LLMesh 🛸🌊

Navigate the Decentralized Cosmos: P2P Chat with LLM Intelligence

• A browser-based P2P chat mesh where LLM joins as a peer via UCEP extensions, creating a living network where AI and humans chat together in a fully decentralized manner.

What It Is 🌌

• Decentralized Chat - No servers, just peers talking directly
• LLM as Extension - AI assistant discoverable via UCEP protocol
• Browser-to-Terminal - Connect browser UI to headless agent node
• GossipSub Mesh - Real-time message broadcasting

Why this exists

• To explore and tinker around the idea that P2P doesn't have to be just limited to boring terminal logs. By combining js-libp2p: universal-connectivity with a local LLM Agent, I tried to create a "living" mesh where your first peer is a cosmic entity(llm-persona) that helps you flow.
• No central servers, just you, your peers, and the Forge of Creation.

Core Features

🛸 UCEP Extension System

• Provider - Terminal node exposes LLM as extension
• Consumer - Browser discovers and uses extensions dynamically
• Protocol-Based - Discover via Identify, execute via direct streams

💬 P2P Chat

• GossipSub - Decentralized message broadcasting
• Encrypted - Noise protocol for security
• Real-time - Instant message delivery

🤖 LLM Integration

• Ollama - Local LLM (default: llama3.2)
• OpenAI - Cloud fallback option
• UCEP Extension - Discoverable AI service

Architecture Flow

┌─────────────┐                    ┌─────────────┐
│   Browser   │                    │  Terminal   │
│   (p2p.js)  │                    │ (index.js)  │
│             │                    │             │
│  UCEP       │◄──Identify─────────►│  UCEP       │
│  Consumer   │   Protocol          │  Provider  │
│             │                    │             │
│  ┌─────────┐│                    │  ┌─────────┐│
│  │GossipSub││◄───Mesh─────────────►│  │GossipSub││
│  └─────────┘│                    │  └─────────┘│
│             │                    │             │
│             │                    │  ┌─────────┐│
│             │                    │  │   LLM    ││
│             │                    │  │ Service  ││
│             │                    │  └─────────┘│
└─────────────┘                    └─────────────┘

graph LR
    User[Browser Node] -- WebSocket Dial --> Agent[Terminal Node]
    Agent -- GossipSub Mesh --> User
    Agent -- Loopback --> Ollama[Local LLM]
    Ollama -- Reply --> Agent
    Agent -- PubSub Message --> User

Loading

1. Browser spawns a libp2p node (WebSocket transport).
2. Terminal Agent listens on TCP + WebSocket and bridges the LLM.
3. Connect: Browser dials Agent's /ws multiaddr.
4. Mesh: Gossipsub mesh forms; peers sync.
5. Chat: Messages flow over the mesh; Alien X replies via the Agent logic.

UCEP Extension Flow

1. Terminal Node
   └─> Registers LLM Extension
       Protocol: /uc/extension/alien-x-llm/1.0.0

2. Browser Connects
   └─> Identify Protocol Exchange
       Discovers: /uc/extension/alien-x-llm/1.0.0

3. Browser Requests Manifest
   └─> Fetches extension metadata (commands, description)

4. User Sends Message
   └─> Browser executes: chat <message>
       └─> Direct protobuf stream to terminal
           └─> LLM processes → Response
               └─> Returns to browser

• Discovery - Via Identify protocol
• Manifest - Extension metadata exchange
• Execution - Direct protobuf streams
• Decoupled - Extensions independent of chat

Quick Start

LLMesh.Demo.1.mp4

Prerequisites

• Node.js (v18+)
• Ollama (Download) - Optional but recommended
• ChatGPT API key
• Model: llama3.2 (default) (auto-downloaded on first use)

Step 1: Start Terminal Agent

cd libp2p-ai/p2p-X/web/app
npm install
node index.js

Copy the /ws multiaddr from output:* [SYSTEM] /ip4/127.0.0.1/tcp/xxxxx/ws/p2p/Qm...

Note the ws:// address and port number, e.g., /ip4/127.0.0.1/tcp/57704/ws/p2p/...

Tip 💡:

local running: use `ws` supported multiaddr to connect /ip4/127.0.0.1/tcp/56989/ws/p2p..
production url: use `webrtc-direct` addr /ip4/127.0.0.1/udp/56987/webrtc-direct/certhash/../p2p/..

Step 2: Start Ollama (Optional)

ollama serve
# First time? Pull model: ollama pull llama3.2

Step 3: Launch Browser UI

cd libp2p-ai/p2p-X/web
npm install
npm run dev

Open http://localhost:5173

Step 4: Connect & Chat

1. Click "LET'S FLOW"
2. Paste the /ws multiaddr from Step 1
3. Wait for "MESH SYNCED"
4. Alien X greets you - Start chatting! 🛸

Multiaddr Formats

Context	Format	Example
Local	/ip4/127.0.0.1/tcp/XXXX/ws/p2p/...	✅ Use this
Production	/ip4/.../udp/.../webrtc-direct/...	HTTPS required
Wrong	/tcp/... or /tls/ws/...	❌ Not supported

Requirements

Local Setup

• ✅ Ollama running (ollama serve)
• ✅ Terminal agent running (node index.js)
• ✅ Browser UI running (npm run dev)
• ✅ Use /ws multiaddr format

Production (Vercel)

• ⚠️ Requires /webrtc-direct multiaddr (HTTPS)
• ⚠️ LLM disabled (ngrok free tier limitations)
• 💡 Best experience: Run locally with Ollama

Key Components

Terminal Node (app/index.js)

• UCEP Provider - Registers LLM extension
• P2P Gateway - Bridges browser to mesh
• Protocol Hub - TCP, WS, WebRTC, Relay

Browser Node (src/lib/p2p.js)

• UCEP Consumer - Discovers and uses extensions
• GossipSub Client - Chat messaging
• Extension Manager - Handles UCEP & GossipSub extensions

LLM Extension (app/llm-extension-provider.js)

• Extension Definition - Commands, manifest
• Command Handler - Processes chat requests
• LLM Integration - Ollama/OpenAI calls

Extension Commands

LLM Extension (alien-x-llm)

// In browser console
window.listExtension()
window.testExtension('alien-x-llm', 'chat', ['Hello!'])
window.testExtension('alien-x-llm', 'ping')

Echo Extension (echo)

window.testExtension('echo', 'echo', ['test message'])
window.testExtension('echo', 'ping')

What Makes It Special

✨ LLM as Peer - AI joins the mesh, not just responds
✨ UCEP Discovery - Extensions found automatically
✨ No Servers - Pure P2P, no central authority
✨ Extensible - Add new extensions easily
✨ Real-time - Instant message delivery

Resources

• libp2p Docs: https://docs.libp2p.io/
• Universal Connectivity Workshop: https://github.com/libp2p/universal-connectivity-workshop
• js-libp2p: https://github.com/libp2p/js-libp2p

---

Built with Svelte, Libp2p, and Cosmic Energy imbibed from Universal Connectivity Workshop. ❤️‍🔥🚀