All guides
PrivacyProduction·Beginner5 min

Privacy & Zero Data Retention (ZDR)

Zero Data Retention asks OpenRouter to route a request only to endpoints with a zero-data-retention policy — providers that, by OpenRouter's record, store nothing from it — and skips every other. Toggle it per request, or set it as an account-wide default in Settings. It covers single optimizations, not batch runs, and it governs the upstream provider, not what VantagePrompt stores in your own history.

Zero Data Retention sends a prompt only to model endpoints that store nothing from it. Here's what the toggle does and does not do, the per-request switch, the account-wide default, and when to reach for it.

By Andrei Bădulescu, Founder at VantagePrompt·Updated

Some prompts carry things you don't want a third party to keep: an unreleased product spec, internal financials, a customer's PII pasted into a support draft. Zero Data Retention (ZDR) is how you tell VantagePrompt to send that request only to model endpoints that store nothing from it. This guide covers what the toggle does and does not do, the per-request switch, the account-wide default, and when reaching for it is worth it.

What does Zero Data Retention actually mean?

VantagePrompt sends its model requests through OpenRouter, which by default may route them to providers that store user data and may train on it. With ZDR on, VantagePrompt turns on OpenRouter's zero-data-retention filter, which routes only to endpoints with a ZDR policy — in OpenRouter's words, a provider that will not store your data for any period of time — and also excludes providers that collect user data. The safety screening every prompt goes through uses the same filter, and automatic model selection picks only from models that have such an endpoint. If the model you chose has none, the run stops with a message and your credits come back. The guarantee rests on OpenRouter's record of each provider's policy, which it updates when a provider's policy changes.

ZDR is about the upstream model provider's data policy. It does not change what VantagePrompt itself stores in your history. To stop VantagePrompt from keeping your raw prompt text, use the separate 'log prompts' privacy flag in Settings.

How does the per-request ZDR toggle work?

Every optimization has a ZDR control. Flip it on for the runs that carry sensitive input, leave it off for throwaway prompts. Turning it on adds the zero-retention filter to that request, and VantagePrompt only dispatches it to endpoints that pass. This is the right granularity for most people: you are not locked into one mode, you decide per prompt. Batch runs and the optional deep analysis do not use the toggle.

Concretely, picture two prompts from the same session:

Prompt A (ZDR off):
  "Rewrite this tweet to sound more confident: ..."
  -> public marketing copy, nothing sensitive. Route normally.

Prompt B (ZDR on):
  "Draft an internal memo about the Q3 layoff plan, headcount
   numbers attached below: ..."
  -> confidential. Toggle ZDR so it only reaches
     endpoints that store nothing.

How do I make ZDR the default for my account?

If almost everything you run is sensitive, you don't want to remember the toggle every time. In Settings, under your profile privacy options, there is a ZDR default. When you turn it on, every optimization is forced to ZDR automatically, so the per-request toggle no longer has to be set by hand. It is a one-time switch that makes the filter the baseline for every optimization you run.

Set the ZDR default ON if you work with regulated, proprietary, or customer data as a rule. Leave it OFF and use the per-request toggle if only a minority of your prompts are sensitive. The account default always wins, so an enabled default keeps ZDR on even when the per-request switch looks off.

When should I use ZDR, and when can I skip it?

Turn ZDR on whenever it would matter that an outside company retained this exact text — confidential, personal, or contractually restricted input. Skip it for public-facing copy, where the wider provider pool is worth more than the filter.

InputZDRWhy
Confidential or proprietary — internal docs, unreleased plans, source you cannot expose, financials.OnA third party retaining this text is the risk you are managing.
Personal or regulated — anything resembling PII, health, or customer records.OnRetention by an outside processor is a compliance problem, not just a preference.
Contractually restricted — a policy says no third party may keep it.OnThe run only reaches endpoints that OpenRouter lists with a zero-retention policy.
Public-facing copy, generic brainstorming, sample prompts.OffRetention is harmless, and you keep the wider provider pool.
ZDR narrows the eligible provider pool, so switch it on where the input warrants it.

One practical trade-off: ZDR narrows the pool of eligible providers to endpoints with a zero-retention policy. That can mean fewer routing options for a given model, and a few models have none at all. If a run needs to reach the widest set of providers, weigh that against the sensitivity of the input. For most sensitive work the filter is worth the smaller pool. For routing strategy, fallbacks, and provider lists, see the model selection guide.

What is the quickest way to decide?

Ask one question before you hit optimize: would it be a problem if this exact text were logged by an outside company? If yes, ZDR on. If you'd answer yes most of the time, set the account default and stop thinking about it.

Frequently asked questions

What does Zero Data Retention actually guarantee?
That the request, and the safety screening of your prompt, go only to endpoints that by OpenRouter's record store nothing from it; every other endpoint is skipped rather than silently used. If the chosen model has no such endpoint, the run stops and the credits come back. Batch runs and the optional deep analysis do not use it.
Does ZDR stop VantagePrompt from storing my prompt?
No. ZDR is about the upstream provider's data policy. To stop VantagePrompt itself from keeping your raw prompt text in history, use the separate 'log prompts' privacy flag in Settings.
Should I use the per-request toggle or the account default?
Set the account default ON if you work with regulated, proprietary, or customer data as a rule. Leave it OFF and use the per-request toggle if only a minority of your prompts are sensitive. The account default always wins, so an enabled default keeps ZDR on even when the per-request switch looks off.
What is the downside of always leaving ZDR on?
It narrows the pool to endpoints with a zero-retention policy, which can mean fewer routing options for a given model, and a few models have no such endpoint at all. For most sensitive work the filter is worth the smaller pool.
What is the quickest way to decide?
Ask one question before you hit optimize: would it be a problem if this exact text were logged by an outside company? If yes, ZDR on. If the answer is usually yes, set the account default and stop thinking about it.

Sources

Put it into practice.

Run this technique in the optimizer.

Open the optimizer

Keep reading