Memory & Cognition

Claude Memory

Claude Memory is Anthropic's feature that lets Claude retain context across separate conversations by building an editable, project-scoped summary of your preferences and ongoing work, complemented by a "search past chats" tool on paid plans. It is optional, controllable from Settings, and can be paused, reset, or bypassed with incognito chats.

What is Claude Memory?

Claude Memory is Anthropic's feature that lets Claude remember context across separate conversations, so you do not have to re-explain your preferences, projects, and goals each time you start a new chat. Instead of treating every conversation as a blank slate, Claude builds a running summary of what it has learned about you and your work, and pulls that summary into future chats. As of mid-2026 the memory summary is available on every Claude plan, including the free tier, and Anthropic describes it as fully optional with granular controls.

Memory in Claude is two related capabilities. The first is memory itself: an automatically generated, editable summary of key details Claude infers from your conversations, such as how you like responses formatted or which programming languages and frameworks you use. The second is chat search, marketed as the ability to "search past chats", which lets you ask Claude to find and reference specific things you discussed before. Anthropic's help documentation notes that this search runs as a retrieval step and appears as a visible tool call inside the conversation.

A defining design choice is that Claude's memory is project-scoped. Anthropic states that Claude creates a separate memory for each project, so context from one initiative does not bleed into another. Work projects, personal projects, and standalone chats each maintain isolated memory, which doubles as a privacy and safety guardrail for people who keep sensitive or unrelated workstreams in the same account.

Lets Claude carry context (preferences, projects, goals) across separate conversations.
Two parts: an editable memory summary, plus "search past chats" retrieval on paid plans.
Memory is project-scoped, so context stays compartmentalized per project.
Anthropic describes it as fully optional, with controls to view, edit, pause, or reset it.
Distinct from the Claude Code CLAUDE.md memory files used by developers in a repo.

How does Claude Memory work?

Claude generates memory automatically as you chat. Rather than storing raw transcripts, it produces a synthesis of key insights: it infers details about your preferences, ongoing projects, and context, and writes them into a summary you can open and edit at any time. Anthropic's help center indicates that for standalone chats this summary is refreshed on a recurring cadence (roughly every 24 hours), so newly stated preferences are folded in over time rather than instantly after every message.

Because the memory is a human-readable summary, you are not locked out of what Claude knows. You can review exactly what it has stored, correct anything that is wrong or outdated, and edit the summary directly. This transparency is part of the design: the goal is a memory you can audit and steer, not a hidden profile. Each project keeps its own summary, and Anthropic notes that searches inside a project are limited to that project's history.

Chat search is the retrieval half of the system and is separate from the memory summary. When you ask Claude to find what you discussed before or to continue where you left off, it searches your prior conversations and brings the relevant pieces into the current context. Anthropic's documentation describes this as a retrieval-augmented step, surfaced as a tool call so you can see when Claude is pulling from past chats. Searching past chats is restricted to paid plans, while the memory summary itself is available across all plans.

Memory is an auto-generated, editable summary of inferred preferences and context, not raw transcripts.
Standalone-chat summaries refresh on a recurring cadence (about every 24 hours).
You can open, edit, and delete what Claude has stored at any time.
Chat search retrieves from prior conversations and appears as a visible tool call.
Project memory and project search are isolated to each individual project.

Which Claude plans have memory, and when did it roll out?

Claude Memory reached users in stages over roughly six months. Anthropic first brought automatic memory to Team and Enterprise plans on September 11, 2025, positioning it for collaborative and organizational use. On October 23, 2025, Anthropic announced that memory was rolling out to Pro and Max plan subscribers, after additional safety testing and targeted adjustments to how memory behaves.

The final stage closed the paywall on the memory summary. In early March 2026 Anthropic made the memory summary available to free-tier users as well, so cross-conversation recall is no longer gated behind a paid subscription. Anthropic also offers a tool to import or export memory between Claude and other AI assistants, which it documents as a way to back up or migrate stored memory.

One distinction persists across tiers. The memory summary is available to everyone, but the "search past chats" capability remains a paid feature, available on Pro, Max, Team, and Enterprise plans across the web, the Claude desktop app, and the Claude mobile apps. Incognito chats, covered below, have been available to all users since the feature launched. Note that Claude Memory in the consumer apps is separate from the project memory files (CLAUDE.md) that the Claude Code developer tool reads from a repository.

Team and Enterprise: automatic memory launched September 11, 2025.
Pro and Max: memory rolled out October 23, 2025.
Free plan: memory summary added in early March 2026.
"Search past chats" stays paid-only (Pro, Max, Team, Enterprise).
Memory can be imported from or exported to other AI tools for backup or migration.

How do you view, manage, or turn off Claude Memory?

Both memory and chat search are controlled from the same place. In the Claude apps, go to Settings and then Capabilities, where you will find toggles for memory and for searching and referencing past chats. You can turn either off and re-enable it later, and Enterprise administrators can switch memory off for an entire organization.

Claude gives you more than a single on/off switch for memory. Anthropic's help center describes a pause option, which keeps your existing memory but stops Claude from using it or creating new summaries, and a reset option that permanently deletes stored memory (including project memories) and cannot be undone. To review what is stored, click "View and edit memory" to open the manage-memory view, where you can see exactly what Claude remembers, edit the summary directly, or tell Claude what to remember.

