← Back
GitHub
GitHub Copilot for JetBrains gains agentic capabilities: custom agents, hooks, and auto-model selection now GA
GitHub Copilot · featurereleaseapi · github.blog ↗

Agentic Capabilities Now Generally Available

GitHub Copilot's core agentic features are now fully available in JetBrains IDEs. Custom agents, sub-agents, and plan agents move from preview to general availability, enabling developers to tailor Copilot to their specific workflows and collaborate with specialized agents directly within their IDE.

New Agent Hooks and Configuration Features

Agent hooks are available in public preview, allowing developers to run custom commands at key points during agent sessions. This enables workflow automation, policy enforcement, and integration with external tools while maintaining control over agent behavior. Supported events include userPromptSubmitted, preToolUse, postToolUse, and errorOccurred hooks, defined via hooks.json in the .github/hooks/ folder.

Auto-approve support for MCP reduces manual approvals during agent sessions by allowing configuration at both the server and tool level. This is accessible via Settings > GitHub Copilot > Chat > MCP Server and Tool Auto-approve Configuration.

Improved Model Selection and Reasoning Visibility

Auto model selection is now generally available across all Copilot plans, allowing Copilot to automatically choose the best available model based on real-time performance and availability. For extended-reasoning models like Codex, a dedicated thinking panel provides visibility into the model's reasoning process with configurable Anthropic thinking budgets. A new context window usage indicator in the chat panel shows how much context is being consumed during conversations.

Customization and Instruction Files

Support for AGENTS.md and CLAUDE.md instruction files makes it easier to tailor Copilot's behavior to project conventions. Copilot automatically discovers and loads these files into context during agent workflows from the current workspace or globally. The /memory slash command provides quick access to manage instruction file settings. Developers can generate an initial AGENTS.md using the Generate Agent Instructions action in chat.

User Experience and Quality Improvements

The sign-in experience has been streamlined with automatic chat panel opening after authentication and improved device code prioritization for GitHub Enterprise users. Quality improvements address stability issues with terminal output reading, file update application via the replace_string_in_file tool, and various UI inconsistencies. Platform support for Windows ARM has also been enhanced. Edit mode has been marked as deprecated in the chat mode dropdown.