Running local models is good now
Summary
The author argues that local models have become practical on consumer hardware, sharing hands-on experiences with various models (Mistral 7B, Gemma 3/4, GPT-OSS, Qwen variants) and multiple local setups (llama.cpp, Open WebUI, Ollama, LM Studio). They describe running agentic workflows in Docker with a local inference server, highlight practical tasks like code refactoring, proofreading, and building two-tower recommendations, and discuss the current limitations and rapid patching in the local-LM ecosystem.