Skip to main content
Integrate AI agents directly into your BPMN processes to automate document processing, make intelligent decisions, and generate content without user interaction.

​
Overview

BPMN integration allows you to:
  • Automate document processing in the background
  • Make AI-powered decisions at process gateways
  • Generate content automatically within workflows
  • Validate data using AI analysis

​
Integration methods

​
Method 1: Start Integration Workflow action

Use the Start Integration Workflow action to call an AI agent from any BPMN node.
1

Add a Service Task

Add a Service Task node to your process where you want to trigger the agent.
2

Configure the action

Add a Start Integration Workflow action and select your agent workflow.
3

Map input data

Configure which process data to send to the agent (documents, text, parameters).
4

Handle the response

Map the agent’s response back to process variables.

Start Integration Workflow

Detailed configuration reference

​
Method 2: Custom Agent Node

For complex agent interactions, use the Custom Agent Node in Integration Designer:
FeatureStart Integration WorkflowCustom Agent Node
Setup complexitySimpleMore complex
FlexibilityStandard call/responseFull workflow control
Error handlingBasicAdvanced
Best forSingle agent callsMulti-step AI workflows

Custom Agent Node

Integration Designer agent configuration

​
Common patterns

​
Document extraction

Extract data from uploaded documents and continue the process with structured data. 
User uploads document → Service Task calls extraction agent →
Agent returns structured data → Process continues with data
Configuration:
  • Input: Document file ID or content
  • Output: Extracted fields mapped to process variables
  • Error handling: Route to manual review if confidence is low

​
Automated decisions

Use AI to make or recommend decisions at process gateways. 
Process reaches gateway → Service Task calls decision agent →
Agent returns recommendation → Gateway routes based on result
Configuration:
  • Input: All relevant process data
  • Output: Decision result and confidence score
  • Human override: Optional review step for low-confidence decisions

​
Content generation

Generate documents, emails, or reports automatically. 
Process collects data → Service Task calls generation agent →
Agent produces content → Content stored/sent
Configuration:
  • Input: Template reference and data to include
  • Output: Generated content (text, HTML, or document)
  • Review: Optional human approval before sending

​
Input/Output mapping

​
Sending data to agents

Map process variables to agent input: 
// Example input mapping
{
  "documentId": "${processData.uploadedDocument.fileId}",
  "applicantName": "${processData.application.fullName}",
  "extractionType": "loan_application"
}

​
Receiving agent responses

Map agent output back to process variables: 
// Example output mapping
{
  "extractedData": "processData.extraction",
  "confidence": "processData.extractionConfidence",
  "processingTime": "processData.agentMetrics.duration"
}

​
Error handling

Configure timeouts for agent calls. If an agent doesn’t respond in time, route to a fallback path or retry.
Check confidence scores in gateway conditions. Route low-confidence results to human review.
Use boundary error events to catch agent failures and handle them gracefully.

​
Best practices

Async processing

Use async patterns for long-running agent tasks to avoid blocking the process

Idempotency

Design agent calls to be safe to retry in case of failures

Monitoring

Track agent call metrics (latency, success rate, confidence scores)

Fallbacks

Always have a fallback path for when agents fail or return uncertain results

Chat interface

User-facing agent interaction

Integration Designer

Build complex integration workflows

Agent Builder

Create custom agents

Service Tasks

BPMN task configuration
Last modified on February 12, 2026