Anthropic merges Claude chat and Cowork memory so agents stop forgetting prior context

Anthropic merges chat and Cowork memory

Anthropic has unified the memory system behind Claude's chat interface and its Cowork agent product, according to TechCrunch. The change, announced on Tuesday, means information Claude picks up in ordinary conversation now carries over when a user switches to Cowork to get work done, removing a routine annoyance for people who use the agent regularly.

Until now, the two surfaces kept separate memories. A user could spend weeks refining a project in chat, only to find that Cowork knew none of it and had to be told every detail from scratch. TechCrunch characterizes the update as making Claude behave like a single assistant rather than two disconnected tools sold under one brand.

Memory updates during the conversation, not after

Beyond merging the two stores, Anthropic has changed when memories get written. As TechCrunch reports, Claude now records topics into memory as a chat unfolds, instead of producing a summary once the conversation ends. That matters because it lets new context flow to Cowork quickly — even while the chat is still going — rather than waiting for the session to close.

Anthropic's own illustration, relayed by TechCrunch, is a user drafting a status update for a manager or an agenda for a conference. Because earlier chats covered the subject, Cowork already knows details such as the headcount, the city and the speaker lineup, so none of it has to be entered again.

Users can inspect and edit what Claude remembers

The company is also exposing the memory itself. According to TechCrunch, users can now read, edit or delete any information Claude has retained on a given topic, giving them direct visibility into what the assistant actually knows about them.

The privacy defaults are comparatively strict. By default Claude will not store personal or sensitive material — TechCrunch lists health data, race, ethnicity, religious beliefs, politics and gender identity among the excluded categories. A toggle labelled "include sensitive topics in memory" lets users opt in if they want those covered, and the app will notify them whenever a sensitive topic is actually saved.

Some categories are never stored regardless of settings: government-issued ID numbers, Social Security numbers, criminal history and immigration status, along with anything that violates Anthropic's acceptable use policy.

Rolling out on all plans

The unified memory feature is enabled by default on Free, Pro and Max plans across web, desktop and mobile, TechCrunch reports. On iOS and Android, users will need to update to the latest version of the app to get it.

Why it matters

Context continuity is one of the defining problems for AI agents. An agent that forgets everything between tasks forces users to rebuild context by hand each time, which turns a supposed productivity tool into a source of clerical work. By sharing memory across chat and Cowork, Anthropic removes a large part of that tax and lets users move from discussing a plan to executing it without a seam.

The move also signals how memory is becoming shared infrastructure across a vendor's product line rather than a feature scoped to a single app. And the decision to make stored memories inspectable and editable, with sensitive categories excluded by default, speaks to the retention concerns that inevitably follow products designed to remember their users.