KNIME Documentation

Get Started with AI in Knime ​

The AI Extension nodes let you work with large language models (LLMs), build AI agents, and implement Retrieval-Augmented Generation (RAG) pipelines in Knime workflows.

You can connect to a model provider directly, or use models from the GenAI Gateway, which manages access for your team.

Typical use cases include:

  • Classifying text into categories
  • Summarizing long documents into key points
  • Extracting names, dates, or other details from unstructured text
  • Translating text between languages
  • Answering questions about your own data

The example below classifies customer reviews. The workflow sends each review to an OpenAI model asking it to assign a sentiment category (positive or negative). The model's response is added to the table as a new column:

A Knime workflow that sends customer reviews to an LLM and receives sentiment classifications

Prompt a model ​

Send a text instruction to a language model and receive a response. A prompting workflow connects three nodes: an authenticator (API credentials), a model selector, and the LLM Prompter:

Get structured output ​

By default, a model returns free-form text. Structured output mode maps the response directly to table columns with defined types.

Create embeddings ​

Embeddings are numerical vectors that capture the meaning of text. They are used for similarity search, clustering, and as the basis for RAG pipelines.

Generate or edit images ​

Use AI models to create new images or modify existing ones from text prompts.

Agents and RAG ​

  • Agents: Build workflows where a language model decides what to do next, calling tools like database lookups or APIs across multiple steps.
  • Retrieval-Augmented Generation (RAG): Ground model answers in your own documents by retrieving relevant passages before prompting.