🧩  About the Project

This project is a deep dive into the ReAct (Reasoning and Acting) pattern — built from scratch, without relying on high-level AI frameworks like LangChain.

The goal was to create an agent that doesn't just talk, but actually executes actions to find information and perform tasks. It uses a manual implementation of the reasoning loop, where the LLM decides which tool to invoke, processes the output, and continues reasoning until it reaches a final answer.

Now integrated with the Model Context Protocol (MCP), the agent can leverage professional-grade tools via MCP-SERVER-PRO for database management, spreadsheet manipulation, and more.

⚡  How It Works

         User Question
               ↓
      LLM Reasons (Thought)
               ↓
   Action Selection (Tool Call)
               ↓
  Tool Execution (Local / MCP)
               ↓
     Observation (Tool Output)
               ↓
      LLM Final Answer (Result)

✨  Features

🛠️  Tech Stack

📦  Prerequisites

Before getting started, make sure you have the following installed:

• Python 3.11+
• Node.js & npm
• Ollama — running locally on your machine

🚀  Getting Started

1 — Pull the required model

ollama pull llama3.2   # or your preferred model

2 — Clone the repository

git clone https://github.com/EduhxH/IA-agent-with-tools---.git
cd IA-agent-with-tools---/agente-ia-local

3 — Set up the backend

cd backend
python -m venv .venv

# Activate the virtual environment
source .venv/bin/activate    # Linux / Mac
.venv\Scripts\activate       # Windows

pip install -r requirements.txt

4 — Set up the frontend

cd ../frontend
npm install

5 — Run the project

# Terminal 1 — Backend (inside /backend)
uvicorn api.app:app --reload --port 8000

# Terminal 2 — Frontend (inside /frontend)
npm run dev

⚙️  Configuration

The backend is configured via environment variables. Create a .env file inside backend/:

OLLAMA_BASE_URL=http://localhost:11434   # default Ollama address
OLLAMA_MODEL=llama3.2                    # model used for inference
WORKSPACE_ROOT=C:/your/project/path      # configurable path for IDE reader

No API keys required. Everything runs locally through Ollama.

☁️  Deployment

📁  Project Structure

agente-ia-local/
│
├── backend/
│   ├── agent/
│   │   ├── agent.py            # ReAct loop logic
│   │   └── ollama_client.py    # Ollama HTTP client
│   ├── api/
│   │   └── app.py              # FastAPI routes & app setup
│   ├── tools/                  # Core tool implementations (file_reader.py, etc.)
│   └── requirements.txt
│
├── frontend/
│   ├── src/                    # React components with Upload/IDE UI
│   └── vite.config.js
│
└── .gitignore

🧠  What I Learned

• The inner workings of the ReAct pattern and how agentic reasoning loops are structured.
• Manual tool-calling implementation using structured system prompts.
• Integration with the Model Context Protocol (MCP) for professional toolsets.
• Implementing File Upload logic to allow agents to process user-provided data.
• Creating a Dynamic IDE Reader to allow AI interaction with any local workspace.
• Designing a clean, decoupled architecture with FastAPI and React 19.