Which metadata is stored, and how content logging and retention work.
Planned. By default Parezen is designed to store metadata: organisation, project, environment, key id, route, task, agent, workflow, trace, model, token counts, cost, latency, status and which levers ran. That would be enough for the ledger. The hosted service does not store this. Sample screens show fixture rows.
Planned. Prompt and response content are designed to stay off. Full prompt capture is a paid extra on Inference in the plan and stays off unless you turn it on. The change is designed to be written to the audit log. Observability metadata and OTLP export are part of the Inference design. Pages that would show a prompt show a locked state instead when logging is off. One sample project has logging on so the switch can be demonstrated. The others do not.
Replay samples would stay with the gateway that ran them. On a self-hosted gateway that would be your database. This website does not run that gateway, so it does not keep replay audio, images or video. A retention period for a hosted gateway is not published. Human review of a sample would happen only after a written opt-in, and voice or audio would also need its own opt-in. That review is not running. Self-host keeps the ledger in your Postgres. Do not treat this page as a data-processing agreement.
Questions go to [email protected]. The public policy is the privacy policy.
