Workflow Agent

Overview

The Workflow Agent is the most advanced and flexible type of agent on the Evo AI platform. Using LangGraph as the execution engine and React Flow for visual interface, it allows creating complex workflows with multiple node types, conditions, delays, and sophisticated flow control. Unlike other agent types (Sequential, Parallel, Loop), the Workflow Agent offers granular control over each step of the process, enabling the creation of complex business logic with conditional branching, custom loops, and advanced orchestration.

Based on LangGraph: Implementation using LangGraph for workflow execution and React Flow for visual construction interface.

Key Features

Visual Interface

Drag-and-drop construction with React Flow for intuitive workflows

LangGraph Engine

Robust execution based on LangGraph for complex workflows

Multiple Node Types

Agent, Message, Condition and Delay nodes for maximum flexibility

Integrated Testing

Individual agent testing and complete workflow testing

Node Types

1. Agent Node

Agent Node - Workflow Core

Function: Executes agents (new or existing) as part of the workflowSettings:

Select Existing Agent: Choose from agents already created on the platform
Create New Agent: Inline creation of agent specific to the workflow
Edit Agent: Modification of existing agents
Input Mapping: Input data mapping
Output Key: Key to store result in context

Usage example:

Agent Node: "Sentiment Analyzer"
- Agent: sentiment_analyzer (existing)
- Input: text from previous output
- Output Key: sentiment_analysis

Configuration form:

2. Message Node

Message Node - Step Control

Function: Sends direct messages as agent in the flow, ideal for step controlSettings:

Type: Message type (currently only “Text” available)
Content: Message content to be sent

Available type:

Text: Message in plain text format

Configuration example:

Type: Text
Content: Starting quality analysis of received content

Practical use cases:

Instructions for next nodes: Specific guidance for subsequent agents
Status messages: Information about workflow progress
Structured data: Formatted information for processing
Context control: Messages that define context between steps
Notifications: Alerts about specific states or conditions
Process logs: Records of important workflow steps

Practical examples:Analysis Instruction:

Type: Text
Content: Analyze the document focusing on quality and compliance aspects

Progress Status:

Type: Text
Content: Validation step completed successfully, proceeding to approval

Context for Next Step:

Type: Text
Content: Document classified as high priority, apply accelerated process

Configuration form:

3. Condition Node

Condition Node - Flow Control

Function: Checks output from previous node and directs execution based on conditionsCondition Type:

Previous output: Validates the result returned by the previous node (only available type)

Settings:

Output Field: Field from previous output to check (Content or Status)
Operator: Comparison operator
Comparison Value: Value for comparison
Preview: Preview of configured condition

Available Output fields:

Content: Content of the response from previous node
Status: Execution status of previous node

Available operators:Existence Verification:

is defined: Checks if field is defined
is not defined: Checks if field is not defined

Equality Comparison:

is equal to: Field equals specified value
is not equal to: Field does not equal specified value

Content Verification:

contains: Field contains specified text
does not contain: Field does not contain specified text
starts with: Field starts with specified text
ends with: Field ends with specified text

Numeric Comparison:

is greater than: Field is greater than numeric value
is greater than or equal to: Field is greater than or equal to value
is less than: Field is less than numeric value
is less than or equal to: Field is less than or equal to value

Regular Expressions:

matches the regex: Field matches regular expression
does not match the regex: Field does not match regex

Configuration examples:Complete Status Verification:

Condition Type: Previous output
Output Field: Content
Operator: contains
Comparison Value: completed
Preview: Content contains "completed"

Quality Score Verification:

Condition Type: Previous output
Output Field: Content
Operator: contains
Comparison Value: "score": 8
Preview: Content contains "score": 8

Execution Status Verification:

Condition Type: Previous output
Output Field: Status
Operator: is equal to
Comparison Value: success
Preview: Status is equal to "success"

Regex Verification:

Condition Type: Previous output
Output Field: Content
Operator: matches the regex
Comparison Value: "confidence":\s*([0-9.]+)
Preview: Content matches the regex "confidence":\s*([0-9.]+)

How to configure:

Select “Previous output” as Condition Type
Choose the Output Field (Content or Status)
Select appropriate Operator
Enter the Comparison Value
Check the condition Preview
Configure True/False paths by connecting outputs

Configuration form:

Condition addition dialog:

4. Delay Node

Delay Node - Time Control

Function: Adds simple delays in workflow for timing controlSettings:

Delay Value: Numeric delay value
Time Unit: Time unit (seconds, minutes, hours, days)
Description: Optional description of delay purpose

Available Time Units:

Seconds: For short delays (1-59 seconds)
Minutes: For medium delays (1-59 minutes)
Hours: For long delays (1-23 hours)
Days: For very long delays (1+ days)

Configuration examples:Rate Limiting Delay:

