Player not loading? Watch on YouTube
This tutorial connects OpenCode to Ollama as a local coding assistant on Windows. The speaker uses Qwen 3 8B on an ASUS TUF A14 with an Nvidia GeForce RTX 5060 laptop GPU and about 8 GB of VRAM. The video includes a disclosed Nvidia partnership. The speaker says the downloaded model can continue to answer offline; installation and downloads precede that claim.
The setup covers installing Ollama, checking its version, downloading Qwen 3 8B and launching OpenCode. In the OpenCode JSON configuration, the speaker points the provider at localhost:11434, enables tool calls and specifies a context limit of 16,384 tokens with 4,096 output tokens. Troubleshooting includes a Windows execution policy change and correcting a configuration key from model to models so the selected model appears.
The coding tests create a Next.js task manager with a TypeScript table, sorting and pagination, then request a POST API route with Zod validation. The speaker demonstrates adding a task but also acknowledges interface bugs. A separate prompt asks the model to explain a React useDebounce hook.
The final section demonstrates GPU detection and CUDA training in Jupyter Notebook. The speaker considers the setup useful for routine coding and learning, while judging Claude stronger for harder reasoning and multi-agent work. These are the speaker's assessments from the examples, rather than a controlled comparison.
