Open Source Friday with Copilot SDK

Key Concepts

• Copilot SDK: A versatile tool for embedding Copilot’s intelligence into applications beyond coding, supporting Python, TypeScript, Go, and .NET officially, with community support for others.
• Agentic Tooling: The SDK facilitates building AI agents capable of complex tasks and “wiggling their way out of corners,” moving beyond simple code generation.
• Community Driven Development: The SDK is open source, heavily reliant on community contributions, and actively maintained by GitHub employees in collaboration with the community.
• /fleet Command: An experimental feature enabling parallel processing by deploying “an army of agents” to tackle tasks concurrently.
• Practical Application & Experimentation: The focus is on solving real-world problems, encouraging users to explore use cases and provide feedback.

Introduction of the Copilot SDK

The recently released Copilot SDK is a comprehensive tool designed to extend Copilot’s capabilities beyond traditional coding scenarios. Unlike the Copilot CLI, which is interactive and offers a headless mode via the -DASP switch, the SDK is built for seamless integration into applications, internal tools, and even non-coding environments. Billing is handled through the user’s existing Copilot subscription, with ongoing improvements to enterprise identity support. The SDK currently officially supports Python, TypeScript, Go, and .NET, but community-contributed SDKs exist for languages like C++, Rust, and Closure, with a strong push for official Rust support.

Community Contributions and Future Development

The Copilot SDK is fundamentally open source, fostering a collaborative environment. GitHub employees actively contribute to and maintain community-created SDKs, ensuring their quality and relevance. The team prioritizes responding to issues within 24 hours and actively engages with user feedback on GitHub. Future development will focus on decoupling components to increase flexibility in SDK support across various languages. Over 6,000 stars have been accumulated on the Copilot SDK GitHub repository as of the recording.

Practical Applications and Use Cases

The SDK’s versatility is demonstrated through numerous examples. Integrating with Microsoft Office applications (PowerPoint, Excel, Word) allows for automation of tasks like slide creation, content updates, and formula generation. A demonstration showcased using the SDK within PowerPoint to research and add API details to a presentation, including web searches and content styling. Another example involved analyzing positive feedback from Microsoft Teams channels and generating a summary. The “copilot tasks” project illustrates the potential for personalized AI assistants, enabling task delegation and updates via voice calls (using 11 Labs integration). Users are actively experimenting with automating PowerPoint, and running long-duration jobs.

Expanding Beyond Code Generation: Agentic Capabilities

The SDK team positions Copilot as a versatile “coding agent” capable of adapting to diverse scenarios, going beyond simple code generation. These agents can be empowered with tools to enhance their capabilities. The introduction of the experimental /fleet command represents a significant evolution, allowing users to deploy “an army of many agents” to tackle tasks in parallel, with defined boundaries to prevent disruption. A 24-hour job experiment using /fleet demonstrated its potential for long-running tasks, while users are also exploring jobs lasting one or two weeks.

Addressing Limitations and Future Vision

While direct scheduled task support is currently absent, workarounds exist using the Copilot CLI in headless mode or integration into custom applications leveraging the operating system scheduler. The long-term vision includes proactive agents delivering updates (e.g., weather at 9:00 AM), aligning with integrations like Work IQ. The team acknowledges the importance of documentation and is committed to investing in improvements based on user feedback. The SDK utilizes Prompt-Based Units (PRUs) for cost, potentially minimizing subscription impact.

Conclusion

The Copilot SDK represents a significant step towards democratizing AI, empowering users beyond developers to leverage AI capabilities in their workflows. Its open-source nature, coupled with a strong emphasis on community contributions and experimentation, positions it as a rapidly evolving tool with immense potential. The introduction of features like /fleet signals a shift towards more sophisticated agentic workflows, and the team’s commitment to user feedback ensures the SDK will continue to adapt and improve. The core takeaway is to experiment with the SDK, solve real-world problems, and contribute to the growing community.