Skip to content

Pay for runs with a plan you already have

If you pay for ChatGPT, Claude or GitHub Copilot, that plan can pay for the model behind your QuickAgent runs. You sign in once, in your browser, and Agentry stops asking for an API key (a long secret code that lets a program use your account). Your plan pays and your plan's limits apply.

These work for runs on your own computer only: the desktop app and the agentry command. A run In Agentry (hosted) always uses Agentry's own model access. It never uses your plan or any account you connected on your computer (Running on Agentry).

Google's Gemini plans cannot be used this way.

How it works in short

In the desktop app:

  1. Open Settings → Model provider. The first-run setup has the same choice.
  2. Choose the provider: OpenAI, Anthropic or GitHub Copilot.
  3. Choose how it pays: the sign-in, or an API key.
  4. Click the sign-in's button. Your browser opens.
  5. Approve on the provider's own page in your browser.
  6. Choose a model, then Save.

On the command line, run agentry setup and choose the sign-in when it asks what pays for the model. Or sign in directly with agentry model-signins add chatgpt, agentry model-signins add claude or agentry model-signins add copilot. agentry model-signins list shows what is signed in on this computer. agentry model-signins remove <name> signs one out (Removing a sign-in).

You can keep a sign-in and an API key side by side. One is the active choice, and you can switch in the same place. Each provider has one sign-in per computer.

ChatGPT

The button reads Sign in with ChatGPT. It opens OpenAI's own page, where you approve Agentry.

Which plans. Only ChatGPT plans that let other apps use them. OpenAI's page names Plus and Pro. If your account's plan does not allow it, the sign-in stops, says so, tells you to use an API key instead and keeps nothing.

What it uses. Agentry runs OpenAI's Codex program for these runs. The models you can pick are the ones your plan lists.

No browser on the computer? The command prints the address instead and waits ten minutes. Open the address on another computer, approve there, and paste back the address your browser ends on.

Limits. OpenAI may limit how much of your plan other apps can use in a week, on top of the plan's own limits. When a limit is reached, the run ends with plan limit reached. Agentry does not know when the limit resets and does not guess. Check ChatGPT, or wait and try again.

Claude

The choice reads exactly Use your Claude subscription (runs Claude Code). "Runs Claude Code" means these runs use Anthropic's Claude Code program, not Agentry's usual one.

How you sign in. Claude Code signs you in itself, on Anthropic's own page in your browser. Agentry starts it and waits. You do not copy or paste anything. Claude Code keeps the sign-in in a place of Agentry's own on your computer. Agentry never reads that place. It is never your own Claude Code settings, so a Claude Code you already use is left alone.

If something stops the browser route, such as another program using the port Claude Code needs, the desktop app ends the sign-in and says why. The command line can ask you to paste a code from Anthropic's page instead.

The first time. Your computer builds a Claude Code image, a packaged copy of the program that Agentry runs in. That takes a few minutes and downloads roughly 110 MB. It happens once, and again if Agentry later updates the image.

Limits. When your plan's limit is reached, the run ends with plan limit reached and Claude Code's own reset time.

Another way to use your plan. Anthropic says its Max and Team plans include monthly API credits. You can spend them with an API key from the Console organization linked to your plan. Choose the API key way for Anthropic instead of the sign-in, and paste that key. Anthropic's article on it has the current terms.

On the command line, agentry model-signins add claude --installed signs in the Claude Code already installed on your computer, with its own browser. The desktop app runs the image, so that sign-in does not count there.

GitHub Copilot

The button reads Sign in with GitHub. For GitHub Copilot this is the only way to pay. There is no key to choose.

What happens. GitHub's page opens and asks for a short code. The desktop app copies the code for you. Paste it there and continue. On the command line the code is printed, with the address of GitHub's page. The code works for 15 minutes.

Your plan. Choose your Copilot plan: Free, Pro, Pro+ or Max; Business; or Enterprise. On the command line the answers are individual (Free, Pro, Pro+ and Max), business and enterprise, or --copilot-plan sets it.

The model. You choose from the models your plan offers, as Copilot CLI lists them. auto, which lets Copilot choose, comes first. On Copilot Free it is the only choice. A model your plan does not include is not offered, and neither is one your organization has turned off. Copilot CLI makes the list in the same kind of container a run uses. Agentry never asks Copilot's servers for it itself.

  • In the desktop app the model starts on auto. Click List my plan's models to see the others. The first list can take a few minutes, because it prepares the images runs use.
  • agentry setup lists the models after it asks your plan. If the list cannot be made, it says so and you type a model name instead. --model names one directly.

Your organization's rules. If you use Copilot through work or school, your organization or enterprise at GitHub may block Copilot CLI, the program these runs use, or may require someone to approve every tool call. The run then ends and names the policy. Only your organization's administrators can change it. You can pay another way instead.

Not available yet. Agentry's GitHub sign-in apps are still being registered with GitHub. Until they are, Sign in with GitHub says it is not available yet and keeps nothing.

Removing a sign-in

Choose Remove next to the sign-in in Settings → Model provider. On the command line, run agentry model-signins remove chatgpt, claude or copilot. What happens differs by provider:

  • ChatGPT. The sign-in is cancelled at OpenAI and forgotten on your computer. Agentry keeps a small note of how it registered with OpenAI, which OpenAI asks apps to keep, so a later sign-in reuses it. To sign in with a different ChatGPT account, remove the sign-in, then run agentry model-signins add chatgpt --new-account.
  • Claude. Claude Code signs itself out, then its sign-in is deleted from your computer. If Claude Code cannot run (no image yet, or no internet), the sign-in is deleted anyway and Agentry says the session at Anthropic ends when it expires or when you sign out of your devices at claude.ai.
  • GitHub. Agentry sends the sign-in to GitHub to be cancelled. GitHub answers before it acts, so Agentry says "submitted", not "cancelled". A sign-in a run was already given stops working by itself within eight hours. Agentry's app also stays in your list of authorized apps until you remove it at github.com/settings/applications.

What a run on a plan costs and records

Your plan at the provider pays for the model's work, and the provider's own limits and prices for the plan apply. How much of your plan a run uses depends on the model and the task.

A run on a plan records tokens, never dollars. Tokens are the small pieces of text a model reads and writes. So a cost limit (--max-cost) cannot apply to a run on a plan, and Agentry refuses it. The run's receipt names the kind of sign-in, never your account.

What has been tested

Each sign-in works end to end against stand-ins, practice copies of the vendors' services, and against the real Codex, Claude Code and Copilot CLI programs. None has been tried yet with a real ChatGPT, Claude or GitHub account. That live test is still to come. What this page says about plans, limits and wording comes from the vendors' own documentation, so what you see may differ.