Roboflow Inference

Roboflow Inference is a self-hosted computer vision server for teams building camera and image analysis systems. It runs on your own computer, server, or edge device and combines model predictions with workflows for tracking, counting, measuring, and responding to events. Roboflow also offers hosted servers and a Serverless Cloud API, where processing runs on its infrastructure.

Workflows let you combine object detection, classification, and segmentation with OCR, barcode and QR reading, or template matching. You can use foundation models such as Florence-2, CLIP, and SAM2, or serve your own fine-tuned models. Applications include license plate reading, face blurring, background removal, and measuring how long an object stays in a zone. Custom code and models can extend the workflows.

For live video, it manages RTSP streams and webcams, with hardware acceleration and GPU batching. A Python SDK and REST API connect predictions to other software; workflows can also send email or Twilio notifications and call webhooks. You can record and analyze predictions alongside the video processing.

It supports Docker deployments on Linux, Windows, and macOS, plus NVIDIA Jetson and Raspberry Pi devices. NVIDIA CUDA provides GPU acceleration where available. Pre-trained and foundation models, public workflows, and video stream management don't require an API key. A Roboflow API key provides access to fine-tuned models, private workflows, and hosted compute; an account also enables remote stream management through the Roboflow UI. Self-hosted commercial model licensing is a separate enterprise add-on.