One thing I’ve been tracking, and posting about on LinkedIn, is the ability to bring Github Copilot functionality into environments where there is not ability to connect to the internet. This is a key requirements of environments at higher impact levels. Recently there have been some big announcements on this topic.
Problem: We have had the ability to do Bring Your Own Model (BYOM) for a while now (about CY end of Q1 2025), the problem in IL4+ and for the DIB has consistently been that when you try to leverage the VSCode extension or the CLI, the tool performs two steps:
1 – First it makes an authentication call against GitHub.com or GHE.com. 2 – It performs a licensing check to confirm your license.
When it couldn’t perform these checks, customers had no recourse to get GitHub copilot to load to support using BYOM against AOAI in gov.
Current Solution: With the recent announcements, GitHub Copilot CLI has now enabled the ability to use Bring-Your-Own-Key to bypass both of these calls. By doing this we now no longer have to reach out to the internet for the tool to work.
You can read the steps here.
You can read the blog post here
Limitations: Currently this model is only supported in GitHub Copilot CLI, but there are several PRs to the VSCode extension that are working their way through to enable this in VSCode.
What did I build? That being said, I’ve built a solution accelerator that shows this. It deploys to azure government, and creates a vnet that blocks all outbound connectivity creates a dev vm, and an AOAI instance with gpt-5.1 and gives full instructions to get it up and running so you can demo, it also supports you giving an existing vnet and it will deploy it with bastion.