README.md

Thronekeeper

Universal AI Model Routing for Claude Code — Use any AI provider (OpenRouter, OpenAI, Together, Deepseek, GLM) with Claude Code and Anthropic-compatible clients.

🎉 v1.5.62 — Enable dynamic model fetching for Deepseek/GLM providers! ✨

🚀 Quick Start

Prerequisites

• VS Code 1.85.0 or higher (or Cursor, Windsurf, etc.)
• Claude Code installed
• API key from your chosen provider(s) (OpenRouter, OpenAI, etc.)

Installation

1. Download the Extension

• Go to Releases and download the latest thronekeeper-{version}.vsix

2. Install in VS Code

# Command line
code --install-extension thronekeeper-1.5.62.vsix

Or via UI: Extensions (Cmd/Ctrl+Shift+X) → "…" → Install from VSIX…

First-Time Setup

1. Open the Thronekeeper panel
   • View → Panel → Thronekeeper
   • Or Command Palette: Thronekeeper: Open Panel
2. Configure your provider
   • Select provider (OpenRouter recommended for 400+ models)
   • Click "Store API Key" and enter your key
   • Choose models or use recommended pairings
3. Start the proxy
   • Click "Start Your AI Throne"
   • Extension auto-configures Claude Code if enabled
4. Start coding
   • Claude Code now uses your selected models

Notes:

• When you click "Stop Proxy" your Claude Code settings revert to Anthropic defaults.
• If you enter an Anthropic API Key, Thronekeeper refreshes the default Anthropic models list; the key is used only to fetch defaults, not for proxy coding tasks.
• Thronekeeper works per-project. To run multiple instances, set different ports in settings.

Recommended: OpenRouter Setup

1. Get free API key: https://openrouter.ai/keys
2. Select "OpenRouter" in Thronekeeper
3. Store your API key
4. Browse 400+ models or use pairings, e.g.:
   • Speed: qwen/qwen-2.5-coder-32b-instruct
   • Quality: deepseek/deepseek-r1
5. Start proxy

✨ Key Features

• Multi-Provider Support — OpenRouter, OpenAI, Together, Deepseek, GLM, custom endpoints
• Secure Storage — API keys in VS Code keychain, never plaintext
• Three-Model Mode — Separate reasoning/completion/value models for optimal performance
• Real-Time Model Loading — Browse and search available models
• Dynamic Model Loading — Deepseek & GLM fetch models via OpenAI-compatible /models endpoints
• Proxy Lifecycle — Start/stop/monitor from the panel
• CLI Available — Headless proxy management via throne command

⌨️ CLI (Headless Usage)

The throne CLI lets you run Thronekeeper without VS Code:

npm install                    # Install dependencies
throne status                  # Check proxy status
throne config set provider openrouter
throne keys set openrouter    # Store API key
throne models list            # Browse available models
throne start                  # Start proxy daemon
throne stop                   # Stop daemon
throne setup                  # Interactive setup wizard

For full CLI reference, see docs/cli.md.

📖 Documentation

• Advanced Configuration — docs/advanced-setup.md
• Deepseek/GLM Setup — docs/deepseek_glm.md

🔧 Configuration

Key settings in VS Code Settings or settings.json:

{
  "claudeThrone.provider": "openrouter",
  "claudeThrone.proxy.port": 3000,
  "claudeThrone.autoApply": true,
  "claudeThrone.twoModelMode": false
}

Note: claudeThrone.twoModelMode enables “three-model” selection (reasoning/completion/value) in the UI.

Troubleshooting

• Extension won’t install: ensure VS Code is up to date, or run code --install-extension path/to/file.vsix --force.
• Proxy won’t start: verify the configured port is free; check Output → Thronekeeper for logs.
• Models not loading: confirm provider API key is stored; for Deepseek/GLM the panel shows “Enter an API key to see models.” when unauthorized.

📦 Building from Source

git clone https://github.com/KHAEntertainment/thronekeeper.git
cd thronekeeper
npm install
npm run ext:package  # Creates .vsix in extensions/thronekeeper/

🙏 Attribution

Thronekeeper evolved from anthropic-proxy by Max Nowack. While the architecture has been rebuilt, we’re grateful for the inspiration.

License: MIT
Version: 1.5.62