[Edaily Reporter Han Gwang-beom] NVIDIA announced on the 18th that it unveiled “PUBG Ally,” an AI teammate for “PUBG: Battlegrounds” (hereinafter “Battlegrounds”), developed in collaboration with #Krafton, at the game developers’ conference “Unreal Fest.”

Battlegrounds’ new “Ally Duo” mode will be available as a beta on Steam until the 30th (local time). In this mode, players can play alongside PUBG Ally, an adaptive, situationally aware AI teammate.

PUBG Ally is designed to understand the player’s commands and the battlefield situation, dynamically adjusting its playstyle accordingly. It supports the player by independently deciding on actions—such as collecting items, engaging in combat, and navigating routes—without the need for constant prompts. It understands both voice and text inputs, including casual conversation and tactical instructions, and collaborates naturally with the player based on a deep understanding of game-specific terminology, player slang, map locations, and item attributes.

The “NVIDIA ACE Game Agent SDK,” unveiled alongside PUBG Elai by NVIDIA, is a lightweight, open-source C/C++ agentic framework designed for seamless native integration within games. It is fully customizable, optimized for small models, and supports NVIDIA RTX hardware acceleration.

Through three core API categories—Agents, Chat, and RAG—developers can enable NPCs to dynamically perceive, reason about, and act within the game world based on real-time player input. The SDK has undergone performance validation through its application in features such as the experimental in-game AI advisor in “Total War: PHARAOH.”

A new “NVIDIA ACE Plugin Suite” for Unreal Engine 5 developers and technical artists was also unveiled. Unlike cloud-based services, it features RTX-optimized local workflows that reduce latency and operational costs. It natively supports Blueprint and C++ integration.

In the Automatic Speech Recognition (ASR) field, it provides the ready-to-use English model “nemo-conformer-ctc-120m,” along with download options for seven additional language models and Blueprint examples for rapid deployment. The Small Language Model (SLM) category supports the local .gguf file format and offers low-latency text generation and function call capabilities. It is optimized for character dialogue and decision-making and includes the Qwen 3.5 4B model, which is ready to use out of the box. For Text-to-Speech (TTS), it brings characters to life with high-quality Chatterbox Turbo 350M TTS models, example voices, and sample levels.

Meanwhile, NVIDIA RTX technology is tightly integrated with Unreal Engine 5 through the NVIDIA Unreal Engine RTX branch and the DLSS Unreal Engine plugin, allowing developers to directly leverage advanced rendering, frame generation, and ray-traced lighting capabilities. Detailed information is available on the NVIDIA blog.