langgroup 0.2.3

A multiagent system framework built on LangChain and LangGraph with supervisor-based coordination

LangGroup

A multiagent system framework built on LangChain and LangGraph with supervisor-based coordination.

LangGroup provides a flexible architecture for creating groups of specialized AI agents that collaborate on complex tasks under the guidance of a supervisor agent.

Features

• Supervisor Architecture: Intelligent task routing and coordination
• Hierarchical Supervisors: Supervisors can manage other supervisors, enabling nested group structures
• Extensible Agent System: Easy-to-extend base classes for custom agents
• LangGraph Integration: Stateful workflows with LangGraph
• Type-Safe: Full type hints and Pydantic models

Installation

pip install langgroup

For development:

pip install -e ".[dev]"

Quick Start

1. Set up your environment:

cp .env.example .env
# Edit .env and add your OPENAI_API_KEY

2. Create your custom agents by extending BaseAgent:

from langgroup import BaseAgent
from typing import List, Callable

class MyAgent(BaseAgent):
    @property
    def description(self) -> str:
        return "Description of what this agent does"
    
    @property
    def tools(self) -> List[Callable]:
        return [my_tool_function]
    
    @property
    def system_prompt(self) -> str:
        return "System prompt for the agent"

3. Set up the agent system:

from langgroup import AgentSystem
from langchain_openai import ChatOpenAI

llm = ChatOpenAI(model="gpt-4o-mini", temperature=0)
agents = [MyAgent(llm), AnotherAgent(llm)]
system = AgentSystem(llm, agents)

result = system.run("Your task here")

Hierarchical Supervisors

💡 Key Feature: SupervisorAgent can be used as a regular agent within another AgentSystem, enabling powerful hierarchical group structures.

Create groups of agents for complex workflows:

from langgroup import AgentSystem, SupervisorAgent
from langchain_openai import ChatOpenAI

llm = ChatOpenAI(model="gpt-4o-mini", temperature=0)

# Create specialized groups
research_group = [ResearchAgent(llm), AnalysisAgent(llm)]
research_supervisor = SupervisorAgent(
    llm, research_group, 
    name="ResearchGroupSupervisor"
)

content_group = [WritingAgent(llm), EditingAgent(llm)]
content_supervisor = SupervisorAgent(
    llm, content_group, 
    name="ContentGroupSupervisor"
)

# Top-level supervisor coordinates the groups
top_system = AgentSystem(llm, [research_supervisor, content_supervisor])
result = top_system.run("Research AI trends and write a comprehensive report")

How it works:

1. The top-level supervisor receives the task
2. It intelligently routes to the appropriate group supervisor (e.g., ResearchGroupSupervisor)
3. The group supervisor manages its specialized agents
4. Results flow back up to coordinate between groups
5. Complex multi-stage tasks are handled seamlessly

This architecture allows you to build sophisticated agent organizations with clear separation of concerns.

Usage

Single Agent (Basic)

Run the basic agent:

python main.py

The agent includes two tools:

• Calculator: Performs mathematical calculations
• Weather: Returns weather information (mock implementation)

Multiagent System with Supervisor

Run the multiagent system:

python multiagent_supervisor.py

The multiagent system includes:

• Supervisor Agent: Orchestrates and routes tasks to specialized agents
• Research Agent: Handles research and information gathering
• Analysis Agent: Analyzes data and provides insights
• Writing Agent: Writes and formats content
• Math Agent: Performs calculations

The supervisor dynamically decides which agents to invoke and coordinates their work, allowing agents to collaborate back and forth on complex tasks.

Architecture

The multiagent system uses LangGraph to create a stateful workflow where:

1. Supervisor receives the task
2. Supervisor routes to appropriate sub-agent
3. Sub-agent completes its work
4. Control returns to supervisor
5. Process repeats until task is complete

Customization

Add your own tools by creating functions and registering them with the Tool class in either main.py or multiagent_supervisor.py.