Skip to content

AI Agent

Delegates a step in the workflow to an AI model or agent with tool access.

Ai Agent node on the canvas

The AI Agent node represents a workflow step handled by an AI model. The editor captures the agent’s configuration - which model to use, what tools it can access, its system prompt, and memory settings. Your backend is responsible for invoking the actual AI service at runtime.

When to use

Section titled “When to use”

Use an AI Agent node when a workflow step requires language understanding, generation, or decision-making that cannot be expressed as simple rules or conditions.

Common examples:

  • Document processing - extract structured data from unstructured text (invoices, contracts, support tickets)
  • Content generation - draft emails, summaries, or reports based on workflow data
  • Classification and routing - analyze incoming requests and route them to the right team or process
  • Tool-augmented agents - let the AI call external tools (Gmail, Jira, Slack, Airtable) to gather information or take action as part of a larger workflow
  • Conversational steps - generate context-aware responses in customer support or onboarding flows
  • Data enrichment - look up and synthesize information from multiple sources before passing it downstream

The AI Agent node connects to other nodes like any other step. It receives input from upstream nodes and passes its output to downstream nodes. The difference is that the “logic” is delegated to an LLM rather than coded explicitly.

Chat model

Section titled “Chat model”

Select the AI model that powers this agent step. The models listed below are the defaults shipped with Workflow Builder - they are illustrative examples. You can replace them with any model your backend supports by editing the node schema.

ModelValueIcon
GPT-5.4gpt5.4OpenAiLogo
Gemini 3.1 Progemini3.1proGeminiLogo
Claude Sonnet 4.6claudeSonnet4.6ClaudeLogo

Tools

Section titled “Tools”

Tools give the AI agent access to external services. Each tool is a named integration the agent can invoke during its execution. Tools are configured as an array - you can add as many as needed. The presets below are examples bundled with the default configuration; you can define your own tools to match whatever services your backend integrates with.

PropertyTypeDescription
ToolTextName of the tool
DescriptionTextWhat the tool does (passed to the model as context)
API KeyTextAuthentication key for the tool’s service

Available tool presets:

ToolIcon
GmailGoogleLogo
ExcelMicrosoftExcelLogo
AirtableAirtableLogo
JiraJiraLogo
SlackSlackLogo
HubspotHubspotLogo

Memory

Section titled “Memory”

Controls whether the agent retains context across invocations within the same workflow run.

OptionDescription
Window-based MemoryRetains a sliding window of recent conversation context

System prompt

Section titled “System prompt”

A free-text field for the system prompt that defines the agent’s behavior, persona, and constraints. This is passed directly to the model at runtime.

Properties

Section titled “Properties”
PropertyTypeDescription
LabelTextDisplay name shown on the node card
DescriptionTextShort description of the node’s purpose
StatusDropdownActive / Draft / Disabled

Schemas

{
  "type": "VerticalLayout",
  "elements": [
    {
      "type": "MessageOnError",
      "scope": "#/properties/missingPreviousVariable",
      "text": "plugins.validation.missingDependency"
    },
    {
      "type": "Accordion",
      "label": "General Information",
      "elements": [
        {
          "type": "Text",
          "scope": "#/properties/label",
          "label": "Title",
          "placeholder": "Node Title..."
        },
        {
          "type": "Select",
          "scope": "#/properties/status",
          "label": "Status"
        },
        {
          "type": "Text",
          "scope": "#/properties/description",
          "label": "Description",
          "placeholder": "Type your description here..."
        }
      ]
    },
    {
      "type": "Accordion",
      "label": "Operational Settings",
      "elements": [
        {
          "type": "VerticalLayout",
          "elements": [
            {
              "type": "Select",
              "scope": "#/properties/chatModel"
            },
            {
              "type": "Select",
              "scope": "#/properties/memory"
            }
          ]
        }
      ]
    },
    {
      "type": "Accordion",
      "label": "Tools",
      "elements": [
        {
          "type": "AiTools",
          "scope": "#/properties/tools"
        }
      ]
    },
    {
      "type": "Accordion",
      "label": "System Prompt",
      "elements": [
        {
          "type": "TextArea",
          "scope": "#/properties/systemPrompt",
          "placeholder": "Type your prompt here...",
          "minRows": 5
        }
      ]
    }
  ]
}