CKEditor AI Service: Cloud Provider Support, Observability, and More
CKEditor AI launched in November 2025 as the industry’s first AI co-writer built natively into a rich text editor. No switching between standalone tools or wrestling with half-built components that eat engineering hours. Since then, it has grown past the initial set of AI Chat, AI Review, AI Translate, and AI Quick Actions features. These features are just the surface. What makes CKEditor AI different is what runs underneath them.
CKEditor AI is not just a set of writing features. It also provides a full AI integration layer built specifically for rich text editing—one that gives LLMs the tools to work with rich text correctly: generating valid HTML output, understanding document structure, and leveraging CKEditor’s own capabilities like state management for AI Chat History. If you have not explored it yet, What is CKEditor AI? guide and the feature demos are good starting points.
The integration layer that powers it all is the CKEditor AI Service. It sits between the LLMs and your users, and shapes the entire experience. It handles multiple users editing while AI is responding, avoids unnecessary data transfer when only a single word needs fixing, keeps costs manageable, and normalizes behavior across any AI model you connect. For organizations running on their own infrastructure, the service goes further: connect any AI model or cloud provider, keep data within your security perimeter, and deploy under your own brand.
Five months in, the pace of development at the service level is accelerating. From cloud provider integrations and MCP support to programmatic access and expanded model options, organizations have more control than ever over their AI deployment. This is the first post in a dedicated series covering updates to that layer: on-premises releases, cloud improvements, and new capabilities for product teams building AI content workflows.
What’s new in the CKEditor AI Service?
Updates in this section are available for both cloud and on-premises deployments.
NEW Conversation and file management endpoints
Two new API endpoints expand what integrators can build around conversations:
-
GET /v1/conversations/:conversationId/documentsreturns a paginated listing of all documents belonging to a conversation. -
GET /v1/conversations/:conversationId/files/:fileIdretrieves file metadata without downloading the full file content.
Increased file size limit
The maximum size for a single non-image file attached to a conversation has been raised from 5 MB to 25 MB. The total limit for all files remains at 30 MB. The stricter 5 MB cap now applies only to images when using Anthropic models, not to all file types. This is particularly useful for workflows involving larger PDFs, including scanned documents.
What’s new in CKEditor AI On-premises?
CKEditor AI on-premises deployment is available only on the Custom plans. Contact us to learn more.
NEW Dedicated cloud provider support
Connecting the on-premises AI service to a managed cloud AI is now significantly simpler. We now offer dedicated first-class support for Azure OpenAI, Amazon Bedrock, and Google Vertex AI – no custom adapter work required. If your organization already runs on one of these platforms, plugging CKEditor AI into your existing AI infrastructure is now a configuration step, not an engineering project.
NEW OpenTelemetry observability
The service now supports OpenTelemetry instrumentation out of the box. Monitor AI interactions, token usage, response quality, HTTP calls, and database operations through your existing observability stack.While Langfuse has dedicated support, if you use a different observability platform, you can contact us. We are happy to extend our support with additional tooling.
NEW LLM circuit breaker
When an AI model starts experiencing failures, the service now detects this automatically and temporarily stops sending requests to it. After a cooldown period, a single probe request checks whether the model has recovered before traffic resumes. This replaces the previous health-check mechanism and makes failover more reliable in production environments without any manual intervention required.
Other improvements
The following on-premises improvements also shipped in recent releases:
-
Custom models support, allowing you to configure and use your own AI models.
-
Conversation sorting by
pinnedstatus, with pinned conversations appearing first -
Improved handling of data synchronization with RTC server in more complex scenarios
-
GET /modelsendpoint now includes removed models -
Updated default model configuration to Claude 4.6 Sonnet, GPT-5.4, GPT-5.4 Mini, and Gemini 3.1 Pro
What’s new in CKEditor AI Cloud (SaaS)?
CKEditor AI is available as an add-on in Essential and Professional plans. Start a 14-day free trial or contact us to learn more.
Apart from features specific to on-premises, such as MCP support, all features and updates land in the cloud first, with on-premises following shortly after.
NEW Extended model lineup
The default model configuration has been extended with new models to reflect the latest available options:
-
Claude 4.6 Sonnet – replaces Claude 4.5 Sonnet in the recommended list
-
GPT-5.4 – replaces GPT-5.2 in the recommended list
-
GPT-5.4 Mini
-
Gemini 3.1 Pro
Other improvements
Recent cloud updates also resolved several stability and reliability issues: optimizations resulting in smaller token consumption and more precise billing data, reasoning mode engagement for GPT-5.2, and a chunking issue in the make-longer review action.
