JetBrains IDE 新增 Codex 代理提供程序及代理功能增强

本次更新将 Codex 作为新的代理提供程序以公开预览形式推出，扩展了定制编辑器，新增了 Hooks 支持与更丰富的 MCP 服务器管理功能，并引入了由 GitHub Copilot Business 和 Enterprise 管理员配置的自定义模型支持。

此外，还为 Copilot CLI 会话增加了审批设置，为 Claude 代理新增了权限模式选择和调试日志支持，并将 Inline Chat 提升为正式可用版本，同时进行了一系列性能和可靠性改进。

Codex 现已作为代理提供程序推出公开预览

Codex 现已作为代理提供程序以公开预览形式提供，让您可以更灵活地选择最适合任务的代理，而且无需离开 JetBrains IDE。

要使用该功能，请先在您的机器上安装 Codex CLI。然后前往 设置 > 工具 > GitHub Copilot > 聊天，启用 Codex 并设置 Codex CLI 路径。配置完成后，从 Copilot Chat 面板的代理选择器中选择 Codex 即可启动会话。

如果您是 Copilot Business 或 Copilot Enterprise 的订阅用户，需要管理员启用编辑器预览功能策略后方可使用此功能。

• Hooks 支持：现在可以在「代理定制」中直接管理 Hooks，适用于本地和 Copilot CLI 会话。
• MCP 服务器管理：现在可以在 Copilot CLI 会话的「代理定制」中直接管理 MCP 服务器。您可以通过 浏览市场 浏览可用的服务器，或在此视图中直接添加 MCP 服务器（支持命令和 HTTP 类型）。对于已配置的 MCP 服务器，您可以查看每个服务器的状态，并执行启动、停止、重启和卸载等关键操作。我们还支持工作区级别的 MCP 服务器。您可以在项目的 .github/mcp.json 中定义和管理它们。专门的用户体验改进将在未来的更新中推出。
• 使用 AI 生成定制文件：在「概览」页面中，您可以点击 新建 按钮快速创建每个定制文件。您还可以通过聊天中的 /create-instruction、/create-prompt、/create-skill、/create-agent 或 /create-hook 来生成脚手架文件。

为 Copilot CLI 会话配置审批设置

您现在可以为 Copilot CLI 会话配置审批设置，从而更精确地控制代理任务期间的工具调用和审批处理方式。

要使用此功能，请在 Copilot CLI 会话的聊天输入区域从权限下拉菜单中选择一个权限级别。

• 默认审批：Copilot CLI 代理遵循您配置的审批设置，并根据当前策略提示确认。
• 绕过审批：所有工具调用自动批准，无需确认对话框。必要时，Copilot CLI 代理仍会提出澄清问题。
• 自动驾驶（预览）：所有工具调用自动批准，Copilot CLI 代理会自动回复澄清问题，以便持续迭代直至任务完成。

Claude 代理会话支持权限模式

Claude 代理会话现在支持 权限模式 选择，让您可以选用适合工作流程的审批行为。要开始使用，请在 Claude 会话的聊天输入区域从权限下拉菜单中选择一种权限模式。详细了解权限模式。

Claude 会话支持调试日志

代理调试日志现已支持 Claude 会话，便于您查看和调试会话活动。

本地模型选择器增强

本地模型选择器现在包含模型视图管理功能，让您对跨会话模型的列出和选择方式拥有更多控制权。在 本地代理 中，点击模型下拉菜单，然后选择 管理模型… 打开模型管理视图。

自定义模型支持

JetBrains IDE 的 GitHub Copilot 现在支持由 GitHub Copilot Business 和 Enterprise 管理员在 GitHub 设置中配置的自定义模型。管理员完成配置后，自定义模型将自动对成员可用。有关设置详情，请参阅使用您自己的 API 密钥。

性能与可靠性改进

我们对 JetBrains IDE 在性能和日常可靠性方面做出了若干改进：

• 增加了基于用量计费的最佳实践提示，帮助您更好地理解和管理工作中的 AI 信用消耗。
• 提升了加载 Copilot 面板时的稳定性和性能。
• 通过建议缓存改进了下一步编辑建议（NES），以获得更快、更灵敏的补全。
• 改进了文件监视器的性能并降低了内存占用。