Delay Value: 30
Time Unit: Seconds
Description: Rate limiting between API calls

Processing Delay:

Delay Value: 5
Time Unit: Minutes
Description: Wait for external processing to complete

Prioritization Delay:

Delay Value: 1
Time Unit: Hours
Description: Delay for low priority leads

Scheduling Delay:

Delay Value: 1
Time Unit: Days
Description: Wait for next business day for processing

Practical use cases:

Rate limiting: Avoid overloading external APIs
Wait for processing: Time for external systems to process
Simulate human time: Realistic delays in automations
Throughput control: Manage processing speed
Scheduling: Delays for specific times
Prioritization: Different delays based on priority

Configuration form:

Creating a Workflow Agent

Step by Step on the Platform

1. Start creation

On the Evo AI main screen, click “New Agent”
In the “Type” field, select “Workflow Agent”
You will see the React Flow visual interface

2. Configure basic information

Name: Descriptive workflow name

Example: content_approval_pipeline

Description: Complex workflow summary

Example: Complete workflow for content approval with 
automatic analysis, conditional review and final approval

Goal: Workflow objective

Example: Automate content approval process ensuring 
quality through multiple verifications and conditional approvals

3. Build visual workflow

React Flow Interface:Node Palette: Drag nodes from palette to canvas

🤖 Agent Node - To execute agents
💬 Message Node - For direct messages
🔀 Condition Node - For conditional logic
⏱️ Delay Node - For time delays

Connections: Connect nodes by dragging from outputs to inputsConfiguration: Double-click any node to configure

Content Nodes Menu:

Logic Nodes Menu:

4. Configure individual nodes

For each node in the workflow:Agent Nodes:

Select existing agent or create new
Configure input mapping
Define output key

Message Nodes:

Write message content
Use context variables
Configure message type

Condition Nodes:

Define verification logic
Configure true/false paths
Test conditions

Delay Nodes:

Configure duration
Define skip conditions
Configure dynamic delays

5. Connect and validate flow

Connections between nodes:

Drag from one node’s output to the next’s input
Validate that all connections make sense
Configure multiple outputs for Condition Nodes

Automatic validation:

Infinite loop verification
Data type validation
Orphan node checking
Complete path verification

Start and End nodes:

Every workflow needs a Start node
Configure End nodes for different scenarios

Practical Examples

1. Content Approval Pipeline

Workflow Structure

Objective: Automate content approval with multiple verificationsWorkflow flow:

START
  ↓
[Message: Initial Instruction]
  ↓
[Agent: Quality Analyzer]
  ↓
[Condition: Score >= 8.0?]
  ├─ TRUE → [Agent: Automatic Approval] → [Message: Content Approved] → END
  └─ FALSE → [Agent: Manual Review]
               ↓
             [Condition: Approved in Review?]
               ├─ TRUE → [Message: Content Approved] → END
               └─ FALSE → [Agent: Correction Request]
                           ↓
                         [Delay: 1 hour]
                           ↓
                         [back to Quality Analyzer]

Detailed configuration:1. Message Node: “Initial Instruction”

Type: Text
Content: Starting quality analysis of received content

2. Agent Node: “Quality Analyzer”

Agent: content_quality_analyzer
Input: content and quality criteria
Output Key: quality_analysis

3. Condition Node: “Score >= 8.0?”

Condition Type: Previous output
Output Field: Content
Operator: contains
Comparison Value: score >= 8.0
Preview: Content contains "score": 8

4. Agent Node: “Automatic Approval”

Agent: auto_approver
Input: content and quality analysis
Output Key: approval_result

5. Agent Node: “Manual Review”

Agent: manual_reviewer
Input: content and identified issues
Output Key: manual_review

6. Condition Node: “Approved in Review?”

Condition Type: Previous output
Output Field: Status
Operator: is equal to
Comparison Value: approved
Preview: Status is equal to "approved"

7. Delay Node: “1 hour”

Delay Value: 1
Time Unit: Hours
Description: Wait before new analysis

2. Lead Analysis Workflow

Workflow Structure

Objective: Process leads with intelligent analysis and automatic routingWorkflow flow:

START
  ↓
[Agent: Data Enricher]
  ↓
[Agent: Lead Scorer]
  ↓
[Condition: Score > 70?]
  ├─ TRUE → [Condition: Score > 90?]
  │           ├─ TRUE → [Agent: VIP Router] → [Delay: 5 min] → [Agent: Notifier] → END
  │           └─ FALSE → [Agent: Normal Router] → [Delay: 30 min] → [Agent: Notifier] → END
  └─ FALSE → [Message: Lead Discarded] → END

Main nodes:1. Agent Node: “Data Enricher”

