Which settings you recommend so I don’t burn too many tokens?

I’m using Cursor as my main IDE for full-stack projects (frontend + backend + database), and I’d love some guidance on which settings you recommend so I don’t burn too many tokens, while still getting high-quality results.

Specifically, could you please share:

  1. Recommended model & settings for full-stack work

    • Which model(s) should I use by default for:

      • Day-to-day coding (Composer / Ctrl+L)

      • Big refactors / architecture changes

      • Chat / planning / code reviews

    • Any suggested values for things like max tokens, context size, and temperature for large repositories?

  2. How to reduce token usage without losing context

    • Best practices for:

      • Organizing projects and chats so the model doesn’t need to read the whole repo every time

      • When to start a new chat vs continue an old one

      • How to use repo indexing / project context efficiently

      • Which features I can safely disable (for example: web search, long auto-fixes, etc.) if I want to save tokens.

  3. Tips, tricks, or recommended workflow changes

    • Any “tricks” or lesser-known features that full-stack developers should enable or change in settings

    • How you recommend handling:

      • Mono-repos or larger codebases

      • Environment files, generated code, or logs so they don’t pollute the context

    • Anything else I should change in my workflow to get better results with fewer tokens.

If you have a sample configuration or screenshot of ideal settings for a typical full-stack project, that would be super helpful as well.

Thank you in advance for any suggestions or best practices you can share!