Kili Technology has three deployments options:
- Kili on Cloud (SaaS) hosted on our infrastructure
- Kili on Cloud (SaaS) deployed as a managed application in Azure Marketplace
- Kili On-Premise
The easiest solution is to use Kili Technology cloud platform, with a Software as a Service model.
Both the platform and your data are hosted on our infrastructure.
The SaaS version of Kili is deployed with two different cloud providers:
- Google Cloud, using a data center located in Belgium (Europe)
- Azure, East US region located in Virginia
You can deploy Kili as a managed application in Azure Marketplace, on a private Azure subscription. The deployed Kili instance is based on these associated Azure services:
- Kubernetes cluster: Azure Kubernetes Services (three
Standard_D2as_v5machines and one
- Object storage: two Blob storage buckets
- Relational database: Azure Database for PostgreSQL (
Standard_D4ds_v4: 4 CPUs, 16 GB RAM, 128 GB disk space)
Updates to newer versions of the Kili application are managed automatically through the Azure marketplace, following the pace of the Long Term Support versions of Kili (updated twice a year).
To deploy Kili on Azure Marketplace, you must first create the resources that will interact with Kili.
The required resources are:
- An Azure Machine Learning workspace
- A compute instance to execute your notebook code
The required version of the compute instance depends on the scope of tasks that you're planning to execute using Kili. For simple data management tasks like project setup, labeled data access and management, a basic compute instance will be enough. For more advanced applications, like model-assisted labeling, you will need a a slightly more advanced instance.
For details and available plans, refer to the information in Kili Technology's overview page on Azure Marketplace.
The third solution is to have Kili Technology on your own servers: on-premise.
For details, refer to Kili architecture.
Minimum configuration for 20 labelers or 5 reviewers:
For the services:
- 2 CPU
- 4 GB RAM
- 500 GB storage
For the PostgreSQL database:
- 2 CPU
- 4 GB RAM
On this machine:
- An updated version of Linux, for example Ubuntu 16.04
- Kubernetes or Docker-compose
During the installation:
- Internet access
- Root access
There is no need for Internet/root access after installation.
The S3 bucket is accessible from collaborators' desktops.
Virtual machine exposes port 80 and a domain name is set up under its IP.
Updated 3 days ago