🔀 Chat Switchboard

The Plugin-First, Multi-Model AI Platform

Chat Switchboard is a next-generation AI interface that works offline or managed, with a unique plugin architecture and visual workflow builder.

---

🎯 What Makes Us Different

Feature	Chat Switchboard	Others
Works Offline	✅ Full-featured unmanaged mode	❌ Backend required
Plugin System	✅ Core features ARE plugins	🟡 Limited or none
Visual Workflows	✅ Chain multiple AI models	❌ Single-shot only
Multi-Model Routing	✅ Auto-select best/cheapest	🟡 Manual only
Channels	✅ User + AI collaboration	🟡 Users only
Self-Hosted	✅ Easy Docker setup	🟡 Complex

Unique Selling Points:

1. Workflows - No competitor has visual AI orchestration (like n8n for LLMs)
2. Dual-Mode - Privacy-first offline mode OR full collaboration backend
3. Plugin-First - Chat, Channels, Notes are ALL plugins (proves extensibility)
4. Smart Routing - Automatic model selection for cost/quality optimization

---

🚀 Quick Start

Option 1: Offline Mode (No Backend)

# Clone and build
git clone https://git.gobha.me/xcaliber/chat-switchboard.git
cd chat-switchboard
./build.sh

# Open in browser
xdg-open standalone/index.html

Configure API:

1. Click ⚙️ Settings
2. Enter API endpoint (OpenAI, OpenRouter, Venice.ai, Ollama, etc.)
3. Add your API key
4. Start chatting!

Option 2: Full Backend (Collaboration + Workflows)

# Clone repo
git clone https://git.gobha.me/xcaliber/chat-switchboard.git
cd chat-switchboard

# Configure
cp .env.example .env
# Edit .env with your settings

# Start with Docker
docker-compose up -d

# Access at http://localhost:3000

See GETTING_STARTED.md for detailed instructions.

---

✨ Core Features

1. 💬 Chat (User → AI)

• Multi-model support (OpenAI, Anthropic, Ollama, etc.)
• Per-conversation model switching
• Streaming responses with stop button
• Export (Markdown, JSON, Plain Text)
• Auto-routing to best/cheapest model (managed mode)

2. 👥 Channels (User → User + AI)

Managed mode only

Multi-user chat rooms where you can @mention AI models:

#general
  @alice: What do you think about this design?
  @claude: I'd suggest a darker color scheme for better contrast...
  @bob: Great idea! @gpt4 can you review the implementation?
  @gpt4: I found a potential issue in the error handling...

• Public/private/DM channels
• Real-time updates (WebSocket)
• Threaded conversations
• Reactions and formatting

3. 📝 Notes & Knowledge Bases

• Markdown notes with folders
• Full-text and semantic search
• RAG (Retrieval Augmented Generation) in managed mode
• Link notes to chats

4. 🔄 Workflows ⭐ (UNIQUE FEATURE)

Managed mode only

Visual workflow builder for chaining AI models and tools:

[User Query] → [Web Search] → [GPT-4 Summarize] → [Claude Verify] → [Save to KB]

Use Cases:

• Research Assistant: Search → Summarize → Verify → Save
• Code Review: Fetch PR → Find Bugs → Security Check → Report
• Multi-Model Consensus: Run through 3 models → Vote → Best answer
• Content Factory: Outline → Draft → Edit → SEO → Publish

See WORKFLOWS.md for details.

---

🔌 Extension System

Everything is a Plugin

Chat Switchboard proves its extensibility by implementing core features as plugins:

Core (Minimal)              Plugins (Modular)
├── HTTP Router             ├── Chat Engine (Python)
├── WebSocket Hub           ├── Channels (Go)
├── Extension Manager       ├── RAG Engine (Python)
├── Auth/Users              ├── Workflows (Go/Python)
└── PostgreSQL              └── Your Custom Plugin...

Frontend Plugins (JavaScript)

// Simple UI extension
window.ChatSwitchboard.registerExtension({
  name: 'token-counter',
  hooks: {
    onMessageSend: (msg) => {
      const tokens = estimateTokens(msg);
      showToast(`~${tokens} tokens`);
    }
  }
});

Backend Plugins (Python, Go, Node.js)

# Full-featured extension
from fastapi import FastAPI

app = FastAPI()

@app.post("/tools/web_search")
async def search_web(query: str):
    results = duckduckgo_search(query)
    return {"results": results}

Create a plugin:

# Use template
cp -r extensions/_template-python extensions/my-plugin
cd extensions/my-plugin
# Edit extension.json, main.py
python main.py

See PLUGIN_SPEC.md for complete guide.

