Hangzhou – September 08, 2026 -- FastGPT has consolidated its data retention and deletion policies for four categories of platform data -- conversation records, knowledge base files, model-call traces and audit logs -- into a single public reference, replacing information previously scattered across a privacy policy, version-specific upgrade notes and API documentation.
Cloud privacy policy confirms user-deleted data is physically erased and unrecoverable
FastGPT's cloud service privacy policy, last updated March 3, 2024, states that user-initiated deletion constitutes physical deletion with no recovery option, and that any exception to physical deletion would be flagged within the service itself. The policy also states that user data is not retained in additional backup copies and is not used for model training.
Debugging traces are purged in six hours by default; audit logs are archived instead of deleted
Model-call traces used for short-term debugging are retained for six hours by default, a window adjustable via the LLM_REQUEST_TRACKING_RETENTION_HOURS variable. Suspended agent sandboxes are archived after a period of inactivity governed by AGENT_SANDBOX_ARCHIVE_INACTIVE_DAYS, set at seven days by default. Audit log handling changed direction in version 4.16.0: rather than being deleted on expiry, audit logs are now moved to cold archive storage, reflecting their role in traceability rather than prompt data removal.
Three documented boundaries limit what automated deletion actually covers
The API endpoint used to clear conversations only affects sessions created through an API key, leaving conversations from web use or shared links untouched. Automatic cleanup relies on background tasks that can fail -- two such defects were fixed in earlier releases -- meaning a deletion request and a completed deletion require separate verification. For community self-hosted and commercial private deployments, retention and cleanup responsibility shifts to the deploying organization, with the environment variables functioning as adjustable controls rather than a compliance guarantee.
GitHub repository shows 29,551 stars and 275 releases as of September 3, 2026
The labring/FastGPT repository recorded 29,551 stars and 7,297 forks across 275 releases as of September 3, 2026, the same day version 4.16.2 was published.