An integration is a first-party connection between your agent and a system you already use outside the Platform — a CRM, a spreadsheet, a store, a payment method. Unlike connecting “whatever you want” with no control, every integration exposes only a curated, concrete set of actions: never free-form access to the provider.
What your agent can actually do with a given integration always goes through four levels of control, in this order:
| Level | What it decides | Who controls it |
|---|---|---|
| 1. Existence | Whether the integration is available for your account and whether you’ve connected it (sections 2-3). | Your provider (availability by vertical) and you (connect or not). |
| 2. Agent | Which of your account’s agents can see this integration as a resource. | You, via tags (Manual 23 · AI Agents). |
| 3. Audience | Whether a specific action or tool is Internal (only your team) or Both (also your end customers) — Internal by default. | You, from “Agent control” (Manual 52). |
| 4. Permissions | Whether the person or key currently using the Platform has permission to manage integrations. | Your provider (roles) and you (API keys, Manual 50). |
Note — Nothing is on by default
Connecting an integration never activates anything on its own: every panel action and every agent tool is born off — you turn them on, one by one, as you need them.