Manas is a sophisticated Multi-Agent System Framework designed for Large Language Model (LLM) applications, enabling the development of intelligent and interactive applications with ease. Its modular and extensible architecture allows developers to create applications with autonomous agents, integrate various tools, and structure complex tasks into manageable workflows.

Key Features

• Intelligent Agents: Build autonomous agents that utilize a think-act-observe cycle, coupled with robust state management, enabling seamless interaction and decision-making.
• Flexible Tool System: Incorporate a dynamic registry for tools that supports automatic parameter discovery and comprehensive documentation, enhancing operational flexibility.
• Task Decomposition: Simplify complex tasks by breaking them into manageable subtasks, ensuring clarity and execution efficiency.
• Retrieval Augmented Generation (RAG): Augment LLM responses with relevant contextual information via document chunking, leading to more informed outputs.
• Dynamic Flows: Create and modify workflows dynamically with integrated visualization and dependency management, accommodating evolving requirements.
• Provider Architecture: Support for multiple LLM providers including OpenAI, Anthropic, HuggingFace, and Ollama, enabling versatile application development.
• Vector Store Integration: Seamless integration with various vector stores like FAISS, Chroma, and Pinecone, providing consistent interfaces for data retrieval.
• Memory and Middleware: Implement built-in memory middleware alongside an extensible middleware architecture, facilitating advanced functionality and customizability.
• Asynchronous Design: Fully asynchronous architecture that emphasizes proper error handling, ensuring the robustness of applications.
• Formal Verification: Core components undergo rigorous validation through formal verification processes, promoting reliability and trustworthiness.

Quick Start Examples

Simple Question-Answering Agent

A straightforward example of an agent that answers questions:

import os
from core import LLM, Agent

model = LLM.from_provider(
    "openai",
    model_name="gpt-4",
    api_key=os.environ.get("OPENAI_API_KEY")
)

agent = Agent(llm=model, system_prompt="You are a helpful assistant.")
response = agent.generate("What is the capital of France?")
print(response)

Multi-Agent Flow Creation

Creating a workflow with multiple specialized agents is easy:

from core import Flow
from core.nodes import QANode

model = LLM.from_provider("openai", model_name="gpt-4")

researcher = QANode(name="researcher", llm=model, system_prompt="You are an expert researcher who provides factual information.")
writer = QANode(name="writer", llm=model, system_prompt="You are a skilled writer who creates engaging content.")

flow = Flow()
flow.add_node(researcher)
flow.add_node(writer)
flow.add_edge(researcher, writer)

result = flow.process("Explain quantum computing.")
print(result)

Comprehensive Documentation

Complete guidance and detailed information are available at the Manas Documentation, covering:

• Getting Started
• Core Concepts
• API Reference
• Examples
• FAQ

Manas represents an advanced solution for developers looking to leverage the power of LLMs within dynamic, agent-based workflows, streamlining the process of creating intelligent applications.