Skip to main content

Bring Your Own Key (BYOK)

If you have your own API key for an external AI provider (OpenAI, Google Gemini, Anthropic Claude, Groq, xAI Grok, OpenRouter, or local engines like Ollama and LM Studio), you can plug it into CraftyAI. Your server's prompts route directly to your chosen provider or local endpoint, using your own quota, billing, and custom model choices.

Why use BYOK​

Key advantages:

  1. Independent Rate Limits: Your chat limits are governed by your chosen AI provider's quota rather than default cloud tiers.
  2. Direct Routing: Prompts and responses communicate directly with your configured AI endpoint.
  3. Full Model Freedom: Use any state-of-the-art model or open-source local LLM (e.g., Claude 3.7 Sonnet, GPT-5.6 Sol, Grok-2, DeepSeek R1, Llama 3.3, Qwen 2.5).
  4. Custom Cost Control: Manage token billing, spend caps, and API keys directly in your provider's developer console.

What BYOK doesn't change​

Your CraftyAI configuration and server features (such as in-game permissions, safety checks, Discord integration, and admin tools) remain active. When custom provider mode is enabled, chat completions route directly to your custom endpoint without consuming default cloud tier request quotas.

Supported Providers & Setup Tutorials​

CraftyAI connects to any AI provider that exposes an OpenAI-compatible HTTP REST endpoint (/v1/chat/completions).

1. Google Gemini​

Step 1: Get an API Key​

  1. Go to Google AI Studio.
  2. Sign in and click Create API Key.

Step 2: Configuration​

Plugin (config.yml):

server:
  custom_provider:
    enabled: true
    url: "https://generativelanguage.googleapis.com/v1beta/openai"
    api_key: "AIzaSy..."
    model: "gemini-3.6-flash"

Mod (craftyai.json):

{
  "custom_provider_enabled": true,
  "custom_provider_url": "https://generativelanguage.googleapis.com/v1beta/openai",
  "custom_provider_key": "AIzaSy...",
  "custom_provider_model": "gemini-3.6-flash"
}

2. OpenAI (GPT-5.6 Sol / GPT-4o)​

Step 1: Get an API Key​

  1. Navigate to the OpenAI API Platform.
  2. Create a new secret key.

Step 2: Configuration​

Plugin (config.yml):

server:
  custom_provider:
    enabled: true
    url: "https://api.openai.com/v1"
    api_key: "sk-proj-..."
    model: "gpt-5.6-sol"

Mod (craftyai.json):

{
  "custom_provider_enabled": true,
  "custom_provider_url": "https://api.openai.com/v1",
  "custom_provider_key": "sk-proj-...",
  "custom_provider_model": "gpt-5.6-sol"
}

3. Groq (Ultra-Fast Inference)​

Groq delivers sub-second response times for Llama and DeepSeek models.

Step 1: Get an API Key​

  1. Go to the Groq Console.
  2. Create an API key.

Step 2: Configuration​

Plugin (config.yml):

server:
  custom_provider:
    enabled: true
    url: "https://api.groq.com/openai/v1"
    api_key: "gsk_..."
    model: "llama-3.3-70b-versatile"

Mod (craftyai.json):

{
  "custom_provider_enabled": true,
  "custom_provider_url": "https://api.groq.com/openai/v1",
  "custom_provider_key": "gsk_...",
  "custom_provider_model": "llama-3.3-70b-versatile"
}

4. xAI Grok (Grok-2 / Grok-3)​

Connect directly to xAI's native Grok models via OpenAI-compatible endpoints.

Step 1: Get an API Key​

  1. Visit the xAI Console.
  2. Generate an API Key under Account Settings.

Step 2: Configuration​

Plugin (config.yml):

server:
  custom_provider:
    enabled: true
    url: "https://api.x.ai/v1"
    api_key: "xai-..."
    model: "grok-2"

Mod (craftyai.json):

{
  "custom_provider_enabled": true,
  "custom_provider_url": "https://api.x.ai/v1",
  "custom_provider_key": "xai-...",
  "custom_provider_model": "grok-2"
}

5. Anthropic Claude (Claude 3.7 Sonnet / Claude 3.5 Haiku)​

Anthropic Claude models can be connected via two methods:

