Roboflow + Kubernetes
Run the Roboflow Inference server in your own Kubernetes cluster, from a single pod and service to an autoscaled deployment with the Enterprise Helm chart.
How to connect Kubernetes to Roboflow
4 stepsSave the starter manifest
Copy the pod and ClusterIP service manifest from Roboflow's Kubernetes docs into roboflow.yaml. It runs the roboflow/roboflow-inference-server-cpu image on port 9001.
Apply it to your cluster
Run `kubectl apply -f roboflow.yaml`. The cluster must be able to pull images from Docker Hub.
Call the service in the cluster
Other workloads reach Inference at http://rf-service.default.svc:9001. Use that address as the `api_url` with your Roboflow API key.
Move to the Helm chart for production
Enterprise customers can install the chart from the inference repository with `helm upgrade --install` and values-cpu.yaml or values-gpu.yaml, which set the image, GPU resources, autoscaling and ingress.
What teams build with Kubernetes
Add vision to services already on Kubernetes
Run Inference as an in-cluster service that your existing applications call over HTTP.
Scale inference with load
Use the Helm chart's horizontal pod autoscaler to add Inference pods as CPU usage rises.
Schedule models on GPU nodes
Place GPU Inference pods on GPU node pools with node selectors and NVIDIA GPU resource requests.
Questions
Yes. Roboflow publishes a Helm chart for the Inference server in the roboflow/inference repository, offered to Enterprise customers. It supports the CPU and GPU images.
Yes. Use the GPU image and schedule the pods on GPU nodes; the Helm chart's values-gpu.yaml requests an NVIDIA GPU per pod.
The starter manifest uses a ClusterIP service. Roboflow's docs suggest a NodePort or LoadBalancer service, or an ingress controller for TLS, to serve it externally.
Enterprise customers can integrate Secure Gateway and offline mode with their Inference deployment.
Why are companies choosing Roboflow?
The end-to-end platform for vision AI: create datasets, train models and deploy to production.
Flexible deployment modes
Move from a scalable cloud API to edge hardware like NVIDIA Jetson without managing complex dependencies.
Integrated workflow builder
Combine multiple AI models, custom business logic and third-party integrations into one production-ready pipeline in a low-code editor.
Model monitoring and alerts
Dashboards track confidence, inference time and request volume for your deployed models, and email alerts tell your team when confidence suddenly drops.
