Claude's memory is an optional feature that generates entries from a user's chats and organizes them into categories, so context carries forward across conversations without being re-explained.

Memory is offered by 📝Anthropic in the 📝Claude app on web, Claude Desktop, and Claude Mobile. Anthropic describes the feature as changing Claude from a stateless chat interface into a collaborator that accumulates understanding over time. It is currently rolling out in two forms: a newer experience found under Settings → Memory, and a legacy experience found under Memory in Settings → Capabilities. Free, Pro, and Max users are being migrated to the newer experience, while Team and Enterprise accounts remain on the legacy one pending their own rollout.

Key Facts

Category: Optional personalization and context-persistence feature
Surfaces: Claude on web, Claude Desktop, and Claude Mobile — not currently available in Claude Cowork
Availability: New experience on free, Pro, and Max plans; Enterprise remains on the legacy experience
Setting: Generate memory from chats, under Settings → Memory
Storage model: Individual entries grouped into categories, written and updated as the conversation happens
Scope: Separate memory space and dedicated summary for each project, kept apart from other projects and from non-project chats

How It Works

Memory is built as a set of discrete entries rather than a single running transcript, and Claude reads, writes, and updates those entries in real time during a chat rather than on a fixed daily schedule. Anthropic states that safeguards and evaluations are applied to memory to help keep users safe. In the legacy experience the mechanism differs: a memory summary is synthesized from chat history and refreshed every 24 hours, providing context to each new standalone conversation.

Memory is distinct from chat search. Chat search retrieves past conversations on demand using retrieval-augmented generation, appears in the conversation as a visible tool call, and is available on paid plans. Memory carries forward what Claude has already recorded about the user; chat search goes back and finds the original conversation.

The feature is scoped by container. Each project maintains its own memory space and summary, so project context does not mix with other projects or with standalone chats. Incognito chats are excluded entirely — they are not saved to chat history and do not contribute to memory.

Why It Matters

Memory addresses the re-explanation cost that accumulates when an assistant starts every session without prior context. Anthropic frames what the feature retains as work-oriented: a user's role and professional context, communication preferences and working style, technical and coding preferences, and details of ongoing projects.

The controls are as material as the capability. Memory can be paused, which preserves existing entries while halting both use and creation of new ones, or reset, which permanently deletes all entries including project memories and cannot be undone. Individual entries can be edited or deleted from the Memory panel, or changed conversationally by telling Claude directly, with edits applying to the next conversation. On Enterprise plans, an Owner disabling memory organization-wide permanently deletes all memory data for every user in that organization.

FAQ

How do I turn Claude's memory on or off?

Open Settings → Memory and use the Generate memory from chats toggle. Users on the legacy experience find the same control under Settings → Capabilities. Enterprise members can only enable it individually after an Owner has enabled it for the organization.

What does Claude remember about me?

Anthropic lists four categories: role, projects, and professional context; communication preferences and working style; technical preferences and coding style; and details of ongoing work. The full set of entries is viewable in the Memory panel, grouped by category.

Does deleting a conversation delete the memory it created?

In the newer experience, no. When a conversation expires or is deleted, entries generated from it are not automatically removed, though any individual entry can be deleted manually. In the legacy experience, deleted conversations are removed from the memory synthesis.

How is memory different from searching past chats?

Memory is context Claude already holds and applies automatically; chat search is a retrieval action Claude performs when asked, visible as a tool call, and limited to paid plans. The two features are toggled separately.

Can I move my memory to or from another AI tool?

Yes. Anthropic offers an import and export path for transferring memory between Claude and other AI services, documented separately and described as experimental and under active development.

📝Claude Code Memory — the separate, file-based persistence layer in the terminal client
📝AI Memory Systems — where platform-native memory sits in the wider landscape
📝MemPalace vs Obsidian vs MythOS as Claude Memory — third-party alternatives to the native feature
📝Importing Your ChatGPT History into Claude — migrating prior context in from another assistant