May 2026 monthly summary for NVIDIA/TensorRT-LLM focusing on delivering extensibility features and ensuring high-impact customer value. Key feature delivered: AGSI Middleware Support for FastAPI Server, enabling dynamic middleware registration to extend request handling and simplify integration of custom workflows. This aligns with strategic goals to modularize middleware and accelerate onboarding of third-party components.

March 2026 focused on reliability, deployment hygiene, and developer guidance for nebius/soperator. Key features delivered zip-based jail uploads across CI/CD and E2E workflows, updated workflow steps from Tar Jail to Zip Jail to reflect the compression method, and implemented related stability improvements. Docker configuration and installation were enhanced to ensure correct docker-cli setup, restoration of the real docker binary, creation of required Docker config directories, and alignment of the package name to docker.io in Ansible. User-facing messaging for nested job submissions was clarified to improve observability and configuration guidance. A fix was applied to mount path resolution in complement_jail.sh using absolute paths, increasing reliability in jail mounting across environments. Overall, these changes reduce CI/CD flakiness, improve deployment consistency, and shorten remediation time for related incidents.