Memory Bank feature for your Cursor

Showcase Built for Cursor
1

Hello, Cursor Forum members!

I’m excited to introduce the Cursor Memory Bank, a rules designed to enhance AI assistants by providing persistent memory between sessions. This system addresses the common issue of AI forgetting context after each interaction by organizing documentation into structured files that serve as a continuous memory source.

Getting Started:

To quickly set up the Cursor Memory Bank in your project, you can use the following command:

npx cursor-bank init

This command performs the necessary setup automatically, creating the required directory structure for you.

Alternative Installation Methods:

If you prefer, you can install the memory bank globally or download the files directly. Detailed instructions are available in the GitHub repository.

Usage:

Once set up, you can utilize the memory bank with commands like:

• PLAN – Enter or return to planning mode.

• ACT – Approve the plan and switch to implementation mode.

• update memory bank – Trigger an update to the documentation.

These commands facilitate a structured workflow, enhancing your AI assistant’s efficiency.

Key Features:

• Persistent Memory: Ensures your AI assistant retains context across sessions, reducing the need for repetitive explanations.

• Structured Documentation: Organizes information into specialized files, making it easily accessible and manageable.

• Seamless Integration: Easily incorporate the memory bank into your existing projects with minimal setup.

I welcome any feedback, suggestions, or questions you may have. I hope the Cursor Memory Bank proves to be a valuable tool in your development process with Cursor as for us.

10 Likes
Cursor Feature Request – cq Integration
4

I like this, I find it very useful.

4 Likes
5

I can see that, when appropriate, Cursor reports “Start a new chat for better results”.

So logically, when this recommendation appears, I should first ‘update memory bank’, and then start a new chat, yes?

Once I start the new chat, the first thing I should say is “(Please) get fully up to speed with this project by reading memory bank”, yes? That’s a bit of a mouthful! What’s the shortest command to achive the same pls?

Finally… please remind me what are the key resource constraints that determine when Cursor recommends starting a new chat? Is there anything I can do on my local development PC to improve this so that existing chats can go on for longer before a new chat is recommended by Cursor, i.e. because everything is slowing down?

Thanks!

6

If you’re into the Memory Bank pattern, check out Lore — same core idea (structured markdown files for persistent context) but extended into a full harness:

  • Gotchas get captured as reusable “skills” that auto-load in future sessions — not just a static memory file, but knowledge that grows as you work
  • Coding conventions are reinforced via hooks before every file write, not just loaded at session start
  • Same knowledge base works across Cursor, Claude Code, and OpenCode — no platform lock-in
  • Delegates complex tasks to focused worker agents with curated context

Cursor support is experimental right now (Claude Code is the most mature), so feedback from Cursor users is especially welcome.

Demo from a fresh install: https://youtu.be/u2rkR1XeHZk
GitHub: GitHub - lorehq/lore: Your coding agent forgets everything between sessions. Lore fixes that.