MiniGen (5-line AI Agent Framework)

MiniGen is a lightweight, Python-native framework designed to help you learn and build AI agents without getting lost in boilerplate code. It's built for simplicity, experimentation, and understanding the core concepts that make AI agents work.

Benefits

MiniGen offers several key advantages:*Simple & Intuitive: Minimal boilerplate, maximum functionality*Tool Integration: Easily extend agents with custom capabilities*Workflow Orchestration: Chain prompts and coordinate multiple agents*Parallel Execution: Run agents concurrently for better performance*Multi-Agent Networks: Build complex systems with intelligent routing*Provider Agnostic: Works with OpenAI, Anthropic, and other LLM providers

Use Cases

MiniGen can be used in various scenarios:*Code Review Agent: Analyze Python code for potential issues and provide specific, actionable feedback.*Research and Writing Pipeline: Create an article about the environmental impact of electric vehicles by verifying information accuracy and creating engaging, well-structured content based on verified facts.*Parallel Processing: Analyze the potential impact of quantum computing on the cryptocurrency market from multiple perspectives.

Installation

You can install MiniGen using pip:

pip install minigen

Or install from source:

git clone https://github.com/devdezzies/minigen.gitcd minigenpip install -e .

Quick Start

1. Environment Setup

Create a.envfile in your project root:

# Required: Your API keyOPENAI_API_KEY="your-api-key-here"# Optional: Specify your preferred modelDEFAULT_MODEL="gpt-4"# Optional: For non-OpenAI providersBASE_URL="https://api.anthropic.com" # Example for Claude

2. Create Your First Agent

from minigen import Agent# Create a specialized agentpirate_agent = Agent(name="Captain Sarcasm",system_prompt="You are a sarcastic pirate captain. Answer all questions with sarcasm and pirate slang.")# Start chattingresponse = pirate_agent.chat("How does a computer work?")print(f"[{pirate_agent.name}]: {response}")

Core Concepts

Agents

Think of anAgentas a specialized AI personality with a specific role. Each agent has:* Anamefor identification* Asystem promptthat defines its behavior and expertise* Optionaltoolsfor extended capabilities*Memoryof the conversation context

Tools

Toolsare Python functions that agents can call to interact with the external world:

from minigen import Agent, tool@tool(description="Convert temperature from Celsius to Fahrenheit")def celsius_to_fahrenheit(celsius: float) -> float:return (celsius * 9/5) + 32@tool(description="Get current weather for a city")def get_weather(city: str) -> str:return f"The weather in {city} is sunny and 250C"# Create agent with toolsweather_agent = Agent(name="Weather Assistant",system_prompt="You help users with weather information and temperature conversions.",tools=[celsius_to_fahrenheit, get_weather])response = weather_agent.chat("What's the weather in London and convert 250C to Fahrenheit?")

Chains

Chainsexecute a sequence of prompts, where each step builds on the previous output:

from minigen import Agent, Chainagent = Agent()research_chain = Chain(agent=agent, verbose=True)research_chain \.add_step("Generate a comprehensive technical explanation of {input}") \.add_step("Simplify the following explanation for a beginner: {input}") \.add_step("Create a practical example to illustrate: {input}")result = research_chain.run("machine learning")print(result)

Agent Networks

Agent Networkscoordinate multiple specialized agents to solve complex problems:

from minigen import Agent, AgentNetwork# Create specialized agentsplanner = Agent(name="Planner",system_prompt="You excel at breaking down complex tasks into actionable steps.")researcher = Agent(name="Researcher",system_prompt="You find accurate information and cite sources.")writer = Agent(name="Writer",system_prompt="You create well-structured, engaging content.")# Build the networknetwork = AgentNetwork()network.add_node(planner)network.add_node(researcher)network.add_node(writer)# Set up intelligent routingfrom minigen import create_llm_routerrouter = create_llm_router(network.nodes)network.set_router(router)network.set_entry_point("Planner")# Execute the workflowresult = network.run("Write a blog post about the benefits of renewable energy",max_rounds=8)

Examples

Example 1: Code Review Agent

from minigen import Agent, toolimport ast@tool(description="Analyze Python code for potential issues")def analyze_code(code: str) -> str:try:ast.parse(code)return "Code syntax is valid. Ready for detailed review."except SyntaxError as e:return f"Syntax error found: {e}"code_reviewer = Agent(name="Code Reviewer",system_prompt="""You are an expert Python developer who reviews code for:- Best practices and conventions (PEP 8)- Performance optimizations- Security vulnerabilities- Code maintainabilityProvide specific, actionable feedback.""",tools=[analyze_code])# Review some codecode_to_review = '''def fibonacci(n):if n <= 1:return nreturn fibonacci(n-1) + fibonacci(n-2)'''review = code_reviewer.chat(f"Please review this code:{code_to_review}")print(review)

Example 2: Research and Writing Pipeline

from minigen import Agent, AgentNetwork, create_llm_router# Define specialized agentsfact_checker = Agent(name="FactChecker",system_prompt="You verify information accuracy and find reliable sources.")content_creator = Agent(name="ContentCreator",system_prompt="You create engaging, well-structured content based on verified facts.")editor = Agent(name="Editor",system_prompt="You polish content for clarity, flow, and professionalism.")# Create the pipelinepipeline = AgentNetwork()pipeline.add_node(fact_checker)pipeline.add_node(content_creator)pipeline.add_node(editor)router = create_llm_router(pipeline.nodes)pipeline.set_router(router)pipeline.set_entry_point("FactChecker")# Generate contentfinal_content = pipeline.run("Create an article about the environmental impact of electric vehicles",max_rounds=6)

Example 3: Parallel Processing

from minigen import Agent, AgentNetwork, Parallel, create_llm_router# Create domain expertstech_expert = Agent(name="TechExpert",system_prompt="You analyze technology trends and innovations.")market_expert = Agent(name="MarketExpert",system_prompt="You analyze market conditions and business implications.")synthesizer = Agent(name="Synthesizer",system_prompt="You combine different perspectives into comprehensive insights.")# Set up parallel processingparallel_analysis = Parallel(name="ParallelAnalysis",agent_names=["TechExpert", "MarketExpert"])network = AgentNetwork()network.add_node(tech_expert)network.add_node(market_expert)network.add_node(synthesizer)network.add_node(parallel_analysis)router = create_llm_router(network.nodes)network.set_router(router)network.set_entry_point("ParallelAnalysis")# Analyze from multiple perspectivesanalysis = network.run("Analyze the potential impact of quantum computing on the cryptocurrency market",max_rounds=5)

Advanced Usage

Custom Routing Logic

from minigen import AgentNetwork, NetworkStatefrom typing import Optionaldef custom_router(state: NetworkState) -> Optional[str]:last_message = state.messages[-1]# Route based on content keywordscontent = last_message.get('content', '').lower()if 'research' in content:return "Researcher"elif 'write' in content or 'draft' in content:return "Writer"elif 'review' in content or 'edit' in content:return "Editor"elif 'done' in content or 'complete' in content:return None # End the workflowreturn "Planner" # Default fallbacknetwork.set_router(custom_router)

Parameters:*name: Agent identifier*system_prompt: Instructions that define agent behavior*tools: List of functions the agent can call*model: LLM model to use (defaults to environment setting)*max_retries: Maximum retry attempts for failed requests

Contributing

We welcome contributions! Please see our for details.

License

This project is licensed under the MIT License - see the file for details.

Support

• 
  F33A
• 
  F41B
• 
  F34C

**Built with
F49A by the MiniGen community