GitHub Copilot

If you pay for GitHub Copilot, you can use it in Empryo. The Claude, GPT, and Gemini models your Copilot plan includes work in Empryo at no extra cost.

SetupLink to this section

1. 1

   Sign in

   /login copilot

   Empryo shows a short code and opens github.com/login/device. Enter the code there and approve. Empryo stores the sign-in in your system keychain. You do not need an editor installed.

2. 2

   Use it

   Press Ctrl+L, pick copilot, and choose a model.

   empryo --headless --model copilot/claude-sonnet-4.6 "explain this file"

Company GitHub (Enterprise)Link to this section

Some companies run Copilot on their own GitHub instead of github.com. This is either GitHub Enterprise Cloud with data residency (an address like octocorp.ghe.com) or a self-hosted GitHub Enterprise Server. A sign-in on one GitHub does not work on another, so you give the host when you sign in:

/login copilot octocorp.ghe.com

/login copilot enterprise asks you for the host instead. In the /login list, this is the GitHub Copilot (Enterprise) row, in both the terminal UI and the desktop app. Headless mode takes the host in the same place:

empryo --headless --login copilot octocorp.ghe.com

Give the main host name of your company's GitHub. Empryo works out the other addresses from it:

Empryo saves the host with your sign-in. To use a different host for a single run or in a CI image, set COPILOT_GITHUB_DOMAIN. Signing out switches back to github.com.

Business and Enterprise seats on github.com need none of this. Sign in with /login copilot as usual.

Reuse a sign-in from your editorLink to this section

If you already signed in to Copilot in VS Code or a JetBrains IDE, you can reuse that token:

jq -r '."github.com".oauth_token' ~/.config/github-copilot/apps.json
empryo --set-key copilot <token>

On Windows, the file is at ~\AppData\Local\github-copilot\apps.json. You can also paste the token in /keys in the terminal UI.

Which models your plan can useLink to this section

Copilot's model list shows every model GitHub offers, not only the ones your plan includes. Copilot Free can only use the base GPT models, gpt-4.1 and gpt-4o. Any other model returns:

The requested model is not supported.

GitHub's message does not say why, so Empryo adds the reason and your plan name to the error. Claude, Gemini, and the GPT-5 models need a paid Copilot plan.

Some new models require you to accept their terms before first use. Empryo accepts them for you the first time you pick the model, the same way the official editor extensions do.

Each model has its own reasoning effort levels in Copilot. /effort shows the levels your model supports. For example, Copilot's Claude models take low, medium, high, and max.

Usage and quotaLink to this section

Copilot gives you a monthly allowance instead of charging per token, so what matters is how much of the month's allowance is left. GitHub reports it, and Empryo shows it in the same kind of meter it uses for other subscriptions:

Subscription · copilot
● pro                    octocat
  ▸ premium        [▰▰▰▰▰▱▱▱▱▱]   50%  ↺ 12d
    chat           [▰▱▱▱▱▱▱▱▱▱]   10%  ↺ 12d
    completions    [▱▱▱▱▱▱▱▱▱▱]    0%  ↺ 12d  editor only

GitHub reports three allowances:

That is why completions is dimmed and marked *editor only*. Running out of completions does not affect Empryo.

Where you can see it:

The status bar shows one plan meter, for the model you are using. A copilot/* model shows your Copilot allowance. A proxy/* model shows your proxy plan.

Empryo checks your allowance at most every five minutes. The numbers change slowly, and GitHub's Acceptable Use Policies ask clients not to send unnecessary automated requests.

Context size depends on your planLink to this section

Copilot often offers models with a smaller context window than the model supports natively. For example, a standard seat gets Claude Opus with 200k tokens, while Anthropic offers 1M. Seats with extended context get more. GitHub reports the numbers for your account.

Copilot reports two numbers for each model, and Empryo uses both:

If your seat has 1M-token context, open the model picker once. Empryo then loads your account's real numbers and shows the larger window. Until then, it uses the numbers for a standard seat. Guessing low is safer: if Empryo overestimated the window, it would compact too late and Copilot would reject the turn.

Where the model list comes fromLink to this section

Empryo builds the list from three sources, in this order. Each one only fills gaps the previous one left:

When you sign in with /login copilot, Empryo drops the previous account's list and loads the new one right away. Numbers from an old Free plan never linger after you upgrade or switch accounts.

Empryo shows only the models GitHub marks as available in its model picker, the same as the official client.

How Empryo identifies itselfLink to this section

GitHub can answer differently depending on which client asks. For example, extended context windows are offered to specific clients. So Empryo sends the same editor version headers a current Copilot client sends.

Empryo does not pretend to be GitHub Copilot Chat. Its User-Agent is `Empryo/<version>`. It also never sends invented machine or session ids.

You can override each value with an environment variable:

For example, COPILOT_INTEGRATION_ID=copilot-cli empryo --list-models copilot asks GitHub for the model list as a different client. The source: line in the output tells you which of the three sources answered.

Request limitsLink to this section

GitHub's terms do not require a particular client. They do limit volume: the Acceptable Use Policies forbid "excessive automated bulk activity" and "undue burden on our servers through automated means", and GitHub can suspend API access for request rates it considers excessive.

Empryo keeps its traffic close to the official clients:

In practice:

• The model list is fetched at most once a minute. Signing in is the only exception.
• A 429 or 403 error stops requests instead of retrying. Empryo waits for the time GitHub asks for (or 15 minutes), uses your last saved model list, and tells you once.
• A rejected token is checked only once, never in a loop.
• Usage is checked at most every five minutes, and only when you use a copilot/* model.

Legal positionLink to this section

GitHub changed its terms on 27 April 2026. Which terms apply depends on your plan:

Section J covers who owns input and output, model training and how to opt out, disclaimers, and indemnity. It does not restrict which client you use. The limits that do apply come from other parts of the agreement:

The official routeLink to this section

GitHub now offers a supported way to build on Copilot:

If using an approved integration matters more to you than using Copilot models inside Empryo, use that SDK. Empryo talks to Copilot's chat API directly instead. That is what lets Copilot models use Empryo's tools, the Genome (Empryo's map of your code), and compaction, like every other provider.

Check this page again if GitHub publishes new terms after 2026-08-02.