> ## Documentation Index
> Fetch the complete documentation index at: https://docs.prisminference.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Zero data retention
> How Prism handles API inputs and outputs without storing them.
Zero data retention (ZDR) is the default for every Prism inference request on
every service tier. You do not need to enable a setting or enter a separate
agreement.
Prism processes API content only for the time needed to complete and return a
request. It is discarded after processing and is never used for model
training.
## What ZDR covers
ZDR applies to the content sent to and returned from every Prism inference
endpoint:
| API content | Examples |
| ----------- | ----------------------------------------------------------------------------------------------------------- |
| Inputs | Prompts, messages, code, system and developer instructions, tool definitions and results, images, and files |
| Outputs | Model-generated text, code, reasoning, tool calls, and other inference results |
Prism does not write this content to persistent storage, application logs,
analytics systems, or backups. It is not used to train, fine-tune, evaluate, or
otherwise improve models.
This applies to Chat Completions, OpenAI Responses, Anthropic Messages, and
token-count requests.
## What Prism retains
ZDR covers API inputs and outputs. It does not cover the non-content information
needed to operate the service:
| Information | Examples |
| -------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| Service metadata | Account or organization ID, API-key ID, endpoint, model, request time, token counts, response status, latency, rate-limit information, and source IP address |
| Account information | Name, email address, authentication identifiers, organization membership, plan, and API-key records |
| Billing information | Credit balance, transaction amount, payment status, and invoice history |
| Website and dashboard data | Pages viewed, features used, cookie or device identifiers, and browser details |
Service metadata is designed not to contain the substance of API inputs or
outputs. Prism retains non-content information only as described in the
[Privacy Policy](https://prisminference.com/privacy).
If you voluntarily copy API content into a support request, that copy becomes
part of the support communication and may be retained with the support record.
## Build stateless applications
Because Prism does not retain API content:
* Keep conversation history and application state in your own system.
* Resend the required context with each request.
* Do not rely on Prism to retrieve or export past inputs or outputs.
* Use `store: false` with the Responses API. Stored Responses and
`previous_response_id` are not supported.
The API still returns token usage and request status in each response so your
application can observe the current request without Prism storing its content.
Read the complete policy for API content, metadata, accounts, and billing.
Learn how to use Prism's stateless Responses-compatible endpoint.