WatsonX AI Agent icon

WatsonX AI Agent

A custom agent that connects to WatsonX.

WatsonX AI Agent

AI-generated

Summary

Run a tool-calling AI agent for each input item using a connected chat model and tools. Supply the user prompt from a connected Chat Trigger’s chatInput field or define it directly; optionally add memory, an output parser, and execution settings.

Inputs

  • Prompt source and prompt (required) — Choose Connected Chat Trigger Node to read chatInput from the current item, or Define Below to enter text or an expression. The prompt must resolve to a value.
  • Chat Model (required) — Connect a chat model that supports tool calling. The node accepts configured chat model types, including WatsonX models.
  • Tool (required) — Connect one or more tools for the agent to call. Tool names must be unique.
  • Output Parser — Enable Require Specific Output Format and connect an output parser to validate and parse the final response.
  • Memory — Optionally connect memory for conversation context.
  • Options — Set the system message (default: “You are a helpful assistant”), maximum iterations (default: 10), whether to return intermediate steps (default: off), and whether to pass binary images to the agent (default: on). Batch Processing sets parallel items per batch (default: 1) and delay between batches in milliseconds (default: 0).

Output shape

One main output item per input item, containing the agent response as JSON.

When an output parser is connected and memory is also connected, the parsed response is returned as the output value. With Return Intermediate Steps enabled, the result can include those steps. Processing errors stop execution unless Continue On Fail is enabled, in which case that item contains an error field.

Links

Discussion