microAgents 1.3.4

A lightweight LLM orchestration framework for building Multi-Agent AI systems.

microAgents Framework

A lightweight LLM orchestration framework for building Multi-Agent AI systems. The framework provides an easy way to create and orchestrate multiple AI agents with XML-style tool calls.

Key Features

🚀 Universal Tool Calling Support

• Works with ANY LLM API that follows OpenAI-compatible format
• Unique Feature: Enables function/tool calling even with models that don't natively support it
• XML-based tool calling format that's intuitive and human-readable

Framework Comparison

Framework	Core Abstractions	Size & Complexity	Dependencies & Integration	Key Advantages	Limitations/Trade-offs
LangChain	Agent, Chain	405K LOC +166MB	Many vendor wrappers (OpenAI, Pinecone, etc) Many app wrappers (QA, Summarization)	Rich ecosystem Extensive tooling Large community	Heavy footprint Complex setup JSON schema based
CrewAI	Agent, Chain	18K LOC +173MB	Many vendor & app wrappers (OpenAI, Anthropic, etc)	Role-based agents Built-in collaboration	Complex hierarchies Heavy dependencies
SmolAgent	Agent	8K LOC +198MB	Some integrations (DuckDuckGo, HuggingFace)	Simplified agent design	Limited tool ecosystem Large package size
LangGraph	Agent, Graph	37K LOC +51MB	Some DB integrations (PostgresStore, SqliteSaver)	Graph-based flows DAG support	Complex DAG definitions JSON schema based
AutoGen	Agent	7K LOC +26MB (core)	Optional integrations (OpenAI, Pinecone)	Lightweight core Modular design	Limited built-in tools
microAgents	Agent, Tool	~2K LOC <1MB	Minimal (requests, urllib3)	✓ Universal tool calling ✓ XML-based format ✓ Ultra lightweight ✓ Simple integration ✓ Any OpenAI-compatible LLM	Bring your own tools No built-in vendors

Key Differentiators

• Ultra Lightweight: microAgents is <1MB, compared to hundreds of MB for other frameworks
• Universal Compatibility: Works with any OpenAI-compatible API endpoint
• XML Tool Calls: More readable and intuitive than JSON schemas
• Minimal Dependencies: Only core HTTP libraries required
• Simple Integration: Direct function integration without wrapper classes
• LLM Agnostic: Works with any LLM that follows OpenAI's API format, including those without native function calling

Installation

You can install microAgents directly from PyPI:

pip install microAgents

Or install from source for development:

git clone https://github.com/prabhjots664/MicroAgents.git
cd MicroAgents
pip install -e .

Quick Start

Here's a complete example showing how to create a multi-agent math system:

from microAgents.llm import LLM
from microAgents.core import MicroAgent, Tool, MessageStore

# Initialize LLM with your API
llm = LLM(
    base_url="https://api.hyperbolic.xyz/v1",
    api_key="eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiJrYW1hbHNpbmdoZ2FsbGFAZ21haWwuY29tIiwiaWF0IjoxNzM1MjI2ODIzfQ.1wZmIzTZUWLzr-uP7Qtib_kkXNZmH_yQtSn1lP9S2z0",
    model="Qwen/Qwen2.5-Coder-32B-Instruct",
    max_tokens=4000,
    temperature=0.8,
    top_p=0.9
)

# Define tools for basic math operations
def add_numbers(a: float, b: float) -> float:
    return a + b

def multiply_numbers(a: float, b: float) -> float:
    return a * b

# Create specialized agents
math_agent = MicroAgent(
    llm=llm,
    prompt="You are a math assistant. Handle basic arithmetic operations.",
    toolsList=[
        Tool(description="Add two numbers", func=add_numbers),
        Tool(description="Multiply two numbers", func=multiply_numbers)
    ]
)

# Create message store for conversation history
message_store = MessageStore()

# Use the agent
response = math_agent.execute_agent(
    "First add 3 and 5, then multiply the result by 2", 
    message_store
)
print(response)

Multi-Agent Orchestration Example

Here's an example of creating multiple specialized agents and orchestrating them:

from microAgents.llm import LLM
from microAgents.core import MicroAgent, Tool, MessageStore

# Initialize LLM
llm = LLM(
    base_url="https://api.hyperbolic.xyz/v1",
    api_key="eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiJrYW1hbHNpbmdoZ2FsbGFAZ21haWwuY29tIiwiaWF0IjoxNzM1MjI2ODIzfQ.1wZmIzTZUWLzr-uP7Qtib_kkXNZmH_yQtSn1lP9S2z0",
    model="Qwen/Qwen2.5-Coder-32B-Instruct",
    max_tokens=4000,
    temperature=0.8,
    top_p=0.9
)

# Define tools for different agents
def add_numbers(a: float, b: float) -> float:
    return a + b

def multiply_numbers(a: float, b: float) -> float:
    return a * b

def factorial(n: int) -> int:
    if n == 0:
        return 1
    return n * factorial(n - 1)

# Create specialized agents
simple_math_agent = MicroAgent(
    llm=llm,
    prompt="You are a simple math assistant. Handle basic arithmetic operations.",
    toolsList=[
        Tool(description="Add two numbers", func=add_numbers),
        Tool(description="Multiply two numbers", func=multiply_numbers)
    ]
)

advanced_math_agent = MicroAgent(
    llm=llm,
    prompt="You are an advanced math assistant. Handle complex math operations.",
    toolsList=[
        Tool(description="Calculate factorial", func=factorial)
    ]
)

# Create an orchestrator agent
class Orchestrator(MicroAgent):
    def __init__(self):
        super().__init__(
            llm=llm,
            prompt="""You are a math query analyzer. For each query:
1. If it contains basic arithmetic, output exactly: SIMPLE_MATHS NEEDED
2. If it contains advanced math, output exactly: ADVANCED_MATHS NEEDED
3. If unsure, output exactly: UNKNOWN_MATH_TYPE""",
            toolsList=[]
        )
        self.simple_math_agent = simple_math_agent
        self.advanced_math_agent = advanced_math_agent

    def execute_agent(self, query: str, message_store: MessageStore) -> str:
        # Get initial analysis from orchestrator
        analysis = super().execute_agent(query, message_store)
        
        if "SIMPLE_MATHS NEEDED" in analysis:
            result = self.simple_math_agent.execute_agent(query, message_store)
            return f"Simple Math Agent: {result}"
        elif "ADVANCED_MATHS NEEDED" in analysis:
            result = self.advanced_math_agent.execute_agent(query, message_store)
            return f"Advanced Math Agent: {result}"
        else:
            return "Unable to determine the appropriate agent for this query."

# Use the orchestrated system
message_store = MessageStore()
orchestrator = Orchestrator()

# Example queries
queries = [
    "What is 15 plus 27?",  # Will use simple_math_agent
    "Calculate 5 factorial",  # Will use advanced_math_agent
    "First add 3 and 5, then multiply the result by 2"  # Will use simple_math_agent
]

for query in queries:
    response = orchestrator.execute_agent(query, message_store)
    print(f"Query: {query}")
    print(f"Response: {response}\n")

This example demonstrates:

• Creating multiple specialized agents with different tools
• Building an orchestrator agent to route queries
• Using a message store to maintain conversation history
• Coordinating multiple agents to handle different types of tasks

Examples

• math_demo.py: Basic math operations using tool calls

License

MIT License