How authentication works
The connection uses OAuth 2.1 with PKCE. You sign in to Prominara in your browser, and your client receives a token scoped to your organization. There is no key to copy into a config file and nothing long-lived on your machine that cannot be revoked from the product.
Consent
Every new connection shows a consent page naming the client that is asking, the address it will send you back to, and the organization you are connecting. Nothing is approved silently from a previous grant, so a second client cannot inherit the first one’s access.
Scopes
| Scope | What it allows |
|---|---|
| read | Every read tool: sites, visibility, prompts, answers, missions, sources, measurement, suggestions, usage. |
| write | Create prompts, run a check, log a content change, set a mission status, enable or disable prompts. |
Write is a checkbox on the consent screen. Leave it unchecked for a connection that can read everything and change nothing. A write attempt on a read-only connection returns a permission error naming the scope it needs, so your client can ask you to re-authorize.
What no tool can do
- Delete anything
- Change billing or plans
- Invite, remove or edit team members
- Change organization or site settings
- Reach a site or prompt outside the organization you consented to
Quotas
Agent calls draw on the plan you already pay for. A check run by your agent counts exactly like pressing Run now in the app, against the same monthly allowance. Creating prompts consumes the same tracked-prompt slots. When a limit is reached the tool says which limit, what remains and when it resets rather than spending silently.
What we store about a connection
- The client name and the addresses it registered
- The scopes you granted and the organization you chose
- Hashed tokens, never the tokens themselves
- When the connection was created and when it was last used
Tokens are stored as hashes and never written to logs. The privacy policy covers this alongside the rest of the data we hold.
Revoking
Open Settings, then Connected agents. Each connection lists the client, the scopes, the organization, when it was created and when it was last used. Revoking takes effect on the next call the agent makes. Reconnecting later creates a fresh consent rather than reviving the old one.
Answers are untrusted content
Text an AI engine produced about your brand is third-party content. When a tool returns it, it is wrapped and labelled so your agent treats it as evidence to read, never as instructions to follow.