🎛️ OmniBridge

A sleek, highly optimized hybrid LLM chat client built with Next.js (App Router), TypeScript, and Tailwind CSS. It is architected for the ultimate geek setup, allowing you to seamlessly switch between private, low-spec local edge computing (e.g., llama.cpp GGUF instances on old CPUs) and powerful remote cloud endpoints (DeepSeek, OpenRouter, OpenAI) within a single, unified interface.

---

✨ Core Features

• 🔀 Hybrid Dual-Mode Switch:
  • Local Mode: Connects directly to your local llama.cpp instances. Uses dual-port thread splitting (Chat vs. Auto-Title) to squeeze maximum performance out of legacy hardware (tested smoothly on 4th-Gen i3, 4GB RAM) without system freeze.
  • Remote Mode: Connects to any cloud-based OpenAI-compatible API. Features customizable BaseURL, masked API Key, and a dynamic model selector dropdown at the bottom area.
• 🧠 Context-Aware Auto-Title Generation: Automatically distills conversation intent into crisp, 2-word engineering topics (e.g., Nginx API Configuration) instead of raw text copying. Empowered by a tailored 200 tokens physical context cache and specialized summary prompt routing.
• 🛡️ Industrial-Grade Robustness:
  • Aggressive Endpoint Normalization: Automatically sanitizes messy input URLs (removes accidental copy-pasted trailing slashes or duplicate /chat/completions suffixes) to ensure zero 404 Bad Request errors.
  • Anti-Repetitive Guard: Built-in string sanitization that filters out local model infinite loops, token bleeding, and provides quick length-capped fallbacks.
• ⚙️ High-Granularity UX Control: A frosted, semi-transparent Server Settings panel featuring instant visual state feedback (disables and dims local inputs when switching to cloud endpoints, and vice versa).

---

🛠️ Tech Stack

• Frontend: Next.js 14+ (App Router), TypeScript, Tailwind CSS, Lucide Icons
• Backend Runtime Support: llama.cpp (Server Edition) or any standard OpenAI-Compatible API gateway.
• Recommended Edge Models: Qwen2.5-0.5B-Instruct (GGUF Format, Q4_K_M / Q8_0)

---

🚀 Getting Started

1. (Optional) Run your local LLM instances

If you want to use the Local Mode, boot up your GGUF models using llama-server.exe with strict thread allocation to prevent resource racing:

# Terminal 1: Main Chat Model (Port 8080, N Threads)
llama-server.exe -m qwen3.6-35b.gguf -t N -c XXXX --port 8080

# Terminal 2: Title Summary Specialist (Port 3000, 1-4 Thread, 128-512 Context)
llama-server.exe -m qwen2.5-0.5b-instruct.Q8_0.gguf -t N -c xxxx --port 3000

2. Clone & Launch the Client

# Clone the repository
git clone [https://github.com/haliChina/VibeChat.git](https://github.com/haliChina/VibeChat.git)

# Install dependencies
npm install

# Start the dev server
npm run dev

Open http://localhost:3000, hit the top-right settings icon, flip the switch to your preferred mode, and enjoy your privatized, lightning-fast AI workstation!

---

🔒 Security & Privacy

This client is architected to perform Direct Browser-to-Endpoint Fetching. All chat payloads and API credentials communicate straight from your browser to your specified localhost or remote cloud endpoints. Your secret tokens are securely cached inside your local storage (localStorage) and never pass through any third-party middleman servers.

---

📝 License

MIT License. Free to fork, mod, and deployment to your own private servers/Docker containers!