---
title: "Claude Memory: How It Works and What It Stores"
description: "How Claude memory works in September 2026: what gets stored, project scoping, plan availability, user controls, and the limits Anthropic documents."
canonical: https://past.dev/blog/claude-memory
date: 2026-09-02
category: Research
authors: The past.dev team
---
# Claude Memory: How It Works and What It Stores

Claude memory, as of September 2026, is a set of individual topic entries that Claude writes while you chat and reads back in later conversations ([Anthropic help center](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context)). It is on by default on Free, Pro, and Max plans across web, desktop, and mobile, and off by default for Team and Enterprise organizations until an owner enables it ([Anthropic](https://claude.com/blog/claudes-memory-works-everywhere-and-you-decide-whats-in-it), [help center](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context)). Every entry is listed under Topics in Settings > Memory, where you can read, edit, or delete it ([Anthropic](https://claude.com/blog/claudes-memory-works-everywhere-and-you-decide-whats-in-it)). Each project keeps a separate memory of its own, and incognito chats keep a conversation out of both history and memory ([help center](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context)).

## Three layers that do different jobs

Claude has three persistence mechanisms, and they answer different questions.

1. **Model context.** Within one conversation, Claude sees that conversation's transcript and attachments, bounded by the model's [context window](/glossary/context-window). A new chat starts with none of it.
2. **The memory feature.** [Cross-session memory](/glossary/cross-session-memory): Claude saves "individual topics as you chat, rather than summarizing conversations after they end" ([help center](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context)). These entries are read back in future conversations.
3. **Projects.** A [project](https://support.claude.com/en/articles/9517075-what-are-projects) is a workspace holding uploaded knowledge, project instructions, and its own chats, and per the launch announcement Claude "creates a separate memory for each project" ([Anthropic](https://claude.com/blog/memory)).

Chat search sits alongside these three. On Pro, Max, Team, and Enterprise plans, Claude can search your past chats and answer with citations linking back to the original conversations ([help center](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context)). The help center lists chat search for those paid plans; the memory feature itself runs on Free as well. The general map of these layers, for any assistant, is in [what is agent memory](/what-is-agent-memory).

## What the Claude memory feature stores

The current design shipped on August 25, 2026: "Everything Claude remembers is in a list of files under Topics in Memory settings, where you can read, edit, or delete each one" ([Anthropic](https://claude.com/blog/claudes-memory-works-everywhere-and-you-decide-whats-in-it)). Entries are added in real time during the chat rather than distilled after it ends ([help center](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context)). The feature predates that design: memory launched on September 11, 2025 for Team and Enterprise plans with a single editable "memory summary", and rolled out to Pro and Max on October 23, 2025 ([launch announcement](https://claude.com/blog/memory)).

What lands in memory:

- Work context Claude judges useful later, plus anything you ask it to remember. You can also tell Claude in plain language to change or forget a specific memory ([help center](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context)).
- Excluded by default: health, race, ethnicity, religious beliefs, politics, and gender identity. The "Include sensitive topics in memory" toggle opts those categories in ([help center](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context)).
- Never, whatever the settings: government ID numbers, financial account numbers, criminal history, and immigration status ([help center](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context)).

One memory serves claude.ai on the web, the desktop app, mobile, and Claude Cowork ([Anthropic](https://claude.com/blog/claudes-memory-works-everywhere-and-you-decide-whats-in-it)).

## Controls: view, edit, pause, reset, incognito

- **View and edit.** Settings > Memory > Topics lists every entry; open one to read it, edit it, or delete it ([help center](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context)).
- **Pause.** Pausing memory keeps existing entries while Claude stops reading and writing them ([help center](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context)).
- **Reset.** Resetting deletes all memories permanently; the help center marks it irreversible ([help center](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context)).
- **Incognito chats.** Available on every plan, these conversations "don't appear in your conversation history or save to memory" and are excluded from chat search ([Anthropic](https://claude.com/blog/memory), [help center](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context)).
- **Export and import.** You can export memory by copying it from settings or by asking Claude to write its memories out verbatim, and you can import another assistant's export by pasting it into Settings > Memory > Start import ([help center](https://support.claude.com/en/articles/12123587-import-and-export-your-memory-from-claude)). Anthropic labels imports "experimental and still in active development" and says Claude may not always incorporate them ([same article](https://support.claude.com/en/articles/12123587-import-and-export-your-memory-from-claude)).

Deletion controls of this kind carry regulatory weight for anyone who ships an assistant; the design questions are collected in [GDPR memory deletion](/guides/gdpr-memory-deletion).

## Claude projects memory: scope boundaries

- Each project has "its own separate memory space and dedicated project summary", separate from other projects and from non-project chats ([help center](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context)).
- Chat search inside a project is limited to that project ([help center](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context)). A project does not search your other chats.
- Project knowledge is a different mechanism from memory: documents, text, and code you upload deliberately, plus written project instructions ([projects article](https://support.claude.com/en/articles/9517075-what-are-projects)). It changes only when you change it.
- Projects exist on every plan, and free accounts are capped at five projects ([projects article](https://support.claude.com/en/articles/9517075-what-are-projects)). On paid plans, a project whose knowledge approaches context limits switches to retrieval (RAG) so it can hold more content than fits in context at once ([projects article](https://support.claude.com/en/articles/9517075-what-are-projects)).

The boundaries run in both directions: general memory serves non-project chats, each project's memory serves that project, and the official documentation describes no path for entries to move between them.

## Limits the documentation states

- Chat search requires a paid plan: Pro, Max, Team, or Enterprise ([help center](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context)).
- On Team and Enterprise, memory exists only after an owner enables it in organization settings, and admins can disable it for the organization at any time ([help center](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context), [Anthropic](https://claude.com/blog/memory)).
- Memory import is experimental and can fail to incorporate entries ([import article](https://support.claude.com/en/articles/12123587-import-and-export-your-memory-from-claude)).
- Expiry: none documented. The help center describes manual editing, deletion, and reset; no automatic aging or expiry of entries appears in it. Whether an entry is still true is yours to notice.

## For builders: a memory layer for your own agent

Claude's memory is a consumer feature inside Claude; it does nothing for the assistant or agent you are building yourself. For that layer, past.dev is a memory API: POST /api/v1/ingest accepts raw text with its original timestamp, entity resolution links references to the same person across sources, and stored facts carry event time, validity windows, and supersession links. POST /api/v1/recall returns ranked, dated evidence with sources and one of four status values (Supported, Conflicted, NoKnownSupport, UnknownBecauseDegraded), deployed managed or self-hosted in your own Postgres. Start with the [Memory API quickstart](/docs/memory-api/quickstart) and the [benchmarks](/benchmarks).