Zero Data Retention

Your prompts. Never stored. ZDR is an enterprise-grade privacy feature: when it's on, every request routes only through upstream providers that retain nothing - no prompts, no completions, no training. Turn it on with one flag.

How it works.

Enterprise-grade, on every request: ZDR enforces zero data retention and no prompt training with the upstream provider that serves your request.
One flag to enable: Run the CLI with CMD_ZDR=1, or send the x-cmd-zdr: 1 header on the Provider API.
ZDR-capable upstreams only: Requests route only to ZDR-capable providers with available capacity. If none are available, the request fails instead of falling back to a provider that retains data.
Pass-through pricing: ZDR capacity is priced per provider. We pass on exactly what we're charged. No markup.
Works on GOAT and Pro: Both plans allow ZDR usage from their standard plan credits. Details below.

Enable ZDR in Command Code

In the Command Code CLI, set the environment variable:

CMD_ZDR=1 cmd

On the Provider API, send the header on any request:

curl https://api.commandcode.ai/provider/v1/chat/completions \
  -H "Authorization: Bearer <CMD_API_KEY>" \
  -H "Content-Type: application/json" \
  -H "x-cmd-zdr: 1" \
  -d '{
    "model": "deepseek/deepseek-v4-flash",
    "messages": [{"role": "user", "content": "Write a haiku about race conditions."}]
  }'

That's it. Every request in that session is ZDR-enforced.

Note

99% of our models have ZDR-capable upstreams, and most use ZDR by default even without the flag. New models can take time to gain coverage as provider agreements are renewed monthly. If ZDR is enforced and no ZDR-capable upstream is available, the request fails with a 422 (cmd_zdr_no_providers) instead of routing through a non-ZDR provider. We’re adding coverage for the remaining models.

Routing

A ZDR request routes only to providers that support ZDR for the requested model and have capacity at the time of the request. The provider can vary from one request to the next.

This means:

The upstream may change: The provider serving your ZDR request may differ from the one serving the same model without ZDR.
The price follows the upstream: Each ZDR provider sets its own rates. We pass through the provider’s rates with no markup.

Why ZDR costs more

ZDR capacity is more limited than regular capacity, so providers often charge more for it.

A ZDR request may also route to a more expensive provider. Because of that, ZDR requests usually cost more than regular requests.

The Usage page in Studio shows the actual price charged per request, ZDR or not, as it happens.

ZDR on GOAT and Pro

Both GOAT and Pro plan usage is allowed under ZDR. No separate tier, nothing extra to buy:

Same credit pool: ZDR requests draw from the same monthly credits as any other request. There's no separate ZDR balance.
Metered at the default allowance: ZDR requests use your plan’s default allowance, even if a model normally has a higher boosted allowance. That means $20 on GOAT, and $30 for open-source models or $20 for premium models on Pro.
Higher per-request cost: Between default-allowance metering and routing that can land on a pricier upstream, the same credits buy fewer ZDR requests than regular ones.
Most models already covered: Most models on both plans use ZDR by default. The flag makes it strict: use ZDR or fail, with no fallback.

Need ZDR at scale? Let's talk!

High-volume ZDR usage can get expensive at pass-through rates. For Team and Enterprise plans, we can negotiate lower rates with upstream ZDR providers based on committed usage. Those plans use the rates in your contract.

Email support@commandcode.ai to discuss Team and Enterprise ZDR terms, or to request ZDR coverage on a specific model that doesn't have it today.

FAQs

What exactly does ZDR guarantee?

With ZDR enforced, your request is served only by upstream providers with zero-data-retention and no-training agreements: prompts and completions are not stored and are never used for training. If no ZDR-capable upstream exists for a model, the request fails rather than silently falling back to a provider that retains data. See our Privacy Policy for the full picture.

Why does a ZDR request cost more than a regular one?

ZDR capacity is priced per provider, and a ZDR request routes to whichever ZDR-capable upstream can serve it at that moment - we can't predict which one. We pass through exactly what that provider charges us, with no markup. Since that upstream isn't always the cheapest one serving the model, ZDR usage generally costs more. The Usage page shows the real per-request price.

Which plans can use ZDR?

ZDR enforcement via CMD_ZDR=1 works across all plans, and both GOAT and Pro plan usage is allowed under ZDR from standard plan credits, metered at the plan's default allowance rather than boosted per-model allowances. On the Provider API, send x-cmd-zdr: 1 on any request. For committed ZDR volume with contracted rates, email support@commandcode.ai about Enterprise plans.

What happens if a model has no ZDR-capable upstream?

The request fails - in the Provider API with a 422 (cmd_zdr_no_providers) - rather than route through a provider that retains data. Only a small handful of models are affected, and we're actively adding ZDR coverage to them. Need ZDR on a specific model that doesn't have it today? Email support@commandcode.ai.

Can I get better ZDR pricing?

Yes - at committed volume. If you need ZDR usage for your team or organization, contact us and we'll negotiate rates with our upstream ZDR providers on your behalf. Committed team and enterprise plans bill the rates in their contract instead of pass-through spot pricing. Email support@commandcode.ai to get started.

Resources