FastGPT 4.16.0 Updates Code Sandboxing and Enterprise Data Handling

The latest iteration moves code sandboxes from a conversation-based model to a user-application binding. By reusing sandbox instances across multiple sessions, the platform streamlines resource usage while maintaining file isolation within session-specific directories. Administrators now gain granular control over CPU, memory, and storage limits via environment variables. With the removal of E2B support, deployments are transitioning to OpenSandbox or Sealos Devbox.

Knowledge base management receives a significant upgrade with custom metadata support. Users can now import metadata via JSON, CSV, or Excel, ensuring that classification labels and versioning persist through retrieval processes. This change allows business systems to maintain clearer context when processing AI-generated answers.

Operational security also sees hardening measures. Audit logs now migrate to cold storage rather than facing automatic deletion, and tool keys are encrypted at rest. Furthermore, the system initialization endpoint has been stripped of sensitive configuration data. FastGPT advises teams to deploy the sandbox preview proxy on a separate origin, noting that current same-origin configurations lack strict isolation. Given these architectural shifts, the company recommends a careful review of migration paths before scheduling implementation.