ML engineers at Baseten and elsewhere who want a simple, framework agnostic way to serve models in production.

Truss is an open source Python library that packages a trained model plus its dependencies into a deployable, Docker based serving container with minimal boilerplate.

Used locally to define and test a model server, then deployed either to Baseten's cloud or any other Docker environment.

Released by Baseten in 2022 and maintained with regular updates through 2026.

Standardizes model serving across frameworks like PyTorch, TensorFlow, and Hugging Face so teams avoid writing custom Flask or FastAPI wrappers.

Built in Python and generates Docker images, with optional one command deployment to Baseten's hosted infrastructure.

Open source under MIT on GitHub, no account required for local packaging, a Baseten account is only needed for their hosted deployment target.