Tools and scopes

A connected AI client can only do what you approved on the consent screen. Access comes in six scopes, and each scope unlocks a specific set of actions — nothing more.

​

The six scopes

Scopes are all read-only except Create & manage AI agents. Without that scope, a connected assistant cannot change anything in your workspace.

​

What the assistant can actually do

Each scope maps to a set of named actions — “tools” in AI-client language. You will occasionally see these names in your client’s activity log, so here is what each one is.

​

Workspace and identity

Available to every connection, no scope required.

​

Channels — requires Read channels

​

AI agents — requires Read AI agents

​

AI agents — requires Create & manage AI agents

An assistant creating an agent can set the same things you would set in the app, including the Effort level — the low, medium or high tier that decides how much reasoning the agent applies and how many credits a conversation costs. It cannot pick a provider or a model; the platform handles that. See Creating an agent.

​

Conversations — requires Read conversations

​

Billing — requires Read billing

​

Automations — requires Read automations

​

Things worth knowing

• Long lists arrive in pages. An assistant asking for hundreds of conversations receives them in batches and fetches the next page as needed. You may see it work through a large result set in steps.
• A missing scope is a clean refusal, not a failure. Calling something outside the granted scopes returns a “forbidden” answer, and a well-behaved assistant will tell you it lacks the access rather than retrying.
• Everything is limited to your own workspace. There is no way for a connected client to reach another customer’s data.
• Scopes cannot be edited after the fact. To change what a client may do, revoke it and connect again — see Connecting an AI client.

​

Next steps