Note: This article is adapted from the official phala.com announcement.
GitHub recently released the Copilot SDK, which lets developers embed Copilot Agent directly into their own apps and services. Until now, if you wanted Copilot’s capabilities, you worked inside GitHub’s interface. The SDK changes that by opening up the agent layer so builders can bring it into whatever they’re building, on their own terms.
That’s useful on its own. But it also opens up a question most developers haven’t had to think about yet: when your agent is running inside your infrastructure, what happens to the context it’s working with? Your repo contents, your prompts, the state the agent holds while it’s executing, where does all of that actually live, and who can see it?
What Phala Built on Top of It
Phala put together a deployment template that runs the Copilot Agent inside a TEE CVM. Without getting into the weeds, that means the agent executes inside a hardware-protected environment where the contents are sealed off, even from the infrastructure provider itself. Your repo context, prompts, and execution state stay inside that environment throughout the entire process. Nothing leaks out, and the execution can be verified rather than just trusted.
If you’re a developer ready to try it, You can be up and running with a private Copilot Agent deployment faster than you’d expect. If you’re part of a team or institution evaluating how to deploy AI agents without exposing sensitive data to third parties, this is a setup worth a serious look. The infrastructure is already there, and it works today.
The template is ready to deploy at https://cloud.phala.com/templates/copilot-sdk, and both the template code and the upstream Copilot SDK are open for anyone to inspect on GitHub.
Template code: https://github.com/Phala-Network/phala-cloud/tree/main/templates/prebuilt/copilot-sdk
Upstream: https://github.com/github/copilot-sdk
Top comments (0)