Personal Google AI Ultra: what exactly does "Enable Telemetry = off" stop? (training, retention, human review)

Hello,

I use the Antigravity CLI (agy 1.4.3, Linux) with a personal Google AI Ultra subscription. I would like to use it for professional work, so I need a clear understanding of how my data is handled. I have read the Additional Terms of Service and several earlier threads on this topic (for example 125236, 121602 and 132828), but these points still seem to be unanswered.

My current settings: “Enable Telemetry” is off in /config, and I unchecked the data-sharing option during first-run setup.

  1. Training: The terms say Interactions are used to “evaluate, develop, and improve … machine learning technologies” and that I can “navigate to settings to change your preference”. Is “Enable Telemetry” the setting this refers to? With it off, are the prompts, code and model responses that are sent for inference excluded from model training and product improvement?

  2. Retention: With telemetry off, are those prompts and responses stored on Google’s servers after the response is returned? If so, for how long, and for what purpose?

  3. Human review: With telemetry off, can Google employees or contractors still access or review my prompts and responses? If yes, only in limited cases (for example abuse or safety investigations), or more generally?

  4. Contractual guarantee: Is there a way for an individual (not a Google Workspace organization) to get a documented no-training commitment? For example, if I sign in with my own Google Cloud
    project via AGY_ADC_AUTH, do the Google Cloud terms apply instead of the consumer terms? And in that case, is the Google AI Ultra quota still used, or is it billed separately?

  5. Deletion: When I email antigravity-support@google.com to request deletion of my Interactions, will I receive a confirmation when the deletion is complete?

An official answer, or a link to documentation that covers these points, would be greatly appreciated.

Thank you.

Hi @KEITARO_MARUNO,

Here is how each of those works with your current setup:

  • Training: Yes, Enable Telemetry in /config (along with the first-run data-sharing checkbox) is the exact setting referenced in Section 5 of the Antigravity Additional Terms. When it is off, your prompts, code context, tool outputs, and model responses are not collected or used to train models or improve Google products.
  • Retention: With telemetry off, persistent interaction logging is disabled—your prompts and responses are only processed in-memory to serve the request. Because a personal account falls under the standard Google Privacy Policy and Generative AI Additional Terms, routine operational metadata (like quota accounting) and short-term safety/abuse logs may still be retained briefly for security and abuse prevention.
  • Human review: Turning telemetry off also opts you out of human review for model training and product evaluation. Your prompts and responses aren’t reviewed by employees or contractors except in narrow legal, security, or abuse-investigation cases covered by the Privacy Policy.
  • Contractual commitment (AGY_ADC_AUTH) & Ultra quota: Authenticating with AGY_ADC_AUTH=true against a Google Cloud project switches your usage to the Google Cloud Terms of Service and Cloud Privacy Notice, which include a contractual no-training commitment. However, it does not use your personal Google AI Ultra quota (Ultra is tied to consumer Google One sign-in). Using AGY_ADC_AUTH requires a Gemini Enterprise license or Google Cloud billing setup on that project. Alternatively, as an individual without Workspace, you can link a Cloud Billing account in Google AI Studio and use a paid GEMINI_API_KEY, which is governed by the Gemini API Paid Services terms (no training on your data, billed pay-as-you-go).
  • Deletion confirmation: Yes. If you email antigravity-support@google.com to request deletion under Section 3 of the terms, the support team will reply to confirm once it is completed (though since you unchecked data sharing at setup and kept telemetry off, no interaction logs were stored for training in the first place).

Hope this helps clarify things!