Skip to main content

Security

Read-only by default. Human approval always.

What Crescade can access, what the AI model sees, where your data lives and how to take it with you. Questions: hello@crescade.com.

Read-only Google measurement access

For Analytics and Search Console, Crescade requests only analytics.readonly and webmasters.readonly, plus your Google identity, and tagmanager.readonly only if you turn on Tag audit. That connection cannot change tracking, tags, properties or Search Console settings. Disconnecting deletes the stored tokens and asks Google to revoke the grant.

Google Ads changes only after two approvals

Google Ads is a separate, optional connection. Google's only Ads permission allows changes, so Crescade limits itself: it creates a Search campaign paused after an owner or admin approves it, turns it on only after a second approval, keeps it within the budget and cap you set, and never edits campaigns it did not create. Anyone in the workspace can pause at any time.

Tokens encrypted at rest

Google access and refresh tokens are encrypted with AES-256-GCM before they are stored. Exports never include connector secrets.

Nothing ships without a person

Crescade publishes to your website or social accounts, or changes your ad account, only after a workspace owner or admin approves exactly what will be sent, and only when publishing is switched on for your workspace. It never changes your analytics settings. Every published post keeps a receipt and can be undone: taken off your site, or deleted from LinkedIn or X.

Your Google data stays out of AI models

Connected Google Analytics, Search Console, Tag Manager and Google Ads data is never sent to an AI provider. Measurement reports are computed without a model.

What the AI model sees

Sent to the model

  • Public text captured from your website
  • Sources saved in your workspace
  • Company context you confirmed

Never sent

  • Google Analytics, Search Console, Tag Manager or Google Ads data
  • Passwords, tokens or payment details
  • Anything from another customer's workspace

Every agent run is recorded in your workspace with the model used and its credit cost. Website text is treated as untrusted evidence: instructions found on a page are ignored by the agents.

Infrastructure and accounts

  • HTTPS everywhere with HSTS; the database is reachable only from the application network, not from the internet.
  • Accounts require a verified email or Google or Apple sign-in. Repeated failed sign-ins lock the account temporarily.
  • Public forms and the free scan are rate limited, and the scanner refuses private and internal network addresses.
  • Workspace roles (owner, admin, member, viewer) are checked on the server for every change.

Your data, your call

  • Export a JSON copy of your workspace from Account settings at any time.
  • Delete your account from the same page: you are signed out, connections are revoked and running work stops immediately; customer content is erased after 30 days.
  • Anonymised billing and security records are kept only for the retention period the law requires.

Service providers

ProviderPurposeLocation
OVHcloudApplication server and databaseUnited Kingdom
CloudflareDNS and bot protection on public formsGlobal network
DeepSeekAI model for agent tasks and the scan's first-impression readChina
GoogleAnalytics, Search Console, Tag Manager and Google Ads data you connect; site analytics only after consentGlobal
ResendAccount and task emailsUnited States
PostHogProduct usage events, sent from our serversUnited States
SentryError monitoring, with personal data strippedUnited States

Report a vulnerability to hello@crescade.com. See also the privacy policy.