KNIME Documentation

Bring your own K-AI model ​

Knime features an AI assistant (K-AI) that efficiently answers questions about Knime and helps users building customized workflows, simplifying their data analysis tasks. On Knime, the AI service provides these AI features to your users, including the Knime AI Assistant and Code Generation.

The following parts for the AI service can be configured:

  • LLM Backend
  • Disclaimer & Welcome Messages
  • AI Service Q&A Content
  • AI History Service

INFO

The AI service is available to all Enterprise users, and to licensed users on the Basic and Standard editions.

Configure the LLM backend ​

Currently, it is possible to configure the AI service to use either OpenAI or Azure OpenAI as backend.

To configure OpenAI as LLM Backend do the following steps:

  1. Create an OpenAI account
  2. Generate an API key on the API keys page. It is recommended to create a fresh API key that is used exclusively by the AI service.

To configure Azure OpenAI as LLM Backend do the following steps:

  1. Create an Azure account and get access to Azure OpenAI
  2. Create an Azure OpenAI resource to be used by the service and enter it in the KOTS Admin Console. It is recommended to create a separate resource that is used exclusively by the AI service to avoid any interference with other applications.

See the Azure OpenAI documentation to learn how to create a resource.

  1. Deploy the following models with API version 2024-08-01-preview or newer:
    • Lightweight model: gpt-4.1-mini
    • High intelligence model: gpt-4.1
    • Embedding model: text-embedding-3-small

See the Azure OpenAI documentation for more details.

K-AI is not hosted by Knime. Instead, it connects directly to the OpenAI API, which processes the requests and sends back responses. For more information, see the OpenAI Platform Privacy Policy.

Which models K-AI uses ​

The Knime AI Assistant uses different models depending on the interaction mode.

For Q&A mode:

  • gpt-4.1-mini (lightweight model) for response generation
  • text-embedding-3-small for embeddings

For Build mode and code generation:

  • gpt-4.1 (high intelligence model)
  • text-embedding-3-small for embeddings

This distinction exists because gpt-4.1 is more capable than gpt-4.1-mini, but also much slower to respond and more expensive. If response time and cost are not constraints, gpt-4.1 can also be used for Q&A mode.

INFO

Models with reasoning enabled by default, such as gpt-5, are currently not supported. Other models in the GPT-5 series (gpt-5.1, gpt-5.2, gpt-5.3, gpt-5.4) are supported since reasoning is disabled by default.

Configure the disclaimer and welcome messages ​

The KOTS Admin Console also allows you to customize the initial messages displayed by the Knime AI Assistant. The disclaimer is not shown by the Knime AI Assistant if it is left empty.

In addition to the initial disclaimer, users will be prompted to grant permission when K-AI requests to fetch workflow data in Build mode. This permission is requested per workflow and cannot be configured by administrators. For more information, see the K-AI Build mode documentation.

Include workflows and components in Q&A ​

Enable these options to include workflows and/or components in the Q&A mode of the AI Assistant. The AI Assistant will then be able to answer questions about the public workflows and components available on your Knime instance.

Only public workflows and components are searched, and their descriptions are shared with the selected GenAI provider.

Access the AI history endpoint ​

The AI service exposes an endpoint that allows you to retrieve the requests that were made to the AI service via the Knime AI Assistant. The KOTS Admin Console allows you to configure which Keycloak groups can access the endpoint.

It is recommended to only give access to admins.

The address depends on your Knime instance but it is typically located at https://api.<base-url>/ai-history/kai.

GET requests to this endpoint need to have the authorization header set with a valid bearer token from your Knime instance.

You have the possibility to filter data by date directly within the endpoint, for instance:

https://api.<base-url>/ai-history/code/python?start_time=2023-12-31T00:00:00&end_time=2024-04-10T00:00:00

The format of the parameters start_time and end_time needs to be yyyy-mm-ddThh:mm:ss as in the example above.

The Gen AI Monitoring data app gives you four perspectives on GenAI-related activity:

  • Users: Identify who is using K-AI.
  • Conversations: Understand how users are interacting with K-AI.
  • Cost: Analyze the costs associated with Gen AI usage.
  • All: A comprehensive analysis of all attributes and metrics.

Only global admins with appropriate access rights can run this workflow.

Follow the general data app deployment steps to deploy it.

How K-AI handles your data ​

To power our AI features, Knime connects to providers like OpenAI or Microsoft Azure. Here is exactly what gets shared:

  • Your Chats: Anything you type into the AI Assistant is sent to the provider to generate a response.
  • Workflow Metadata: In Build mode, K-AI automatically sees your column names and data types. This allows it to suggest relevant logic and configurations for your specific nodes.
  • Actual Data (Optional): K-AI cannot see your actual data values unless you explicitly toggle the "Fetch Data" permission. This is off by default and used only to provide highly specific troubleshooting or analysis.