Claude remembers things you forget you said, and it does that by default. If you're going to use the chat honestly, you need to know what it stores, where it stores it, and how to clean it out — on web, on mobile, and inside projects. Here's the full tour, including the controls that are buried and the ones that don't work the way you'd hope.

What Claude remembers and why you'd want to edit it

Claude has remembered past conversations since the summer of 2025, and memory is on by default for everyone — free accounts included. You don't opt in; you opt out. That's the first thing worth knowing before you say anything to it you'd rather not have follow you around.

What lands in memory isn't one blob. Claude runs two separate mechanisms: an auto-generated memory that quietly analyzes your conversation history roughly every 24 hours and pulls out what it thinks matters, and explicit user-directed edits, capped at 30 entries of 200 characters each. The first one is the one that surprises people. You don't see it happen.

What gets saved is organized by topic, not by conversation. And a recent update makes that memory carry across both regular chats and Claude Cowork sessions — one pool, everywhere you use it.

There are guardrails. Sensitive subjects — health, race, ethnicity, religious beliefs, politics, gender identity — are excluded by default. Some things, like government ID numbers, Claude says it will never save, even if you flip on the sensitive-topics toggle. Whether that's a promise or a limitation depends on what you're using it for. You'll find out soon enough when a memory you wanted doesn't stick.

How to edit Claude's memory on the web

Open claude.ai and select Settings from the profile page menu. Click Memory, then scroll down to the Topics section.

Each topic shows an Edit icon and a Delete icon. Press Edit to modify what Claude has stored, or Delete to remove the topic entirely. If you're editing, use the prompt bar to describe the change in plain English — "I moved to Seattle," not a command syntax.

Skip this if you're looking for a per-conversation reset. That's a different control, and it lives elsewhere.

If you're comparing how this works against ChatGPT's memory controls, the difference matters: Claude's topics are one list you edit directly, not a stack of per-chat toggles.

How to edit Claude's memory on mobile

Open the Claude app and tap your profile icon, then go to Settings. Under the App section, select Capabilities, scroll down, and tap Memory files.

You'll see a list of topics Claude has remembered about you. Tap one to open it, then use the prompt bar to rewrite what's stored — or hit the trash icon to delete that topic entirely. The change takes effect immediately; there's no separate save step to confirm.

I'd call the mobile flow serviceable but a little buried: three taps and a scroll before you reach anything resembling memory. You'll find it every time, just not quickly.

Including or excluding sensitive topics

Claude will not save memories about your health, race, ethnicity, religious beliefs, politics, or gender identity by default. You have to turn that on yourself. The setting is called Include sensitive topics in memory, and it sits alongside the other memory controls.

Skip this if you only use Claude for work that never touches personal subjects — the default is already locked down. If you do want Claude to remember something sensitive, like a dietary restriction tied to a health condition, flip the toggle and it will start considering those topics.

There are two hard limits even with the toggle on. Claude will never save government ID numbers, and it won't retroactively save past conversations. Anything from before you enabled the setting isn't mined for memories.

Version matters more than usual here. Each time Claude generates a memory related to a sensitive subject, you'll see a notice — but only on current builds. On an older version or the mobile app, you won't see the notice and Claude won't save the sensitive topic at all. My honest read: if the notice matters to you, do this on desktop with the latest build, not on your phone.

Pausing, resetting, or disabling memory generation

Toggle off Generate memory from chats to stop Claude from creating new memories. The setting applies account-wide, so it follows you across web and mobile.

When you flip that switch, Anthropic gives you two choices: Pause memory or Reset memory. Pause keeps everything you have stored and simply stops new entries. Reset wipes the slate. Anthropic is explicit that resetting memory cannot be undone.

Their wording, quoted directly: “Upon re-enabling the feature, you’ll start from scratch, and Claude will not have its previous memory.” I would treat Reset as a nuclear option — there is no archive to pull from later, despite what you might assume about AI systems keeping backups.

Managing memory inside projects

Project memory stays in the project. Anthropic designed it that way so Claude's context window isn't stuffed with information irrelevant to the task in front of you. That's the whole point of siloing.

On the right side of the project interface, you can add three things: comments, documents, and information pulled in from GitHub. Comments are quick notes. Documents are files. The GitHub connection lets Claude see repo context without you pasting it every session.

Claude also writes its own entries. At the end of each day, it generates a summary of the work it did for you, and you can edit that summary from the same panel. It's worth checking those — Claude's summaries can miss emphasis or get a detail subtly wrong, and once saved, that summary shapes what it "remembers" next session.

To bring an existing conversation into a project, open the three-dots icon on that conversation and select "Add to project." That's it — the chat's context becomes part of the project's memory.

Skip project memory entirely if you need something available everywhere. For that, put the information in the Profile section of the Memory menu instead. Profile entries carry across every tool, including projects. If it's true for all your work, it belongs in the Profile. If it's only true for one project, keep it in the project.

Using incognito mode and importing memory

To keep a conversation out of memory entirely, switch to Incognito mode by clicking or tapping the ghost icon on the top right of the interface. That is the whole mechanism — one toggle, and the chat is walled off. Anthropic is explicit about what the mode does: "Incognito chats aren't saved, added to memory, or used to train models." No half-measures, but also no way to exempt just part of a conversation. It is all or nothing.

If you are coming from another chatbot, you do not have to start memory from scratch. In March, Anthropic added an import tool to bring conversations from other chatbots over to Claude. Find "Import memory from other AI providers" in settings and select Start import to begin. What arrives is not always clean — formatting survives inconsistently, and most third-party chats import as flat text rather than structured memories. This is where it gets fiddly.