🤖 Agentic_AI

Agentic_AI is a multi-agent orchestration framework powered by LangChain + LangGraph + MCP and accelerated by Groq (LLaMA-3-70B).
It integrates specialized tools like Math, Translation, Gmail, Weather, Web Search, and features memory persistence with FAISS embeddings.

---

✨ Demo

🎥 Demo GIF or screenshot placeholder
(Add a quick terminal demo GIF using asciinema or a screenshot of agents collaborating.)

---

🧠 Architecture

Here’s how the system is designed:

graph TB
    User[👤 User Input] --> IntentRouter{🎯 Intent-Based<br/>Router}
    
    IntentRouter -->|Conversational| DirectAgent[💬 Direct Response<br/>Agent]
    IntentRouter -->|Task-Based| Planner[📋 Task Planner<br/>LLaMA-3.1-8B]
    
    Planner -->|JSON Plan| Executor[⚙️ Plan Executor<br/>LLaMA-3.1-8B]
    
    Executor --> Tools[🛠️ MCP Tool Suite]
    Tools --> Math[🧮 Math Server<br/>stdio]
    Tools --> Translate[🌍 Translator<br/>stdio]
    Tools --> Gmail[📧 Gmail API<br/>stdio]
    Tools --> WebSearch[🔍 Web Search<br/>stdio]
    Tools --> Weather[🌦️ Weather<br/>HTTP]
    
    Executor -->|Results| Verifier{✅ Rule-Based<br/>Verifier}
    
    Verifier -->|PASS| FinalAnswer[📤 Final Answer]
    Verifier -->|RETRY<br/>Max 2x| Executor
    Verifier -->|FAIL| ErrorHandler[❌ Error Handler]
    
    DirectAgent --> Memory[(🧠 FAISS Vector DB<br/>HuggingFace Embeddings)]
    Executor --> Memory
    Memory -->|Top 3 Context| Executor
    Memory -->|Context| DirectAgent
    
    FinalAnswer --> User
    DirectAgent --> User
    ErrorHandler --> User
    
    subgraph "🌐 Web Interface"
        API[FastAPI Server<br/>CORS Enabled]
    end
    
    User -.->|HTTP| API
    API -.-> IntentRouter
    
    style IntentRouter fill:#ff9999
    style Planner fill:#99ccff
    style Executor fill:#99ccff
    style Verifier fill:#99ff99
    style Memory fill:#ffcc99
    style Tools fill:#e6b3ff

Loading

🚀 Features

🎯 Core Intelligence Features

Intent-Based Query Routing: Automatically detects conversational vs. task-based queries, bypassing unnecessary tool planning for greetings and casual chat for more natural interactions

Plan-Verify-Execute Architecture: Sophisticated 3-stage pipeline with Planner (breaks complex tasks into JSON plans), Executor (executes steps with context), and Verifier (validates outputs with smart retry logic)

Smart Rule-Based Verification: Multi-layer validation including step count validation, tool compliance checking, goal satisfaction analysis, freshness detection for news queries, and generic failure detection

Adaptive Retry Mechanism: Self-correcting system with up to 2 retries, providing specific retry hints to executors for automatic error recovery without user intervention

🛠️ Tool & Integration Features

⚡ Fast & Scalable: Powered by Groq's LLM for blazing fast inference

🧮 Math Agent: Handles calculations & symbolic tasks

🌍 Translator Agent: Supports multilingual conversations

📧 Gmail Agent: Reads & interacts with Gmail API

🔍 Web Search Agent: Searches online data for better answers

🌦️ Weather Agent: Provides current weather, air quality, and environmental reports

Multi-Transport MCP Integration: Supports multiple MCP servers with stdio and streamable_http transports for mixing local and remote tools seamlessly

🧠 Memory & Context Features

Persistent Conversational Memory: FAISS vector database with HuggingFace embeddings stores and retrieves conversation context semantically

Context-Aware Responses: Retrieves top 3 relevant past interactions for maintaining conversation continuity

Cross-Session Memory: Maintains conversation history across sessions with persistent storage

🧹 Memory Reset: Use clear to reset past memory when needed

🌐 Deployment Features

Web Interface Ready: FastAPI integration with CORS support for web deployment and frontend interfaces

RESTful API: Production-ready API endpoints for chat interactions

🗂 Project Structure

Agentic_AI/ ├── main.py # Main entry point
├── mathserver.py # Math agent (MCP) ├── translate.py # Translator agent (MCP) ├── websearch.py # Web search agent ├── gmail.py # Gmail integration ├── rag_model.py # Optional RAG pipeline ├── mcp_use.py # MCP agent utilities ├── requirements.txt # Dependencies ├── .env # API keys & config └── README.md # This file

⚙️ Getting Started

1️⃣ Clone the repository git clone https://github.com/sobhan2204/Agentic_AI.git cd Agentic_AI

2️⃣ Setup environment python3.10 -m venv venv source venv/bin/activate pip install -r requirements.txt

3️⃣ Configure .env GROQ_API_KEY=your_api_key_here HF_TOKEN=your_huggingface_token_here (Optional: add Gmail API credentials if using Gmail Agent)

4️⃣ Run the agent python main.py

📈 Roadmap

Add finance/news/calendar agents

Memory expiration + relevance scoring

Web dashboard UI for interactions

Dockerized deployment

🧰 Tech Stack

LangChain + LangGraph + MCP – multi-agent orchestration

Groq (LLaMA-3-70B) – blazing fast inference

FAISS + HuggingFace embeddings – vector memory store

Python 3.10+ – backend

🤝 Contributing

##💡 Contributions are welcome!

Fork the repo & create a feature branch

Submit a PR with clear description

For new MCP agents, follow modular design