AI steps in the panel are fed from one of two sources: system credit or your own key. System credit comes with your plan and works without you doing anything; your own key is your provider account, and you pay the provider directly. This article explains the difference and which one is right for you.
The two sources
System credit is the AI allowance included in your plan. When a flow reaches an AI step, the panel makes the call from its own provider account and deducts the usage from you as credit. No setup is needed: you connect your account, build the flow, and it works.
Your own key is a secret string from your Claude, Gemini or ChatGPT account. Add it to the portfolio and the AI steps run on it, with the provider billing you for the tokens. The panel is only the intermediary in that case.
💡 This is not an either/or choice. If your own key runs out of quota or fails, the panel quietly falls back to system credit. Your own key is a priority, not a replacement.
Which one, when
System credit is better when:
You use AI lightly. Opening a separate provider account and tracking another invoice is not worth it to classify a few hundred messages a month.
You are starting out. Until your flows are built and tested you don't need to think about the source.
You want one invoice. Everything stays inside your plan: no separate payment, no separate currency, no separate tax record.
You don't want to deal with technical detail. There is no model to pick, no quota to watch, no key to rotate.
Your own key is better when:
You use AI heavily. At volume, a provider's own price can come out cheaper than the credit equivalent.
You want a specific model. With your own key you can type a model name by hand, including one the provider just shipped that isn't in the list.
You already have a provider account. If your company pays a Claude or OpenAI invoice for other work, using it here simplifies your bookkeeping.
You want to track usage on your side. You get the token breakdown in the provider's own dashboard.
You want to let the model think. Thinking spends tokens and is billed to the key's owner, so that decision is yours with your own key.
The service fee
Using your own key can carry a charge on the panel's side. If your plan defines one, a few credits are deducted as a service fee per reply, and the number is printed in the info box at the top of the AI keys screen.
That corrects the expectation "with my own key I spend no credit at all". There are two line items:
The provider's token charge → you pay it, directly to the provider.
The panel's service fee → deducted from your credit.
If no fee is defined, that sentence never appears on the screen and your own key spends no credit.
Numbers to look at before deciding
Looking at figures beats guessing. Two screens provide them:
Portfolio settings › Usage. Shows this portfolio's AI usage for the period in two totals: System credit and Own keys. Each row names the account, the model, the tokens and the credits spent. Read your monthly volume there. See Usage.
My user settings › Plan and invoices › Credits. Shows your credit pool across all your portfolios and this month's consumption. If credit keeps running out, it may be time to consider your own key. See Plan and invoices.
A concrete route to the decision
A shop owner decides whether to switch to their own key like this:
Runs on system credit for a month. They build their flows and run them with real customers.
Looks at the Usage screen at the end of the month. The "System credit" row shows 1,800 credits over 900 calls. That tells them their volume.
Checks the credit balance. Their plan gives 2,000 credits a month, so they are near the ceiling and will have to buy extra credit if they grow.
Opens the provider's pricing page and works out the token cost of 900 calls.
Adds the panel's service fee. The info box at the top of the AI keys screen states how many credits are charged per reply; they multiply that by 900.
Compares the two totals, sees their own key comes out cheaper, and adds it.
The point of this route: the decision is made on a month of real usage, not on a guess. If your volume is low, system credit is almost always more practical; if it is high, your own key is both cheaper and frees your hand on model choice.
The scope decision
When adding your own key you also pick a scope: the portfolio's all accounts or selected accounts.
For an agency that choice enables per-client billing: you add each client's provider key specifically to that client's account, and the token cost lands on that client's invoice. For a single business, All accounts is usually the right pick.
Limits and known situations
Using your own key requires use your own key to be enabled on your plan. If it isn't, the screen doesn't appear in the menu at all.
A key belongs to the portfolio, not to an account. Move an account to another portfolio and keys picked for that account do not move; the target portfolio's keys or system credit are used instead. See Moving an account to another portfolio.
Once saved, a key is never shown again in the panel; keep it in your own records.
If a key runs out of quota or becomes invalid, the panel moves to the next key, or to system credit if there is none. Your flow doesn't stop — but if your provider invoice comes in lower than expected, that may be why.
Not every provider generates images. A text model reads images; generating them needs a different model family, and the panel uses the provider's own image model for that.
Picking the right source decides both the cost and the flexibility of your AI steps; revisit the decision as your volume changes.