Support local models via ollama or similar

[AzisK]

Hello, do you plan to add support for local models via ollama or similar platforms?

[github-actions]

Thanks for opening this issue, @AzisK!

This is an interesting feature request. I've labeled this as both enhancement and question since it's asking about future support for local models via Ollama or similar platforms.

The Copilot SDK currently communicates with GitHub Copilot through the Copilot CLI, which is designed to work with GitHub's Copilot service. Supporting local models would be a significant architectural change.

The team will review this request. In the meantime, could you provide more details about your use case? For example:

• What specific local model providers are you interested in? (Ollama, LM Studio, LocalAI, etc.)
• What's your primary motivation? (Privacy, cost, offline access, custom models, etc.)
• Would you expect the same API surface to work with local models, or would you be open to a separate configuration?

This additional context would help the team evaluate the feasibility and priority of this enhancement.

AI generated by Issue Triage Agent

[john-mckillip]

Hi @AzisK I was just looking at the documentation for BYOK for a separate issue and is says Ollama is supported already.

[gn00678465]

By using the copilot-sdk in conjunction with ollama, you cannot use models that do not support tool calls, such as deepseek-coder-v2.
You can use models that support tool calls, such as ministral-3:14b, ministral-3:8b.

const session = await client.createSession({
model: "ministral-3:14b",
streaming: true
provider: {
baseUrl: "http://localhost:11434/v1"
type: "openai"
}
});

Perhaps the development team could add a parameter to control whether tool calls are supported

[mfbrki]

Hk

[mfbrki]

Uh