As part of Unreal Fest 2026, NVIDIA has announced that its new NVIDIA ACE Game Agent SDK for creating and building in-game AI companions that run on GeForce RTX hardware is now available in Beta. This suite of tools includes Unreal Engine 5 plugins for the AI-powered Automatic Speech Recognition (ASR), Small Language Models (SLM), and Text-to-Speech (TTS) components of NVIDIA ACE.

Examples of the technology in action are already available, with KRAFTON recently launching its new PUBG Ally in-game AI teammate as part of a limited-time Beta to test and gauge feedback on a new Ally Duo Mode that will pair humans with AI. Another in-development example is Creative Assembly adding an AI advisor to Total War: PHARAOH.
Naturally, the NVIDIA ACE Game Agent SDK will allow any game developer to play around with in-game AI companions, with NVIDIA noting that it has been designed as a "lightweight, C/C++ agentic framework" for seamless integration thanks to the inclusion of a ready-to-use Qwen 3.5 4B model for decision-making and an ultra-lightweight Chatterbox Turbo 350M model for text-to-speech. As seen with the new PUBG Ally, ACE Game Agents can run on GPUs with only 8GB of VRAM, and cards like the GeForce RTX 3060. So yeah, you won't need a GeForce RTX 5080 to run them.

The NVIDIA ACE Game Agent SDK comprises three key APIs, and, without getting too technical, they can be summarized as follows. The Agent API handles chat history and drives multi-step reasoning with minimal coding. The Chat API gives developers complete control over inference, while the RAG (retrieval-augmented generation) API focuses on game-specific knowledge, using developer-built databases and up-to-date information to retrieve and answer player questions and act.
With Unreal Engine 5 plugins, all of these tools and features can be stored within the game and then run on local RTX hardware without any cloud-based assistance. There's a lot more coming too, with NVIDIA partnering with Epic to bring NVIDIA ACE to Unreal Engine's MetaHuman NPC technology to leverage models for voice synthesis and Audio2Face technology for facial animation that takes into account emotion and personality traits of characters.




