Ollama, OpenCode and Qwen3.6: local setup tutorial

This tutorial walks through Qwen3.6 35B A3B in LM Studio, then Ollama and OpenCode. The presenter explains the mixture-of-experts design and demonstrates a local LLM with everyday writing and Spring AI study prompts.

The LM Studio walkthrough covers downloading the app for Mac M Series or Windows, finding the model and starting a chat. The model shown occupies 22.7 GB. The presenter recommends at least 16 GB of VRAM, preferably 32 GB, and says smaller configurations will run more slowly. These are his hardware recommendations rather than measured requirements. He also reports using the model offline.

The Ollama section covers downloading, listing, removing and running models, plus checking which models are loaded. In the terminal examples, the model says it cannot retrieve current weather or write directly to the filesystem.

OpenCode supplies the AI agent workflow demonstrated next. After connecting the Ollama model, the presenter repeats the Barcelona weather query and requests a Markdown guide saved under Documents/Notes. The session returns weather information and creates a file, which he opens afterward. The transcript does not independently verify the weather result or establish complete privacy for web requests.

The examples also expose limits: initial loading takes longer, some responses contain unexpected characters, and the generated guide slightly exceeds the requested 200 lines.