Crush on FreeBSD: Gemini, Charm Hyper and Ollama setup

This tutorial sets up Crush as a coding assistant on FreeBSD and connects it to Charm Hyper, Google Gemini and an Ollama server running on a separate Mac. The speaker starts with pkg install Crush, then configures the cloud providers through API keys and environment variables. Both providers answer a request to identify the operating system; the reported result is FreeBSD 15.1 release.

The command examples show Crush asking for permission before execution. The speaker also demonstrates Ctrl+Y to enable YOLO mode, which lets commands run without those prompts, and advises against using it. With Gemini, Crush writes a Python number guessing game for values between 1 and 100, tries to run it and corrects an error. The speaker then plays the generated game using Python 3.12.

For the local LLM setup, the speaker adds an Ollama provider to Crush's JSON configuration with a server base URL and an OpenAI provider type. The presenter first accidentally empties the configuration while reformatting it, then restores its contents from terminal output. The resulting configuration lets the FreeBSD client use a model hosted on the Mac. The model loads and performs some work, but it fails to complete a request to explain the game file. The speaker reports almost ten minutes of waiting and attributes the bottleneck to hardware load while recording on the same machine. The example demonstrates the connection setup, but does not establish usable performance for that hardware.