Agents

Understanding Agents in PraisonAI

Understanding Agents

Agents are the core building blocks of PraisonAI. Each agent is an autonomous AI entity with specific roles, goals, and capabilities.

Single Agent vs Multi-Agent

Use Case	Class	Code
One agent, one task	Agent	Agent(instructions="...").start()
Multiple agents working together	AgentTeam	AgentTeam(agents=[...]).start()

---

Agent Architecture

Key Components

Role & Goal

Defines the agent’s purpose and objectives through role definition and specific goals

Capabilities

Tools and functions available to the agent for task execution

Memory

Context retention and learning capabilities across interactions

Language Model

The underlying AI model powering the agent’s intelligence

Component Details

Role and Goal

Clear role and goal definitions are crucial for optimal agent performance.

Component	Description	Example
Role	Agent’s function and expertise	Research Analyst, Code Developer
Goal	Specific objectives to achieve	Analyze market trends, Generate reports
Backstory	Contextual background	Expert with 10 years of experience

Capabilities

Basic

agent = Agent(
    name="Researcher",
    role="Senior Research Analyst",  # Part of System Prompt
    goal="Uncover cutting-edge developments in AI",  # Part of System Prompt
    backstory="You are an expert at a technology research group",  # Part of System Prompt
    llm="gpt-4o"
)

1. Install PraisonAI

   Install the core package:

   Terminal

   pip install praisonaiagents
   
2. Configure Environment

   Terminal

   export OPENAI_API_KEY=your_openai_key
   

   Generate your OpenAI API key from OpenAI Use other LLM providers like Ollama, Anthropic, Groq, Google, etc. Please refer to the Models for more information.

3. Create Agent

   Create app.py:

   Single Agent

   from praisonaiagents import Agent
   
   agent = Agent(instructions="Your are a helpful AI assistant")  # System Prompt: defines WHO the agent is
   agent.start("Write a movie script about a robot in Mars")  # User Prompt: defines WHAT to do
   

   from praisonaiagents import Agent, AgentTeam
   
   research_agent = Agent(instructions="Research about AI")  # System Prompt
   summarise_agent = Agent(instructions="Summarise research agent's findings")  # System Prompt
   
   agents = AgentTeam(agents=[research_agent, summarise_agent])
   agents.start()  # User Prompt (uses task descriptions)
   
4. Start Agents

   Execute your script:

   Terminal

   python app.py
   

   You should see:

   • Agent initialization
   • Agents progress
   • Final results
   • Generated report

Agent Types

Basic Agent

Overview

Perfect for straightforward tasks and direct interactions

Features

• Single-purpose focus
• Direct user interaction
• Limited tool set
• Ideal for simple tasks

Specialized Agent

Overview

Experts in specific domains with advanced capabilities

Features

• Domain expertise
• Advanced capabilities
• Custom tools
• Deep knowledge base

Collaborative Agent

Overview

Designed for team-based operations and complex workflows

Features

• Team interaction
• Task delegation
• Shared context
• Coordinated actions

Best Practices

Always implement proper error handling and resource management in your agent configurations.

Agent Design

1. Role Definition

   Define clear, specific roles for each agent

2. Goal Setting

   Set specific, measurable goals

3. Tool Selection

   Choose relevant tools for the task

4. Memory Setup

   Configure appropriate memory settings

Agent Interaction

1. Communication

   Establish clear communication protocols

2. Delegation

   Define explicit delegation rules

3. Error Handling

   Implement robust error handling

4. Resource Management

   Set up efficient resource allocation

Async Capabilities

Key Features

Async Support

async def main():
    agent = Agent(name="AsyncAgent")
    result = await agent.aprocess_task()

• Full async/await support
• Non-blocking operations
• Enhanced performance

Parallel Execution

async def parallel_tasks():
    tasks = [agent1.task(), agent2.task()]
    results = await asyncio.gather(*tasks)

• Parallel task execution
• Efficient resource usage
• Improved throughput

Integration

@agent.async_tool
async def custom_tool():
    async with aiohttp.ClientSession() as session:
        # Async operations
        pass

• Async tool integration
• Callback support
• Mixed sync/async operations

Advanced Features

Memory Management

• Short-term conversation memory
• Long-term knowledge retention
• Context preservation

Tool Integration

• Custom tool development
• External API integration
• Resource access control

Async Support

Agents now support asynchronous operations through the following methods:

• achat: Async version of the chat method
• astart: Async version of start method
• aexecute_task: Async version of execute_task method
• arun_task: Async version of run_task method
• arun_all_tasks: Async version of run_all_tasks method

Example Usage:

import asyncio
from praisonaiagents import Agent, Task, AgentTeam

async def main():
    # Create an async agent
    async_agent = Agent(
        name="AsyncAgent",
        role="Async Task Specialist",
        goal="Perform async operations",
        backstory="Expert in async operations",
        tools=[async_search_tool],  # Your async tool
        
    )

    # Create an async task
    async_task = Task(
        description="Perform async operation",
        expected_output="Async result",
        agent=async_agent,
        async_execution=True  # Enable async execution
    )

    # Create and start agents with async support
    agents = AgentTeam(
        agents=[async_agent],
        tasks=[async_task],
        
    )
    
    # Start async execution
    result = await agents.astart()
    print(result)

# Run the async main function
if __name__ == "__main__":
    asyncio.run(main())

Key Features:

• Full async/await support
• Parallel task execution
• Async tool integration
• Async callback support
• Mixed sync/async operations

Next Steps

Create Your First Agent

Follow our quickstart guide to create your first agent

API Reference

Explore the complete Agent API documentation