Why can't GitHub Copilot on Visual Studio customize AI model service providers like it can on VS Code?

[MCerror318]

🏷️ Discussion Type

Question

💬 Feature/Topic Area

Visual Studio

Body

Why can't GitHub Copilot on Visual Studio customize AI large model service providers like it can on VS Code or IntelliJ IDEA? Because of this, I can't use any LLM other than the ones listed on Visual Studio.

[lucaasporto]

Bro, I just asked an AI, and here is the answer. It makes sense:

"In short, VS Code allows you to use custom LLMs in Copilot because its open API lets third-party extensions hack new models into the Copilot UI. Visual Studio is a closed system without that API, meaning you are restricted to the official, pre-approved list of models hosted by GitHub. If you want to use custom LLM providers in Visual Studio today, you cannot use the GitHub Copilot UI to do it; you must uninstall or disable Copilot and install a completely separate third-party AI extension."

If it makes sense to you too, please mark this topic as answered.

[antoniorotundo2]

There is an important distinction here: Visual Studio is not limited exclusively to the GitHub-hosted models, but its BYOM support is currently much narrower than VS Code's.

According to Microsoft's Visual Studio documentation, Copilot Chat in Visual Studio can already use your own API key for OpenAI, Anthropic, and Google. From the model picker you can select one of those providers, enter your API key, and even select a supported model from that provider that is not in the default Copilot list.

So the issue is not simply that “Visual Studio is closed and cannot use custom LLMs.”

The main difference is the provider extensibility.

VS Code currently has a much broader language-model layer. It supports:

• built-in BYOK providers;
• language-model providers contributed by extensions;
• a Custom Endpoint provider for arbitrary compatible endpoints;
• OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages-compatible APIs;
• self-hosted/local models and enterprise gateways.

Visual Studio's current Copilot BYOM implementation does not expose an equivalent general-purpose custom-endpoint/provider mechanism. Its local BYOM support is currently limited to the providers Microsoft explicitly supports.

There are also a couple of limitations worth mentioning:

• In Visual Studio, custom models apply to Copilot Chat, not to code completions or every other AI-assisted feature.
• Microsoft's current documentation says local custom-model support in Visual Studio is not available to Copilot Business/Enterprise users.
• VS Code's BYOK implementation has evolved further and now also supports Business/Enterprise users when the relevant organization policy allows it.

So if what you mean by “customize the service provider” is something like OpenRouter, Ollama, a self-hosted OpenAI-compatible endpoint, an enterprise gateway, or another arbitrary provider, then yes: at the moment VS Code exposes capabilities that Visual Studio's Copilot integration does not.

But you should not need to disable Copilot merely to use your own OpenAI/Anthropic/Google key in Visual Studio — those are already supported in Copilot Chat.

Microsoft documentation:
https://learn.microsoft.com/en-us/visualstudio/ide/copilot-select-add-models

VS Code model-provider documentation:
https://code.visualstudio.com/docs/agent-customization/language-models

[mhmdwaelanwr]

This is currently a client capability difference rather than an issue with your Copilot account.

Visual Studio supports switching between the GitHub Copilot models that are available for that client through the model picker, but it currently does not expose the same “add additional models/providers” capability that VS Code does.

GitHub's documentation explicitly notes that VS Code can add models beyond those available by default with a Copilot subscription, while the supported-model matrix lists model availability separately for Visual Studio and the other Copilot clients.

So at the moment:

• Visual Studio: you can select from the Copilot models supported by GitHub for Visual Studio.
• VS Code: in addition to the supported Copilot models, it has broader support for adding/configuring additional models.

Therefore there isn't a setting you are missing in Visual Studio. If you specifically need a custom LLM provider today, the practical options are to use VS Code for that workflow or use a separate Visual Studio extension that supports your provider.

It would make sense to submit this as a feature request for Visual Studio Copilot parity rather than treating it as a configuration problem.