Agentflows - Flowise

What are Agentflows?

Agentflows are Flowise’s advanced workflow system designed for multi-agent systems, workflow orchestration, and complex AI automation. Unlike chatflows (single-agent systems), agentflows enable multiple AI agents to work together, each with specialized roles and capabilities.

Agentflows are ideal for complex workflows requiring coordination between multiple specialized agents, sequential processing, or advanced decision-making logic.

Agentflow Versions

Flowise supports two versions of agentflows:

V2 (Recommended)

The latest version with improved performance, better agent coordination, and enhanced features:

Enhanced agent communication protocols
Better memory management across agents
Improved error handling and recovery
Advanced workflow orchestration
New agent types and capabilities

All new agentflow projects should use V2. V1 is maintained for backward compatibility but is deprecated.

V1 (Deprecated)

The original multi-agent system. While still functional, it’s recommended to migrate to V2:

// V1 Deprecation Notice
if (agentflowVersion === 'v1' && showDeprecationNotice) {
  // Warning: "V1 Agentflows are deprecated. 
  // We recommend migrating to V2 for improved performance 
  // and continued support."
}

Why Agentflows Matter

Agentflows unlock capabilities beyond simple chatbots:

Specialized Agents: Each agent focuses on specific tasks (research, writing, code review)
Workflow Orchestration: Define complex, multi-step processes
Parallel Processing: Multiple agents work simultaneously
Advanced Decision Making: Agents can route tasks based on context
Enterprise Scale: Handle complex business processes

Agentflow vs Chatflow

Feature	Chatflow	Agentflow
Agent Count	Single	Multiple
Complexity	Simple to Moderate	Moderate to Complex
Use Case	Q&A, RAG, Support	Research, Analysis, Automation
Coordination	N/A	Built-in orchestration
Best For	Direct interactions	Multi-step workflows

Core Structure

Type Definitions

enum EnumChatflowType {
  CHATFLOW = 'CHATFLOW',      // Single agent
  AGENTFLOW = 'AGENTFLOW',    // Multi-agent V2
  MULTIAGENT = 'MULTIAGENT',  // Multi-agent V1 (deprecated)
  ASSISTANT = 'ASSISTANT'     // OpenAI Assistant
}

FlowData Schema

Agentflows extend the basic chatflow structure:

{
  id: string
  name: string
  flowData: string            // Contains agent nodes and orchestration logic
  type: 'AGENTFLOW' | 'MULTIAGENT'
  category: string
  deployed: boolean
  workspaceId: string
  createdDate: Date
  updatedDate: Date
}

Agent Node Types

Agentflows use specialized node types:

1. Sequential Agents

Agents execute tasks in a defined order:

2. Supervisor Pattern

A supervisor agent coordinates worker agents:

3. State Machine Agents

Agents transition through defined states:

// Example state transitions
states: {
  'research': ['analyze', 'gather_more'],
  'analyze': ['report', 'research'],
  'report': ['complete', 'revise']
}

Creating an Agentflow

Step 1: Choose Version

Select V2 for new projects:

// Navigation routing based on version
const addNew = () => {
  if (agentflowVersion === 'v2') {
    navigate('/v2/agentcanvas')
  } else {
    navigate('/agentcanvas')  // V1 - deprecated
  }
}

Step 2: Design Agent Roles

Define what each agent will do: Example: Content Creation Pipeline

Research Agent: Gathers information from multiple sources
Writer Agent: Creates initial draft based on research
Editor Agent: Reviews and improves content
SEO Agent: Optimizes for search engines
Fact-Checker Agent: Verifies claims and citations

Step 3: Build the Flow

Step 4: Configure Agent Communication

Agents share context through:

Shared Memory: All agents access common conversation history
State Passing: Explicit data transfer between agents
Tool Results: Output from one agent becomes input to another

Real-World Example: Customer Support Automation

// Agentflow configuration
{
  "name": "Multi-Tier Support System",
  "type": "AGENTFLOW",
  "agents": [
    {
      "name": "Triage Agent",
      "role": "Classify incoming support requests",
      "tools": ["ticket_classifier", "urgency_detector"],
      "nextAgents": ["FAQ Agent", "Technical Agent", "Escalation Agent"]
    },
    {
      "name": "FAQ Agent",
      "role": "Handle common questions",
      "tools": ["knowledge_base_search"],
      "condition": "confidence > 0.8"
    },
    {
      "name": "Technical Agent",
      "role": "Debug technical issues",
      "tools": ["log_analyzer", "system_checker"],
      "condition": "issue_type == 'technical'"
    },
    {
      "name": "Escalation Agent",
      "role": "Route to human support",
      "tools": ["ticket_creator", "email_notifier"],
      "condition": "requires_human == true"
    }
  ]
}

Agent Icons & Visualization

Agentflows use special icons to distinguish agent types:

// From source code
const AGENTFLOW_ICONS = [
  { name: 'supervisorAgent', icon: '👔' },
  { name: 'workerAgent', icon: '👷' },
  { name: 'sequentialAgent', icon: '📋' },
  { name: 'stateAgent', icon: '🔄' }
]

// Icon resolution logic
const foundIcon = AGENTFLOW_ICONS.find(
  (icon) => icon.name === node.data.name
)
if (foundIcon) {
  icons[agentflowId].push(foundIcon)
} else {
  // Fallback to standard node icon
  const imageSrc = `${baseURL}/api/v1/node-icon/${node.data.name}`
  images[agentflowId].push({ imageSrc, label: node.data.label })
}

