Hey! One of the maintainers of Ollama. 8GB of VRAM is a bit tight for coding agents since their prompts are quite large. You could try playing with qwen3 and at least 16k context length to see how it works.