Newsletter
Join the Community
Subscribe to our newsletter for the latest news and updates
Source: TechCrunch's independent report, which credits the image to Anthropic. The same controls are described in Anthropic's official announcement.
Claude shared memory fixes a familiar AI problem: you explain a project in chat, switch to an action-oriented workspace, and have to explain it again. Anthropic now says Claude Chat and Claude Cowork use one memory, with topic-level controls that users can inspect and edit.
That continuity is genuinely useful. It also means a remembered preference, client detail, or project assumption can influence later tasks. The right question is not simply whether memory saves time. It is whether you can see what persists, keep it current, and limit the consequences when it is wrong.
Claude shared memory is Anthropic's new single memory store for Claude Chat and Claude Cowork. Context saved while you discuss a project in chat can now be available when Cowork starts a related task, so the handoff does not begin from zero. Anthropic's official announcement says the memory is organized as topic entries that users can review, edit, or delete, and the official product thread says entries can update while a conversation is still happening. TechCrunch independently reported that the feature is enabled by default on Free, Pro, and Max plans across web, desktop, and mobile. The practical tradeoff is continuity versus retained context: a useful old preference can prevent repeated briefing, while a stale or overly broad memory can shape later work in ways the user may not notice. The sensible first step is therefore an audit, not blind trust.
The evidence supports a narrower story than “Claude remembers everything everywhere.”
Anthropic's launch thread gives a simple example: discuss a project, a manager's preferences, or a client in Chat, then hand a related task to Cowork without restating that context. TechCrunch's report confirms the product intent and notes that users can also ask Claude to remember a specific detail.
The important change is not raw model intelligence. It is continuity across product surfaces. Memory becomes part of the workflow architecture instead of a convenience tied to one conversation.
Source: TechCrunch's article, which identifies this as an Anthropic image. The visible list demonstrates topic-level memory and deletion controls; it does not prove that every remembered detail is accurate.
Claude's current product model appears to have three layers:
The screenshot matters because it makes those layers visible as separate settings. A user can allow search across chats, allow generated memory, and separately decide whether sensitive topics may enter memory. The topic list then gives the user a correction surface.
That is better than an invisible profile, but it creates a maintenance job. A memory about a preferred writing style may stay useful for months. A memory about a deadline, employer, customer, dependency, or security policy can become stale quickly. If Cowork takes actions based on old context, the cost of staleness is higher than it is in a casual chat.
Developers and technical teams should treat memory as mutable application state, not as a magical extension of the model's context window.
Create a harmless project fact, confirm Claude remembers it, then change the fact and inspect the topic entry. The useful metric is not “Did it remember?” It is “Did the correction replace the stale value before the next task used it?”
Stable preferences such as tone, output format, or coding conventions are good memory candidates. Fast-moving task state—deployment status, deadlines, branch names, customer approvals—should still live in an authoritative system that the agent reads at task time.
The official announcement covers Chat and Cowork. Do not assume Claude Code, local Cowork, API sessions, Team workspaces, or Enterprise policies share the same state until the relevant product documentation confirms it for your account.
Memory can reduce repeated briefing, but it should not silently widen permissions. A remembered client or project should help prepare a draft. It should not substitute for confirming recipients, file scope, credentials, or external actions.
Review Settings → Memory after major project changes, role changes, or sensitive conversations. Delete entries that no longer serve a recurring workflow. The feature is easier to trust when the retained state is small enough to understand.
The useful middle ground is not “memory on for everything.” It is memory for stable, inspectable context while authoritative systems continue to hold facts that must be current.
The public evidence does not yet answer several practical questions:
There is also no independent population-wide study showing that shared memory saves more time than it adds in review. TechCrunch confirms the feature and its controls; it does not establish a reliability rate.
Bottom line: Claude shared memory is worth trying for recurring, reviewable work, but the first action should be opening Settings → Memory and treating the saved topics like any other data that can go stale.
Is Claude shared memory available on the free plan?
TechCrunch reports that memory is enabled by default on Free, Pro, and Max plans across web, desktop, and mobile. Mobile users may need the latest app version. Cowork access and memory availability are separate questions, so confirm which Cowork features your plan includes.
Can I see and edit what Claude remembers?
Yes. Anthropic's official thread says saved information appears as topic entries in Settings, where users can read, edit, or delete each entry. You can also explicitly ask Claude to remember a detail. Review the list after major project or preference changes.
Does Claude save sensitive topics by default?
Anthropic says categories such as health information and religious beliefs stay out of memory unless the user enables sensitive topics in Settings. Organizations should still apply their own data-handling rules.
Does the new memory work with Claude Code?
The August 25 announcement confirms one memory across Claude Chat and Claude Cowork. It does not announce Claude Code integration. Treat Claude Code, local Cowork, API sessions, Team, and Enterprise as unconfirmed surfaces until current official documentation says otherwise.
Should developers use shared memory as project storage?
No. Use it for stable preferences and helpful background, not as the authoritative store for deadlines, approvals, credentials, deployment state, or customer records. Consequential tasks should read current facts from a system designed to own them.
Find practical AI products and workflow ideas at AIToolHunt.