Agentflow Features

1. Version Toggle

Switch between V1 and V2 agentflows:

<ToggleButtonGroup value={agentflowVersion}>
  <ToggleButton value='v2'>
    <Chip label='NEW' /> V2
  </ToggleButton>
  <ToggleButton value='v1'>
    V1
  </ToggleButton>
</ToggleButtonGroup>

2. Agent Memory Management

Agentflows support advanced memory patterns:

Individual Memory: Each agent maintains its own context
Shared Memory: All agents access global state
Hierarchical Memory: Parent agents pass context to children

3. Conditional Routing

Route to different agents based on conditions:

if (intent === 'technical_question') {
  routeToAgent('TechnicalExpertAgent')
} else if (intent === 'billing') {
  routeToAgent('BillingAgent')
} else {
  routeToAgent('GeneralSupportAgent')
}

Performance Considerations

Agentflows involve multiple LLM calls, which can impact:

Latency: Sequential agents increase response time
Cost: More agents = more API calls
Complexity: Debugging multi-agent interactions is harder

Optimization Tips

Parallel Processing: Run independent agents simultaneously
Smart Routing: Skip unnecessary agents with conditions
Caching: Store intermediate results
Model Selection: Use faster/cheaper models for simple agents

Common Agentflow Patterns

Pattern 1: Research Pipeline

Query Agent → Search Agent → Synthesize Agent → Fact Check Agent → Report Agent

Pattern 2: Code Review System

Code Parser → Linter Agent → Security Agent → Performance Agent → Style Agent → Summary Agent

Pattern 3: Content Moderation

Content Intake → Toxicity Agent → Bias Agent → Factuality Agent → Decision Agent → Action

Migration from V1 to V2

To migrate an existing V1 agentflow:

Export your V1 agentflow JSON
Create a new V2 agentflow
Recreate agents with updated V2 nodes
Test thoroughly - agent behavior may differ
Update any API integrations

API Reference

List Agentflows

# V2 Agentflows
GET /api/v1/chatflows?type=AGENTFLOW

# V1 Agentflows (deprecated)
GET /api/v1/chatflows?type=MULTIAGENT

Get Specific Agentflow

GET /api/v1/chatflows/:id

Determine Canvas Route

const goToCanvas = (selectedAgentflow) => {
  if (selectedAgentflow.type === 'AGENTFLOW') {
    navigate(`/v2/agentcanvas/${selectedAgentflow.id}`)
  } else {
    navigate(`/agentcanvas/${selectedAgentflow.id}`)  // V1
  }
}

Chatflows - For single-agent systems
Nodes and Edges - Building blocks of flows
Credentials - API keys for agent tools
Variables - Dynamic configuration

Troubleshooting

Agent Not Executing

Check agent dependencies are met
Verify tool configurations
Ensure proper routing conditions
Check agent execution order

Infinite Loops

Review state transitions
Add maximum iteration limits
Implement proper exit conditions

Memory Issues

Clear shared memory between runs
Limit conversation history length
Use memory summarization for long interactions

Get Started

Core Concepts

Building Workflows

Integrations

Features

Deployment

Development

​What are Agentflows?

​Agentflow Versions

​V2 (Recommended)

​V1 (Deprecated)

​Why Agentflows Matter

​Agentflow vs Chatflow

​Core Structure

​Type Definitions

​FlowData Schema

​Agent Node Types

​1. Sequential Agents

​2. Supervisor Pattern

​3. State Machine Agents

​Creating an Agentflow

​Step 1: Choose Version

​Step 2: Design Agent Roles

​Step 3: Build the Flow

​Step 4: Configure Agent Communication

​Real-World Example: Customer Support Automation

​Agent Icons & Visualization

​Agentflow Features

​1. Version Toggle

​2. Agent Memory Management

​3. Conditional Routing

​Performance Considerations

​Optimization Tips

​Common Agentflow Patterns

​Pattern 1: Research Pipeline

​Pattern 2: Code Review System

​Pattern 3: Content Moderation

​Migration from V1 to V2

​API Reference

​List Agentflows

​Get Specific Agentflow

​Determine Canvas Route

​Related Concepts

​Troubleshooting

​Agent Not Executing

​Infinite Loops

​Memory Issues

Build docs developers (and LLMs) love

What are Agentflows?

Agentflow Versions

V2 (Recommended)

V1 (Deprecated)

Why Agentflows Matter

Agentflow vs Chatflow

Core Structure

Type Definitions

FlowData Schema

Agent Node Types

1. Sequential Agents

2. Supervisor Pattern

3. State Machine Agents

Creating an Agentflow

Step 1: Choose Version

Step 2: Design Agent Roles

Step 3: Build the Flow

Step 4: Configure Agent Communication

Real-World Example: Customer Support Automation

Agent Icons & Visualization

Agentflow Features

1. Version Toggle

2. Agent Memory Management

3. Conditional Routing

Performance Considerations

Optimization Tips

Common Agentflow Patterns

Pattern 1: Research Pipeline

Pattern 2: Code Review System

Pattern 3: Content Moderation

Migration from V1 to V2

API Reference

List Agentflows

Get Specific Agentflow

Determine Canvas Route

Related Concepts

Troubleshooting

Agent Not Executing

Infinite Loops

Memory Issues