NobodyWho is a Godot Engine plugin enabling developers to integrate local Large Language Models (LLMs) into their games, facilitating AI-driven characters and interactive narratives without internet dependency. It targets game developers seeking to embed sophisticated NPC interactions and dynamic storytelling directly within their projects.

How It Works

The plugin leverages llama.cpp for efficient, local LLM inference, supporting GPU acceleration via Vulkan or Metal. It provides two core nodes: NobodyWhoModel to load and manage GGUF-formatted LLM weights, and NobodyWhoChat for conversational interaction, allowing developers to send prompts and receive streamed or complete responses. This approach ensures fast, on-device processing and simplifies LLM integration into game logic.

Quick Start & Requirements

• Install: Via Godot's AssetLib (search "NobodyWho") or by importing a zip from the releases page.
• Prerequisites: Godot 4.3+, a GGUF-formatted LLM file (e.g., Gemma 2 2B), and sufficient RAM for the LLM.
• Setup: Two nodes (NobodyWhoModel, NobodyWhoChat) added to a Godot scene.
• Docs: https://github.com/nobodywho-ooo/nobodywho

Highlighted Details

• Enables LLM-driven characters locally without internet.
• Fast inference on GPU via Vulkan or Metal.
• Cross-platform support (Windows, Linux, macOS).
• Includes example scripts for chat and embeddings.

Maintenance & Community

• Active development with community contributions welcomed.
• Discord and Matrix communities available for support and discussion.
• Issue tracker for bug reporting and feature requests.

Licensing & Compatibility

• Permissive license allowing use in proprietary and commercial projects.
• Modified versions of the plugin's code must be open-sourced.

Limitations & Caveats

Exported projects require manual copying of the GGUF model file to the export directory due to current packaging limitations. HTML5, Android, and iOS exports are not yet supported, though progress is being made on mobile platforms.