Agent Types

Learn about all the agent types available on the Evo AI platform and choose the ideal one for your use case

Overview

The Evo AI platform offers different types of agents, each optimized for specific use scenarios. From simple LLM-based agents to complex workflows with multiple steps, you can choose the approach that best adapts to your needs.

Smart Choice: Each agent type has been designed to solve specific problems. Use this guide to identify which type best serves your use case.

Available Types

Fundamental Agents

LLM Agent

Base agent with LLM

Fundamental agent based on Large Language Models, ideal for conversations, text analysis, and tasks requiring natural language understanding.

When to use:

• Intelligent conversations
• Text analysis
• Content generation
• NLP tasks

Features:

• Chat interface
• Multiple LLM models
• Personality configuration
• Sub-agent system

A2A Agent

Integration with A2A protocol

Integrates external agents that implement Google’s Agent-to-Agent (A2A) protocol, enabling interoperability with external systems.

When to use:

• Integration with legacy systems
• External specialized agents
• AI microservices
• Third-party APIs

Features:

• JSON-RPC 2.0 protocol
• Standardized communication
• Automatic health checks
• Circuit breakers

Orchestration Agents

Sequential Agent

Ordered sequential execution

Executes multiple sub-agents in ordered sequence, where each agent receives the output of the previous one as input.

When to use:

• Processing pipelines
• Linear workflows
• Dependencies between steps
• Data transformation

Features:

• Step-by-step execution
• Data passing between agents
• Failure control
• Detailed logs

Parallel Agent

Simultaneous parallel execution

Executes multiple sub-agents simultaneously, ideal for independent tasks that can be processed in parallel.

When to use:

• Independent analyses
• Distributed processing
• Time optimization
• Unrelated tasks

Features:

• Simultaneous execution
• Result aggregation
• Concurrency control
• Significant speedup

Loop Agent

Iterative execution with refinement

Executes sub-agents in iterative loops, refining results with each iteration until reaching convergence criteria.

When to use:

• Iterative refinement
• Parameter optimization
• Progressive improvement
• Result convergence

Features:

• Controlled iterations
• Stop criteria
• Convergence monitoring
• Infinite loop prevention

Workflow Agent

Complex visual workflows

Creates complex workflows with visual interface using LangGraph and React Flow, offering maximum control and flexibility.

When to use:

• Complex business logic
• Conditions and branching
• Visual workflows
• Granular control

Features:

• Drag-and-drop interface
• Multiple node types
• Conditions and delays
• Integrated testing

Specialized Agents

Task Agent

Specific and structured tasks

Inspired by CrewAI, allows assigning specific tasks to agents with structured prompts and well-defined expected outputs.

When to use:

• Well-defined tasks
• Single responsibility
• Structured output
• Agent specialization

Features:

• One task per agent
• Structured prompt
• Defined expected output
• Automatic validation

Quick Comparison

Type	Complexity	Use Cases	Execution	Configuration
LLM Agent	Low	Chat, text analysis	Single	Simple
A2A Agent	Medium	External integration	Single	Moderate
Sequential	Medium	Linear pipelines	Sequential	Moderate
Parallel	Medium	Independent analyses	Parallel	Moderate
Loop	High	Iterative refinement	Iterative	Complex
Workflow	Very High	Complex logic	Visual/Conditional	Very Complex
Task	Low-Medium	Specific tasks	Single	Simple-Moderate

Selection Guide

By Problem Complexity

Simple Problems

For direct tasks and conversations:

• LLM Agent: Chat, text analysis, content generation
• Task Agent: Specific tasks with structured output

Characteristics:

• Quick setup
• Direct execution
• Ideal for getting started

Medium Problems

For processes with multiple steps:

• Sequential Agent: When steps depend on each other
• Parallel Agent: When steps are independent
• A2A Agent: When you need to integrate external systems

Characteristics:

• Multi-agent orchestration
• Basic flow control
• Reuse of existing agents

Complex Problems

For sophisticated business logic:

• Loop Agent: When you need to refine results iteratively
• Workflow Agent: When you need granular control and conditions

Characteristics:

• Advanced conditional logic
• Sophisticated flow control
• Detailed monitoring

By Execution Pattern

Single Execution

Agents that execute once:

• LLM Agent: Direct response to a question
• A2A Agent: Call to external system
• Task Agent: Execution of specific task

Ideal for:

• Point queries
• Simple analyses
• Direct integrations

Sequential Execution

Agents that execute in order:

• Sequential Agent: Pipeline with dependencies

Ideal for:

• Data transformation
• Linear workflows
• Step-by-step processing

Parallel Execution

Agents that execute simultaneously:

• Parallel Agent: Independent analyses

Ideal for:

• Time optimization
• Comparative analyses
• Distributed processing

Iterative Execution

Agents that execute in loops:

• Loop Agent: Progressive refinement

Ideal for:

• Parameter optimization
• Iterative improvement
• Result convergence

Conditional Execution

Agents with complex logic:

• Workflow Agent: Flows with conditions and branching

Ideal for:

• Complex business logic
• Conditional decisions
• Visual workflows

Use Case Examples

E-commerce

Product Analysis

Scenario: Complete product analysis for optimization

Recommended solution: Parallel Agent

• Price analysis (specialized sub-agent)
• Review analysis (sentiment sub-agent)
• Specification analysis (technical sub-agent)
• Competition analysis (market sub-agent)

Why Parallel? Independent analyses that can be executed simultaneously to optimize time.

