Web chat stores messages to provide conversation history.
Signed-in chat needs conversation content to render and continue prior threads. Users should avoid entering secrets or regulated data unless their organization has approved the service, account controls, and retention implications for that use.
Direct API prompts are not saved as web-chat conversations.
A direct request to the public chat-completions route is processed for inference but is not written into BrokenGPT conversation history. The application does record operational and metering metadata such as the public model alias, token counts, latency, key identifier, request status, workspace attribution, and request ID.
Minimize sensitive data before it enters any model workflow.
- Remove credentials, private keys, authentication cookies, and unnecessary personal data.
- Use server-side authorization before attaching internal documents or tool access.
- Limit logs to the identifiers and metrics needed for operations.
- Configure deletion and backup policies together so restored data does not violate expectations.
- Review external search, reader, or tool providers separately when those features are enabled.
Use the legal policy for commitments and this page for the data-flow map.
The system page explains the public product contract and data flow. The privacy policy and terms are the governing customer-facing documents. Enterprise or regulated deployments should confirm retention periods, regions, subprocessors, incident terms, backups, and deletion workflows before sending production data.