---

📚 Documentation

• Getting Started - Installation and setup
• Architecture - System design and data flow
• Workflows - Visual workflow builder guide
• Plugin Spec - Extension development
• Roadmap - Development timeline and features

---

🏗️ Architecture

┌─────────────────────────────────────┐
│  Frontend (Vanilla JS)              │
│  - Works offline (LocalStorage)     │
│  - Switches to backend if available │
└─────────────┬───────────────────────┘
              │
    ┌─────────┴─────────┐
    │                   │
[Unmanaged]       [Managed Mode]
LocalStorage           │
                       ▼
              ┌────────────────┐
              │  Go Backend    │
              │  - Auth/Users  │
              │  - WebSocket   │
              │  - Extensions  │
              └────────┬───────┘
                       │
         ┌─────────────┴──────────────┐
         │                            │
    PostgreSQL                   Extensions
    - Chats, users              - Chat (Python)
    - Channels                  - RAG (Python)
    - Knowledge                 - Workflows (Go)
    - pgvector                  - Custom tools...

---

🛠️ Tech Stack

Frontend

• Vanilla JavaScript - No framework bloat
• LocalStorage - Offline-first
• WebSocket - Real-time updates (managed)

Backend (Managed Mode)

• Go - Core API, routing, WebSocket
• PostgreSQL - Primary storage
• pgvector - Vector embeddings for RAG
• Redis - WebSocket pub/sub (optional)
• Python - AI/ML extensions
• Docker - Easy deployment

---

🎨 Screenshots

(Coming soon - will add workflow builder, channels, chat interface)

---

🗺️ Roadmap

Current: Phase 1 - Backend Core ✅

• Frontend (unmanaged mode)
• Go backend with PostgreSQL
• User authentication
• WebSocket server
• Extension manager

Next: Phase 2 - Core Features as Plugins

• Chat Engine (Python)
• Channels (Go)
• Notes (Go)
• RAG Engine (Python)

Future: Phase 3+

• Visual Workflow Builder
• Desktop app (Tauri)
• Extension marketplace
• Mobile PWA

See ROADMAP.md for complete timeline.

---

🤝 Contributing

We welcome contributions! Here's how:

1. Pick a task from ROADMAP.md or GitHub Issues
2. Fork the repo
3. Create a feature branch
4. Submit a PR

Good first issues:

• Frontend UI improvements
• Backend handler implementations
• Example extensions
• Documentation

---

📖 Example Use Cases

Personal (Unmanaged)

• Privacy-focused AI assistant
• Offline research tool
• Model comparison testing

Team (Managed)

• Collaborative AI workspace
• Shared knowledge bases
• Automated workflows
• Code review pipelines

Enterprise

• Self-hosted AI platform
• Custom model routing
• Compliance and audit logs
• SSO/SAML integration

---

🆚 Comparison

vs Open WebUI

• ✅ Works offline (unmanaged mode)
• ✅ Visual workflows (they don't have)
• ✅ Plugin-first architecture
• 🟰 Similar RAG features

vs ChatGPT/Claude Desktop

• ✅ Multi-model (not locked to one provider)
• ✅ Self-hosted option
• ✅ Open source
• ✅ Extensible (closed systems)
• 🟰 Similar UX quality

vs LangChain

• ✅ Visual workflow builder (no-code)
• ✅ Multi-model orchestration
• 🟰 Similar capabilities
• ❌ Less Python ecosystem (for now)

Unique Position: LangChain for non-coders + n8n for LLMs + privacy-first design

---

📄 License

MIT License - build anything, including commercial products.

---

🙏 Credits

Built with inspiration from:

• Open WebUI (knowledge bases, channels)
• Claude Desktop (thinking blocks)
• n8n (workflow concepts)
• VSCode (plugin architecture)

---

🔗 Links

• Repository: https://git.gobha.me/xcaliber/chat-switchboard
• Documentation: /docs
• Issues: GitHub Issues
• Discussions: (Coming soon)

---

Ready to build the future of AI interfaces? Star the repo and let's go! 🚀