Connector

A connector is a permissioned bridge that lets an AI agent search, read, or sometimes act inside another app or data source, such as email, a calendar, a meeting recorder, or a document store.

A connector is a permissioned bridge that lets an AI agent search, read, or sometimes act inside another app or data source, such as email, a calendar, a meeting recorder, or a document store. You authorize it once through the provider's sign-in, the provider hands back a scoped key, and from then on the agent can reach that account within the limits you granted. Say you want an assistant to summarize this week's meetings. You add a connector to your calendar so the agent can list the events, and a second connector to your notes app so it can write the summary where you will read it. Each connector covers one source and carries only the access you approved, so the agent sees your calendar without touching your bank or your photos.

Builder example

Connectors decide what your agent can reach, so which ones you grant and at what access level shapes everything it can do. An assistant that drafts replies needs read access to your inbox, but giving it send access too means a bad draft can leave before you see it. Add connectors one source at a time, start each at read-only, and tell the agent to draft into a holding place you review rather than acting directly until you trust the results.

Common confusion: A connector grants access to a source; it does not decide what the agent is allowed to do with that access. The narrower permission scope you select during sign-in sets the limits, so a connector with a read-only scope can never send or delete even if the agent tries.