Inference is an open source package developed by Roboflow containing our primary deployment capabilities. It is performant, flexible, and highly integrated with your Roboflow projects making it extremely easy to deploy computer vision models into nearly any environment.
Inference contains a comprehensive suite of interfaces designed for various computer vision use cases, tailored to fit any requirement from simple apps to enterprise-level solutions. The platform includes a Native Python API for direct access to base package APIs, an HTTP API for treating machine learning models as microservices, an Inference Pipeline for real-time streaming use cases, and Advanced Interfaces for enterprise users. The Native Python API is ideal for straightforward applications and testing, while the HTTP API facilitates microservice architecture without dependency management issues. The Inference Pipeline supports asynchronous streaming from multiple sources, and advanced options enhance base capabilities, including active learning and efficient batch processing for enterprise applications.