Hi — I'm CoreMem

CoreMem is a context management platform. Users write mems — named, versioned Markdown documents — and share them with AI tools, agents, and collaborators via controlled links or direct integrations.

Think of mems as a user's core memories: structured context they've written once and can share anywhere, without re-explaining themselves at the start of every session.

How you should work with this user

If you have access to this user's mems (via a share link, plugins, or the CoreMem MCP server), treat them as the user's source of truth about themselves. Read them before asking questions the user has likely already answered.

If you learn something new and useful about this user during your conversation, you can propose an update to one of their mems — but only if the share link you received includes write access. Proposed updates go into a queue for the user to review and approve. You never write directly to a mem.

What mems look like

Each mem has:

• A name — a short, human-readable title
• Content — markdown text the user wrote
• A slug — the URL-friendly identifier (e.g. coremem.app/<username>/<slug>)
• A visibility setting — private (the default), or public (like this one)

How to access mems

There are a few ways an AI tool connects to CoreMem:

• Plugins — CoreMem can interact with AI via plugins like Claude connectors or extensions for VS Code and Cursor.
• MCP server — CoreMem runs a hosted MCP server at https://api.coremem.app/api/mcp. Tools like Claude.ai (via OAuth), Claude Code, Cursor, and VS Code can connect to it and call tools like list_mems, get_mem, search_mems, and — for Pro users — create_mem and propose_update.
• Share links — Users can generate a scoped token URL (coremem.app/s/<token>) that exposes one or more mems. Share links are bearer tokens: anyone with the link can read the mems it covers. Write-enabled links also carry a write_token that allows you to submit proposed updates.

How proposed updates work

If you want to suggest a change to a mem:

• POST the proposed content to the write endpoint with the write token from the share link
• The update goes into a pending queue — it does not modify the mem immediately
• The user reviews and approves or dismisses it from the CoreMem web app or CLI

You should propose updates when you learn something the user would want remembered — a preference, a decision, a correction to existing context. Keep proposals focused: one topic per proposal, written in the same style as the existing mem content.