OVHcloud ML Serving

Easily deploy and industrialize your Machine Learning models with an efficient and scalable platform

Going from development to production with Machine Learning has never been so easy yet powerful.

Bring your ML models made from various tools and language such as TensorFlow, PMML or ONNX, and deploy them in production effortless.

We provide in few minutes an API endpoint, Swagger and infrastructure scalability !


Now available! Try it for free in Public Cloud control panel


How it works


You can find the full documentation here :


Step 1 : Connect to OVHcloud control panel

Connect to OVHcloud control panel, and find "Serving Engine" in the Public Cloud menu.

Note : currently only available on EUROPE region (soon in CANADA region)


Serving engine public cloud



Step 2 : Initialize a namespace

This namespace will be linked to your Public Cloud object storage bucket and will contain your deployed models



Step 3 : Deploy machine learning models

You can deploy your own models or pretrained ones in our catalog.


Serving model selection


We are compliant with standard formats,, such as TensorFlow HDF5/SavedModel, PMML, ONNX and so on.

ml tools


Step 3 : create a security token

By default, no one can query neither manage your deployed models. You need to create a token to do so.



Step 5 : Done ! Access your swagger to query your models

You can now access to your swagger in order to know how to interact with your model. As the swagger is generated with your model input, you can easily generate a client library to plug it to your code.

OVHcloud will scale automatically the infrastructure behind.






Features and benefits



Include your model in any framework or languages, you are not bounded to the framework used to create your ML models.



Retrace your steps with integrated versioning.


Auto Scalable

Based on amount of calls you receive and metrics we gather, such as CPU, RAM consumption and latency, we will scale your infrastructure to always provide a reliable and performant service.


API Endpoint

Deployed models are available through and HTTP API Endpoint along with a Swagger tailored to your actual model inputs.


Monitoring / Healthchecks

Powered by OVHcloud Observability, monitor your models usage, load and liveness, never feel caught off guard by an unavailable serving.


Multi-backend : TensorFlow / PMML / ONNX & more to come !

You can bring Machine Learning models made with various tools, such as Tensorflow, Scikit-Learn, PMML language or ONNX hub. The platform is backend agnostic and no format/framework is out of reach and can be added to the platform to fit your needs.


High Availability

Your models will be deployed by default on multiple instances, powered by Kubernetes.


Granular security & security policies

Manage access to models and deployments using Role Based Access Policy (RBAC).


Rolling updates

Deploy your new model version without any service interruption.


Pricing and limitations

For this free lab, we limit the amount of models, scalability and call rates. Once in production, offers will be quite more flexible with Pay-As-you-Go.


Serving Engine - Lab plan
Price 100% Free for lab (except if you use object storage)
Models 10
Automatic scaler Yes, from 1 to 3 instances during lab
Rate limit No
Tools/Languages supported (more to come)

TensorFlow, Pytorch, PMML, ONNX



Lab access : now available for everyone !


Simply connect to OVhcloud control panel, and use "Serving Engine" in the Public Cloud section


Now available! Try it for free in Public Cloud control panel


#gitter         Talk machine learning with us on

Trademark policies :

  • PMML brand and logo are the property of Data Mining Group
  • ONNX brand and logo are the property of ONNX Project Contributors (MIT license)


  • BETA