Skip to main content
Every AI feature in Vector — drafting your product description, finding and researching competitors, rebuilding a comparison, running a monitoring check — runs on API keys you provide. Vector never bills you for the AI itself: your provider does, against your own account and limits. You don’t pick one provider and live with it. You add a key for each provider you want available, and every AI function is then routed to a provider and model of its own.

​
Add a provider key

Open Settings and go to the AI Providers tab. The counter on the tab shows how many providers already have a key. The first section, Provider API keys, holds one card per provider. Paste the key and click Save key. The card then shows a key set badge and a masked value — the last few characters only, so you can tell which key is stored without ever seeing it again in full. To change a key, type the new one into replace the key… and save it. Remove key deletes it from the workspace.
The AI Providers tab in Settings, showing a card per provider with a key set badge, a masked key, an Enabled switch, and Save key and Remove key buttons

Provider API keys: one card per provider, each with its own key and switch

A key belongs to the workspace you’re in, not to your account. Each workspace has its own key and its own AI spend, and only admins and owners can open workspace settings. See Workspaces.

​
Enable and disable a provider

Each card has an Enabled switch, and the providers are independent of one another — several can be enabled at the same time, and functions can be spread across them. Turning a provider off leaves its key stored, so switching it back on doesn’t mean pasting the key again.

​
Which model runs what

Adding keys doesn’t decide anything on its own. That happens in the second section of the same tab, Function routing: a table with a row per AI function, where each row carries its own provider and model. You can route every function at once with a preset, or set a row by hand — Choose a model per function covers the table in full.
A row of the Function routing table with the provider dropdown open, listing the default provider and each provider that has a key

Function routing: one row per AI function, each with its own provider and model

​
If something goes wrong

A function pointed at a provider with no key is flagged in the routing table and blocked when it runs. Vector never quietly re-routes it to a provider that does have a key: you’d have no way of knowing which model produced the result, or what it cost. The same applies to a model that has been retired, or one that can’t do what its function needs. If a key is saved but the provider rejects it, the message says so — “key rejected” is distinguished from “no key set”, without ever revealing the key itself.
Results have the same shape no matter which provider produced them. Competitor lists, research summaries and every other AI output follow the same structure, so routing a function elsewhere changes its cost and quality, not the format of what comes back.

​
Where to go next

Set up your workspace

Where this step sits in the setup checklist.

Describe your product

The next step, and the one this key unlocks.