GemAgent

GemAgent is a lightweight Python SDK to build Gemini-powered agents with tool-calling capability. It provides a simple interface to run Gemini 1.5 models with custom tools and simulated event streaming.

✅ Current Features

• 🎯 Supports Google Gemini 1.5 models (via google.generativeai)
• 🧰 Register and use custom tools easily
• 🔁 Agent and runner system with clean separation
• 📤 Simulated streaming output (run_streamed)
• ✅ Simple tool call parsing using call_tool(tool_name, arg=value) format

🔧 Example Usage

from gemagent import Agent, Runner, tool_custom

@tool_custom
def say_hello(name="world"):
    return f"Hello, {name}!"

agent = Agent(
    name="DemoAgent",
    instructions="Be helpful and use tools if needed.",
    tools=[say_hello],
)

import asyncio
response = asyncio.run(Runner.run(agent, "call_tool(say_hello, name='Rishabh')"))
print(response.final_output)

📦 Installation

pip install gemagent

📍 Environment

Set your Gemini API key in .env file:

GOOGLE_API_KEY=your_api_key_here

🚧 Roadmap (Coming Soon)

• 🔌 Native support for multiple models (OpenAI, Claude, local models)
• 🧠 Tool chaining and multi-step reasoning
• 🔄 Agent memory and context persistence
• 🎥 Real token-level streaming (for terminal and web)
• ⚙️ CLI interface and Web playground
• 📁 Plugin-style tool loading

📄 License

MIT License

💡 Inspiration

OpenAI recently released the Agents SDK, which provides a powerful tool-using agent system. However, it requires a paid API key with at least $5 balance to access.

GemAgent was born out of the need for:

• A free, minimal alternative using Gemini 1.5 Flash
• Similar design and syntax to OpenAI's Agents SDK for easy migration later
• Local dev/testing without paying just to prototype tool-based agents