GitHub Copilot Adds Targeted Model Rules for Organizations

What changed

GitHub Copilot has introduced targeted model rules, a new feature in public preview that allows enterprise owners to dictate which GitHub Copilot models are accessible to specific organizations within their enterprise. Previously, administrators could only set a single, enterprise-wide default for model availability. This new capability enables a more nuanced approach, permitting different models to be made available to different organizations.

Alongside targeted rules, the management interface for default model availability has been refreshed. This updated experience provides a centralized view where administrators can select which default Copilot models are enabled for their organizations. They can set each model's availability to 'Enabled,' meaning it's automatically on for all organizations, or 'Optional,' allowing individual organizations to decide whether to activate it.

Why it matters for builders

This granular control means that builders within an organization can be assured that the Copilot models they have access to are intentionally selected for their team's needs. It moves away from a one-size-fits-all approach, potentially leading to more relevant code suggestions and a more streamlined development workflow tailored to specific project requirements or compliance standards.

Practical impact

Enterprise owners on Copilot Business and Copilot Enterprise plans can now implement more sophisticated deployment strategies for GitHub Copilot. To utilize these new features, administrators should refer to the documentation on 'Managing availability of default models' and 'Managing default models for your organization.' Builders should communicate with their administrators to understand which Copilot models are available to them and how they can be leveraged.

Caveats and source limits

The targeted model rules feature is currently in public preview, indicating that it may be subject to change before its general availability. The source does not provide details on the specific models that can be targeted or any potential limitations on the number of rules that can be created. Furthermore, independent benchmark data or performance metrics for this new control mechanism are not available in the provided source.