AI Agent1 input → 2 outputs

AI Tool

Defines a custom tool that the AI Agent can call, and coordinates between the AI and the tool's execution flow.

Inputs

payload any: Message from upstream node (typically AI Model or function node)
aiagent object: AI configuration passed through the flow
aiagent.toolCall object: When AI calls this tool: {name, arguments}

Outputs

Output 1 - To Next Node (AI Agent)

payload any

Original payload passed through

aiagent.tools array

Updated tools array with this tool's definition added
Output 2 - To Tool Flow (when AI calls tool)
toolInput object

Arguments provided by AI: {param1: value1, param2: value2}

_toolContext object
Internal tracking info:
- identifier - Tool ID
- adapterNodeId - Node ID for Tool Terminate
- requestId - Unique request ID for concurrency
- startTime - Execution start timestamp
- timeout - Configured timeout in ms

Details

The Tool Adapter is the bridge between the AI Agent and custom tool implementations. It:

Registers a tool definition with the AI (via output 1)
When AI calls the tool, routes the request to your flow (output 2)
Coordinates with Tool Terminate to return results to the AI

Tool Definition

Tool ID

Unique identifier (letters, numbers, underscores, hyphens). AI uses this to call the tool.

Description

Tell the AI what this tool does. Be specific—AI uses this to decide when to call it.

Parameters

Define what arguments the AI should provide:

name: Parameter identifier
type: string, number, boolean, object, array
description: Help AI understand what value to provide
required: Whether this parameter is mandatory

Tool Settings

Timeout

Max milliseconds to wait for tool execution. Default: 30000 (30 seconds)

Validation Code

JavaScript to validate tool input before execution:

if (!input.patientId) {
  return { valid: false, error: 'patientId required' };
}
return { valid: true };

Examples

FHIR Search Tool:

Tool ID: search_patient
Description: Search for patients by name or MRN
Parameters:
  - name: searchTerm (string, required)
    "The patient name or MRN to search for"

Appointment Booking Tool:

Tool ID: book_appointment
Description: Schedule an appointment for a patient
Parameters:
  - patientId (string, required)
  - dateTime (string, required)
  - providerId (string, required)
  - reason (string, optional)

Flow Pattern

[AI Model] → [Tool Adapter]
                ↓ output 1 → continues to AI Agent
                ↓ output 2 → [Your Tool Flow] → [Tool Terminate]

Best Practices

Write clear, specific tool descriptions
Mark only truly required parameters as required
Validate inputs to prevent errors downstream
Always end tool flows with Tool Terminate

Tool ID: book_appointment Description: Schedule an appointment for a patient Parameters: - patientId (string, required) - dateTime (string, required) - providerId (string, required) - reason (string, optional)