Gastown with local ollama?

[timohund]

Hi,

since i am interested in agentic coding i had a look into gastown during the last days. I was using it with z.ai but as mentioned token cosumption is very high, so far so clear. Since i have an nvidia rtx 3090 with an instance of ollama, i did a setup with ollama and open code. So far gastown is running. I am using qwen-coder-3:32b and in general it is running.

I can the mayor the task to break down an spec.md into multiple tasks and epics. But somehow, most of the time, the polecats do not work in these issues. I recognize, that issues are sometimes created in the rig, with a wrong prefix.

So my questions would be:

• Has anybody a running local setup that is working?
• Which models are you using?
• Or do i need in general a more strong cloud model?
• If yes how can i maybe delegate tasks partly to local models?

Any help appreciated