Claude Cowork Now Remembers What You Told Claude in Chat

Anthropic has connected the memory systems behind Claude chat and Claude Cowork, removing a familiar frustration for people using AI assistants: having to repeat the same background, preferences and project details every time they switch tools.

The update means information Claude retains from a chat can now be available when a user starts a task in Cowork’s cloud-based workspace. Details learned while Cowork is completing work can also flow back into Claude chat.

For users, that could make the difference between an AI tool that answers individual prompts and one that gradually understands how they work.

The update: one memory across chat and Cowork

Until now, moving from a Claude conversation to Cowork could mean starting again with context. A user might explain their company tone, a manager’s preferred update style or an ongoing project in chat, then need to provide the same instructions when asking Cowork to prepare a document or complete a larger task.

Anthropic says the new shared-memory system connects those experiences. If Claude already knows a user prefers concise updates, for example, Cowork can apply that context when preparing an update in the cloud.

The company says memory is now saved as individual topics while a conversation is happening, rather than being created only as a summary after the chat ends. That is meant to make changes, such as a shifted deadline or updated preference, available more quickly.

What Claude will remember

Claude’s memory is designed to retain useful working context, including preferences, recurring tasks and project-related details. Users can also directly ask the assistant to remember something.

Each Claude Project keeps its own separate memory space, helping prevent unrelated personal chats from mixing with a work project’s context. This separation is important for people using Claude across multiple clients, teams or editorial workflows.

The feature is most useful for repeat work. A content team could discuss its preferred headlines, brand tone and publishing structure in chat, then ask Cowork to turn notes into a draft without repeating every editorial rule from scratch.

The cloud-only limitation

There is a major technical boundary: shared memory works when Claude Cowork runs in the cloud. Cowork sessions running locally on a computer do not use the shared memory system.

That distinction matters because local mode is often chosen when a task needs direct access to files on a user’s device. In those cases, users should not assume Cowork will bring in what Claude knows from earlier chats.

Anthropic has been expanding Cowork beyond its early desktop-focused experience. The company recently brought cloud-based Cowork access to web and mobile platforms, allowing tasks to continue even when a computer is offline. The new memory connection makes that cloud workflow more continuous.

Privacy controls are part of the rollout

Anthropic says users can see saved memories as individual Topics in Settings > Memory, where they can edit or delete them. Memory is enabled by default for Free, Pro and Max users on Claude’s web, desktop and mobile apps.

For Team and Enterprise organisations, memory is off by default and can be enabled by an organisation owner. That gives companies more control over whether a workplace AI assistant should retain context across tasks.

Sensitive subjects, including health and beliefs, are excluded from memory unless a user explicitly turns on the option to include sensitive topics. Users can also pause memory or reset it entirely. A reset permanently deletes saved memories, including project memories.

Why this matters for AI agents

The update addresses a core weakness of AI agents: they can be capable inside one task but forgetful between tasks. An assistant that requires repeated briefings can still save time, but it becomes less useful as a long-term collaborator.

Anthropic’s move reflects a wider industry push to give AI products more persistent, user-controlled context. The goal is not simply to store chat history, but to let an assistant carry forward the details that make future work more relevant.

For Claude Cowork, the change could make it more practical for recurring operations, research, content drafting and file-based tasks. The real test will be whether its memory stays accurate, relevant and easy for users to control as it learns more over time.