Skip to main content

​
Overview

The Agent Node runs a single AI agent that uses tools you configure to complete tasks. The agent receives your prompt (which can use format strings from workflow input), decides when to call which tool, and returns a final answer. Use it for RAG, research, orchestration, and multi-step tasks.
  • Tools — Workflows (invoke other workflows), Node tools (e.g. Web Search, Google Search, REST API, Knowledge Base Search, Human Input), and Knowledge Base Search (search a selected knowledge base). The agent chooses when and how to use each tool.
  • Model — Must support function calling. You can set temperature, top-p, system prompt, and max steps.
  • Output — Final result plus a conversation log of reasoning and tool calls.
RAG and Q&A: For answering questions using your documents, we recommend adding the Knowledge Base Search tool and selecting your knowledge base. The agent will search the knowledge base when needed and ground answers in your content.

​
Configuration Parameters

​
Node Configuration

  • Query The task or question for the agent. Supports format strings from workflow input (e.g. {user_question}). The agent can call tools as needed to complete the task.
  • Tools Select which tools the agent can use. The agent only has access to the tools you enable.
    • Node tools — Expose individual nodes as tools, e.g. Web Search, Document Reader, Code, and others. The agent calls the node with parameters it chooses.
    • Workflow — Invoke another workflow. Configure workflow and input template.
    • Knowledge Base Search — Search a selected knowledge base (semantic search). Recommended for RAG and Q&A over your documents.
    Choose tools that match the task. The agent can only perform actions through the tools you provide.
  • System Prompt: Define the agent’s persona and behavior (e.g. “You are a research assistant. Use the knowledge base to answer questions.”). Supports format strings from workflow input.
  • Model: Language model for the agent. Must support function calling.
  • Temperature: 0–2. Lower for more deterministic tool use; higher for more varied behavior.
  • Top P: 0–1. Nucleus sampling for generation.
  • Max Steps: Maximum number of steps (1–50) the agent can take. Increase for complex multi-tool tasks.
  • PII guardrails for agent operations (excl. tools): Enable PII guardrails on the agent’s reasoning; combine with PII handling in tools as needed.
  • Enable Model Fallbacks: Automatically try alternative models if the primary model fails.

​
Expected Inputs and Outputs

  • Inputs:
    • input: Data that can be referenced in the Query and System Prompt using format strings (e.g. {user_question}, {ticket_id}).
  • Outputs:
    • output: The final result produced by the agent.
    • conversation: A detailed log of the agent’s reasoning and tool calls.

​
Use Case Examples

  1. RAG / Q&A over your docs Add the Knowledge Base Search tool and select your knowledge base. Prompt: Answer the user’s question using the knowledge base. If the answer isn’t there, say so. See the Knowledge Base guide for creating and populating knowledge bases.
  2. Web research Add Web Search and/or Google Search (and optionally Webpage Extractor). Prompt the agent to research companies, find contact info, or gather up-to-date information from the web.
  3. Workflow orchestrator Add Workflow tools that invoke your existing workflows. The agent can decide when to run which workflow and pass inputs, so users can give high-level tasks and the agent orchestrates execution.

​
Error Handling and Troubleshooting

  • Model Compatibility: Ensure the selected model supports function calling. If you receive a compatibility error, switch to a supported model.
  • Tool Configuration: Verify that all configured workflow and node tools exist and are accessible to the agent.
  • Max Steps Limitation: If the agent isn’t completing tasks, consider increasing the max_steps parameter (up to 50) or breaking down the task into smaller components.
If you encounter any issues not covered in this documentation, please reach out to our support team for assistance.

​
Relevant Nodes

Agent Swarm Node

Orchestrate multiple agents for complex tasks

LLM Node

Direct LLM access without the agent framework

Knowledge Base Search Node

Standalone KB search when you need a fixed search step; prefer Agent + KB Search tool for RAG