Claude memory: does Claude remember your chats, and where?
Claude memory is the feature that lets Claude carry context from one chat into the next, saved as individual topics you can read and edit. Does Claude have memory? Yes: across chats on Free, Pro, Max and Ultra, on by default, and on Team and Enterprise when an owner allows it. Each Project keeps its own memory. Claude Code uses a separate system of files. Searching past chats is a paid feature, and incognito chats are never remembered.
As of November 2026. Details come from Anthropic's memory article, updated this week, and OpenAI's help center. Written by Flori, not an independent review.
Does Claude have memory on my plan and surface?
| Where you use Claude | Memory? | Detail |
|---|---|---|
| Chat on Free, Pro, Max, Ultra | Yes, on by default | Web, Claude Desktop and Claude Mobile |
| Chat on Team or Enterprise | If the owner allows it | Most organizations created on or after October 31, 2026 start on; older ones stay off until each member turns it on |
| Searching past chats | Paid plans only | Pro, Max, Ultra, Team, Enterprise; shows as a tool call |
| Claude Projects | Yes, one memory per project | Also uses your account memory unless you turn that off |
| Cowork in the cloud | Yes, shared with chat | What comes up in a task carries back to chat |
| Cowork running on your computer | No | Local sessions don't use memory |
| Claude Code | Yes, its own system | CLAUDE.md and auto memory, described below |
| Incognito chat | No | Not saved to memory or history |
Everything in this table except the Claude Code row comes from Anthropic's memory article. Older advice saying memory is paid-only is out of date, although one of Anthropic's usage tips pages still words it that way.
How does Claude memory work across chats?
Claude saves memory as individual topics while you talk, so a moved deadline can be known in your next conversation. It saves on its own, and you can also say "remember this." Your role, the people you work with, preferences and ongoing project details are the typical content.
Two things are held back. Sensitive subjects such as health or politics are skipped unless you opt in, and government ID numbers, financial account numbers, criminal history and immigration status are never saved even if you ask.
A small number of Team and Enterprise organizations are still on the legacy version, which builds a summary of your chats and refreshes it every 24 hours. If you see Memory under Settings > Capabilities, that is you.
Does Claude have memory like ChatGPT?
Both remember across chats and both let you review and delete. They differ in shape.
| Area | Claude | ChatGPT |
|---|---|---|
| How it is kept | Individual topics you can open and edit | A memory summary, saved memories and references to chat history |
| Other sources | Your chats and Cowork tasks | Depending on plan and region, also files and connected apps such as Gmail |
| Where to see it | Settings > Your Claude > Manage | Settings > Personalization > Memory |
| One chat without memory | + menu, turn off Memory, before the first message | Temporary chat, choosing Unpersonalized |
| Keep a project separate | Turn off Use account memory on the project | Project-only memory |
| Sensitive topics | Off until you opt in | OpenAI says they can appear if you share them |
The practical gap is sources. ChatGPT's memory can draw on connected apps where available, while Anthropic describes Claude's as built from your chats and tasks. Moving between them is covered in how to transfer ChatGPT memory to Claude, and the ChatGPT side of limits in ChatGPT memory full.
Does Claude memory use more tokens?
Yes in principle, by an amount Anthropic does not publish for chat. A Claude Academy tutorial explains that memory notes are placed into context before your first message, and context is what uses up a conversation. But memory is absent from Anthropic's list of factors that affect usage limits, which names message length, attachments, conversation length, tools and model choice.
Three things follow:
- Searching past chats runs as a tool call, and tool use is on that list. Asking Claude to dig through history can cost more than letting memory carry a few facts.
- Connectors matter more. Anthropic calls tools and connectors token-intensive, so switch off unused ones first if you are hitting limits.
- To compare for yourself, open a new chat, select +, turn off Memory, and run the same task both ways.
In Claude Code the cost is documented. Auto memory and CLAUDE.md load into every session, and Anthropic suggests keeping each CLAUDE.md under 200 lines because longer files use more context. Run /context all to see what loaded.
How do I turn Claude memory on, off or delete it?
Turn it on or off in Settings > Your Claude, and delete single memories under Manage next to Your memory. The steps:
- Open Settings > Your Claude. In the older layout, open Settings > Memory.
- Turn on the memory-from-chats toggle. Anthropic's pages label it both Generate memory from chats and Memory from chats.
- Optional: turn on Include sensitive topics in memory.
To turn it off, the same toggle offers two options, and the difference matters:
- Pause memory keeps what is stored but stops using or adding to it.
- Reset memory permanently deletes everything, including project memories, and can't be undone.
Team and Enterprise members can do this unless the owner has turned memory off for the organization, which also deletes all existing memories for everyone.
Does Claude Projects have memory?
Yes. Each project has its own memory, and inside a project Claude uses that plus your memory from outside projects, but not memory from other projects. New memories save to the project, and with memory on, Claude can reference them outside it too. To seal a project, turn off Use account memory under Memory on its page. That matters when one client's details shouldn't leak into another's work.
For facts that must hold exactly, memory is the wrong container, since Claude decides what to keep. Put them in project knowledge or instructions. Our guide to giving Claude project memory covers that setup.
Does Claude Code have memory across sessions?
Yes. Each Claude Code session starts with a fresh context, and two mechanisms carry knowledge over:
| Area | CLAUDE.md | Auto memory |
|---|---|---|
| Written by | You | Claude |
| Holds | Rules and instructions | Your corrections, preferences and project notes |
| Loaded | Every session | Every session, first 200 lines or 25KB |
| Lives in | Your repository or ~/.claude/ | A memory folder under ~/.claude/projects/, one per repository |
Auto memory is on by default in local sessions, and /memory opens the files or toggles it off. For the difference between instruction files, see CLAUDE.md and AGENTS.md.
Where does Flori fit?
Assistant memory suits what Claude should know about you. A client project's tasks, decisions and status are a shared record, and more than one person or assistant may need them. Flori holds that record and lets Claude read it over MCP, with access set per workspace or project. See Flori for Claude.
FAQ
Does Claude remember between chats on the free plan? Yes. Anthropic says memory is on by default for Free, Pro, Max and Ultra plans. The separate ability to prompt Claude to search your past conversations is limited to paid plans, so on Free you rely on what memory has saved.
Where does Claude memory live and can I see it? Go to Settings > Your Claude and select Manage next to Your memory. Everything Claude remembers is listed under Topics, where you can read, edit or delete each one. Memory data is also included in data exports, so you can keep a copy.
If I delete a chat, does Claude forget it? Not automatically. Anthropic says memory entries generated from a chat aren't removed when that chat is deleted. Delete the individual memory under Manage, or tell Claude in a chat what to forget. Legacy memory users work differently, since their summary updates when chats are deleted.
Can I get Claude's memory back after a reset? No. Reset memory permanently deletes all memories, including project memories, and Anthropic says this can't be undone. If you only want a break, choose Pause memory, which keeps what is stored. Export your memory first if you may want it later.
Why does Claude seem to have no memory in some chats? Common causes are an incognito chat, a chat started with Memory off in the + menu, Cowork running locally, or memory turned off for your organization. Check Settings > Your Claude first. A chat that used linked accounts in Personal Finance is also left out of memory.
Start your workspace