Coding agents send whole repositories. Support bots send customer messages. ZDR answers the question that follows: what happens to that content after the answer comes back?
How it works
Under ZDR, the provider holds prompt and output content only in memory while serving the request. Afterward it is not written to:
- request or debug logs;
- abuse-monitoring stores;
- persistent caches or queues;
- training or evaluation datasets.
Records that are normally still kept, without content, include request IDs, timestamps, status codes, model, token counts, cost and client IP addresses.
ZDR is not a default everywhere. OpenAI’s API, for example, keeps abuse-monitoring logs for up to 30 days by default and offers ZDR to approved customers. It also says API data is not used for training unless you opt in. Read each provider’s terms rather than assuming.
Why it matters
- Compliance. Prompts often carry personal data, health details or trade secrets. Content that is never stored cannot leak from logs or backups.
- Fewer parties. Every system that keeps a copy is another place to secure. A router or reseller adds parties to the data path.
- Speech and images. Voice notes and scanned documents are among the most sensitive inputs. ZDR should cover audio and transcripts too.
ZDR protects what happens after your request arrives. You still need to protect your API keys, lock them to known IPs and redact what the model does not need.
On BroadNet Inference
BroadNet’s docs say ZDR is “built into how we serve requests, not only stated in policy.” BroadNet Inference does not retain prompts, images, audio, answers or transcripts on any request. It keeps content-free traffic and usage records, such as timings, token counts and cost, for as long as billing, tax, security and legal obligations require. An automated retention audit runs every hour.
Per BroadNet’s privacy statement, the models it hosts run on BroadNet-owned or leased hardware in the United States, with partner capacity as backup; partner-served models run under zero-data-retention terms with US inference partners. Keys can be locked to your servers with an IP allow list.
The full list of what is and is not stored is on the zero data retention page. To compare providers, use the 10-question ZDR checklist.