MASLibPy: Multi-Agent System Python Library

maslibpy is a lightweight, modular Python library designed to facilitate the creation and management of multi-agent systems with enhanced reasoning capabilities. It offers a structured framework for building agents, managing message flows, integrating with various LLM providers, and leveraging advanced reasoning modules.

---

Features

• Modular Agent Design: Create and manage agents with customizable roles, goals, and LLM integrations.
• Message Flow Management: Built-in support for system, user, and assistant communications.
• Advanced Reasoning: Utilizes Prompt-based and Mathematical-based reaoning with various prompt templates like chain-of-thought (CoT), REACT, ReWOO, Reflection and Reflexion prompt for improved decision-making.
• LLM Integration: Seamless connectivity with multiple providers (e.g., OpenAI, Together, Groq, Replicate).
• Dynamic Collaboration: Orchestrate multi-agent interactions with integrated reasoning and scoring modules.

---

Installation

Clone the repository and install the dependencies using Poetry:

git clone https://github.com/bayeslabs/maslibpy.git
cd maslibpy

pip install poetry
poetry install

---

Directory Structure

maslibpy/
├── assets/
│   ├── reasoning_workflow.png
│   └── MathematicalReasoning_Experimentations.pdf
├── prompts/
│   ├── cot/
│   │   └── cot_prompts.py
│   └── react/
│       └── react_prompts.py
├── reasoning/
│   ├── scorer.py
│   ├── prompt_based.py
│   └── mathematical.py
├── agent/
│   ├── baseagent.py
│   └── agent.py
├── messages/
│   ├── base.py
│   ├── system.py
│   ├── user.py
│   └── assistant.py
├── patttern/
│   └── sequential.py
├── llm/
│   ├── llm.py
│   └── constants.py
├── pyproject.toml
└── README.md

---

File Descriptions

Prompts

• prompts/cot/cot_prompts.py: Contains chain-of-thought (CoT) prompt templates that guide agents through step-by-step reasoning.
• prompts/react/react_prompts.py: Provides REACT prompt templates (REACT, Reflection, Reflexion, ReWoo) to help agents integrate reasoning and decision-making within their interactions.

Reasoning

• reasoning/scorer.py: Implements functionality to evaluate and score the reasoning outputs of agents.
• reasoning/prompt_based.py: Offers prompt-based reasoning modules to assist in structured problem solving.
• reasoning/mathematical.py: Contains utilities for mathematical reasoning and computations during agent interactions.

Agent

• agent/baseagent.py: Defines the base agent class with core attributes (e.g., id, name, role, goal) and LLM integration.
• agent/agent.py: Extends the base agent class with additional behaviors and advanced interaction capabilities.

Messages

• messages/base.py: Establishes the foundational BaseMessage class for managing message flows.
• messages/system.py: Implements the SystemMessage class for system-level communications.
• messages/user.py: Implements the UserMessage class for user-initiated communications.
• messages/assistant.py: Implements the Assistant (AI) message class for agent responses.

Pattern

• pattern/sequential.py: Provides a sequential coordination mechanism for orchestrating multi-agent collaborations.

LLM

• llm/llm.py: Manages integration with LLM providers, handling model invocations and API key validations.
• llm/constants.py: Defines environment variables, supported LLM providers, and available model configurations.

---

Example Usage

To see how maslibpy works, instantiate an Agent and invoke a query. Below is a simple example of creating an agent and asking it about AI Agents.

from maslibpy.agent.agent import Agent
from maslibpy.llm.llm import LLM

# Initialize an agent with specific attributes
agent = Agent(
    name="TestAgent",
    role="AI Assistant",
    goal="Assist users effectively",
    backstory="An advanced AI designed to provide helpful insights.",
    prompt_type="cot",
    max_iterations=2,
    critique_llm=LLM(provider="together", model_name="together_ai/meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo"),
    generator_llm=LLM(provider="together", model_name="together_ai/mistralai/Mistral-7B-Instruct-v0.1"),
    score_type="mathematical"
)

# Invoke the agent with a user query
response = agent.invoke([{"role": "user", "content": "What are AI Agents?"}])

# Print the response
print(response)

Customization & Flexibility

The Agent class extends BaseAgent, which comes with several configurable attributes. The example above demonstrates a basic use case, but you can modify additional parameters to fine-tune the agent’s behavior.

Available Attributes

The following attributes can be customized:

Attribute	Description
name	Name of the agent.
role	The agent’s assigned role.
goal	The main objective of the agent.
backstory	Background information about the agent.
generator_llm	LLM used for generating responses.
critique_llm	LLM used for critique (if applicable).
prompt_type	Type of prompting strategy (cot or react).
prompt_pattern	Specific prompt pattern for reasoning.
max_iterations	Number of reasoning iterations before stopping.
score_type	Scoring mechanism (mathematical or prompt_based).
entropy_threshold	Threshold for entropy-based scoring.
conciseness_weight	Weight factor for response conciseness.
max_plateau_count	Limit for consecutive non-improving iterations.

How to Modify Attributes

If you need to adjust any of these attributes, simply pass them as keyword arguments when instantiating the agent.

Example: Modifying Additional Parameters

agent = Agent(
    name="CustomAgent",
    role="Scientific Researcher",
    goal="Analyze complex datasets and generate insights.",
    backstory="An AI model specialized in data science and reasoning.",
    prompt_type="react",  
    max_iterations=5,  # Increased iterations for deeper reasoning
    entropy_threshold=0.2,  # Adjusted entropy for stricter evaluation
    conciseness_weight=0.5  # Prioritizing concise responses
)

---

Configuration

Environment Variables

Ensure the appropriate API keys are set in your environment variables:

• TOGETHER_API_KEY
• GROQ_API_KEY
• REPLICATE_API_KEY

Supported LLM Providers and Models:

• Refer to maslibpy/llm/constants.py for the list of supported providers and models.

---

Contributing

Contributions are welcome! Please follow these steps:

1. Fork the repository.
2. Create a feature branch.
3. Submit a pull request with a detailed description.

---

Contact

• For questions or support, please reach out to the repository maintainers: [email protected]

---