可靠性修复

此版本包含重要的可靠性修复：

• 修复了 BYOK 会话可能触发 Copilot 子代理并消耗 Copilot 用量的问题。子代理逻辑现在会遵循当前活动的 BYOK 提供商。
• 修复了多个 UI 冻结问题。

Inline Chat 正式可用

Inline Chat 现已正式可用。

我们鼓励您试用最新版本的 GitHub Copilot 插件并分享您的反馈。您的意见对我们完善和改进产品至关重要。

您的反馈推动着我们的进步。我们期待通过以下渠道听到您的使用体验：

This update brings Codex as a new agent provider in public preview, expands the Customizations editor with Hooks support and richer MCP server management, and introduces custom model support configured by GitHub Copilot Business and Enterprise administrators.

It also adds approval settings for Copilot CLI sessions, permission modes selection and debug logs support for the Claude agent, and takes Inline Chat to general availability, alongside a set of performance and reliability improvements.

Codex is now available as an agent provider in public preview, giving you more flexibility to pick the agent that best fits your task. And you can do this without leaving your JetBrains IDE.

To use it, first install the Codex CLI on your machine. Then go to Settings > Tools > GitHub Copilot > Chat, enable Codex and set the Codex CLI path. Once configured, select Codex from the agent picker in the Copilot Chat panel to start a session.

If you are a Copilot Business or Copilot Enterprise subscriber, an administrator will have to enable the editor preview features policy before you can use this feature.

• Hooks support : Hooks can now be managed directly in “Agent Customizations” in both local and Copilot CLI sessions.

• MCP servers management : MCP servers can now be managed directly in “Agent Customizations” for Copilot CLI sessions. You can browse available servers through Browse Marketplace or add an MCP server directly from this view (both command and HTTP types are supported). For configured MCP servers, you can view each server’s status and perform key actions such as start, stop, restart, and uninstall. We also support workspace-level MCP servers. You can define and manage them with .github/mcp.json in your project. Dedicated UX improvements are coming in a future update.

• Generate customization files with AI : In the “Overview” page, you can quickly create each customization file by clicking the New button. You can also scaffold files from chat with /create-instruction, /create-prompt, /create-skill, /create-agent, or /create-hook.

You can now configure approval settings for Copilot CLI sessions, giving you more control over how tool calls and approvals are handled during agentic tasks.

To use this feature, select a permission level from the permissions dropdown in the chat input area for Copilot CLI sessions.

• Default Approvals : Copilot CLI Agent follows your configured approval settings and prompts for confirmation based on your current policy.
• Bypass Approvals : All tool calls are auto-approved without confirmation dialogs. When needed, Copilot CLI Agent still asks clarifying questions.
• Autopilot (Preview) : All tool calls are auto-approved, and Copilot CLI Agent auto-responds to clarifying questions so it can continue iterating until the task is complete.

Claude agent sessions now support Permission modes selection, letting you choose the approval behavior that fits your workflow. To get started, select a permission mode from the permissions dropdown in the chat input area for Claude sessions. Learn more about permission modes.

Claude sessions are now supported in agent debug logs, making it easier to review and debug session activity.

The local model picker now includes model view management, giving you more control over how models are listed and selected across sessions. In Local agent , click the model dropdown, then select Manage models… to open the model management view.

GitHub Copilot for JetBrains IDEs now supports custom models configured by GitHub Copilot Business and Enterprise administrators in GitHub settings. Custom models are automatically available to members once configured by an admin.

For setup details, see Use your own API keys.

We’ve made several improvements to performance and day-to-day reliability across JetBrains IDEs:

• Added best practice tips for usage-based billing, so you can better understand and manage AI credits consumption as you work.
• Improved stability and performance when loading the Copilot panel.
• Improved Next Edit Suggestions (NES) with suggestion caching for faster and more responsive completions.
• Improved file watcher performance and reduced memory usage.

This release includes important reliability fixes:

• Fixed an issue where BYOK sessions could trigger Copilot subagents and consume Copilot usage. The subagent logic now respects the active BYOK provider.
• Fixed multiple UI freeze issues.

Inline Chat is now generally available.

We encourage you to try out the latest version of the GitHub Copilot plugin and share your feedback. Your input is invaluable in helping us refine and improve the product.

Your feedback drives improvements. We’d love to hear about your experience in the following channels: