Fireworks AI
Fireworks says it does not use prompts, training data, or API inputs to train or improve its AI models without explicit opt-in, and inference has zero data retention by default.
Customer content is not used to train Fireworks-owned or shared models by default. Training datasets are used solely to fine-tune the customer's models. Explicit opt-in can change logging and improvement treatment.
No training opt-out is required for the stated baseline. Users can keep inference logging off, set Response API store=false, and delete training datasets or artifacts when no longer needed.
By default, prompts and generations for open models exist only in volatile memory during the request. Bring Your Own Bucket can let managed training read a dataset in place without copying it onto Fireworks-managed storage.
Default inference has no persistent prompt or generation storage. Response API conversations with store=true are retained for 30 days unless deleted; store=false prevents that storage. Training artifacts have feature-specific retention.
Fireworks says zero retention means it will not log content for human review except where required by law or to provide service and support.
Foundational reference pages for teams adopting, deploying, and operating production AI with inspectable controls.