How to build interactive experiences with canvases

GitHub has launched canvas extensions within the GitHub Copilot app, providing shared, interactive surfaces where developers and AI agents can collaborate in real time. These canvases allow users to visualize data, explore workflows, and take direct actions—such as triaging issues or cleaning up worktrees—without relying solely on text-based prompts. The interface updates dynamically as agents process information, enabling continuous iteration and refinement of the workspace. Developers can shape the canvas by requesting modifications, adding new features, or refining existing ones through natural language instructions.

To create a canvas, users initiate the process with the command /create-canvas in an agent session, then describe the desired functionality and capabilities. The resulting canvas can take various forms, such as a card-based interface for reviewing GitHub Issues or an interactive diagram of a codebase’s architecture. Each canvas is generated from a prompt and evolves alongside the user’s workflow, adapting to specific needs like visualizing project structures or managing active sessions. The flexibility allows for both practical and creative applications.

One example demonstrates a canvas designed to quickly triage GitHub Issues by displaying them in a swipeable card format. Users can approve or reject issues with a simple swipe, with the canvas tracking decisions in real time and organizing items accordingly. Another example generates a colorful, interactive diagram of a codebase, where nodes represent different components and users can hover, drag, or filter to explore relationships. These visual tools aim to make complex information more accessible and actionable.

Additional use cases include a worktree management canvas that visually distinguishes active and stale sessions, allowing users to clean up resources with a single click. Another canvas functions as an interactive prompt coach, reviewing past interactions to suggest improvements such as correcting spelling errors or adding missing context. Developers can also create canvases to search across multiple platforms like Slack, Teams, email, and documentation to find individuals with relevant expertise on specific files or topics.