News
VS Code Expands AI Flexibility with Bring Your Own Key
Microsoft announced expanded AI model support in Visual Studio Code through a new Bring Your Own Key (BYOK) capability that lets developers connect models from different providers by entering their own API keys. The feature allows developers to use AI models beyond GitHub Copilot's built-in options, integrating services such as OpenAI, Google, Ollama, and OpenRouter directly into the editor.
BYOK was initially launched as a configuration-based option that let users manually connect third-party models through environment variables. The October update transforms it into a fully extensible system powered by the new Language Model Chat Provider API, which lets extension developers register and manage model providers programmatically. This change enables a plug-in ecosystem where model vendors can create extensions to make their models available inside VS Code without manual setup.
The VS Code team said the new API makes it possible for providers to integrate their models securely while maintaining a consistent experience for developers. Built-in models remain supported, but BYOK expands the framework to include external and custom AI sources, supporting both local and cloud deployments. The team described it as a shift toward an open, scalable AI ecosystem within the editor.
Among the first examples of BYOK-based integrations is the AI Toolkit for Visual Studio Code, which connects to models hosted in Azure AI Foundry or locally through Foundry Local. Another extension, the Hugging Face Provider for GitHub Copilot Chat, brings access to popular open models such as Kimi K2, DeepSeek V3.1, and GLM 4.5. The Cerebras Inference extension offers high-speed inference performance, delivering token generation rates of up to 2,000 tokens per second for faster experimentation.
Developers using OpenAI-compatible APIs can also configure custom providers through the new github.copilot.chat.customOAIModels setting. This feature, available in VS Code Insiders, lets users define which models to expose in chat, specify default prompts or tools, and manage API endpoints directly within the settings file. Microsoft said this allows fine-grained control for developers who want to tune AI behavior without relying on preconfigured integrations.
The update also sets the foundation for more sophisticated model management. Microsoft plans to introduce a dedicated model configuration interface, easier installation for model-providing extensions, and enhancements to built-in Copilot providers that use the new APIs. These upcoming features are intended to streamline how developers install, configure, and switch between AI models across different environments.
According to the VS Code team, the goal of BYOK is to give developers freedom of choice while maintaining security and performance standards. The API ensures that API keys and model credentials are stored locally and not shared between providers. The approach also supports organizational policies that require developers to use specific AI models or host models within private environments.
BYOK is available now to developers using GitHub Copilot Free, Pro, and Pro+ plans in Visual Studio Code. Additional documentation and example extensions are available on the official Visual Studio Code blog at code.visualstudio.com.
About the Author
David Ramel is an editor and writer at Converge 360.