GitHub published an update for GitHub Copilot for JetBrains on July 14, 2026. The focus is giving development teams more control over model providers, agents, plugins, and sandbox behavior. This is not just a model release; it moves Copilot into a more configurable developer workflow.
The clearest change is expanded bring-your-own-key support with custom endpoints. Users can configure OpenAI-compatible custom endpoints and API keys to use their own models. That gives teams with internal models, proxy services, or multi-model strategies more choice and reduces the need to bind every workflow to one provider.
Customizations now include a fuller plugin-management experience. Users can browse and install plugins from a marketplace or a source repository without jumping between disconnected setup surfaces. The update also supports custom agents, skills, and instructions for the Claude agent provider, although that capability remains in public preview for GitHub Copilot Pro and higher plans.
Local sandbox support is also in public preview, with new sandbox settings and configuration flows. Copilot CLI sessions gain a built-in debugger skill that can guide step-by-step investigation in the development environment. These features extend an agent's operating surface beyond chat into local tools and workspaces.
GitHub also lists UX and reliability improvements across the model picker, custom agents, provider settings, authentication recovery, account switching, and session persistence. More provider flexibility is useful, but it creates governance work around endpoint allowlists, API-key handling, plugin provenance, and sandbox permissions.
BYOK should therefore be treated as more than a model-switching convenience. Teams need to define which endpoints are allowed, which agent instructions can load, and which files and tools a local sandbox can reach. With those boundaries in place, flexibility can become a repeatable workflow instead of another opaque configuration layer.
