Zero data retention

Fireworks has Zero Data Retention by default. Specifically, this means

  • Fireworks does not log or store prompt or generation data for any open models, without explicit user opt-in.
    • More technically: prompt and generation data exist only in volatile memory for the duration of the request. If prompt caching is active, some prompt data (and associated KV caches) can be stored in volatile memory for several minutes. In either case, prompt and generation data are not logged into any persistent storage.
  • Fireworks logs metadata (e.g. number of tokens in a request) as required to deliver the service.
  • Users can explicitly opt-in to log prompt and generation data for certain advanced features (e.g. FireOptimizer).
  • For proprietary Fireworks models (e.g. f1, FireFunction), prompt and generation data may be logged to enable bulk analytics to improve the model.
    • In this case, the model description will contain an explicit message about logging.

Response API data retention

For the Response API specifically, Fireworks retains conversation data with the following policy because the Response API defaults to store=True:

  • Default behavior: Conversation history is stored by default (store=True) for 30 days, then automatically purged to comply with GDPR requirements.
  • Opt-out option: You can disable data storage by setting store=False in your API requests to prevent any conversation data from being retained.
  • Retention period: All stored conversation data is automatically deleted after 30 days.

This retention policy is designed to be consistent with the OpenAI API while providing users control over their data storage preferences.

The Response API retention policy only applies to conversation data when using the Response API endpoints. All other Fireworks services follow the zero data retention policy described above.