Skip to main content

Elys Mind

Local large-language-model inference for Unreal Engine 5. Elys Mind runs GGUF models on-device with a bundled llama.cpp backend — fully offline, no cloud, no API keys. It is Blueprint-first and ships as a fully standalone plugin.

Why Elys Mind​

  • 100% offline — runs on the player's CPU/GPU; nothing leaves the machine.
  • Any GGUF model — Llama, Gemma, Qwen, Phi, SmolLM, Mistral… download from the built-in catalog or point at your own file.
  • Conversation context — system prompts, history, and per-actor contexts for NPCs.
  • Pluggable backend — IERP_LLMBackend for alternative engines.
  • Transparent chaining — PushText (FString) in, OnText (FString) out, so it drops between Elys Listen (STT) and Elys Speak (TTS). See Chaining.

Installation​

Install Elys Mind from the FAB Marketplace (Vault → Add to Project), enable it in Edit → Plugins → Elys'Plugins, then download a model from Tools → Elys Mind → Download LLM Models. See the Setup Guide.

A 30-second example​

UERP_LLMComponent* LLM = NPC->FindComponentByClass<UERP_LLMComponent>();
LLM->SetSystemPromptRuntime(TEXT("You are a gruff but kind tavern keeper."));
LLM->OnGenerationComplete.AddDynamic(this, &ANPC::HandleReply);
LLM->SendMessage(TEXT("Got any rooms for the night?"));

At a glance​

EngineUE 5.7 (Win64; Mac/Linux planned)
Backendllama.cpp — any GGUF model, CPU/GPU offload
Dependencynone (no external Elys plugin)
I/OSendMessage / PushText in; OnGenerationComplete / OnText out