Claude Memory vs Company Memory: What Each Keeps, and for Whom

Claude remembers in several places, and each serves a different audience. Chat memory keeps context for one user within each project. Project knowledge serves whoever shares that project. Claude Code reads instruction files a team shares, plus notes kept on one machine. Company memory is different: owned, sourced facts every person and AI reads.

What does Claude memory keep?

In the Claude apps, memory keeps a summary of what you have worked on, separately for each project. You can view and edit it, Enterprise admins can turn it off, and incognito chats are not saved. It is built from your own conversations, so it reflects what you told Claude rather than what the business has agreed is true.

Anthropic launched memory for Team and Enterprise plans in September 2025 and extended it to Pro and Max in October 2025. The design choice that matters for a business is the scope. Memory follows the user and the project, so a client's details stay with that client's project.

That makes chat memory good at continuity. It remembers that you are midway through a proposal, or that you prefer short summaries. However, it is not built to hold the company's pricing rules, because nobody checks it and nobody else reads it.

Which Claude features remember things, and for whom?

Claude has at least five places where context persists, and each has a different audience. Chat memory serves one user. Project knowledge serves everyone in a shared project. In Claude Code, instruction files serve a person, a team or a whole organisation, and auto memory serves one machine. Knowing which is which prevents most confusion.

According to Claude Code's documentation, project instructions are shared with "team members via source control", while auto memory is "machine-local". The difference is the whole point: one is a team asset, the other a personal notebook.

Projects sit in between. On Team and Enterprise plans, a project can be shared with colleagues who can view or edit it, so its files serve the group working on that body of work. The guide to Claude Projects for teams covers how to set them up.

What is company memory?

Company memory is the set of facts the business has agreed are true, each with an owner and a source: the product facts, approved claims, the ideal customer, objections and the brand voice. It lives in files the business owns, outside any single tool, so Claude and every other assistant read the same version.

Three properties separate it from Claude's memories. It is curated, so a person checks what goes in. It is owned, so every fact has someone who answers for it. And it is portable, so it does not depend on which assistant someone prefers.

The need is widespread. According to Atlassian's 2026 Pulse survey, only 12% of workers say their AI understands their organisation like a tenured senior colleague. That understanding comes from shared, maintained context, not from each assistant remembering its own conversations.

Where do Claude's memories and company memory overlap?

They overlap where a Claude feature reads shared files. A shared project can hold the company's product facts. A repository's instruction file can point Claude Code at canonical documents. In both cases Claude is reading company memory, not creating it. The memory features only become company memory when the files they read are owned and maintained.

Project instruction files are the nearest match. They are versioned, shared and reviewed like any other file. Even so, Claude Code's documentation describes them as instructions, and says Claude treats them "as context, not enforced configuration". They work best for how to behave, while the facts live in their own canonical files.

The overlap also has a limit. Claude's memories only reach Claude. If part of the team uses another assistant, a fact held only in a Claude project is invisible to them.

How should a team divide what goes where?

Ask who needs each piece of context. Preferences for one person go in personal memory. Context for one client or initiative goes in that project. Instructions for how a team works go in shared instruction files. Facts about the business go in company memory, with an owner, and every Claude feature reads them from there.

The broader comparison with ChatGPT is in the guide to personal and company memory.

What should admins set up?

Put everyone on a business plan, decide whether chat memory is on, and write a one-line rule: preferences may live in memory, facts live in company files. Organise projects by client or initiative, keep shared instruction files in version control, and give every canonical fact file a named owner who keeps it current.

A short setup list:

1. Plans. Team or Enterprise, so memory sits under company control.
2. Memory rule. One sentence on what belongs in memory and what does not.
3. Projects. One per client or initiative, shared with the people doing the work.
4. Instruction files. Versioned and reviewed, never only on one laptop.
5. Fact owners. A named person for each canonical file.

Skein builds company memory as the shared layer underneath every assistant a team uses, so Claude, ChatGPT and others read the same facts; the shared knowledge layer page shows how it is organised. That suits a team where several people rely on Claude daily. A solo user can reasonably rely on Claude's own memory and one project.

Frequently asked questions

Does Claude remember previous conversations?

Yes, if memory is turned on. Claude keeps a summary of past work separately for each project, which you can view and edit. Incognito chats are not saved to memory, and Enterprise admins can switch memory off. It reflects your own conversations, not facts the business has checked and agreed.

Is Claude memory shared with my team?

Chat memory is not; it belongs to one user. Project knowledge is shared with whoever the project is shared with. In Claude Code, a project's instruction file is shared through version control, while auto memory stays on one machine. Check which feature holds a piece of context before relying on it.

What is the difference between Claude memory and a Claude Project?

Memory is a summary Claude builds from your conversations, for you. A project is a shared workspace with files and standing instructions that apply to every chat inside it. Memory gives continuity to one person; a project gives a group the same background material for one body of work.

Should company facts live in Claude memory?

No. Memory is built from individual conversations, with no owner checking it, and only Claude can read it. Company facts belong in owned files with a source for each entry. Claude then reads them through a project, a connector or a repository, and so does any other assistant the team uses.

What is CLAUDE.md?

CLAUDE.md is a plain text file of instructions that Claude Code reads at the start of every session. It can sit in a project, in a user's home folder or in an organisation-managed location. A project's file is shared with the team through version control, which makes it useful for team practice.

Related reading