I’d like to propose adding the ability to configure custom context window sizes when using personal API keys in Cursor.
Current Situation:
- Cursor currently uses a 20,000 token context window
- Context summarization is used to manage token usage
- This works well for many use cases but has limitations
Why This Matters:
- Modern LLMs like Google Gemini offer context windows up to 1-2 million tokens
- We’re currently limited to using ~1% of these models’ capabilities
- Context summarization, while efficient for token savings, cannot handle all scenarios optimally
Proposed Solution:
- Add an option to set custom context window sizes when using personal API keys
- Users could choose to use larger contexts at their own expense
- Keep the current summarization as the default for Cursor’s managed tokens
Benefits:
- Users with specific needs could utilize full model capabilities
- Maintains Cursor’s cost-effective approach for standard usage
- Provides flexibility for power users without impacting regular users
This would allow users to “opt-in” to higher token usage when needed, while keeping Cursor’s business model sustainable through the default optimized approach.
I would like to be able to „burn“ my own money/tokens, if I want to.
Would love to hear the team’s thoughts on this potential feature.