Local LLM Home Automation Gains Traction

New walkthroughs: SeloraHomes HA integration with local 1.7B model and LoRA adapters, MCP server; NAS testing finds Qwen3 1.7B only accurate and fast; Google replacing Assistant with LLM-driven Nest layer. Latest: Home Assistant 2026.8 removes 'Advanced' toggle, integrates local LLMs (llama.cpp, LiteLLM) for fully local voice pipeline with sub-3-second response and air-gapped capability — strong signal for privacy-first smart home AI. New: first-person walkthrough of running Gemma-4-E2B on TrueNAS via LocalAI (10.5 t/s on CPU, 6.2GB memory) with nomic-embed-text-v1.5 for RAG on Paperless AI, Open Notebook, Blinko — adds to lightweight self-hosted AI productivity examples. New: first-person walkthrough of turning security cameras into AI assistants using LLM Vision for Home Assistant — replicable, covers local model option, concrete automations (trash cans, package location). New: reflective homelab ranking includes local LLM (Qwen3.6 27B) for JSON grunt work and Telegram bot for Obsidian integration — real implementation details. New: Muse Glimmer demo — agent autonomously discovers local Home Assistant, queries APIs, writes a dashboard. Short but concrete demonstration. Latest: Google's August 12 event is framed as the real test for Gemini as a home OS, with Pixel Tag and on-device Tensor G5 as concrete signals of ambient AI intent. Competitive pressure from Alexa+ makes this a pivotal moment for AI home automation. New: Show HN: Needle2 — a 14MB agentic LLM for edge devices (phones, wearables, smart homes), enabling private on-device assistants.