Inline image generation lets you describe an image in natural language and receive a preview saved directly to your workspace. You can optionally add a “Made with AI” watermark, which admins can disable via policies. Copilot Fusion is now generally available, automatically routing and orchestrating multiple models based on the selected effort level. Fusion offers three effort tiers—Standard, High, Max—and uses cascade or critique workflows to balance performance and cost. Admins can control which models Fusion may use and set it as the default model for their organization.
GitHub Copilot adds Projects, Computer Use, Security Extension preview
Copilot Projects enters public preview, letting you set a goal, attach repos, design mocks or Slack threads, and run AI agents that deliver pull requests. Computer Use, now in public preview for macOS and Windows, lets Copilot read, click, type, and navigate across desktop apps after permission. The new security extension, in public preview, displays code-scanning alerts in Copilot and lets you ask the model to explain or fix them. Run /security-review on local changes to see new findings alongside existing alerts, all within the same canvas.
GitHub Copilot Fusion adds local model routing and Windows download
Local model routing lets Fusion automatically choose between on-device and cloud models in the CLI, Copilot app, and VS Code. Managed Windows download lets users install MAI Code 1.1 Flash via the Windows ML provider in a guided flow. Bring-your-own local models can be connected via BYOK on any OS, allowing Fusion to route requests intelligently.
GitHub Copilot VS Code adds cross-client session continuity and remote Agent Host
Session continuity lets you start a cloud Copilot session in VS Code and pick it up later in the mobile app, web UI or CLI. Agent Host isolates sessions from VS Code windows, letting them continue after closing the folder and run on a remote machine via SSH or dev tunnels. You can monitor progress and give new directions to an agent from the GitHub Mobile app on your phone. In preview, the Agents window can run multiple models side-by-side, let a Judge model recommend a solution, and synthesize the best ideas. Customization Marketplace and Connectors let you add skills, plugins, and external services like Slack or Notion, with admin policy controls.
GitHub Copilot for JetBrains adds Fusion, sandboxes and suggestions
Copilot Fusion now works in JetBrains IDEs, offering Standard, High, or Max tiers and automatic model orchestration, with a 30% discount for the first 30 days. Local sandboxes are generally available, providing a secure boundary that limits filesystem, network, and credential access for agentic tasks on the developer’s machine. Next edit suggestions are GA in JetBrains, delivering higher-quality, context-aware inline suggestions and smoother acceptance. A new session sidebar button and default permission setting let you open sessions side-by-side and avoid repeated permission prompts.
GitHub Copilot adds Intelligence with Memory, Dreaming, Code Search
Memory is now generally available, letting the agent retain personal preferences and repository conventions across sessions and devices. Dreaming preview lets Copilot continuously refine personal and repository memories, proposing updates via pull requests to keep guidance current. Code Search lets Copilot find symbols, patterns, or semantic matches across all organization repositories without needing a local clone.
GitHub Copilot adds Agentic Workflows, Automations, Plugin Controls
GitHub Agentic Workflows are now generally available, letting you define AI-driven actions in natural-language markdown that compile to standard Actions YAML. Shared Copilot Automations let teams create, edit, and run collaborative AI workflows that are billed to the repository and persist after creators leave. Repository plugin defaults and enterprise-managed plugin controls let admins enforce required plugins or block unwanted ones across all contributors.