AI keys
The screen for adding your own provider keys to a portfolio. Provider, model, scope, thinking permission, testing a key and the order keys are tried in.
Last updated: September 30, 2026
Loading…
The screen for adding your own provider keys to a portfolio. Provider, model, scope, thinking permission, testing a key and the order keys are tried in.
Last updated: September 30, 2026
The AI steps in your flows run on the panel's credit — but if you have your own provider account you can use that instead. Portfolio settings › AI keys is where you add your own keys to the portfolio, pick which accounts use them, and test that they work. This guide covers every field on the screen.
⚠️ Owner and Admin can open this screen, and it only appears in the menu if using your own key is enabled on your plan. If it isn't, the item isn't drawn and typing the address won't open it. For why you might want your own key, see Using your own AI key.
A key is a secret string that proves an AI provider account (Claude, Gemini, ChatGPT) is yours. Give the panel a key and the AI steps in your flows run on it — and you pay the provider.
Keys belong to the portfolio. A key can be used across every account in the portfolio, or only on the accounts you pick.
A two-sentence info box sits at the very top and says two things.
First, what happens with no key: "If no suitable key exists for an account, system credit is used." So adding your own key is not compulsory; without one, your plan's AI credit is spent.
Second, that your own key may carry a service charge: if your plan defines one, the box says "With your own key, n credits are charged as a service fee per reply." That is the panel's own share on top of what you pay the provider, and the number is printed in the box. If no fee is defined, that sentence doesn't appear at all.
The small line under the box explains the order keys are tried in: "In an account, keys picked for that account are tried first, then keys open to all accounts. If a key is out of quota, the next one is used."
That sentence is the most important rule on the screen. Broken out:
So adding an account-specific key does not disable the general key on that account; it changes the order.
The Keys section lists your saved keys in the order they are tried. Each row shows:
With no keys the list reads "No keys yet." A long list paginates.
The section's description explains it: "Test connection sends a short request with the saved key; no credit is charged."
Press the button and the panel makes a small trial call to the provider. On success it says "The key works (model name)." On failure it names the reason: invalid key, quota exhausted, unknown model and so on.
Do this after adding every new key. Otherwise you find out the key doesn't work on the first real customer message.
Press Delete key and the panel asks for confirmation and states the consequence: "Delete …? Accounts using this key will fall back to the next key or to system credit."
So deleting doesn't break your flows; it changes the source. To know what the next one will be, look at the list first.
The section below the list. Its title is Add key when adding and Edit key when editing (while editing, the key you are on is named under the title).
The list where you pick which AI service the key belongs to. The panel supports three:
Pick a provider and a link appears beneath it, labelled You can get a key here:. It goes to that provider's key page; if you have no account there you create one.
The list where you pick which of the provider's models to use. Your options:
⚠️ A typo in a hand-typed model name makes the call fail with "unknown model". After typing one, always check it with Test connection.
An optional field. Fill it and the list shows this name instead of the provider's.
If you have several keys from the same provider, treat it as required: three rows all called "Claude (Anthropic)" won't help you. Names like "Claude — main", "Claude — backup", "Claude — test" make the list readable.
Paste the secret you got from the provider here. The hint underneath makes a promise: "Stored encrypted, never shown again."
The second half matters: after saving, you cannot read the key back in the panel. Keep it in your own records.
When editing an existing key the hint changes: "Leave blank to keep the saved key. If you change the provider, enter a new key." So you don't have to retype the key to change the scope or the model; leave the field blank. But if you changed the provider, the old key is invalid there and you must enter a new one.
Where you set the key's scope. Two options:
If the portfolio has no accounts yet, it says "This portfolio has no accounts yet; you can pick once one is connected."
A checkbox that decides whether the key is Active or Off. Turning it off is different from deleting: the key stays saved but isn't used. If you suspect a key is causing trouble, turn it off and watch the result rather than deleting it.
A checkbox. Its hint: "Thinking models reason before writing the answer; the tokens they spend are billed to this key's owner. With it off, the model writes directly."
Two points to help you decide:
On models where thinking cannot be switched off, the panel drops the effort to its lowest — so the setting has an effect on every model.
Press Add key for a new one, or Save when editing. To abandon an edit, the Cancel button next to it returns you to the list.
An agency with three Instagram accounts sets it up like this:
The result: AI steps on A's account use the Claude key (account-specific is tried first). The other two use Gemini. If either key runs out of quota, the panel falls back to system credit.
Adding a key and never testing it. A badly pasted key surfaces on the first customer message.
Trying to read the key back in the panel after saving. It is never shown again.
Changing the provider and leaving the key blank. The old key is invalid on the new provider; you must enter a new one.
Adding an account-specific key and assuming the general key is disabled. The general key stays as a fallback; the order changes, the scope doesn't close.
Picking "Selected accounts" and ticking none. The key is used nowhere; the list reads "No account selected".
Turning thinking on for every key. Your token cost rises with no visible gain on short replies.
Adding three keys from the same provider without naming them. You won't be able to tell which is which in the list.
The difference between your own key and system credit, and how to choose, is in Using your own AI key. To see how much is being spent see Usage, and for your credit balance see Plan and invoices.
Was this article helpful?