100% Private Local AI with Web Search on Mac Mini M4 Pro (LM Studio Setup)

Can you run a smart, tool-calling AI model locally on an M4 Pro Mac Mini with just 24GB of RAM? In this video, we test Liquid AI’s LFM 2.5 (24B Mixture-of-Experts model with 2B active parameters) using 4-bit MLX quantization (~13GB VRAM) inside LM Studio. Tinyfish: https://go.bitdoze.com/tinyfish We also connect the model to the internet for free using the TinyFish MCP for live web search and page fetching, tweak KV cache and sampling settings for reliable tool calling, and put it through real-world tests—including summarizing live articles and coding a complete HTML/SVG website from scratch. 📌 CHAPTERS: 0:00 - First Usable Local LLM 1:17 - Inside LFM 2.5 Architecture 2:34 - Model Benchmarks & Alternatives 3:16 - Setting Up TinyFish MCP 5:49 - First Chat: Memory Stress Test 8:08 - Summarize a Page Live 11:17 - Building a Solar Panel Website 14:39 - Reviewing the Generated Website 15:32 - Final Verdict & Speed Tips