A practical Hacker News thread took off around a gist that condenses an April 2026 setup for running Ollama and Gemma 4 on an Apple Silicon Mac mini. The document is not a benchmark paper or launch announcement; it is the kind of operator note HN tends to amplify when local LLM users are trying to get a stable workstation setup without wasting time on trial and error. The discussion is on Hacker News, while the underlying checklist lives in a public gist.

The gist recommends installing the macOS app with brew install --cask ollama-app, starting the menu bar service, pulling gemma4, and checking GPU usage with ollama ps. The most practical point is the author's sizing note: after trying gemma4:26b on a 24GB unified-memory Mac mini, the system reportedly became barely responsive and swapped heavily under concurrent load, so the guide recommends the default gemma4:latest 8B model instead.

Install Ollama via Homebrew cask and verify the local server with ollama list.
Pull the model with ollama pull gemma4.
Use a LaunchAgent to preload the model every 5 minutes after login.
Set OLLAMA_KEEP_ALIVE=-1 if the goal is to keep the model resident in memory indefinitely.

The guide also treats local deployment as operations work, not just model selection. It walks through launchctl registration, preload logging, and API usage via http://localhost:11434, which is useful for coding agents or local automations that need predictable warm-start behavior. In other words, the interesting part is not simply that Gemma 4 runs on Mac, but how to keep the stack available and responsive on a small Apple Silicon box.

HN commenters immediately turned the thread into a tooling debate. Multiple high-ranked replies argued there is little reason to choose Ollama over llama.cpp, LM Studio, or other local front ends, with critics describing Ollama as slower and overly simplified. That criticism is part of the value of the thread: the gist provides a concrete operational recipe, while the comments expose the tradeoff space around convenience, performance, and control. For local LLM practitioners, the post reads like a compact field note on where today's Apple Silicon defaults work well and where they still hit memory and tooling limits.

#mac-mini

HN Focuses on a Practical Mac mini Setup for Ollama and Gemma 4