Requirements

Before you install CKEditor AI On-Premises, check that you can provide everything on this page: machines with a container runtime, an SQL database, an in-memory data store, file storage, at least one LLM provider, a token endpoint in your application, and outbound network access. To plan how the pieces fit together, see Architecture. To install, see Deployment.

If you want to know what CKEditor AI does and which integration fits your product, see the CKEditor AI guides.

No single infrastructure setup fits every deployment. The recommendation below covers high availability. Monitor your application after you deploy it, and adjust the infrastructure to the load you see.

For high availability, run at least three instances of CKEditor AI On-Premises. You can run them as Docker containers on a single machine, but we recommend at least three machines. The deployment then survives the failure of one machine.

To scale across several machines, use a load balancer such as HAProxy or NGINX. For configuration examples, see SSL communication. You can also scale with a container platform such as Amazon Elastic Container Service, Azure Container Instances, or Kubernetes.

CKEditor AI On-Premises ships as a Linux-based Docker image. Run it on Linux machines, or use Docker Desktop on Windows and macOS.

CKEditor AI On-Premises runs under any Open Container runtime, such as Docker, Kubernetes, OpenShift, or Podman.

On Windows, enable the WSL 2 backend and switch to Linux containers. See the Docker Desktop installation guide for the steps.

CKEditor AI On-Premises connects to OpenAI, Anthropic, Google, Azure OpenAI, Amazon Bedrock, Google Vertex AI, and any provider compatible with the OpenAI Chat Completions API. You need at least one provider. For the supported providers and their options, see LLM providers.

CKEditor AI On-Premises needs PostgreSQL 12.0 or later, or MySQL 8.0 or later. For the privileges the database user needs, and for the creation scripts, see SQL Database on the Deployment page.

For another SQL engine, such as Microsoft SQL Server, contact us.

CKEditor AI On-Premises needs an in-memory data store compatible with the Redis protocol. The REDIS_* options configure it, whichever product you run. Supported implementations:

• Valkey 9.0.0 or newer, as a single instance or in Cluster mode
• Redis 3.2.6 or newer, as a single instance or in Cluster mode

Redis Sentinel is not supported. For the connection options, see Redis database on the Configuration page.

CKEditor AI On-Premises stores uploaded files in one of these:

• Amazon S3
• Azure Blob Storage
• Filesystem
• SQL database

For the connection options, see Storage on the Configuration page.

Every request to CKEditor AI On-Premises carries a JWT that your own application signs. You provide the endpoint that authenticates the user and returns that token. Authentication explains the token, and Create the token endpoint on the Deployment page shows the installation step.

The application containers open every connection themselves. No CKSource host connects to your deployment.

Inbound traffic comes from your own users and systems: the editors in your users’ browsers, and your backend. Both call the API through your load balancer, on the application port (8000 by default). The application port serves plain HTTP. To serve HTTPS, terminate TLS at the load balancer. See SSL communication.

The outbound destinations you must allow depend on your configuration. The hosted endpoints below use HTTPS. Endpoints you provide yourself use the host and port you configure.