Open-Source Enterprise AI Platform FastGPT Releases v4.16.0 with User-Level Code Sandboxes and Knowledge Base Metadata

Audit logs move from scheduled deletion to cold archive in object storage, and system tool keys are now encrypted at rest

FastGPT, an open-source AI application platform for organizations, released version 4.16.0 on Aug. 16, 2026. The release reworks three areas: how code sandboxes are provisioned, what metadata a knowledge base can carry, and how audit logs are retained.

Code sandboxes are now bound to an application and a user rather than to individual conversations. Multiple conversations from the same user in the same application reuse one sandbox instance, with files isolated per session directory. Sandbox HTML and files can be previewed through short-lived read-only links, removing a repeated upload to object storage. Per-instance CPU, memory, storage, auto-suspend and auto-archive limits are now configurable through environment variables. The E2B sandbox provider has been removed; existing deployments migrate to OpenSandbox or Sealos Devbox.

Also Read: AiThority Interview with Gou Rao, co-founder and CEO at NeuBird AI

Knowledge base entries accept custom metadata, importable as JSON through the API or CSV and Excel templates. Retrieval results and backup exports preserve the field, so source, version and clearance labels can travel with a retrieved answer back into business systems.

On the operations side, audit logs are transferred to cold archive storage instead of being deleted on expiry, administrator-configured tool keys are encrypted, and the system initialization endpoint no longer returns model API keys, request addresses or internal configuration. The upgrade guide also states a boundary for deploying teams: the sandbox preview proxy should use an origin separate from the main site, since same-origin deployment places preview scripts inside the main site’s security boundary, and the system does not currently enforce origin isolation.

“Sandboxes, metadata and audit log retention each touch deployment boundaries, so confirm those and the migration order before choosing an upgrade window,” a FastGPT spokesperson said.