Swarm for Local LLMs 🐝

This project demonstrates how to use a Swarm framework to manage and interact with local Large Language Models (LLMs) using Ollama. The Swarm framework allows you to create multiple agents, each with its own instructions and capabilities, and seamlessly transfer control between them during a conversation.

Project Overview

The Swarm for Local LLMs project is designed to simulate a multi-agent system where each agent can handle specific tasks or domains. For example:

• Agent A: A general-purpose assistant that can transfer conversations to specialized agents.
• Agent B: A specialized agent that explains concepts in 1700s English.

This project is particularly useful for:

• Local LLM experimentation: Run and test LLMs locally without relying on cloud-based APIs.
• Multi-agent systems: Create and manage multiple agents with distinct roles and capabilities.
• Agent handoff: Seamlessly transfer conversations between agents based on user requests.

Features

• Local LLM Integration: Uses Ollama to run local LLMs like deepseek-r1:1.5b.
• Swarm Framework: Manages multiple agents and their interactions.
• Agent Handoff: Transfers conversations between agents dynamically.
• Customizable Agents: Each agent can have its own instructions, functions, and behavior.
• REPL Interface: Provides a simple command-line interface for interacting with the Swarm.

Repository Link

🔗 GitHub Repository

---

Getting Started

Prerequisites

1. Python 3.7+: Ensure you have Python installed on your system.
2. Ollama: Install Ollama to run local LLMs. Follow the Ollama installation guide to set it up.
3. Local LLM: Download a local LLM like deepseek-r1:1.5b using Ollama:

   ollama pull deepseek-r1:1.5b

Installation

1. Clone the repository:

   git clone https://github.com/Priyanshu-i/Swarm-for-Local-LLMs.git
   cd Swarm-for-Local-LLMs

2. Install the required Python packages:

   pip install -r requirements.txt

---

Usage

Running the Swarm

1. Start the Swarm by running the agent_handoff.py script:

   python agent_handoff.py

2. Interact with the agents:

   • The conversation starts with Agent A.
   • Request to talk to Agent B by saying: I want to talk to agent B, have him explain LLMs.
   • Agent B will respond in 1700s English.

Example Output

Swarm client initialized
Agent A created
Agent B created
Starting conversation with Agent A
Transferring to Agent B
Final response:
Agent B: Verily, I shall elucidate upon the matter of Large Language Models, which art the marvels of modern computation...

---

Code Structure

The repository is organized as follows:

Swarm-for-Local-LLMs/
├── agent_handoff.py          # Example script to demonstrate agent handoff
├── swarm/                    # Swarm framework implementation
│   ├── __init__.py           # Package initialization
│   ├── core.py               # Core Swarm and Agent classes
│   ├── repl.py               # REPL interface for interacting with the Swarm
│   ├── util.py               # Utility functions
│   ├── types.py              # Custom types and classes
│   └── custom.py             # Custom classes for mimicking OpenAI's API
├── requirements.txt          # Python dependencies
└── README.md                 # This file

---

Customizing Agents

You can create and customize your own agents by modifying the agent_handoff.py script. For example:

agent_c = Agent(
    name="Agent C",
    instructions="You are a math expert. Explain mathematical concepts in simple terms.",
    functions=[],  # Add custom functions if needed
)

---

Contributing

Contributions are welcome! If you'd like to contribute to this project, please follow these steps:

1. Fork the repository.
2. Create a new branch for your feature or bugfix.
3. Commit your changes and push them to your fork.
4. Submit a pull request with a detailed description of your changes.

---

License

This project is licensed under the MIT License. See the LICENSE file for details.

---

Acknowledgments

• Ollama for providing the local LLM infrastructure.
• Inspired by the concept of Swarm Intelligence and multi-agent systems.

---

Contact

For questions or feedback, feel free to reach out:

• Priyanshu Singh
  GitHub: Priyanshu-i

---

Enjoy experimenting with the Swarm for Local LLMs! 🚀