Data and privacy
How the Ertas AI Gateway handles the data you send: 30-day retention for abuse detection, training only with your opt-in, and test data only during the beta.
On this page
This page summarises how we handle what you send to the gateway. The Terms and Conditions (section 6.1) and the Privacy Policy are the binding versions.
What we keep
- Inputs and responses are kept for 30 days to detect and investigate abuse and breaches of the Terms, then deleted. We keep specific data longer only to investigate a suspected breach or security incident, or where the law requires it.
- Usage records with no prompt or response content (token counts, timestamps, model IDs, request IDs and error codes) support billing, security and support.
- Short-lived caching: parts of a request may be held briefly in memory to serve it, including for prompt caching.
Training
We train on your API data only if you opt in, and then only to train models for you. You can withdraw your opt-in at any time.
Medical data during the beta
MedErtas models are research models, not medical devices. Validate them before any use with patients.
Where requests run
Models run on GPU capacity in regions around the world. The Privacy Policy lists our sub-processors and where they operate.