Scopes and permissions for tokens and assistants

Every API token and every connected assistant has scopes: the kinds of thing it may do. On top of that, it can never do more than your role on the account allows: a viewer who connects an assistant gets a read-only assistant, whatever it asked for.

The scopes

ScopeAllowsDefault
readLook at everything: sites, health, logs, deploys, backups, traffic, domains, DNS, email, invoices, the teamAlways
deployDeploy from git, set build steps, approve or roll back releases (with your confirmation)On
manageSite settings (environment variables, PHP or Node.js version), connect a domain, staging, cron, backups and restores, webhooksOn
dnsChange DNS records, import zones, enable DNSSEC (with your confirmation)On
dataRead-only SQL on site databases; write to warehouse databases; dashboards, pipelines and jobsOff: tick it on purpose
billingPropose orders for you to approve (how ordering works); nothing is bought with a tokenOff: tick it on purpose

Who can give which scope

RoleCan grant
OwnerAll of them
Admin, developerread, deploy, manage, dns, data
Billingread, billing
Viewerread
Site team memberread, deploy, manage, data, for their sites only

If someone's role changes, the tokens and assistants they connected change with it straight away.

Actions that ask first

A few actions cannot be undone, so an AI assistant must ask you twice: the first call only explains what will happen and returns a short-lived confirmation; the second call, with your yes, does it. (A script using a personal API token is not asked: you decided when you wrote it.) They are restoring a backup over the live site, enabling DNSSEC, dropping a warehouse database, approving a release and rolling back. Restores started from an assistant are finished by you in the portal.

Things only you can do, in the portal

Deleting a site, a domain or a DNS zone; changing a domain's nameservers; creating API tokens; changing the team; and paying. An assistant will send you a link for these.