Lead Qualification

Scenario: Sales lead qualification pipeline

Recommended solution: Sequential Agent

• Data enrichment (public data)
• Lead scoring (scoring algorithm)
• Priority classification (high/medium/low)
• Salesperson routing (based on specialization)

Why Sequential? Each step depends on the previous one to function correctly.

Content and Marketing

Content Creation

Scenario: Optimized blog article generation

Recommended solution: Loop Agent

• Initial content generation
• SEO analysis and optimization
• Quality and readability review
• Refinement based on feedback

Why Loop? Iterative improvement until reaching desired quality.

Performance Analysis

Scenario: Campaign performance report

Recommended solution: Task Agent

• Specific task: generate metrics report
• Input: campaign data
• Output: structured report with insights

Why Task? Well-defined task with specific structured output.

Customer Service

Smart Support

Scenario: Support system with escalation

Recommended solution: Workflow Agent

• Initial request classification
• Automatic resolution attempt
• Condition: resolved? → End : Escalation
• Routing to specialized human agent

Why Workflow? Complex conditional logic with multiple paths.

Basic Chat

Scenario: Simple chat for frequent questions

Recommended solution: LLM Agent

• Direct conversation with customer
• Integrated knowledge base
• Contextual responses

Why LLM? Direct and simple interaction, no need for orchestration.

Output Keys - State Sharing

New Feature: All agent types now support Output Key for state sharing between agents and processes.

What are Output Keys?

The Output Key is a field available in all agent types that allows saving the execution result in a specific variable in the shared state. This facilitates communication between agents and building complex systems.

LLM Agent

Saves model response

The response generated by the LLM is saved in the specified variable, allowing other agents to access the generated content.

Output Key: "analysis_response"
→ state.analysis_response

Task Agent

Saves task result

The structured result of the task is saved in the variable, allowing use in subsequent tasks or workflows.

Output Key: "processed_data"
→ state.processed_data

A2A Agent

Saves external response

The response from the external system via A2A protocol is saved in the variable, enabling integration with other agents.

Output Key: "external_result"
→ state.external_result

Workflow Agents

Saves workflow result

The final result of the workflow (Sequential, Parallel, Loop, Workflow) is saved in the specified variable.

Output Key: "workflow_result"
→ state.workflow_result

How to Use

Configuration

In the platform interface:

1. Configure the Output Key in any agent
2. Use a descriptive name in snake_case
3. The result is saved automatically in the state
4. Other agents access via placeholders {{output_key_name}}

Example:

Agent 1 - Output Key: "initial_analysis"
Agent 2 - Instructions: "Analyze the request: {{user_input}} and base it on the analysis: {{initial_analysis}}"

Data Flow

How data flows between agents:

Initial state:
{
  "user_input": "Analyze this product"
}

After Agent 1 (output_key: "analysis"):
{
  "user_input": "Analyze this product",
  "analysis": "High quality product..."
}

After Agent 2 (output_key: "recommendation"):
{
  "user_input": "Analyze this product",
  "analysis": "High quality product...",
  "recommendation": "I recommend the purchase because..."
}

Best Practices

Tips for using Output Keys effectively:

• Clear naming: Use names that describe the content
• Snake_case: analysis_result, processed_data
• Be specific: sentiment_analysis_reviews instead of analysis
• Document format: Explain expected structure in instructions
• Avoid conflicts: Don’t use names already existing in state
• Consider hierarchy: Use prefixes to group related data

Agent Management

Export and Import

Exporting agents

Backup and sharing functionality:

In the agent listing, you can export any agent to a JSON file that contains the complete configuration:

What is exported:

• ✅ Complete configurations of the main agent
• ✅ Sub-agents included automatically in the same file
• ✅ All dependencies and nested configurations
• ✅ Hierarchical structure preserved
• ✅ Tools and integrations configured

How to export:

1. In the agent list, locate the desired agent
2. Click on the actions menu (⋯) of the agent
3. Select “Export”
4. JSON file will be downloaded automatically

Example of exported structure:

