Local Models & Home Automation
Key Questions
What Home Assistant automations are covered in the first-person walkthrough?
Five practical automations include a sedentary monitor, procrastination monitor using frontmost app detection, snarky doorbell with LLM Vision, and kitchen tidy reminders via local LLM. These focus on productivity and focus improvements using off-the-shelf tools.
How does the local LLM tool calling with Pi harness compare to Claude Code?
Pi's harness integrates tools like Playwright, SearXNG, Crawl4AI, Qdrant, and mem0, offering greater extensibility than Claude Code's sealed system. It supports full control over setups like Proxmox nodes.
What local voice assistant replaces Google Home and Alexa?
A setup using llama.cpp MoE offloading on GTX 1080 with Gemma-4-26B-A4B and Home Agent integration provides a privacy-focused alternative. It runs entirely locally without cloud dependencies.
What ESP32 projects are recommended over buying smart home gadgets?
Six ESP32 projects offer practical smart home enhancements that make more sense than off-the-shelf devices. Details include voice AI assistants and NFC-based controls for bulbs.
What is WebBrain and its key features?
WebBrain is an open-source, local-first browser agent that reads pages and automates tasks in Chrome and Firefox. It emphasizes privacy and runs without external services.
How does openHAB MCP Server support AI integrations?
The openHAB MCP add-on enables AI assistants like Claude to interact with home automation systems. It provides documentation for system integrations.
What experiments involve giving local LLMs control over infrastructure?
One test gave a local Qwen3.6-35B-A3B model full control over a Proxmox node via Pi agent harness, performing better than expected. Related work covers energy tracking and fine-tuning on scripts.
What local LLM setups are compared for home use?
Comparisons include local LLMs on gaming PCs versus phones, Ollama on low-power CPUs, and MoE models on older GPUs like GTX 1080. Focus is on practical next steps after initial setup.
Local AI home automation gaining depth. New today: first-person walkthrough of 5 weird but practical Home Assistant automations — sedentary monitor, procrastination monitor (using frontmost app detection), snarky doorbell (LLM Vision), kitchen tidy reminder (local LLM). Directly addresses productivity and focus. Replicable with off-the-shelf tools. Also new: first-person walkthrough of local LLM tool calling with Pi harness (Playwright, SearXNG, Crawl4AI, Qdrant, mem0) — Pi's extensibility beats Claude Code's sealed system. First-person experiment giving local LLM (Qwen3.6-35B-A3B) full control over Proxmox via Pi agent harness. WebBrain open-source local-first browser agent. First-person walkthrough replacing Google Home/Alexa+ with local voice assistant using llama.cpp MoE offloading on GTX 1080, Gemma-4-26B-A4B, Home Agent integration. First-person walkthrough of six ESP32 smart home projects. openHAB MCP Server documentation. From previous: local LLM on gaming PC vs phone, Ollama on 15W CPU, ESP32 voice AI assistant, energy tracking beats light automation, 'I replaced my entire browser extension stack with one local LLM', Ollama + n8n + TheMealDB, fine-tuning local Qwen on AutoHotkey scripts, NFC tags for smart bulb control, Qwen 3.6 27B vs Opus 4.8 on 5090.