Enthusiast builds uncensored MCP brain for Home Assistant

This project reflects a shift toward "agentic" local AI where users prioritize instruction compliance over safety alignment for automation. Techniques like orthogonalization (abliteration) are becoming standard for power users who find safety guardrails obstructive to complex technical logic. While 8B models are accessible, their accuracy in tool-use often requires 70B models for reliable reasoning, which demands high-end hardware like the RTX 3090/4090. MCP has evolved from an experimental bridge to a native integration, simplifying how local LLMs interact with IoT devices.