Gemini Simple Agent

A lightweight single-agent orchestration framework using Google Gemini API.

Features

• 🤖 Simple Agent Orchestration: Clean single-agent implementation with Gemini
• 💬 Streaming Responses: Real-time streaming chat responses
• 🔧 Built-in Tools: Support for code execution and web search
• 📝 Conversation History: Automatic conversation management
• ⚙️ YAML Configuration: Easy configuration via YAML files
• 🎨 Rich CLI: Beautiful terminal interface with Rich

How to Run in PowerShell

Setup (First Time Only)

1. Create a virtual environment:

python -m venv gemini-agent

2. Copy and configure the environment file:

cp .env.example .env
# Edit .env and add your Google API key from: https://aistudio.google.com/apikey

3. Activate the virtual environment and install dependencies:

. gemini-agent\Scripts\activate.ps1
pip install -r requirements.txt

Running the Agent

Single Question Mode:

. gemini-agent\Scripts\activate.ps1
python -m gemini_agent.cli --model gemini-2.5-pro "Your question here"

Interactive Mode:

. gemini-agent\Scripts\activate.ps1
python -m gemini_agent.cli --model gemini-2.5-pro

With Configuration File:

. gemini-agent\Scripts\activate.ps1
python -m gemini_agent.cli --config config.yaml "Your question here"

With Code Execution Enabled:

. gemini-agent\Scripts\activate.ps1
python -m gemini_agent.cli --model gemini-2.5-pro --enable-code-execution "Calculate the first 10 prime numbers"

With Web Search Enabled:

. gemini-agent\Scripts\activate.ps1
python -m gemini_agent.cli --model gemini-2.5-pro --enable-web-search "What's the latest news on AI?"

Installation

1. Clone or navigate to this directory:

cd dummy-agent-framework

2. Install dependencies:

pip install -r requirements.txt

Or install in development mode:

pip install -e .

3. Set up your API key:

cp .env.example .env
# Edit .env and add your Google API key

Get your API key from: https://aistudio.google.com/apikey

Quick Start

Interactive Mode

Start a conversation with the agent:

python -m gemini_agent.cli --config config.yaml

Or if installed:

gemini-agent --config config.yaml

Single Question Mode

Ask a single question:

python -m gemini_agent.cli --config config.yaml "What is the capital of France?"

Without Configuration File

Quick usage without config:

python -m gemini_agent.cli "Explain quantum computing in simple terms"

With custom settings:

python -m gemini_agent.cli --model gemini-2.5-pro --enable-web-search "What's the latest news on AI?"

Configuration

The framework uses YAML configuration files. See config.yaml for a basic example.

Basic Configuration

agent:
  id: "gemini-agent"
  
  backend:
    type: "gemini"
    model: "gemini-2.0-flash-exp"
    enable_code_execution: false
    enable_web_search: false
    temperature: 1.0
  
  system_message: "You are a helpful AI assistant."

ui:
  display_type: "rich_terminal"
  logging_enabled: true

Example Configurations

The examples/ directory contains several pre-configured setups:

• config_code_execution.yaml: Agent with Python code execution
• config_web_search.yaml: Agent with web search capability
• config_full_tools.yaml: Agent with all tools enabled

Try them:

python -m gemini_agent.cli --config examples/config_code_execution.yaml "Calculate fibonacci(20)"

Usage Examples

Code Execution

python -m gemini_agent.cli --config examples/config_code_execution.yaml

You: Calculate the first 10 prime numbers

Agent: [executes code and shows results]

Web Search

python -m gemini_agent.cli --config examples/config_web_search.yaml

You: What's the current weather in Paris?

Agent: [searches web and provides current information]

Custom System Message

python -m gemini_agent.cli --system-message "You are a Python expert" "How do I sort a dictionary by value?"

Command-Line Options

python -m gemini_agent.cli --help

Options:

• --config: Path to YAML configuration file
• --model: Gemini model to use (default: gemini-2.0-flash-exp)
• --system-message: Custom system prompt
• --enable-code-execution: Enable code execution capability
• --enable-web-search: Enable web search capability
• --temperature: Sampling temperature (0-2)
• --verbose: Enable verbose logging

Available Models

• gemini-2.0-flash-exp - Fast, efficient model (default)
• gemini-2.5-pro - More capable model for complex tasks
• gemini-1.5-flash - Previous generation flash model
• gemini-1.5-pro - Previous generation pro model

Project Structure

dummy-agent-framework/
├── gemini_agent/
│   ├── __init__.py       # Package initialization
│   ├── agent.py          # Agent class with conversation management
│   ├── backend.py        # Gemini API wrapper
│   ├── cli.py            # Command-line interface
│   └── config.py         # Configuration loader
├── examples/
│   ├── config_code_execution.yaml
│   ├── config_web_search.yaml
│   └── config_full_tools.yaml
├── config.yaml           # Default configuration
├── requirements.txt      # Python dependencies
├── setup.py             # Package setup
├── .env.example         # Environment template
└── README.md            # This file

Key Components

Agent Class

The Agent class manages conversation history and provides a simple chat interface:

from gemini_agent import create_agent

agent = create_agent(
    model="gemini-2.0-flash-exp",
    system_message="You are a helpful assistant",
    enable_code_execution=True
)

# Streaming chat
async for chunk in agent.chat("Hello!"):
    if chunk.type == "content":
        print(chunk.content, end="")

# Simple chat (non-streaming)
response = await agent.chat_simple("What is 2+2?")
print(response)

GeminiBackend Class

Low-level wrapper around the Google Gemini API:

from gemini_agent.backend import GeminiBackend

backend = GeminiBackend(
    model="gemini-2.0-flash-exp",
    enable_code_execution=True,
    enable_web_search=True
)

async for chunk in backend.chat_stream(messages):
    print(chunk)

Environment Variables

Create a .env file with:

GOOGLE_API_KEY=your_api_key_here
# or
GEMINI_API_KEY=your_api_key_here

Interactive Mode Commands

When in interactive mode:

• exit, quit, bye, q - Exit the conversation
• reset - Clear conversation history

Comparison with other frameworks

This framework is a simplified single-agent implementation based on other frameowork's architecture:

Other frameworks (Full Framework):

• Multi-agent coordination
• Binary decision voting system
• Advanced memory management
• Multiple LLM backend support
• Complex orchestration

This Framework (Simple Agent):

• Single agent only
• Gemini-specific
• Streamlined conversation management
• Easy to understand and extend
• Perfect for learning and simple use cases

Development

Install in development mode:

pip install -e .

Run with verbose logging:

python -m gemini_agent.cli --verbose

Troubleshooting

API Key Issues

• Ensure your GOOGLE_API_KEY or GEMINI_API_KEY is set in .env
• Verify the key is valid at https://aistudio.google.com/apikey

Import Errors

• Make sure all dependencies are installed: pip install -r requirements.txt
• If using Python 3.9-3.11, ensure compatibility with google-genai library

Model Not Found

• Check available models in Google AI Studio
• Some models may require different access levels

Contributing

This is a reference implementation. Feel free to:

• Add new features
• Improve error handling
• Add more examples
• Extend with custom tools

License

MIT License - Feel free to use and modify

Learn More

• Google Gemini API Documentation
• Rich Library Documentation