CoPilot doesn't execute actions in Agentic mode with local LLM

taco_truck_tom

I'm playing with local ollama model (qwen2.5-coder:7b) trying to connect it to CoPilot. The local model is detected without issues. Then, in Agent mode I give it a simple task like "Add a string field named XXX to this class" (C# project). Model runs, provides some answer which look sensible, but CoPilot doesn't execute the action, instead it prints json output into its chat, something like: {"name": "get_file", "arguments": {"filename":...

Originally asked by ElDog on Stack Exchange (stackoverflow) (CC BY-SA 4.0).

uphilltech

Great question — answering as the UphillTech team:

I was playing around with Ollama and Visual Studio 2026 today using the qwen3.5-9b model. Using this model I only got the "ask mode" to work in the Copilot Chat window using the local Ollama running instance. Then I switched to gemma4:12b-it-q8_0 , asked it the same question, in the same Visual Studio instance, where I just switch to this model and the agent mode worked. The task I asked it to do was to "implement the single instance pattern" for an application. Maybe this could be also worth a try for you? So to me it looks like the model is not working with tools of Visual Studio 2026. Maybe it was trained for other tools?

Full answer by Xargas: https://stackoverflow.stackexchange.com/a/79979894 (CC BY-SA 4.0).

Need it done for your business? UphillTech.