Zero Data Retention
What Runware keeps and deletes when Zero Data Retention is enabled on your organization, which models it covers, and how to work with outputs that expire.
Introduction
Zero Data Retention (ZDR) is an organization-level option for enterprise accounts. When it is on, Runware processes your prompts and input media to run each request and keeps none of that content afterwards. Generated outputs stay available only until their TTL expires.
ZDR is enabled per organization by our team and applies from the moment it is turned on. Requests processed before that keep the retention they had.
What is retained
Only the data needed to bill and operate the service is kept. Everything you send is discarded, and everything a model generates is deleted once its ttl expires.
| Data | Retention |
|---|---|
| Prompts and text generations | Not retained. Processed in memory for the duration of the request. |
| Input media | Not retained. This covers input and reference images, masks, video frames and audio. |
| Generated media | Deleted when the ttl of the request expires, 60 seconds by default. |
| Technical metadata | Retained for billing, debugging, reliability monitoring and security auditing. |
| Billing records | Retained as required by tax and accounting law. |
| Account and authentication data | Retained for the life of the account plus any statutory retention period. |
For language models with prompt caching, part of a prompt and its KV cache can stay in memory between requests. That cache is never written to disk, is evicted automatically, and is discarded when the replica restarts.
Runware never uses data sent through the API to train or improve models, with or without ZDR.
Covered models
ZDR covers every model that runs on Runware's own infrastructure:
- Runware-hosted models of every modality, marked Runware Optimized in the model catalog.
- All Black Forest Labs models, both the open-weight FLUX models we host and the ones served through the Black Forest Labs API.
- Community and user-uploaded models, which run on the same infrastructure as the hosted ones.
Other models are served by their providers under the provider's own retention terms, and ZDR does not extend to them.
Working with outputs
Without ZDR, output URLs stay available for 7 days. With ZDR, the file behind an output URL is deleted when the ttl expires, which is 60 seconds unless the request sets its own. Download the result before then, or raise ttl on the request when your pipeline needs more time.
Asynchronous requests follow the same clock. getResponse keeps returning the finished task after the ttl expires, but the URL in it no longer resolves, so poll and download within the window. Webhooks deliver the result as soon as it is ready, and the same deadline applies to the URL they carry.
To skip the URL entirely, request base64Data or dataURI as the outputType. The output comes back inside the response and no file is stored, so there is nothing to fetch or expire.
What changes in the dashboard
The usage log keeps the metadata of each task, such as its UUID, model, date and cost. Request and response payloads are not stored, so the log can't show what you sent and reusing a past request is not available. The Playground still generates, but past generations are not saved.
Enabling ZDR
ZDR is enabled per organization on request.
Contact Sales to enable Zero Data Retention on your organization.