How I Set Up Local LLMs in Cursor IDE for Offline Coding Success
June 19, 2025How I Discovered DeepSeek v3 Could Cut My AI Costs in Half with Cursor – And Why You Should Care
June 19, 2025
As an AI enthusiast using Cursor IDE daily, I kept hitting the same frustrating roadblock. Repeating custom prompts for git commits or input refinements became a real productivity killer. I desperately needed a way to save and reuse them without endless copy-pasting.
The Core Problem
Cursor IDE didn’t offer built-in custom prompt support. My workflow felt awkward and inefficient.
Take git commits: I’d retype instructions every single time. This wasted energy and created inconsistent results.
The ‘rules for AI’ feature didn’t solve it either. Being stuck with one prompt made context switching painfully slow.
My Breakthrough Solution
I discovered a clever trick using Cursor’s existing features. The magic lies in the @ symbol for referencing prompt files. Here’s exactly how I made it work:
- Set up prompt storage: Make a directory like
.cursor/prompts/in your project root. This keeps prompts organized and version-controlled. - Define your prompt files: I created
git-commit.mdwith dynamic slots. Here’s my template:
// .cursor/prompts/git-commit.md
Generate a concise, clear git commit message based on these changes:
Changes:
${user_input}
- Reference prompts in chat: Type
@git-commit [your changes]– it automatically inserts the template with your input. - Level up with context: I link external files like this:
Refine considering our stack: [@docs/tech-stack.md]. It pulls in extra details when needed.
Key Insights for Success
This method revolutionized my AI workflow. No more retyping prompts, more consistent results, and easy project-specific adjustments.
Storing prompts in .cursor/prompts/ makes them team-friendly too – anyone can use or improve them.
Since implementing this, I’ve reduced prompt setup time by 80%. My coding feels lighter, letting me focus on actual problem-solving rather than administrative busywork.
