Skip to main content
  1. Add the Glade Core Component:
    • Open your player character Blueprint.
    • Click + Add in the Components panel and search for GladeCoreComponent. Add it to the character. This will automatically add all the necessary sub-components that are configurable within this component.
  2. Configure the LLM Component:
    • In the Components panel, select the GladeCore LLMComponent that was just added.
      • Under the Chat category, assign your UChatInputWidget Blueprint to the ChatInputWidgetClass property.
      • Under the LLM category, you will find the LLMServiceManager instance. This is where you configure the core AI service.
  3. Set your TTS Provider
    • Set the TTSProvider to Piper (local option) or ElevenLabs (cloud option).
    UE Piper TTS Provider Selection
    • If using Piper, you will need to import Piper models. Navigate to Window → Project Settings and you will find Glade Core Settings under the Game tab.
    UE Project Settings Glade Core Settings
    • Under Piper Voice Downloads you can download preset piper models or import custom models.
    UE Piper Voice Download
    • Below that there is a section to show the status of your current download in progress or idle state. Lastly, there is a custom voice importer for any other voices you may way to import online whether it be custom trained voices from yourself or other open source projects.
    UE Custom Piper Voice Importer
    • So long as you have the 2 files for the model in the format of .onnx and config.json, you can import the custom model.
    • If using ElevenLabs, select ElevenLabs as the TTS Provider and enter the specific Eleven Labs Voice ID string here. Note that using this would be a cloud approach requiring an API instead of a fully local approach.
    UE Eleven Labs TTS Manager
  4. GPU Layer Offloading
    • In the LLMServiceManager, the GPU Layers integer controls how many model layers are offloaded to the GPU. This lets you tune VRAM usage vs. performance.
    • Values:
      • 999 (default): All layers on GPU - maximum performance, requires enough VRAM for the full model
      • 0: CPU only - no GPU used, slowest but works on any machine
      • 1-N: Partial offload - first N layers on GPU, remainder on CPU (useful for limited VRAM)
Below is an example image of a configured LLMComponent. LLM Component Example Pn Below is an example image of a configured Speech Chat Component. Speech Chat Component Example Pn Below is an example image of where to enable or disable Speech to Text and Text to Speech. Enable Disable TTS STT Pn