Anthropic Merges Claude Chat and Cowork Agent Memory

Anthropic has integrated the memory systems of its Claude chat interface and its computer-use agent, Claude Cowork. Previously, Cowork sessions kept memory isolated, forcing users to manually re-explain context when moving from chat-based planning to agent execution. Now, when Cowork runs in the cloud, it automatically inherits the context, project details, and preferences established during standard chat conversations. However, this shared memory feature does not apply to Cowork sessions running locally on a user's computer.

The update also changes how Claude stores information. Instead of summarizing conversations after they end, Claude now updates its memory in real time during active chats, saving details as individual, categorized topic files. Users can view, edit, or delete these files under the Topics section in their Memory settings, allowing for easy corrections that propagate across all future sessions. Additionally, Anthropic has introduced an opt-in toggle to Include sensitive topics in memory. This allows Claude to remember personal details like health conditions or religious beliefs, though the system will still refuse to store highly sensitive data such as government IDs or criminal histories.

The rollout strategy varies by tier. Memory is enabled by default for Free, Pro, and Max plans across the web, desktop, and mobile apps. For Team and Enterprise plans, the feature is disabled by default and must be turned on by an administrator. For developers and practitioners, this shift dramatically improves agent ergonomics. Instead of repeatedly pasting system prompts or re-explaining metrics, users can brainstorm an agenda in chat and have Cowork immediately build a corresponding budget. The new topic-file system acts like a manageable configuration file, making the AI's memory transparent and easily editable.