KNIME Release Notes

Build new tool workflows from scratch or convert existing workflows into agent tools:

• You can parameterize tool workflows with configuration nodes you already use (e.g., String Configuration), and define agent–tool communication using the Tool Message Output node.
• Use Workflow Input and Workflow Output nodes to handle data transfer in agent workflows. Agents can now pass large data tables to and from tools.
• The new Workflow to Tool node reads your tool workflows and connects them directly to agents via the Agent Prompter or Agent Chat View nodes.

Why and for whom?

If you are familiar with building workflows in KNIME, you can create agents from scratch using the same concepts with no new learning curve. Even completely new users can use KNIME's visual workflow building to create agents without coding.

If you're part of an organization with extensive KNIME workflow libraries, or you're a workflow builder who wants to make your work available to AI agents, you can now do it without any complex integration work.

You now get a new Message data type that structures conversation data with AI agents with clear roles and content:

• A type: User, AI, or Tool
• Tool calls: AI messages can contain a "Tool calls" section showing which tools the model decided to call
• Tool name and Tool call ID: Tool messages always include these properties, specifying which tool generated the message and in response to which tool call
• Content composed of parts: text or images

The new Message Creator node lets you build these structured messages without any scripting, while the Message Part Extractor node extracts specific message components without custom code.

All Prompter nodes now operate with Messages.

Why and for whom?

If you're building agentic AI workflows, the Message data type gives you a structured, no-code way to work with conversation data and understand exactly what's happening in your agent interactions, improving transparency and explainability.

A new Agent Prompter node to select tools, call them, and iterate to call multiple tools.

Why and for whom?

Building AI agents used to require agent builders to set up multiple workflow steps for tool selection, calling, and iteration. We've consolidated all that complexity into a single Agent Prompter node that selects tools, calls them, and iterates to call multiple tools. This makes it much easier to build agents.

A new Agent Chat View node gives you a conversational interface where you can chat with your AI agent directly and ask it to perform the tasks you need.

Why and for whom?

If you are an agent builder, you can use this node in your agentic data apps as a readymade chat interface. You can also use it to test and debug your agentic workflows through natural conversation.

New connector nodes that let you use Anthropic Claude, Gemini models from Google AI Studio and Google Vertex AI, and models from IBM watsonx.ai.

Why and for whom?

Different models excel at different things. Some are better for performance, others for cost, others for specific tasks. You need flexibility to choose the right model for your requirements. Whether you're building AI workflows or you're in an enterprise managing AI via Google Vertex AI or watsonx.ai, you now have more options to optimize for your needs.