Kite - Simple AI Agent Framework for Groq

Build with Groq
1

Hey everyone,

I’ve been working on Kite, a Python framework for building AI agents. It’s designed to be simple and includes some production safety features I found myself needing repeatedly.

Repo: https://github.com/thienzz/Kite

What it does

  • Simple API for creating agents with tools

  • Built-in circuit breakers and retry logic

  • Works with Groq (and other providers)

  • Multiple reasoning patterns (ReAct, ReWOO, Tree-of-Thoughts, Plan-Execute)

Example

Here’s a basic customer support agent:

import asyncio
from kite import Kite

async def main():
    # Initialize
    ai = Kite()
    ai.config['llm_provider'] = 'groq'
    ai.config['llm_model'] = 'llama-3.3-70b-versatile'
    ai.config['GROQ_API_KEY'] = ''

    # Define business logic
    def search_orders(order_id: str) -> str:
        """Search for order by ID"""
        return f"Order {order_id} is shipped"
    
    def process_refunds(order_id: str) -> str:
        """Process refund for order"""
        return f"Refund processed for {order_id}"
    
    # Create tools
    search_tool = ai.create_tool("search_orders", search_orders)
    refund_tool = ai.create_tool("process_refunds", process_refunds)
    
    # Create agent
    agent = ai.create_agent(
        name="CustomerSupport",
        system_prompt="You are a helpful e-commerce support agent.",
        tools=[search_tool, refund_tool]
    )
    
    # Run it
    result = await agent.run("Where is order ORD-12345?")
    print(result['response'])

if __name__ == "__main__":
    asyncio.run(main())

Why I built it

I kept writing the same safety patterns (circuit breakers, idempotency) for production agents. Figured I’d package it up properly and add support for different reasoning strategies.

Groq’s speed makes it particularly nice for agent workflows where you’re chaining multiple LLM calls.

Status

  • v0.1.0 released

  • MIT licensed

  • Actively maintained

  • Looking for feedback

Would love to hear thoughts from folks building with Groq. What features would be useful? What’s missing?

Also happy to answer questions about the architecture or design choices.

Thanks!

2

This is awesome! thanks for sharing Kite. Really nice to see a simple agents+tools API with production safety features built in, and great to see Groq supported.

1 Like