Ambra
Compliance

Data and Compliance

How the Ambra inference API handles the data sent to it. In short: prompts and completions are processed in memory and never stored, and nothing is used for training.

Prompt and completion retention
None. Content is held in memory for the life of the request only.
Training on customer data
Never.
Request logging
Metadata only: timestamps, token counts, model, status and latency.
Processing region
North America.
Encryption in transit
TLS on every connection to the API.
Subprocessors
RunPod (compute), Cloudflare (network edge). Stripe for billing only, never request content.
SOC 2
No report yet.
Security and privacy contact
contact@ambracompute.com

Documents

Last updated 13 November 2026