Tech • AI • Robotics
Anthropic Just Dropped Claude Memory 2.0 (Full Breakdown)
7/10TL;DR
Anthropic has launched a shared Claude memory system that syncs between Chat and Co-work, lets users inspect and edit saved details, and extends to free, pro and max accounts across web, desktop and mobile, with one key limit: shared memory applies only when Co-work tasks run in the cloud.
KEY POINTS
Unified memory across Chat and Co-work
Claude now uses one memory across Chat and Co-work, addressing a longstanding weakness in Co-work, which often failed to retain user context outside specific projects. The update is designed to reduce repeated explanations and let tasks pick up from prior conversations, including ongoing priorities, project status and personal preferences.
Memory is built during conversation
The system now adds topics to memory as a conversation unfolds rather than waiting until the end to summarize. That shift means users no longer need to repeatedly prompt the assistant to remember specific facts, and relevant details can be captured in real time while work is in progress.
Users can inspect, edit and delete what is stored
A major change is the ability to view saved memory topic by topic. Claude organizes memory into categories such as profile, preferences and specific topics, and users can open individual entries to revise or remove them. That turns memory into a more transparent, controllable layer rather than a hidden background feature.
Sensitive topics are excluded by default
Anthropic says Claude does not store subjects considered sensitive, including health and beliefs, by default. Users can opt in if they want those details remembered to avoid re-explaining them later. The setting is presented as a privacy control, allowing people to decide how much personal context they want included.
Available across plans and devices
The shared memory system is turned on by default for Free, Pro and Max plans. It is available on web, desktop and mobile, meaning a memory created in one interface can be used in another, provided the task is running under the supported mode.
Memories can be managed in settings
Inside account settings, users can access a dedicated Memory section. There, saved information is broken into categories and topics, ranging from personal style preferences to work relationships and recurring interests. The interface also includes an option to import memory from other AI providers such as ChatGPT or Gemini.
Examples show unusually granular recall
In testing, Claude retained highly specific details, including clothing preferences such as avoiding boxy hoodies, disliking stretched cuffs, preferring shirts that are more oversized around the arms but not too long, and favoring cotton fabric. It also saved a later instruction that pants should be styled slightly oversized, then recalled that detail in Chat after it had been entered through Co-work.
The biggest caveat is cloud-only syncing for Co-work
The most important limitation is that shared memory works with Co-work only when a task is run in the cloud. If Co-work is running locally on a user’s computer, the shared memory does not apply. That distinction matters because Co-work offers both cloud execution and local execution, and only the cloud mode carries the synchronized memory back and forth with Chat.
Cloud mode still allows local file access
Even with that restriction, cloud-run Co-work tasks can still work with folders selected from a local computer. In practice, that means users can get synchronized memory benefits without giving up access to local project files, as long as the task itself is launched through the cloud option.
No shared memory yet with Claude Code
The new synchronization currently covers Chat and Co-work only. There is no shared memory with Claude Code, leaving one part of the broader Claude ecosystem outside the new system for now.
CONCLUSION
The update makes Claude more persistent, customizable and practical for day-to-day use, especially for people switching between conversation and task modes. Its usefulness, however, depends heavily on understanding that Co-work memory sync is tied to cloud execution rather than local runs.
Explain this