security numbers, criminal history, immigration status, or information that violates its acceptable-use policy. The company’s description therefore combines persistent context with explicit limits rather than treating memory as an unrestricted record of everything a user says.

Why it matters for agents

The update addresses a practical weakness in AI agents: continuity. A capable model can still feel unreliable if every stage of a workflow starts with a blank page. Shared memory reduces the cost of moving between brainstorming, research, and execution, especially for work that develops over several sessions.

It does not remove the need for oversight. A system that extracts topics from conversations can still misunderstand what is temporary, current, or important. The ability to inspect, edit, and delete memory will be as important as the ability to create it. Claude’s memory feature is enabled by default for Free, Pro, and Max plans on web, desktop, and mobile. iOS and Android users need the latest app version.

Source: TechCrunch AI