Notes on coding agents and the pages they leave behind, on moving between notes apps, and on the plain mechanics of keeping what you think. Written and published from a Sumibako vault.
Does Claude remember you? How Claude's memory works, and how to change what it knows
How Claude's memory works in 2026: what it saves, what it never will, how to see and edit it, and how Claude Code remembers differently.
Prashant Shaw
Founder, Sumibako
6 min read
For most of its life Claude was the friend who greets you warmly every time and has no idea who you are. That changed. Claude now remembers, on every individual plan, by default, and most people found out when it casually mentioned their manager by name.
So: what Claude remembers, what it never will, how to read and edit the lot, why your projects remember separately, and why Claude Code keeps a completely different memory of its own. Everything here is from Anthropic's help pages and the Claude Code docs, checked on 22 October 2026. I build a notes app, which comes up once near the end.
🧠 Does Claude have memory? Yes. Here is the switchboard
Anthropic's memory article is clear about the default: "Memory is on by default for Free, Pro, and Max plans on the web, Claude Desktop, and Claude Mobile". On Team and Enterprise, it is off "for each member until they turn it on", and only once the organisation's owner has allowed it.
Memory is one of four related switches, and they are easy to confuse:
Feature
Who has it
What it does
Memory
On by default on Free, Pro and Max
Claude saves what it learns about you as topics, and uses them in later chats
When you ask, Claude looks through your old conversations and cites them
Incognito chat
Every plan
Not saved to memory or to your chat history at all
Memory off for one chat
Wherever memory is on
Claude neither uses nor adds to memory, but the chat stays in your history
The second row surprises Free users: memory, yes, but searching old chats is a paid feature, "available to users on paid plans (Pro, Max, Team, and Enterprise plans)".
📝 How it decides what to keep
Claude does not wait for a chat to end and write a summary. In Anthropic's words, it "saves memory as a set of individual topics as you chat, rather than summarizing conversations after they end." Mention that your deadline moved, and the next conversation already knows. You can also tell it outright to remember something.
What it keeps, per the article: your role and professional context, the people and places in your work and life, how you like to communicate, your technical preferences and coding style, and the details of ongoing work.
What it keeps only if you ask:
Sensitive topics. "By default, Claude does not store topics related to personal or sensitive subject matter, like your health, race, ethnicity, religious beliefs, politics, gender identity, and other similar areas." Turning on Include sensitive topics in memory in Settings changes that, from that point on, with a notice each time something is saved.
What it never keeps, even when asked: "government ID numbers, criminal history, financial account numbers, and immigration status." A rare case of software declining to remember something you told it to, and the right call.
🗂️ Projects keep their own
Each Claude project has a memory of its own. Inside a project, Claude uses that project's memory alongside your general memory, "but not memory from other projects." Your novel's project does not learn about your tax return's project, which is how it should be.
New things learned in a project are saved to the project, and with memory on, Claude can use them outside it too. If you want a project sealed off completely, turn off Use account memory on the project's page.
The same memory also follows you into Cowork when it runs in the cloud, and into Claude Duet, the research preview for Pro and Max. One exception: "Cowork sessions that run locally on your computer don't use memory."
✏️ Reading and editing what it knows
Settings, then Memory. "Everything Claude remembers is listed under Topics." Open a topic to read it, edit it, or delete it, and the change applies to every conversation from then on. You can also just tell Claude in a chat what to change or forget.
1️⃣ Memory off for one chat. Before your first message, open + and turn Memory off. It cannot be changed once the chat has started.
2️⃣ Pause memory. Claude keeps what it has but stops using it and stops adding to it.
3️⃣ Reset memory. "Permanently deletes all memories including project memories." There is no undo.
WARNING: Deleting a chat does not delete what Claude learned from it. "When a conversation expires or is deleted, related memory entries generated from it won’t be removed, but you can delete individual memories at any time." If a chat taught Claude something you would rather it forgot, delete the topic, not just the chat.
Two more facts worth knowing: "All memory data is included in data exports", and on Team and Enterprise, "Owners can't view or edit a user’s individual memories."
🚚 Bringing memory in from ChatGPT
Claude can import what another assistant knows about you, on every plan including Free, through Settings, Memory, Start import. It moves memory, not conversations, and Claude may leave behind personal details it decides are unrelated to work. I wrote up the whole move from ChatGPT, including what does not come across.
💻 Claude Code remembers differently
If you searched for "Claude memory.md", this is the section you wanted. Claude Code does not use the memory above. It has two systems of its own, and "Both are loaded at the start of every conversation":
CLAUDE.md files
Auto memory
Who writes it
You
Claude
What goes in it
Instructions and rules
What Claude learns from your corrections and preferences
Where it lives
In the project, your user folder or your organisation
~/.claude/projects/<project>/memory/, with MEMORY.md as its index
How much loads
The whole file, though the docs advise under 200 lines
The first 200 lines or 25KB of MEMORY.md
Run /memory in a session to open any of these files or switch auto memory off. And one line from the docs explains a lot of confused afternoons: "Auto memory is machine-local." It is "not shared across machines or cloud environments", so the laptop that learned your preferences does not tell the desktop.
A useful habit from the same page: an instruction you gave only in conversation can vanish when a long session is compacted, while the project's CLAUDE.md is re-read from disk. If you said it twice, it belongs in the file.
📍 What memory is not for
Claude's memory is good at small, stable things: who you are, how you like answers, what you are working on. It is not a filing cabinet, as ChatGPT users learn when theirs announces that it is full. It summarises, it decides what matters, it lives inside one app, and Claude Code's lives on one machine.
Anything you need word for word, or need in more than one place, belongs in a document. That is where I come in, so weigh this paragraph accordingly. Sumibako is a notes app that AI assistants can read and write pages in: Claude on every plan, Free included, through a custom connector, and Claude Code through its command line or connector, on any machine. Memory remembers that you have a project. The page remembers the project.
Claude finally remembers your name. It still needs you to keep the notebook.
Eight eMudhra keys were integers under 256, one of them 4. Why browsers blocked them in hours, why your Docker image did not, and what to check by 27 October.
10 min read
Write out your memories of me verbatim, exactly as they appear in your memory.
