Agent Prompter

This node builds and executes an agent that responds to a single user prompt using a language model and a set of user-defined tools.

Each tool is represented as a KNIME workflow with configurations (e.g., strings, numbers, column selection) and can optionally accept input data tables. While the agent does not have access to raw data, it is informed about available tables through metadata — including column names and types — enabling it to select suitable data for each tool as needed.

When the node is executed, the agent reasons step-by-step to fulfill the user’s prompt. It may call one or more tools in sequence, using the output of one tool as input for another. The model autonomously selects tools and input data based on the prompt and the tools’ names and descriptions. Including clear tool descriptions and example use cases significantly improves the agent’s decision-making.

The entire internal reasoning process — including tool calls and decisions — is captured and available via the conversation output table. This node is designed for non-interactive, one-shot execution. For interactive, multi-turn conversations, use the Chat Agent Prompter node.

Options

System message

Instructions provided by the workflow builder that guide the agent's behavior. It typically defines the agent’s role, its tone, boundaries, and behavioral rules.This message is prioritized over the user message and should not contain any information that the user can inject in order to prevent prompt injection attacks.

User message

Message from the end user, prioritized behind the developer message.

Tool column

The column of the tools table holding the tools the agent can use.

Conversation column

The column containing the conversation history if a conversation history table is connected.

Conversation column name

Name of the conversation column if no conversation history table is connected.

Recursion limit

The maximum number of times the agent can repeat its steps to avoid getting stuck in an endless loop. If not provided, defaults to 25.

Debug mode

In debug mode, tool executions are displayed as meta nodes in the agent workflow and the meta node is kept in case of an error in the tool.

Data message prefix

Prefix for the data message shown to the agent. You can use this to customize the instructions about the data repository.

Input Ports

Icon

The chat model to use.

Icon

The tools the agent can use.

Icon

The table containing the conversation held so far.

Icon

The data inputs for the agent.

Output Ports

Icon

The conversation between the LLM and the tools reflecting the agent execution.

Icon

The data outputs of the agent.

Popular Predecessors

  • No recommendations found

Popular Successors

  • No recommendations found

Views

This node has no views

Workflows

  • No workflows found

Links

Developers

You want to see the source code for this node? Click the following button and we’ll use our super-powers to find it for you.