Enriches lead data with public information
Output: enriched_data

2. Agent Node: “Lead Scorer”

Calculates score based on criteria
Input: enriched_data
Output: lead_score

3. Condition Node: “Score > 70?”

Filters low quality leads
Condition: score greater than 70

4. Condition Node: “Score > 90?”

Identifies VIP leads
Condition: score greater than 90

5. Agent Nodes: “Routers”

VIP Router: For senior salespeople
Normal Router: For standard salespeople

6. Delay Nodes: “Prioritization”

VIP: 5 minutes (high priority)
Normal: 30 minutes (standard priority)

3. Document Processing Workflow

Workflow Structure

Objective: Process documents with validation and automatic approvalComplex flow with loops:

START
  ↓
[Agent: Data Extractor]
  ↓
[Agent: Format Validator]
  ↓
[Condition: Valid Format?]
  ├─ FALSE → [Message: Format Error] → END
  └─ TRUE → [Agent: Content Analyzer]
              ↓
            [Condition: Complete Content?]
              ├─ FALSE → [Agent: Data Requester]
              │            ↓
              │          [Delay: Wait for Response]
              │            ↓
              │          [Condition: Data Received?]
              │            ├─ TRUE → [back to Content Analyzer]
              │            └─ FALSE → [Message: Timeout] → END
              └─ TRUE → [Agent: Compliance Checker]
                          ↓
                        [Condition: Compliance OK?]
                          ├─ TRUE → [Agent: Final Approver] → [Message: Document Approved] → END
                          └─ FALSE → [Agent: Rejecter] → [Message: Document Rejected] → END

Testing Features

Individual Agent Testing

Agent Form Test

Testing individual agents within the workflow:How to use:

Click on any Agent Node in the workflow
Select “Test Agent” from context menu
Configure test inputs
Execute isolated test
Analyze outputs and performance

Benefits:

Quick validation of individual agents
Debug specific issues
Test different inputs
Verify expected outputs

Test interface:

Agent Test: content_quality_analyzer

Input:
{
  "content": "Sample text for analysis...",
  "criteria": ["grammar", "clarity", "engagement"]
}

Expected Output:
{
  "score": 8.5,
  "issues": [],
  "recommendations": [...]
}

Complete Workflow Testing

Test Workflow

Complete workflow testing with real data:How to use:

Click “Test Workflow” in the top bar
Configure initial workflow inputs
Execute complete test
Monitor real-time execution
Analyze final results

Test features:

Step-by-step execution: See each node being executed
Data visualization: Track data flowing between nodes
Breakpoints: Pause execution at specific nodes
Detailed logs: Complete execution records
Performance metrics: Execution time per node

Test interface:

Workflow Test: content_approval_pipeline

Status: Running
Current Node: Agent - Quality Analyzer
Progress: 3/8 nodes completed

Execution Log:
├── [14:32:15] Start - Workflow initiated
├── [14:32:16] Message - Instruction sent
├── [14:32:17] Agent - Analysis started
└── [14:32:20] Agent - Analysis in progress...

Data Flow:
├── input_content: "Article about AI..."
├── start_message: "Starting analysis..."
└── quality_analysis: { score: 7.5, ... }

Workflow test interface:

Monitoring and Debugging

Execution Dashboard

Real-time Monitoring

Visual execution tracking:React Flow Visualization:

Active nodes highlighted in green
Completed nodes marked with ✓
Error nodes marked with ✗
Data flow visualized in connections

Metrics per node:

Node Performance:
├── Agent: Quality Analyzer
│   ├── Executions: 1,247
│   ├── Avg Duration: 3.2s
│   ├── Success Rate: 98.5%
│   └── Last Error: 2 days ago
├── Condition: Score >= 8.0
│   ├── True Path: 65% (812 times)
│   ├── False Path: 35% (435 times)
│   └── Avg Evaluation: 0.1s
└── Delay: 1 hour
    ├── Skipped: 15% (delays skipped)
    ├── Executed: 85%
    └── Avg Delay: 3600s

Alerts and notifications:

Execution failures
Node timeouts
Infinite loops detected
Performance degradation

Advanced Debugging

Specific debugging tools:1. Visual Breakpoints

- Click any node to add breakpoint
- Execution pauses before node
- Inspect current context state
- Continue step-by-step execution

2. Data Inspector

- Visualize data at each workflow point
- Transformation history
- Data type validation
- Corrupted data detection

3. Execution Replay

- Replay previous executions
- Compare between executions
- Identify regressions
- Analyze failure patterns

4. Performance Profiler

- Identify bottlenecks
- Resource usage analysis
- Optimize slow nodes
- Improvement suggestions

Advanced Settings

Performance Optimization

Execution Optimization

Strategies for efficient workflows:Intelligent Parallelization:

- Identify nodes that can execute in parallel
- Configure parallel branches in workflow
- Use merge nodes to synchronize results

