Skip to content

Security and privacy

  • Your agent signs in with OAuth 2.1, in your browser. AdsGPT never sees your agent’s credentials, and you never copy an API key.
  • An authorization works on one AdsGPT organization, the one you choose when you sign in, with your role’s permissions.
  • It has two scopes: ads:read, to read, and ads:write, which your agent asks for the first time it needs to change something.
  • AdsGPT’s server talks to each ad platform with the access your organization granted when it connected it. Your agent’s token never reaches an ad platform.

The calls your agent makes to AdsGPT’s tools, with their inputs: account IDs, report periods, new campaigns with their budget, targeting and ads, ad texts, image URLs, YouTube links and landing page URLs. It never receives your conversation.

In your AdsGPT organization, available to its members as their roles allow:

  • a log of every change made to your ad accounts, including the campaigns created: what changed, what it had before, the member and the agent it came from, and the tool call that made it.
  • The ad platforms you connected, such as Google Ads: the reports and changes your agent asks for, and the images the tools upload to your ad account.
  • The sites of the images and landing pages you give by URL: AdsGPT downloads the images to check and upload them, and opens each landing page a change sends to check that it works.

Your agent’s provider, such as Anthropic or OpenAI, handles what the tools return under its own privacy policy.

The AdsGPT plugin contains only text: the skills, a manifest for each client and the server’s address. It has no scripts, hooks or binaries.

Remove an agent’s access from AdsGPT at any time, or disconnect an ad platform from the Plugins screen. In your agent, you can also remove the AdsGPT server or plugin.