{
  "name": "copywriter",
  "description": "Creative writer specialized in innovative copywriting.",
  "role": null,
  "goal": null,
  "type": "llm",
  "model": "openai/gpt-4.1-nano",
  "instruction": "You are a senior copywriter with absolute mastery of persuasion techniques",
  "config": {
    "output_key": "new_copy",
    "tools": null,
    "custom_tools": null,
    "mcp_servers": null,
    "custom_mcp_servers": null,
    "agent_tools": null,
    "sub_agents": null,
    "workflow": null
  },
  "id": "7fd77e46-dfcb-40a2-b71f-d2bac56956ac"
}

Importing agents

Restoring complete configurations:

How to import:

1. In the agent list, click “Import Agent”
2. Select the JSON file exported previously
3. System validates structure and dependencies
4. Main agent and sub-agents are created automatically
5. Configurations applied exactly as exported

Import advantages:

• 🚀 Instant setup of complex agents
• 🎯 Exact replication of tested configurations
• 👥 Sharing between teams
• 🔄 Backup and restore of configurations
• 📦 Migration between environments

Automatic validations:

• JSON structure verification
• Agent type validation
• Dependency checking
• Available model confirmation
• Required permissions verification

Use cases:

✅ Backup of critical agents
✅ Replication in production environment
✅ Template sharing
✅ Migration between organizations
✅ Configuration versioning

Folder System

Organizing agents in folders

Hierarchical organization:

The folder system allows organizing your agents hierarchically and collaboratively:

Main features:

• 📁 Visual organization of agents
• 👥 Sharing with other users
• 🔐 Permission control (view/edit)
• 🏷️ Categorization by project, client, or function
• 🔍 Search within specific folders

How to create folders:

1. In the agents screen, click “New Folder”
2. Folder name: Use descriptive name
3. Description: Summary of folder content
4. Access settings: Define who can access
5. Click “Create” to create

Organization examples:

📁 Sales
├── 🤖 lead_qualifier
├── 🤖 product_demo
└── 🤖 closer_specialist

📁 Technical Support  
├── 🤖 troubleshooter
├── 🤖 documentation_helper
└── 🤖 escalation_manager

📁 Marketing
├── 🤖 content_creator
├── 🤖 social_media_manager
└── 🤖 campaign_analyzer

Folder sharing

Collaboration between users:

How to share a folder:

1. Access the folder you want to share
2. Click the “Share” icon (👥)
3. Sharing form will open

Configuring permissions:

1. Enter the email of the Evo AI user
2. Select access level:
   • 👁️ Viewer: Can see and use agents, but not edit
   • ✏️ Editor: Can see, use, edit and create new agents in the folder
3. Add optional message explaining the sharing
4. Click “Share” to send invitation

Permission levels:

Viewer:

✅ View list of agents in folder
✅ Use agents in conversations
✅ View configurations (read-only)
❌ Edit agent configurations
❌ Create new agents
❌ Delete agents
❌ Share folder with others

Editor:

✅ All Viewer permissions
✅ Edit agent configurations
✅ Create new agents in folder
✅ Delete agents (that they created)
✅ Move agents between folders
❌ Share folder with others (owner only)
❌ Delete the folder

Team collaboration

Use cases for sharing:

By department:

📁 "Company ABC Sales" Folder
Shared with: [e-mail]
Permission: Editor
Use: Entire team can create/edit sales agents

By project:

📁 "Client XYZ Project" Folder  
Shared with: [e-mail]
Permission: Viewer
Use: Manager monitors project agents

By specialty:

📁 "Technical AI Agents" Folder
Shared with: [e-mail]
Permission: Editor
Use: Technical team develops specialized agents

For clients:

📁 "Final Client Agents" Folder
Shared with: [e-mail]
Permission: Viewer
Use: Client tests developed agents

Collaboration advantages:

• 🤝 Teamwork facilitated
• 🔄 Automatic synchronization of updates
• 📊 Visibility of team work
• 🎯 Specialization by area/project
• 📈 Development scalability
• 🔐 Granular access control

Access management

Managing permissions:

As folder owner, you can:

• ✅ View all users with access
• ✅ Change permissions of any user
• ✅ Remove access from specific users
• ✅ Transfer ownership of folder
• ✅ Delete folder (removes everyone’s access)

Automatic notifications:

• 📧 Invitation email when folder is shared
• 🔔 Notification when agent is modified by another user
• 📊 Activity report of shared folder

Audit and control:

• 📝 Activity log of all users
• 👤 Modification history with responsible party
• ⏰ Timestamps of all actions
• 🔍 Complete traceability of changes

Best practices:

• Use Viewer permission by default
• Grant Editor only when necessary
• Document the purpose of folder in name/description
• Review permissions periodically
• Remove access from inactive users

---

Tip: Always start with the simplest type that meets your use case. You can evolve to more complex types as your needs grow.

Next Steps

Start with LLM Agent

Ideal for beginners - create your first conversational agent

Explore Workflows

For complex cases - build advanced visual workflows

A2A Protocol

Understand the protocol for integration with external systems

Quick Guide

Set up your first agent in less than 2 minutes