gemma3:27b or QwQ for beefier local models and qwen2.5:7b for models under 10GB, all of which I run though ollama. I also use AnythingLLM if I want any sort of GUI or to interact with documents. I've had... mixed results with vibe coding, even with copilot.