🤖 Local AI Agent with Gemma 4 (Function Calling & Persistence)

🇪🇸 Leer en español

An interactive artificial intelligence agent developed in Python that runs 100% locally using Ollama and the Gemma4:e2b model.

The system natively implements Tool Use / Function Calling via JSON schemas, manages persistent short- and mid-term memory, and includes an automated timers system for deferred tool execution.

---

✨ Key Features

• 🔒 100% Local & Private Execution: Built on the local Ollama API (compatible with gemma4:e2b).
• 🛠️ Dynamic Function Calling / Tool Use:
  • File System: Modular reading, writing, creation, deletion, and listing of files and folders.
  • Web / API Integration: Real-time lookup of the official dollar price (BCV) via an external API.
  • Timers System: Creation of asynchronous background timers (timer.py) to trigger notifications or tasks after a time interval.
• 🧠 Persistent Memory:
  • Short-Term Memory: Maintains the context of the current conversation.
  • Mid-Term Memory: Persistent storage of key data to remember important information across sessions.
• ⚡ Latency and Performance Measurement: Real-time monitoring of model response times per interaction.

---

🛠️ Technologies Used

• Language: Python 3.14
• LLM Engine: Ollama (ollama-python)
• HTTP Requests: requests
• Schema Format: JSON Schema for tool definition and invocation.

---

📂 Architecture and Code Structure

├── main.py              # Main interaction loop (REPL) and performance measurement
├── chat.py              # Chat engine, Function Calling parsing and tool dispatch
├── memory.py            # Context window manager for short/mid-term memory
├── tools.py             # Native function implementations (Files, Dollar API, Timers)
├── timer.py             # Asynchronous execution of timers in background threads
├── tools_esquema.json   # Formal JSON Schema definitions for available tools
└── dependencias.txt     # Project dependencies

🚀 Installation and Usage Guide

Prerequisites

• Have Ollama installed and running on your system.
• Download the desired model (for example, gemma4:e2b):

Note

About models:
This project used gemma4:e2b, however any Ollama model with tool-use capability should work; you may need to edit chat.py at line 51 and change the model name manually.

Installation Steps

1. Clone the repository:

    git clone https://github.com/Aki-new/agente-de-IA-con-gemma4.git
    cd agente-de-IA-con-gemma4

2. Create and activate a virtual environment:

   • Windows:

         python -m venv .venv
         .venv\Scripts\activate

   • Linux / macOS:

        python3 -m venv .venv
        source .venv/bin/activate

3. Install dependencies:

    pip install -r dependencias.txt

4. Run the agent

    python main.py

📝 Special Agent Commands

During the interactive conversation you can use the following commands in the terminal:

• salir / exit: End the current session.
• limpiar memoria: Clear the current session history from the database.
• memoria: Show the messages currently stored in the context.