Claude Now Shares Its Memory Across Chat and Its AI Agent, No More Rebriefing
Anthropic has merged the memory systems in Claude's chat and its Cowork agent, so the assistant carries what it learns from one into the other. Users can read, edit, or delete exactly what Claude has stored.
Key points
- Anthropic on Tuesday merged the separate memory systems in Claude's chat and Claude Cowork, its AI agent (software that can carry out multi-step tasks on your behalf).
- Users can now view, edit, or delete everything Claude has retained, on any topic.
- Sensitive personal data such as health information and political views is excluded from memory by default, with an opt-in toggle to change that.
- The unified memory is available at no extra charge on Free, Pro, and Max plans across web, desktop, and mobile.
- Certain data, including Social Security numbers and immigration status, will never be stored regardless of user settings.
If you've ever had to explain the same project twice to the same AI tool, once while talking it through and again when asking it to act, Anthropic has a fix for that.
The company announced Tuesday that Claude's chat interface and Claude Cowork, its AI agent for completing tasks like drafting documents and running searches, now share a single memory. Whatever Claude learns in one, it carries into the other.
Why did this feel so broken before?
Chat and Cowork previously held separate memories. Spend weeks refining a conference plan in chat, covering headcount and speakers, then open Cowork to draft an agenda and it knew none of it. You'd start over.
Now Claude treats the two modes as one continuous conversation. Anthropic gives a concrete example: ask Cowork to draft a manager update referencing a project from last month, and it already knows the context. No re-entering details.
A second improvement ships alongside it. Claude used to save a memory only when a conversation ended, writing a summary after the fact. It now adds topics as the conversation unfolds, so the shared pool updates in real time, even if you switch between chat and Cowork mid-session.
We first covered Claude Cowork on 25 August 2026, and this memory change is the most substantive update to how the two surfaces relate to each other since launch.
What does Claude actually remember about you?
The update gives users direct control over stored data. A new screen lets you read every fact Claude has filed away and correct or remove any of it.
By default, Claude skips sensitive categories: health data, race, ethnicity, religious beliefs, political views, and gender identity. A toggle labelled "include sensitive topics in memory" lets users opt into storing those. The app sends a notification whenever a sensitive item is saved.
Some data is blocked entirely regardless of settings. Social Security numbers, government-issued ID details, criminal history, and immigration status will never enter Claude's memory.
| Plan | Memory available? | Platforms |
|---|---|---|
| Free | Yes | Web, desktop, iOS, Android |
| Pro | Yes | Web, desktop, iOS, Android |
| Max | Yes | Web, desktop, iOS, Android |
Mobile users on iOS and Android need to update the Claude app to the latest version. Everyone else gets it automatically. TechCrunch AI reported the feature first.
Common questions
Can I stop Claude from storing anything at all?
Yes. You can turn memory off entirely in Claude's settings, or delete individual stored facts from the memory screen at any time.
Does this cost extra?
No. The unified memory ships free across all current Claude plans, including the free tier.
What if Claude stores something wrong?
The memory screen lets you read and correct any stored fact directly, no support contact needed.
The real question worth watching isn't whether this works on launch day. It's whether Anthropic can keep that "never store" list credible as Claude gets more capable and more deeply embedded in people's daily work.
