Crush with Ollama: setup and coding tests

This tutorial installs Crush, an open source terminal coding assistant, and connects it to Ollama for a local LLM workflow. The presenter uses Ubuntu with an NVIDIA RTX 6000 and 48 GB of VRAM. That is the demonstration hardware, not a stated minimum requirement.

The installation uses npm and assumes Node.js and npm are already installed. Ollama is also already running. The presenter edits crush.json to set an OpenAI-compatible endpoint on port 11434 and chooses the model ID shown by Ollama. Qwen3-Coder is the model used in the test; GPT-OSS 20B is also installed.

After launching Crush, the presenter selects the model, initializes the project and grants permission to create an AGENTS.md file. The first task asks Crush to analyze a tiny app.py containing a single print statement. The output shows a tool call rather than an answer. The presenter attributes this behavior to model function calling, particularly during streaming, but the demonstration does not establish the cause.

A later prompt asks for an HTML landing page, and Crush begins generating it. The presenter ends the test early and recommends Claude Code for web pair programming based on his preferred experience. He points to a separate Ollama setup tutorial for Claude Code; that setup is not demonstrated here.