Retention windows for conversation records, model-call traces and audit logs are now documented with their configuration variables and defaults
HANGZHOU, China, Sept. 4, 2026 /PRNewswire/ — FastGPT, an open-source AI application platform for organizations, has consolidated in its public documentation how four classes of data are retained and removed: conversation records, knowledge base files, model-call traces and audit logs. The items were previously spread across a privacy policy, per-version upgrade notes and API reference pages.
For the cloud service, the privacy policy states that data deletion performed by a user is a physical deletion and is not recoverable, and that any non-physical deletion would be indicated in the service. The same policy states that user data is not kept as additional backup copies and is not used for model training. That page records its own last update as March 3, 2024.
Model-call traces used for short-term debugging are kept for six hours by default, adjustable through LLM_REQUEST_TRACKING_RETENTION_HOURS. Suspended agent sandboxes are archived after a period of inactivity set by AGENT_SANDBOX_ARCHIVE_INACTIVE_DAYS, with a default of seven days. Audit logs moved in the opposite direction in v4.16.0: on expiry they are transferred to cold archive storage rather than deleted, since traceability, not prompt removal, is what that class of data is kept for.
Three boundaries are documented alongside the defaults. The API endpoint that clears conversations affects only conversations created through an API key, and does not clear those from web use or shared links. Automatic cleanup depends on background tasks that can fail; two such defects were fixed in earlier releases, so a request to delete and a completed deletion should be verified separately. For community self-hosting and commercial private deployment, retention and cleanup are governed by the deploying organization, and the environment variables provide adjustable controls rather than a compliance conclusion.
About FastGPT
FastGPT is an open-source AI application platform offering RAG knowledge bases, visual workflows, agent orchestration, Skill, MCP and multi-channel publishing, available as a cloud service, community self-hosted, or commercial private deployment. As of Sept. 3, 2026, the GitHub repository labring/FastGPT has 29,551 stars and 7,297 forks across 275 releases, with v4.16.2 published on Sept. 3, 2026. Repository: github.com/labring/FastGPT
This content was prepared by our news partner, Cision PR Newswire. The opinions and the content published on this page are the author’s own and do not necessarily reflect the views of Siam News Network
