GitHub Copilot for JetBrains Gets BYOK Custom Endpoints and Claude Agent Support
GitHub Copilot for JetBrains now supports BYOK custom endpoints, Claude agent provider, local sandboxing, and a built-in debugger skill for Copilot CLI.
GitHub has updated its Copilot plugin for JetBrains IDEs with a significant expansion of bring-your-own-key (BYOK) support, allowing teams to configure OpenAI-compatible custom endpoints with their own API keys. The release, published July 14 2026, also adds Claude agent provider support for Pro and higher plans, local sandboxing, a built-in debugger skill for Copilot CLI, and a cleaner plugin management experience. All new features targeting agents and sandboxing are currently in public preview.
What happened
| Feature | Status |
|---|---|
| BYOK custom OpenAI-compatible endpoints | Generally available |
| Claude agent provider support | Public preview (Pro and higher) |
| Local sandboxing | Public preview |
| Built-in debugger skill for Copilot CLI | Public preview |
| Plugin management via marketplace | Generally available |
GitHub’s July 14 update to the Copilot JetBrains plugin centres on one big theme: giving teams control over which models power their coding assistant. The BYOK expansion means you can now point Copilot at any OpenAI-compatible API endpoint, not just models GitHub itself hosts. If your organisation already pays for a self-hosted model or a third-party provider, you can bring that key directly into your JetBrains workflow.
What else is new?
Claude agent provider in customizations
Teams on Copilot Pro or higher can now configure Claude as an agent provider inside customizations. This means setting up custom agents, defining skills, and writing instructions specific to your team’s work. The feature is in public preview, so expect rough edges and GitHub is actively seeking feedback through the issues repository.
Plugin management and customization improvements
Previously, setting up Copilot customizations required jumping between multiple setup surfaces. The updated plugin management experience lets you browse and install JetBrains marketplace plugins or source-repository plugins directly from within the Copilot customizations panel. Less context-switching, faster team onboarding.
Local sandboxing
Local sandboxing (a way to run agent actions in an isolated environment on your machine rather than in the cloud) is now configurable from inside the JetBrains plugin. New sandbox settings and configuration flows are part of this public preview. GitHub’s documentation covers the distinction between cloud and local sandboxes for teams deciding which fits their security posture.
Built-in debugger skill for Copilot CLI
Copilot CLI sessions now include a built-in debugger skill that lets the agent walk through issues step by step without requiring you to set up a separate debugging tool. This is aimed at agent-driven debugging flows in public preview.
UX and reliability changes
GitHub also improved the model picker in both chat and inline chat, clarified the UI for custom agents and provider settings, and added message re-editing in inline and CLI experiences so you can iterate on a prompt without retyping from scratch. On the stability side, fixes target authentication recovery, account switching, and session persistence across provider changes.
One policy change worth noting: disabling Copilot CLI by organisational policy no longer silences the Copilot CLI provider inside JetBrains IDEs. If you have policy rules in place, check that this new separation matches your intended setup.
Why it matters
BYOK support closes a real gap. Many engineering teams already have negotiated API contracts with model providers, or run self-hosted models for compliance reasons. Without custom endpoint support, Copilot for JetBrains forced those teams to use GitHub-hosted models or look elsewhere. Now they can keep their existing model investment and layer Copilot’s IDE integration on top.
The Claude agent provider addition is notable because it signals GitHub is treating Copilot as a multi-model agent platform, not just an OpenAI wrapper. Teams that prefer Anthropic’s models for certain tasks can now configure Claude directly inside their IDE rather than switching tools.
For development teams already using AI integration in their workflows, this reduces the friction of running different AI tools for different parts of the stack.
Our take
This is a solid, practical release. BYOK custom endpoints are something enterprise teams have wanted for a while, and the implementation here is straightforward: if your provider speaks the OpenAI API format, you can plug it in. That covers a wide range of self-hosted options like Ollama, vLLM, and most commercial alternatives.
The Claude agent support is interesting but still early. Public preview on features this complex often means incomplete documentation and behaviour that changes between plugin versions. Worth testing in a non-critical project first before rolling it out to an entire team.
The policy change around Copilot CLI is easy to miss in a long changelog but could catch IT teams off guard. If your organisation has a policy disabling Copilot CLI, verify that the CLI provider in JetBrains is still configured the way you expect after updating the plugin.
We covered a similar expansion of GitHub Copilot in Visual Studio last month. JetBrains users are now catching up on model flexibility, though the agent features remain a few steps behind what VS offers in stable release.
What to do about it
- Update the GitHub Copilot plugin in your JetBrains IDE to the latest version.
- If you have an existing model API contract, configure your OpenAI-compatible custom endpoint under Copilot settings and test with a small project before broader rollout.
- If you are on Pro or higher and want Claude agents, enable the provider in customizations and define your team’s skills and instructions in a sandbox environment first.
- Check any organisational policies around Copilot CLI. The new separation between CLI policy and CLI provider means your existing rules may behave differently after the update.
- Submit feedback via the JetBrains issues repository while features are in public preview. GitHub has a track record of responding to concrete bug reports during this phase.
If your team is evaluating how to structure AI tooling across the development workflow, our AI integration service covers model selection, API configuration, and team rollout planning.
Frequently asked questions
What does BYOK mean in GitHub Copilot for JetBrains?
BYOK stands for bring your own key. It means you can configure Copilot to use your own API keys and custom endpoints, as long as they are compatible with the OpenAI API format. This lets teams use self-hosted or third-party models instead of only GitHub-hosted ones.
Which Copilot plans support Claude agent provider in JetBrains?
Claude agent provider support in Copilot customizations is available for GitHub Copilot Pro and higher plans. It is currently in public preview as of the July 14 2026 update.
What is local sandboxing in GitHub Copilot for JetBrains?
Local sandboxing runs Copilot agent actions in an isolated environment on your local machine rather than in GitHub's cloud. It is configurable via new sandbox settings inside the JetBrains plugin and is currently in public preview.
Does the new Copilot CLI policy change affect my organisation's existing rules?
Yes, potentially. GitHub changed how CLI policy is handled so that disabling Copilot CLI by policy no longer affects the Copilot CLI provider inside JetBrains IDEs. If you have existing policies disabling Copilot CLI, you should verify the CLI provider behaviour in JetBrains after updating.
