Feature request for product/service
AI Models
Describe the request
Cursor should quantize and optimize Composer 2.5 so it can run locally on user hardware such as an RTX 4090. Right now Composer is very powerful but requires cloud inference. Providing a properly quantized and fine-tuned local version would let it handle common tasks independently, including bug fixes, documentation, refactoring, and code organization.The local version should also work alongside the cloud version — handling simpler or preparatory work locally while handing off to the full cloud model for more complex steps when needed.This would reduce API costs, improve privacy, enable better offline use, and lower the load on Cursor’s infrastructure by moving routine agent work to user GPUs.Many users want to run capable agents locally. Making Composer 2.5 itself practical on consumer hardware through quantization and optimization would be a big step forward.
Operating System (if it applies)
Windows 10/11