🧠 iLLMs – ReAct Agents with Ease

iLLMs is a minimal Python framework for building ReAct-style LLM agents with tool support. It combines a structured reasoning loop (Thought → Action → Observation → Answer) with simple abstractions for tools and agents.

---

✨ Features

• 🧠 ReAct loop: Thought → Action → Observation → Answer
• 🧩 Define tools via a simple decorator
• 🛠 Easily plug tools into your agent
• 🧪 Clean example in Jupyter Notebook

---

🏗️ Structure

iLLMs/
├── Agent.py         # Core agent logic (ReAct loop)
├── Tool.py          # Tool definition and decorator
├── Examples/
│   └── agent.ipynb  # Notebook showing usage
└── README.md

---

🚀 Quickstart

1. Install dependencies

pip install groq

2. Define tools

from Tool import tool

@tool
def greet(name: str) -> str:
    """Returns a greeting for the given name."""
    return f"Hello, {name}!"

3. Create a ReAct agent

from Agent import Agent
from groq import Groq

client = Groq(api_key="your_api_key")

agent = Agent(
    client=client,
    name="ReActBot",
    role="A helpful assistant that reasons and acts using tools.",
    tools=[greet],
    system_message_file="SYSTEM_PROMPT.txt"
)

4. Run the agent

response = agent("Say hello to Sam.")
print(response)

---

🔁 ReAct Loop Overview

The agent internally loops through:

• Thought: Think about what to do next
• Action: Pick and invoke a tool
• Observation: Receive and reflect on the result
• Answer: Conclude when ready

Structured JSON responses control the loop:

{
  "state": "Action",
  "payload": "greet",
  "args": ["Sam"]
}

---

📓 Example Notebook

Check Examples/agent.ipynb for a full demonstration of defining tools and running your agent.

---

📄 License

MIT License © 2025

Future Work

• Add multi_agentic workflows
• Support for different LLM providers