Cursor

Cursor removed Memories. Here is how to get yours back

You open Cursor Settings to check a memory, and the list is gone. You create a new one, click "View Memories", and still see nothing. Months of corrections, the ones you added every time the agent got something wrong, seem lost.

They are not lost. Cursor removed the feature on purpose, and the memories can be exported. This page shows how, where each one belongs now, and what changes for you.

The short answer

Cursor removed Memories starting with version 2.1.17. A member of the Cursor team confirmed it in the Cursor forum on 24 November 2025. The old Memories page in the Cursor docs now redirects to Rules, which is the replacement.

To keep your memories, run Export memories from the command palette and move the content into rules. If the command is missing, version 2.0.77 still shows them.

Get your memories back

  1. Open the command palette: Cmd+Shift+P on macOS, Ctrl+Shift+P on Windows and Linux.
  2. Type Export memories. Cursor saves your memories to an .mdc file.
  3. Sort the content into rules. The table below shows which kind goes where.

If Export memories does not appear, you are not alone: a user on 2.1.34 reported the same in the forum. The Cursor team pointed people to version 2.0.77, where Memories still show up in Settings. One user got their list back that way, copied it out and updated again.

Old memories still in the context

Removing the screen did not remove the memories. In January 2026 a user on version 2.2 found the agent still reporting access to 32 of them. The Cursor team answered in the thread Can't clear memories: the feature "still works, just without a UI".

Asking the agent to "Delete all memories" did not help in that thread. The agent replied that its memory tool was not available. What worked: install 2.0.77, delete the memories in Settings, then update to the current version.

Where each memory belongs now

The memory was aboutPut it inWhy
How you like to work: “answer briefly”, “explain before you edit”User Rule (Customize → Rules)Applies across all your projects.
A project convention: “we use pnpm”, “tests sit next to the code”Project Rule with alwaysApply: true, or AGENTS.mdLives in the repository, so everyone who clones it gets the same rule.
Something about one part of the code: “migrations need a generated client”Project Rule with globsAttached only when a matching file is in context.
A standard for the whole teamTeam Rule (Team and Enterprise plans)Managed in the dashboard and applied first: Team, then Project, then User.

A project rule is an .mdc file in .cursor/rules. Its frontmatter decides when it applies. With globs and alwaysApply: false, Cursor attaches it when a matching file is in context:

.cursor/rules/migrations.mdc
---
globs: db/migrations/**
alwaysApply: false
---

- Run `pnpm db:generate` after changing schema.prisma, before writing a
  migration. Without it the migration compiles against the old client.

Two traps from the Cursor docs:

  • A plain .md file in .cursor/rules is ignored. Use .mdc, or put plain markdown into AGENTS.md.
  • Keep each rule under 500 lines. Split a long export into several focused rules.

If you also use Claude Code, AGENTS.md is the file both tools read. Claude Code reads it when the project has no CLAUDE.md. Otherwise put @AGENTS.md into your CLAUDE.md to import it.

What rules don't do

Rules hold what you write. Memories were often written by the agent itself, right after you corrected it. That step is now yours: every correction you want to keep, you write into a rule.

Rules also suit a stable convention better than a growing pile of fixes. A bug fix from last Tuesday, the cause behind it and the two approaches that failed make a poor rule: too specific to apply always, too many to keep under 500 lines. For those, five ways to make a fix stick work better, starting with a test that fails when the bug returns.

With cachly

cachly keeps those fixes as lessons and brings back the ones that fit what you are asking:

  • Its setup writes .cursor/rules/cachly.mdc and Cursor hooks. The rule tells the agent to save a lesson after a fix, through cachly's MCP tool.
  • Before each prompt, the hook picks up to three matching lessons. The first tool call of the turn hands them to the agent. How the Cursor hooks work.
  • Lessons live in your instance, not in one editor. A fix saved in Cursor shows up in Claude Code and Copilot, and for your whole team.
Claude Code
/plugin marketplace add cachly-dev/cachly-mcp
/plugin install cachly-brain@cachly

Then turn on auto-update once: /plugin → Marketplaces → cachly → Enable auto-update. Claude Code leaves it off for marketplaces outside Anthropic's own, and the hooks update only with the plugin.

Lessons are stored on servers in the EU. What leaves your machine, and where it goes.

Sources

More: Auf Deutsch lesen: Cursor merkt sich nichts · Agent keeps making the same mistake · cachly in your editor · All docs