bee2bee 3.3.15

Decentralized P2P network for AI model hosting and inference

🐝 Bee2Bee: The Neural Consensus P2P Network

Bee2Bee is a decentralized, peer-to-peer neural consensus engine designed to make AI inference accessible, transparent, and resilient. It allows anyone to contribute compute to a global mesh and anyone to consume it through a unified, high-performance API.

---

🏗️ Architecture

Bee2Bee operates as a Decentralized Mesh where every node is both a consumer and potentially a provider:

• Neural Nodes: Host models (Ollama, HF, Transformers) and register via the Global Registry.
• Consensus Router: Intelligently routes requests to the lowest-latency, highest-reliability nodes.
• API Sidecar: Every node optionally hosts a FastAPI management layer for real-time telemetry.

---

🚀 Quick Start (Production)

1. Installation

pip install bee2bee

2. Launch a Worker Node (The Consumer/Provider)

Run a node that automatically joins the network and hosts a model:

# Deploys a Llama3 provider via Ollama with an API sidecar on port 8000
python -m bee2bee serve-ollama --model llama3 --api-port 8000

3. Connect to the Mesh

To join the global discovery map on the Chatit.cloud Dashboard:

python -m bee2bee config bootstrap_url ws://bootstrap.chatit.cloud:4003

---

🛠️ Developer Guide

Developing Solutions (Python)

Integrate P2P intelligence directly into your apps using our rich-ready SDK:

import asyncio
from bee2bee import P2PNode

async def main():
    # Initialize P2P entrypoint
    node = P2PNode()
    await node.start()
    
    # Discovery
    providers = await node.discover_providers(model="llama3")
    
    # Intelligent Request
    response = await node.request_generation(
        provider_id=providers[0].id,
        prompt="Synthesize a response for decentralized governance."
    )
    print(f"P2P Intelligence: {response['text']}")

asyncio.run(main())

Interactive Learning

Check out our Jupyter Guide for a step-by-step walkthrough of the p2p architecture.

---

🌐 FastAPI Integration

Bee2Bee nodes now support a native FastAPI sidecar. This allows you to monitor your node health and peer list via standard HTTP requests:

• Health Check: GET http://localhost:8000/
• Peers List: GET http://localhost:8000/peers
• Cloud Inference: POST http://localhost:8000/chat

---

🤝 Community & Support

• Developer Support: [[phone]]
• Dashboard: Bee2Bee Live Map

Built with ❤️ by the ConnectIT Team for a decentralized future.

🚀 Quick Start Guide

1. Main End Point (The Supervisor)

This runs the core API server. Every network needs at least one Main Point.

Run Locally:

# Starts the API on Port 4002 and P2P Server on Port 4003
python -m bee2bee api

Output:

• API: http://127.0.0.1:4002 (Docs: /docs)
• P2P: ws://127.0.0.1:4003

---

2. Desktop App (The Dashboard)

A modern UI to visualize the network and chat with models.

Prerequisites: Node.js 20+

Run Locally:

cd electron-app
npm install      # First time only
npm run dev

Usage:

• Open the App.
• It connects to http://localhost:4002 by default.
• Go to "Chat" to talk to available providers.
• See MANUAL_TESTING.md for detailed testing steps.

---

3. Worker Node (The AI Provider)

Run this on any machine (or the same machine) to share an AI model.

Step A: Configure (Tell the node where the Main Point is)

# If running on the SAME machine as Main Point:
python -m bee2bee config bootstrap_url ws://127.0.0.1:4003

# If running on a DIFFERENT machine (LAN/WAN):
python -m bee2bee config bootstrap_url ws://<MAIN_POINT_IP>:4003

Step B: Deploy Model

Option 1: Hugging Face (Default) Uses transformers to run models like GPT-2, Llama, etc. on CPU/GPU.

# Deploys distilgpt2 (CPU friendly)
python -m bee2bee deploy-hf --model distilgpt2

Option 2: Ollama (Universal) Uses your local Ollama instance to serve models like Llama3, Mistral, Gemma, etc. Prerequisite: Install and run Ollama

# Serve a model (e.g., llama3)
python -m bee2bee serve-ollama --model llama3

Note: This creates a separate peer node on your machine.

Option 3: Remote Inference (Cloud) Execute models entirely on Hugging Face's servers via Inference API. No GPU required on your local machine!

# Deploys Zephyr 7B (Runs on Hugging Face Cloud)
python -m bee2bee deploy-hf --model HuggingFaceH4/zephyr-7b-beta --remote --token YOUR_HF_TOKEN

Note: The node acts as a proxy/gateway to the remote model.

---

4. Bee2Bee Cloud (Google Colab)

Run a powerful node on Google's free GPU infrastructure using our Hybrid Tunneling setup.

Notebook Location: notebook/ConnectIT_Cloud_Node.ipynb

How it Works (Hybrid Tunneling): To bypass Colab's network restrictions, we use two tunnels:

1. API Tunnel (Cloudflare): Provides a stable HTTPS URL (trycloudflare.com) for the Desktop App to connect to.
2. P2P Tunnel (Bore): Provides a raw WebSocket URL (bore.pub) for other Worker Nodes to connect to.

Instructions:

1. Open the Notebook in Google Colab.
2. Run "Install Dependencies".
3. Run "Configure Hybrid Tunnels" (Installs cloudflared & bore).
   • Wait for it to output the URLs.
4. Run "Run Bee2Bee Node".
   • It automatically configures itself to announce the Bore address.

Connecting your Desktop App to Colab:

1. Copy the Cloudflare URL (e.g., https://funny-remote-check.trycloudflare.com).
2. Open Desktop App -> Settings.
3. Paste into "Main Point URL".

---

🛠 Advanced Configuration

Environment Variables

You can override settings using ENV vars:

Variable	Description	Default
BEE2BEE_PORT	Port for P2P Server	4003 (Worker) / 4003 (API)
BEE2BEE_HOST	Bind Interface	0.0.0.0
BEE2BEE_ANNOUNCE_HOST	Public Hostname (for NAT/Tunnel)	Auto-detected
BEE2BEE_ANNOUNCE_PORT	Public Port (for NAT/Tunnel)	Auto-detected
BEE2BEE_BOOTSTRAP	URL of Main Point	None

Troubleshooting

• "Connection Refused": Ensure the bootstrap_url is correct and reachable (try ping).
• "0 Nodes Connected": Check if the Worker Node can reach the Main Point's P2P address (WSS).
• Colab Disconnects: Ensure the Colab tab stays open. Tunnels change if you restart the notebook.

---

🤝 Contributing

Contributions are welcome! Please open an issue or PR on GitHub.