For conversations you never want recorded, use an incognito chat. Anthropic states that incognito chats are temporary: they do not appear in your chat history, are not saved to memory, and are not used for training across all plan types. Incognito conversations are retained for 30 days by default (longer under an Enterprise custom retention policy), and once closed they cannot be reopened or converted into a regular chat. For people who specifically want fine-grained control over what an assistant retains, dedicated memory layers such as MemX take a different posture, keeping memory private by architecture with per-user isolation and encryption at rest rather than as an account-level toggle.

Manage both features under Settings then Capabilities in the Claude apps.
Pause keeps existing memory but stops Claude from using it or creating new summaries.
Reset permanently deletes stored memory, including project memories, and is irreversible.
Click "View and edit memory" to see and edit exactly what Claude remembers.
Use incognito chats for sessions that are never saved to history or memory and are not used for training.

How does Claude Memory differ from ChatGPT memory?

The clearest difference is scope. Claude's memory is project-scoped: each project gets its own isolated summary, and standalone chats are kept separate from projects. ChatGPT's memory, by contrast, is global to your account. OpenAI's help docs describe two layers, "saved memories" you ask it to remember plus a "reference chat history" capability that draws on your past conversations, and both apply across your chats rather than being walled off per project. If you want compartmentalization in Claude, projects provide it by default; in ChatGPT you generally manage one shared memory.

A second difference is how the stored memory is built and surfaced. Claude presents memory as an editable summary you can read in full and rewrite, and exposes chat search as a visible retrieval tool call so you can see when it is pulling from past conversations. ChatGPT also lets you view and delete saved memories, but its broader chat-history layer is generated by an asynchronous background process that OpenAI calls "dreaming" (consolidated into a single synthesis process in its June 2026 "Dreaming V3" update), which synthesizes memory from many conversations between chats and is less granularly visible than an editable per-project summary.

The control surfaces are similar in spirit but named differently. Both assistants let you turn memory off and offer a private mode (incognito chats in Claude, temporary chats in ChatGPT) that is not saved and is not used for training. The practical takeaway: Claude leans toward isolated, auditable, per-project memory, while ChatGPT leans toward a single account-wide personalization profile applied everywhere.

Claude memory is project-scoped and isolated; ChatGPT memory is account-global.
ChatGPT splits memory into "saved memories" and "reference chat history"; Claude uses an editable summary plus chat search.
Claude's chat search shows as a visible retrieval tool call; ChatGPT's history synthesis ("dreaming") runs in the background.
Both offer a private mode (incognito vs temporary chats) that is unsaved and not used for training.
Both let you view, edit, and delete what is stored, with slightly different settings labels.

Key takeaways

Claude Memory lets Claude retain context across conversations through an editable, project-scoped summary of your preferences and work, plus a paid "search past chats" retrieval feature.
Anthropic rolled it out in stages: Team and Enterprise on September 11, 2025, Pro and Max on October 23, 2025, and the memory summary to free users in early March 2026.
Memory is generated automatically as a synthesis of key insights, refreshed on a recurring cadence, and you can open, edit, or delete what Claude stores.
You control it under Settings then Capabilities, where you can pause memory (keep it but stop using it and creating new summaries) or reset it permanently; incognito chats are never saved or used for training.
Memory is project-scoped, so context stays isolated per project, which is the main design contrast with ChatGPT's account-wide saved memories and reference chat history.
"Search past chats" remains a paid-plan feature, while the memory summary itself is now available on all plans including free.

Frequently asked questions

Yes. As of mid-2026 Claude can remember context across separate conversations using an automatically generated, editable memory summary, available on all plans including the free tier. Paid plans (Pro, Max, Team, Enterprise) additionally get "search past chats", which lets Claude find and reference specific earlier conversations.

Open Settings and then Capabilities in the Claude apps, where you can toggle off memory and the search-past-chats feature. When managing memory you can pause it, which keeps existing memory but stops Claude from using it or creating new summaries, or reset it, which permanently and irreversibly deletes what Claude has stored, including project memories.

Anthropic describes the feature as fully optional with granular controls, including a viewable and editable summary, the ability to pause or reset memory, and incognito chats that are not saved or used for training. Memory is project-scoped, so context stays isolated per project. Anthropic does not describe Claude memory as end-to-end encrypted, so review its privacy and data-retention policies for the details that matter to you.

Anthropic launched automatic memory for Team and Enterprise plans on September 11, 2025, expanded it to Pro and Max on October 23, 2025, and made the memory summary available to free-tier users in early March 2026. The "search past chats" capability remains limited to paid plans.

Claude's memory is project-scoped, giving each project its own isolated, editable summary, while ChatGPT's memory is account-global and split into "saved memories" plus a "reference chat history" layer applied across all chats. Claude also surfaces chat search as a visible retrieval tool call, whereas ChatGPT synthesizes history in a background process it calls "dreaming". Both offer a private mode that is not saved and not used for training.

An incognito chat is a temporary conversation that does not appear in your chat history, is not saved to Claude's memory, and is not used for model training, across all plans. Anthropic retains incognito chats for 30 days by default (longer under Enterprise retention policies), and once an incognito chat is closed it cannot be reopened or converted into a regular chat.

Put the idea into practice

MemX is an AI memory app built on these ideas: store anything, skip the folders, and find it again by asking in plain English.

Try MemX Free