Support Local Models as a Cost-Optimization Option

[Peppe426]

I would like to suggest support for locally hosted models as an alternative execution option, especially now with the new AI credit subscription model.

Background

With relatively lightweight usage, I burned through 22% of my monthly Copilot pro plus AI credits in less than a day.

I cannot see how we can realistically afford to use the top-tier models as part of our daily workflow.

I'm of the opinion that more advanced tasks such as:

• Large-scale code generation
• Architecture design
• Complex debugging
• Multi-file reasoning
• Agentic workflows

benefit significantly from premium cloud models.

Proposal

Allow users to configure one or more local AI models that GitHub Copilot can use alongside cloud-hosted models.

Possible approaches:

• Select a local model as the default for low-cost interactions.
• Automatically or manually route simple requests to a local model and complex requests to premium cloud models.
• Allow users to choose the model on a per-prompt basis.
• Support common local model runtimes such as Ollama, LM Studio, vLLM, or OpenAI-compatible local endpoints.

Benefits

Developers can reserve AI credits for tasks that truly benefit from premium models while handling routine development work locally.