Option A: Via OpenRouter (Simplest, No Local Setup)​

  1. Get an API key from OpenRouter.
  2. Set the model to anthropic/claude-3.7-sonnet or anthropic/claude-3.5-haiku.

Plugin (config.yml):

server:
  custom_provider:
    enabled: true
    url: "https://openrouter.ai/api/v1"
    api_key: "sk-or-v1-..."
    model: "anthropic/claude-3.7-sonnet"

Mod (craftyai.json):

{
  "custom_provider_enabled": true,
  "custom_provider_url": "https://openrouter.ai/api/v1",
  "custom_provider_key": "sk-or-v1-...",
  "custom_provider_model": "anthropic/claude-3.7-sonnet"
}

Option B: Direct Anthropic API Key (Via LiteLLM Proxy)​

If you have a direct Anthropic API key (sk-ant-...) from the Anthropic Console:

  1. Install and launch LiteLLM (which provides an OpenAI-compatible translation proxy for Claude): 
    pip install litellm
    export ANTHROPIC_API_KEY="sk-ant-..."
    litellm --model anthropic/claude-3-7-sonnet-20250219 --port 4000
  2. Configure CraftyAI to point to your LiteLLM proxy:

Plugin (config.yml):

server:
  custom_provider:
    enabled: true
    url: "http://localhost:4000/v1"
    api_key: "sk-ant-..."
    model: "claude-3-7-sonnet-20250219"

Mod (craftyai.json):

{
  "custom_provider_enabled": true,
  "custom_provider_url": "http://localhost:4000/v1",
  "custom_provider_key": "sk-ant-...",
  "custom_provider_model": "claude-3-7-sonnet-20250219"
}

6. OpenRouter (Access 100+ AI Models)​

OpenRouter provides access to Nemotron, Claude, DeepSeek R1, Llama 3.3, and Mistral through a single unified key.

Step 1: Get an API Key​

  1. Visit OpenRouter Keys.
  2. Generate an API Key.

Step 2: Configuration​

Plugin (config.yml):

server:
  custom_provider:
    enabled: true
    url: "https://openrouter.ai/api/v1"
    api_key: "sk-or-v1-..."
    model: "nvidia/nemotron-3.5-ultra-content-safety"

Mod (craftyai.json):

{
  "custom_provider_enabled": true,
  "custom_provider_url": "https://openrouter.ai/api/v1",
  "custom_provider_key": "sk-or-v1-...",
  "custom_provider_model": "nvidia/nemotron-3.5-ultra-content-safety"
}

7. Ollama (100% Local & Free Offline AI)​

Run open-source LLMs locally on your server hardware without sending data to external APIs.

Step 1: Install & Run Ollama​

  1. Install Ollama.
  2. Pull a model (e.g. ollama run llama3.2 or ollama run qwen2.5).

Step 2: Configuration​

Plugin (config.yml):

server:
  custom_provider:
    enabled: true
    url: "http://localhost:11434/v1"
    api_key: "ollama"
    model: "llama3.2"

Mod (craftyai.json):

{
  "custom_provider_enabled": true,
  "custom_provider_url": "http://localhost:11434/v1",
  "custom_provider_key": "ollama",
  "custom_provider_model": "llama3.2"
}

8. LM Studio (Local GUI Inference)​

Step 1: Start Server​

  1. Open LM Studio and download a GGUF model (e.g. Qwen 2.5 7B or Mistral 7B).
  2. Start the Local Server on port 1234.

Step 2: Configuration​

Plugin (config.yml):

server:
  custom_provider:
    enabled: true
    url: "http://localhost:1234/v1"
    api_key: "lm-studio"
    model: "local-model"

Mod (craftyai.json):

{
  "custom_provider_enabled": true,
  "custom_provider_url": "http://localhost:1234/v1",
  "custom_provider_key": "lm-studio",
  "custom_provider_model": "local-model"
}

Reloading & Verification​

After editing your configuration file:

  • Plugin: Run /crafty reload in-game.
  • Mod: Restart your client or server.

Test connection status by chatting in-game or running /crafty status.

Costs & Monitoring​

Custom providers bill directly through your respective account dashboard. Set usage limits and billing alerts in your provider's cloud dashboard (Google AI Studio, OpenAI, Anthropic/OpenRouter, Groq, xAI) to manage expenditures.

Safety & Fallbacks​

  • CraftyAI's built-in prompt safety filters run prior to forwarding requests to custom endpoints.
  • If a custom provider experiences an outage, requests will fail gracefully in-game. Ensure your provider key and URL are active and accessible.