Noko

🦙 Overview

Noko is a user-friendly Godot plugin that facilitates seamless interaction with Ollama models via API. It empowers developers to enhance their games with interactive Large Language Models (LLMs), enabling dynamic dialogues, intelligent NPCs, and more.​

• Effortless Integration: Quickly connect your Godot project to local Ollama models.
• Versatile Model Support: Compatible with various models like Gemma, LLaMA, Mistral, and others.
• Chat Functionality: Implement chat-based interactions with AI models.
• Model Management: Load, unload, and fetch model information directly within Godot.
• Blob Handling: Efficiently manage model blobs to optimize performance.

📦 Installation

1. Download Noko:

   • Clone or download the repository.
   • Copy the addons/noko directory into your Godot project's addons folder.​

2. Enable the Plugin:

   • Open your project in Godot.
   • Navigate to Project > Project Settings > Plugins.
   • Locate Noko in the list and click Enable.​

🚀 Getting Started

Ensure you have Ollama installed and running with your desired model.​

Here's a basic example of using Noko to interact with a model:

extends SceneTree

# Preload the NokoModel module, which provides
# functions to manage AI models.
const NokoModel = preload("res://modules/NokoModel.gd")

# Preload the NokoPrompt module, which provides functions
# to interact with AI models via prompts.
const NokoPrompt = preload("res://modules/NokoPrompt.gd")

func _init():
    # Defer the execution of _start_request to ensure
    # the scene tree is fully initialized.
    call_deferred("_start_request")

func _start_request():
    # Retrieve the root node of the scene tree.
    var root = get_root()

    # Define the name of the AI model to use.
    var model = "gemma3"

    # Define the conversation history as a list of message dictionaries.
    var prompt = [
        {
            "role": "user",
            "content": "Hello! I am Nathanne."
        },
        {
            "role": "assistant",
            "content": "Hello, Nathanne. I am Gemma, an AI assistant developed by Google DeepMind."
        },
        {
            "role": "user",
            "content": "Can you tell me back my name? (Answer in one sentence only)"
        }
    ]

    # Define the server details where the AI model is hosted.
    var server = {
        "host": "http://localhost", # Server host address.
        "port": 11434               # Server port number.
    }

    # Asynchronously load the specified chat model on the server.
    if await NokoModel.load_chat_model(root, server, model, false):
        print("Successfully loaded model: " + model)
    else:
        print("Something went wrong trying to load model: " + model)
        quit()  # Terminate the application if loading fails.

    # Asynchronously send the conversation prompt
    # to the AI model and await the response.
    var generated = await NokoPrompt.chat(
        root,
        server,
        "gemma3",
        prompt
    )

    # Output the user's last message.
    print("Me:\t" + prompt[2]["content"])

    # Output the assistant's response from the AI model.
    print("Gemma:\t" + generated["body"]["message"]["content"])

    # Asynchronously unload the chat model from the server to free up resources.
    if await NokoModel.unload_chat_model(root, server, model, false):
        print("Successfully unloaded model: " + model)
    else:
        print("Something went wrong trying to unload model: " + model)

    # Terminate the application after the interaction is complete.
    quit()

📚 Documentation

Noko provides several modules to interact with Ollama models:​

• NokoModel.gd: Functions to load, unload, and fetch information about models.
• NokoPrompt.gd: Handles sending prompts and receiving responses from models.
• NokoRunner.gd: Utilities to check the status and version of the Ollama runner.
• NokoBlobs.gd: Manages model blobs for efficient performance.​

Each module is documented with inline comments to guide you through their functionalities.

🛠️ Requirements

1. Godot Engine 4.4.x or higher.
2. Ollama installed and running with your chosen model.

📄 License

This project is licensed under the MIT License. See the LICENSE file for details.