NVIDIA just dropped some massive updates at Unreal Fest, and they are completely changing the game for how AI characters operate. Instead of relying on sluggish, expensive cloud-based AI, NVIDIA is doubling down on local, hardware-accelerated on-device AI tools for developers and players alike.
Here is the breakdown of the biggest announcements.
1. Meet Your Next Duo Partner:Â PUBGÂ Ally
Ever get tired of playing with random teammates who won’t communicate? KRAFTON and NVIDIA have teamed up to build PUBG Ally, a highly adaptive, context-aware AI teammate.
Available right now through June 30 via Steam in the new Ally Duo Mode Beta, this AI partner does not just blindly follow you around. It actually understands casual text or voice commands, knows map locations, item attributes, and specific PUBG lingo.
It can independently decide when to loot, when to engage in a firefight, and how to navigate the map to support you without needing constant micro-management. This is a massive peek into the future of immersive gaming.
2. The NVIDIA ACE Game Agent SDK
For developers looking to integrate native, smart NPCs into their game logic, NVIDIA launched the NVIDIA ACE Game Agent SDK.
This is a lightweight, open-source, and fully customizable C/C++ framework optimized for small models running locally on NVIDIA RTX hardware. Developers interact with the SDK through three core API categories: the Agent API, the Chat API, and the RAG (Retrieval-Augmented Generation) API.
Together, these tools empower NPCs to dynamically perceive, reason, and act within the game world based on real-time player input. The tools have already been battle-tested in real games, such as with the experimental in-game AI advisor in Total War: PHARAOH.
3. Local, No-Cloud NVIDIA ACE Unreal Engine 5 Plugins
Cloud-based AI features for games sound great until you get hit with high latency and unpredictable server bills. To fix this, NVIDIA released a new suite of NVIDIA ACE Unreal Engine 5 Plugins that run entirely on the local GPU, supporting both Blueprints and C++ out of the box across three crucial pillars:
- Automatic Speech Recognition (ASR):Â Features a ready-to-use English language model (
nemo-conformer-ctc-120m) with downloadable packs for seven other languages, allowing NPCs to hear and transcribe player voice commands instantly. - Small Language Models (SLM):Â Provides localÂ
.gguf support for ultra-low latency text generation and function calling. It comes bundled with a ready-to-useÂQwen 3.5 4B model so NPCs can make instant, smart decisions. - Text-to-Speech (TTS): Uses the high-qualityÂ
Chatterbox Turbo 350M TTSÂ model along with sample voices to give NPCs realistic, dynamic speech without lagging out the game.
NVIDIA is making it clear that the next generation of AI in gaming isn’t happening in the cloud. It is happening right on your local RTX graphics card, making games more responsive, immersive, and infinitely more unpredictable in the best way possible.
