What is your current go-to stack for running a fully local AI agent?

Curious to know what quantization level (GGUF/EXL2) you find balances speed and smarts for daily use.