Skip to main content
Every Pullfrog run spends model tokens, and there are four ways to pay for them. This page is the decision: pick a route, then follow the link for setup. If a run fails with a credentials or balance error, the fix is almost always to complete one of these four. This is separate from the Pro plan, which is what lets an organization run on private repositories. Pro does not include model tokens, and paying for model tokens does not require Pro.

​
Pick a route

The last two are both BYOK (bring your own key). BYOK and Router are an account-wide toggle in the Model usage section of the Billing card; free models work under either.

​
Free models

Some models cost nothing to run. Pick one in the model selector and you skip credentials and balance entirely — no card, no key, no subscription. This is the fastest way to get a failing repository running again, and a reasonable permanent choice for smaller tasks.

​
Pullfrog Router

Pullfrog Router reaches every major provider through one billing account, at raw provider cost with no markup. You manage no keys: Pullfrog mints a short-lived proxy key for each run. You hold a prepaid balance, and runs stop when it reaches zero. Add a card and top up in the Billing card of your account console, then turn on auto-reload so the balance refills before it runs out. Without auto-reload, an empty wallet stops every run until you top up by hand. Full detail: Pullfrog Router.

​
Your own API key

Bring a provider API key and the provider bills you directly, at their prices. The key must match the model the repository is set to use — an OpenAI key cannot run an Anthropic model, so check the repository’s model setting alongside the key. Keys live in one of two places:
  • Pullfrog’s encrypted store (recommended) — set on the BYOK tab of the Billing card. Scope them to the whole account, where every repository inherits them, or to a single repository, which overrides the account-level one.
  • GitHub Actions secrets — the workflow already wires the standard variable names, so a secret with the right name is picked up automatically.
Full detail, including Bedrock, Vertex, Azure and OpenAI-compatible endpoints: BYOK.

​
Your existing subscription

If you already pay for a Claude Pro/Max plan or a ChatGPT plan, Pullfrog can run on it instead of buying API tokens. These are not API keys and there is nothing to paste. Each is one command, run from inside the repository:
Do not put a subscription credential into GitHub Actions secrets. The Codex credential is refreshed after every run, and a GitHub secret cannot be written at runtime, so it stops working about an hour later. Both belong in Pullfrog’s store, which is where the commands above put them.
A subscription has usage limits. When it hits one, runs fail until it resets — so if you want runs to keep going through a limit, add an API key as well and Pullfrog falls back to it. Full detail: Claude subscription, Codex subscription.

​
Troubleshooting a failing run

Work down this list; each item rules out the one below it.
  1. Does the repository have a model set that you hold credentials for? Check the repository’s model setting in the console. A repository pinned to an OpenAI model needs an OpenAI credential, whichever route you chose.
  2. On Router: is the balance above zero and is a card on file? An empty balance stops runs even when everything else is correct.
  3. On BYOK: does a key exist at the account, the repository, or in GitHub Actions secrets? A secret referenced by the workflow but never created resolves to an empty value rather than an error.
  4. On a subscription: has it hit its usage limit? Runs resume when it resets, and an API key added alongside it prevents the gap.
  5. Is GitHub itself refusing to start the job? If the run has no steps at all, the problem is your GitHub Actions billing rather than Pullfrog — check your GitHub account’s billing settings.