Pax Historia: Custom AI Backend Hook

Tampermonkey userscript that replaces Pax Historia's default AI backend with multiple providers. Use your own API keys or local proxies (Ollama, LM Studio, Copilot API) for AI chats and actions.

Supported Providers

Provider	API Key	Description
Google AI Studio	Yes	Gemini (including Thinking models)
OpenRouter	Yes	Multiple models via openrouter.ai
OpenAI	Yes	Direct OpenAI API (GPT-4, etc.)
Groq	Yes	Fast inference, free tier available
Ollama	No	Local models at http://localhost:11434
LM Studio	No	Local models at http://localhost:1234
Together AI	Yes	Open and fine-tuned models
Fireworks AI	Yes	Fast inference API
Mistral AI	Yes	Mistral models
Anthropic (Claude)	Yes	Claude models
Copilot API	No	Local proxy via copilot-api
Generic	Optional	Any OpenAI-compatible API (custom Base URL)

Features

• 12 providers: Google, OpenRouter, OpenAI, Groq, Ollama, LM Studio, Together, Fireworks, Mistral, Anthropic, Copilot, Generic
• Connection test: Verifies Base URL for Copilot, LM Studio, Ollama, Generic before saving
• Model selector: Auto-loads models from local proxies (Copilot, LM Studio)
• Thinking Budget: Configurable for Gemini models
• Indicator badge: Shows current provider and model in the header (click to open settings)
• Privacy: Prompts go to your chosen provider, not the game's default backend

Installation

1. Install Tampermonkey in your browser (Chrome, Firefox, Edge).
2. Create a new script in Tampermonkey.
3. Copy and paste the contents of PaxHistoriaAIHook.user.js.
4. Save (Ctrl+S).

Configuration

1. Open Pax Historia.
2. Click Tampermonkey icon and select "Open AI Settings", or click the indicator badge in the header.
3. Choose provider and configure:

Google AI Studio

• API Key: aistudio.google.com/app/apikey
• Model: e.g. gemini-3-flash-preview
• Thinking Budget: 4096 (recommended)

OpenRouter

• API Key: openrouter.ai/keys
• Model: e.g. google/gemini-2.0-flash-thinking-exp:free

OpenAI / Groq / Together / Fireworks / Mistral / Anthropic

• API Key: From each provider's dashboard
• Model: Provider-specific model ID

Ollama / LM Studio (local)

• Base URL: http://localhost:11434 (Ollama) or http://localhost:1234 (LM Studio)
• Model: Name of loaded model. Use "Test" to list available models (LM Studio).

Copilot API (no API key)

• Provider: Copilot API (local)
• Base URL: http://localhost:4141 (default). The [copilot-api](https://github.com/ caozhiyuan/copilot-api) proxy must be running.
• Model: Use "Test connection" to load models and select one (e.g. gpt-4.1, claude-opus-4.6).

Generic (OpenAI-compatible)

• Base URL: e.g. https://api.openai.com/v1 or Azure/custom endpoint
• Model: Model ID
• API Key (optional): Leave empty for local or public endpoints

5. Save and reload the page.

Troubleshooting

• "No events" error: The model did not return valid JSON. Try a more capable model or increase the thinking budget.
• Copilot API: Network error or no connection: Ensure the proxy is running (npx copilot-api@latest start) and the Base URL is correct.
• Script not working: Check that the script is enabled in Tampermonkey and that you have accepted the requested permissions (including GM_xmlhttpRequest for Copilot API).
• API errors: Check the browser console (F12) for logs tagged with [PAX AI].

Disclaimer

This script is for educational and personal use only. Use at your own risk.