Give Your Agent Read-Only Access to a Third-Party System, Safely
How the Connectors page in Settings lets your agent read from third-party systems on a key you supply, using a reviewed catalogue instead of an arbitrary URL.
Giving an AI agent access to a third-party system sounds simple until you think about what "connect anything" actually means: a model deciding what URL to call, with your credential attached. That's a real risk category — an agent pointed at an arbitrary base URL with an arbitrary key is effectively a server-side-request-forgery setup with a credential handed to it. The Connectors page in Settings is built around avoiding exactly that.
How it's different from "connect anything"
Connectors doesn't let you type in any endpoint you want. It's a reviewed catalogue: every connector on the list has a base URL that's fixed in the platform's own code, and a fixed set of read-only endpoints it's allowed to call. You supply a key for a connector that's already on the list — you don't define new connectors or new endpoints yourself. That's a narrower design than "bring your own integration," and it's narrower on purpose.
What you actually do on this page
For each connector in the catalogue, you can add your own API key. Once it's saved, your agent can use that connector to read data from the corresponding system, scoped to whatever read-only endpoints that connector exposes. The catalogue tells you where to find or generate a key for each one — that guidance comes from whoever reviewed and added the connector, not from a generic help article.
The same trust rules as BYOK, on purpose
This page follows three rules copied deliberately from the AI provider key page: it never shows a stored key back to you — there's no endpoint that returns one. It doesn't invent connectors that the backend doesn't actually serve; the catalogue always reflects what's real. And it won't offer you a form it can't honor — if your deployment has no encryption configured, the inputs are disabled and the reason is shown up front, not discovered after you've pasted a key in.
An important caveat: this is not the same as the "Integrations" page
Settings has a separate page, also called Integrations, that is a static placeholder — a single card that says "Integration management coming soon" and does nothing else. It is not the same feature as Connectors, and it isn't wired to anything yet. If you're looking for the real, working way to give your agent access to a third-party system, Connectors is it — Integrations is unrelated and not functional today.
When to use this versus a provider key
If you're wondering whether something belongs on this page or on the AI Keys page instead, see the companion piece comparing the two — they serve genuinely different purposes even though they're built the same way.