Multiverse (Project Jupiter)

🚧 EARLY DEVELOPMENT — ACTIVE BUILD PHASE

✅ Working: Jupiter agent sidebar • TF-IDF knowledge search • 3× LLM providers (WebGPU/Ollama/API) • Mesh gossip protocol • WebRTC cross-location P2P • Network topology visualizer • SearXNG/Brave/DDG cascading search

🔨 In Progress: Real-time dashboard metrics • Plugin ecosystem • CRDT knowledge sync

Multiverse is a decentralized, agent-first browser and knowledge engine. Codenamed Project Jupiter, every instance is both a client and a server — there is no central authority. AI agents are first-class citizens, participating in a P2P mesh to share knowledge, distribute query resolution, and guarantee absolute user privacy.

---

🏗️ Architecture

┌─────────────── Browser Shell (React/Tauri) ───────────────┐
│  AgentSidebar • Dashboard • Topology • Settings • Plugins │
├───────────────── Agent Core ("Jupiter") ──────────────────┤
│  Intelligence Gradient • Tool Registry • LLM Providers    │
├──────────────── Mesh Network (P2P) ───────────────────────┤
│  BroadcastChannel (local) + WebRTC (cross-location)       │
│  Signaling Server • STUN NAT traversal • Agent Protocol   │
├───────────────── Data Layer ──────────────────────────────┤
│  IndexedDB • TF-IDF Search • Anti-poisoning (trust/conf)  │
├───────────────── Tauri Rust Backend ──────────────────────┤
│  Ollama Proxy (CORS-free) • Ed25519 Identity • Crypto     │
└───────────────────────────────────────────────────────────┘

See TECHNICAL_ARCHITECTURE.md for detailed data flows and design rationale.

---

🧠 Intelligence Gradient

Jupiter resolves queries through a 4-tier fallback chain:

Tier	Source	Latency	Privacy
1	Local Knowledge (TF-IDF on IndexedDB)	~1ms	🟢 100% local
2	Mesh Peers (BroadcastChannel + WebRTC)	~50ms	🟡 Peer sees query
3	Web Search (SearXNG → Brave → DDG cascade)	~500ms	🟡 Search engine sees query
4	LLM Generation (WebGPU / Ollama / API)	~2-5s	🟢 Local or 🔴 Cloud

---

🌐 Cross-Location P2P

Two machines anywhere in the world can connect directly. No central server relays data.

• Same WiFi: Automatic via BroadcastChannel (tab-to-tab)
• Different WiFi / Different Countries: WebRTC via signaling server + STUN NAT traversal
• STUN servers used: Google (stun.l.google.com:19302), Cloudflare (stun.cloudflare.com:3478) — free, no account

Node A (Lagos) ──WebSocket──→ Signaling Server ←──WebSocket── Node B (NYC)
                           (handshake only, then:)
Node A ────────────── WebRTC DataChannel (encrypted) ────────── Node B

To run the signaling server:

cd signaling-server && npm install && npm start

---

🚀 Getting Started

Prerequisites

• Node.js v18+
• Rust (for Tauri builds)
• Ollama (optional, for local LLM)
• Browser with WebGPU support (Chrome 113+, Edge 113+)

Installation

git clone https://github.com/HarperKollins/Multiverse.git
cd Multiverse/multiverse
npm install

Run (Web Dev Mode)

npm run dev

Run (Tauri Desktop App)

npm run tauri dev

Connect Ollama

If using Ollama for local LLM, set the CORS environment variable:

Windows (PowerShell):

[System.Environment]::SetEnvironmentVariable("OLLAMA_ORIGINS", "*", "User")
# Restart Ollama after setting

macOS/Linux:

export OLLAMA_ORIGINS="*"
ollama serve

---

⚙️ Configuration

Open the Agent Sidebar → click ⚙️ Settings:

Setting	Options
LLM Provider	WebGPU (in-browser) · Ollama (local) · OpenAI API
Search	SearXNG (self-hosted) · Brave (2000 free/mo) · DuckDuckGo · Google CSE
Signaling Server	Default: ws://localhost:9090

---

📂 Project Structure

multiverse/
├── src/
│   ├── components/        # React UI components
│   ├── lib/
│   │   ├── agent-core/    # LLM providers, search, runtime, tool registry
│   │   ├── plugins/       # Plugin loader and API
│   │   ├── database.ts    # IndexedDB persistent storage
│   │   ├── knowledge.ts   # TF-IDF search engine
│   │   ├── peer-manager.ts # BroadcastChannel mesh
│   │   ├── webrtc-transport.ts # WebRTC cross-location P2P
│   │   ├── mesh-types.ts  # Protocol type definitions
│   │   └── logger.ts      # Structured logging & metrics
│   └── App.tsx
├── src-tauri/
│   └── src/
│       ├── lib.rs         # Ollama proxy + identity commands
│       └── identity.rs    # Ed25519 node identity
├── signaling-server/      # WebSocket peer discovery server
├── docs/
│   └── agent-protocol.md  # Agent communication protocol spec
├── TECHNICAL_ARCHITECTURE.md
├── VISION.md
└── LICENSE                # MIT

---

🤝 Contributing

1. Maintain Layer Boundaries: Frontend components should not directly access IndexedDB — go through the Agent Core.
2. Agent Protocol: All mesh messages must follow the Agent Protocol spec.
3. Open an Issue first before submitting a Pull Request.

---

Invert the architecture. Decentralize the intelligence. Build the Multiverse.