GladeCore — real conversational AI for your game characters

Players talk to NPCs by text or voice and get dynamic, unscripted replies in character, driven by a language model that runs entirely on the player's machine. You define who a character is with in-engine Data Assets (personality, emotions, backstory, world lore). GladeCore handles the inference, the speech, and the memory.

No cloud account. No latency to a remote server. No usage costs. Ship it and it just works offline.

What you get

🧠 Local LLM dialogue: unscripted, in-character conversation powered by an on-device model (any GGUF). Fully offline by default.

🎭 Personality & emotion system: tune 10 built-in emotional dimensions (affinity, sarcasm, trust, empathy, formality, curiosity, playfulness, confidence, cautiousness, flirtatiousness) plus your own custom emotions. Set it all in the editor, no code.

📖 World lore & character knowledge: per-character backstory plus shared world lore you can reuse across every NPC.

🔎 RAG knowledge grounding: feed characters documents and facts so they answer from your world, not the model's imagination. Optional strict/refusal modes keep them on-script.

💾 Conversational memory: automatic summarization of older turns so long conversations stay coherent without blowing the context budget.

🗣️ Text-to-Speech: local neural voices (Piper & Kokoro) or ElevenLabs if you want it. Streaming, sentence-by-sentence, so characters start talking immediately.

🎤 Speech-to-Text: push-to-talk voice input via on-device Whisper. Players literally talk to your NPCs.

👄 Lip-sync: viseme/mouth-shape output driven straight from the generated audio.

🛠️ Agentic actions: let characters do things in your game, not just talk (open doors, give quests, react to the world) via constrained action/tool calling.

☁️ Optional cloud backend: prefer to call OpenAI, Anthropic, Groq, OpenRouter, Ollama, or your own endpoint instead of running local? Flip one setting. Local stays the default.

Built for game devs, not ML engineers

Drop the prefab/component onto a character, assign a personality asset, hit play. The drag-and-drop path means you can have a talking NPC working in minutes without writing inference code, and the full programmatic API is there when you need control.

Why local-first?

💵 $0 running cost: no per-token API bills as your playerbase grows. 📡 Works offline: no internet dependency, no third-party outage taking your NPCs down. 🔒 Privacy: player conversations never leave their device. 🎛️ You own the stack: pick your own model, your own voices, your own pace.

Quick start

1. Download gladecore-1.0.zip and drop the plugin into your project.
2. Grab a GGUF language model (any will work, a small one runs fine on modest hardware).
3. Add the GladeCore character component / prefab and assign a personality Data Asset.
4. Hit Play and start talking.

Full setup docs: https://docs.gladecore.com/use-cases/gaming-setup

FAQ

Does it need an internet connection? No. The default pipeline runs 100% on-device. Cloud providers are an optional opt-in.

What does it cost to run? Nothing per message. It's your player's CPU/GPU doing the work.

Which engines? Unreal Engine and Godot.

Can I use my own voice / model? Yes. Bring any GGUF model and swap TTS voices freely.