Claude's Memory Now Spans Chat and Cowork

Would you want an assistant that takes notes on everything you tell it, all day, across two different apps? Anthropic is betting the answer is yes. As of August 25, Claude's memory works as a single system across Claude Chat and Claude Cowork: whatever the model learns about you in one surface, it now remembers in the other.

Until this update the two products kept separate mental files. You could spend a morning briefing Claude in chat about a client's industry, goals and tone, then open Cowork after lunch and find yourself explaining all of it again. That seam is gone, and the mechanics changed with it: instead of writing a summary when a conversation ends, Claude now records topics while you are still talking, so updates flow between the two apps without a lag. The unified memory is live on Free, Pro and Max plans, across web, desktop and mobile.

The controls, in detail

Anthropic clearly expects the skeptical question, because most of the announcement is about limits. You can read, edit or delete anything memory holds. Sensitive categories such as health information, race, ethnicity and religion are not saved by default, though users can switch that on deliberately. Government-issued IDs and Social Security numbers are never stored under any setting. Persistent memory itself is hardly new territory; ChatGPT has carried a version of it for a long time. What Anthropic is doing differently is merging a chat surface with a work surface into one memory while putting the controls front and center.

Convenience is real, and so is the ledger

The productivity case is straightforward. The biggest hidden cost of working with AI day to day is rebuilding context every session; a shared memory genuinely removes it. Describe your client profiles, your house style and your process rules once, and both surfaces know them. It helps to keep two concepts separate here: the context window is what the model sees within a single session and empties when it closes, while memory is the curated set of notes that now travels across sessions and across apps. This update widens the second layer, not the first.

The governance case deserves equal attention. If your team runs client work through Claude, what lands in memory is now a policy question, not a personal preference. Keep the sensitive-category setting off where data protection rules apply to you, review the memory panel on a schedule, and check that context from one client project does not leak into another. A memory that follows you between apps is a fine servant. Just make sure you are the one keeping its books.

Sources: TechCrunch, Anthropic