Result Caching:

- Cache outputs from deterministic nodes
- Configure TTL based on data nature
- Invalidate cache when necessary

Resource Management:

- Limit resources per node
- Configure connection pools
- Manage memory between executions

Resilience and Reliability

Settings for robust workflows:Error Handling:

- Configure retry policies per node
- Implement fallback paths
- Use circuit breakers for external nodes

State Management:

- Persist state between executions
- Configure automatic checkpoints
- Implement recovery mechanisms

Monitoring and Alerting:

- Configure alerts for failures
- Monitor execution SLAs
- Implement health checks

Output Key - Workflow Result

Output Key field in interface:The Output Key allows the Workflow Agent to save the final result of the complex workflow in a specific variable in the shared state, making it available to other agents or processes.How it works:

Configure the Output Key field with a descriptive name
The final workflow result (last executed node) will be saved automatically
Other agents can access using placeholders {{output_key_name}}
Works in nested workflows, loops and multi-agent systems

Configuration examples:

Output Key: "content_approval_result"
→ Saves final result in state.content_approval_result

Output Key: "lead_processing_complete"
→ Saves final result in state.lead_processing_complete

Output Key: "data_pipeline_result"
→ Saves final result in state.data_pipeline_result

Workflow result:

{
  "workflow_summary": {
    "total_nodes": 12,
    "executed_nodes": 10,
    "skipped_nodes": 2,
    "execution_time": "2m 15s",
    "final_status": "completed"
  },
  "final_result": {
    "decision": "approved",
    "confidence": 0.95,
    "details": {...}
  },
  "execution_path": [
    "start_node",
    "validation_node", 
    "analysis_node",
    "decision_node",
    "end_node"
  ]
}

Usage in other agents:

# In subsequent agent instructions:
"Analyze the request: {{user_input}} and use approval result: {{content_approval_result}}"
"Based on processing: {{lead_processing_complete}}"
"Analyze processed data: {{data_pipeline_result}}"

Best practices:

Use snake_case: workflow_result, process_complete
Be specific: credit_approval_workflow instead of workflow
Document final result structure
Consider including execution metadata (path, timing, status)
Use names that reflect business process
Configure last node to produce structured result

Best Practices

Workflow Design

Principles for effective workflows:

Modularity: Divide complex workflows into sub-workflows
Idempotence: Ensure re-executions are safe
Observability: Add adequate logging and monitoring
Error handling: Handle failures gracefully
Performance: Optimize critical paths

Collaboration and Maintenance

Team practices:

Documentation: Document purpose of each node
Versioning: Use version control for workflows
Testing: Implement automated tests
Code review: Review workflows before production
Monitoring: Monitor performance and reliability

Common Use Cases

Content Approval

Quality Workflows:

Automatic quality analysis
Conditional approval
Correction loops

Lead Processing

Intelligent Qualification:

Data enrichment
Automatic scoring
Intelligent routing

Process Automation

Business Process Automation:

Approval workflows
Document processing
System integration

Data Analysis

Data Pipelines:

Complex ETL
Multi-step analysis
Automatic reports

Next Steps

Sequential Agent

Compare with simple sequential workflows

Parallel Agent

Understand basic parallel execution

Loop Agent

Explore simple iterative loops

A2A Agent

Integrate external agents in workflows

The Workflow Agent is the most powerful tool for creating complex automations. Use it when you need granular control over each step of the process and sophisticated business logic with multiple conditions and execution paths.

Get Started

Documentation Index

​Overview

​Key Features

Visual Interface

LangGraph Engine

Multiple Node Types

Integrated Testing

​Node Types

​1. Agent Node

​2. Message Node

​3. Condition Node

​4. Delay Node

​Creating a Workflow Agent

​Step by Step on the Platform

​Practical Examples

​1. Content Approval Pipeline

​2. Lead Analysis Workflow

​3. Document Processing Workflow

​Testing Features

​Individual Agent Testing

​Complete Workflow Testing

​Monitoring and Debugging

​Execution Dashboard

​Advanced Settings

​Performance Optimization

​Best Practices

​Common Use Cases

Content Approval

Lead Processing

Process Automation

Data Analysis

​Next Steps

Sequential Agent

Parallel Agent

Loop Agent

A2A Agent

Overview

Key Features

Node Types

1. Agent Node

2. Message Node

3. Condition Node

4. Delay Node

Creating a Workflow Agent

Step by Step on the Platform

Practical Examples

1. Content Approval Pipeline

2. Lead Analysis Workflow

3. Document Processing Workflow

Testing Features

Individual Agent Testing

Complete Workflow Testing

Monitoring and Debugging

Execution Dashboard

Advanced Settings

Performance Optimization

Best Practices

Common Use Cases

Next Steps