Truss

Truss, an open-source framework from Baseten Labs, packages machine learning models for deployment — combining runtime configuration, optional Python code, and CLI commands so a model can be served consistently regardless of how it was trained. It supports three levels of control depending on need: configuration-only for standard cases, custom Python for specific logic, or full custom Docker containers for maximum flexibility. A live-reload development workflow lets you sync local changes to a running deployment before publishing an immutable production build, and Truss "Chains" support multi-model systems for more complex pipelines. As a packaging framework rather than a hosted service by itself, it's most naturally paired with Baseten's own deployment platform, though the framework itself is free and open source. Real pricing for actually running models in production depends on which infrastructure you deploy Truss onto. Good fit for an ML engineer who wants a consistent, repeatable way to package models for deployment rather than hand-rolling deployment scripts per model.