NVIDIA ACE Game Agent SDK Guide: Building AI Companions for Your Games
NVIDIA released the ACE Game Agent SDK beta at Unreal Fest 2026. This practical guide shows game developers how to implement AI-powered NPCs that run locally on RTX GPUs, with real-world examples from PUBG Ally Duo Mode.
Understanding the ACE Game Agent SDK
<p>The NVIDIA ACE Game Agent SDK, announced at Unreal Fest 2026 and available in beta as of June 18, is a suite of tools that enables game developers to create AI-powered in-game companions that run entirely on local GeForce RTX GPUs. The SDK includes three core components: Automatic Speech Recognition (ASR) for understanding player speech, a Small Language Model (SLM) for decision-making and response generation, and Text-to-Speech (TTS) for natural voice output. All three run locally on the GPU with no cloud dependency, eliminating latency and privacy concerns. The SDK is designed as a lightweight C/C++ agentic framework with Unreal Engine 5 plugins for seamless integration. It ships with a ready-to-use Qwen 3.5 4B model for decision-making and an ultra-lightweight Chatterbox Turbo 350M model for TTS. NVIDIA demonstrated that ACE Game Agents can run on GPUs with as little as 8GB of VRAM, making them accessible to the vast majority of GeForce RTX users. The SDK is available now through NVIDIA’s Developer Program.</p>
Setting Up the SDK in Your Unreal Engine 5 Project
<p>Integrating ACE into your Unreal Engine 5 project requires joining NVIDIA’s Developer Program and downloading the ACE Game Agent SDK from the NVIDIA Developer portal. The SDK includes Unreal Engine 5.8 plugins that handle the heavy lifting: an ASR plugin for speech-to-text, an SLM plugin for the reasoning model, and a TTS plugin for voice synthesis. Installation follows a standard Unreal plugin workflow: copy the ACE plugins to your project’s Plugins folder, enable them in Edit > Plugins, and configure the model paths in the Project Settings under NVIDIA ACE. The SDK downloads and caches the required models on first launch—expect approximately 8GB of model data for the full suite. Runtime memory usage is approximately 4-5GB VRAM for the combined models, leaving room for the game itself on an 8GB GPU. The SDK provides sample Blueprint nodes and C++ classes for common integration patterns: spawning an ACE-controlled character, sending player speech to the ASR system, receiving an AI response, and triggering the TTS output. NVIDIA provides several sample projects including a simple NPC interaction demo and a combat companion example based on PUBG’s Ally system.</p>
Designing AI Companion Behaviors with ACE
<p>Designing ACE companion behaviours requires a fundamentally different approach from traditional NPC AI. Instead of scripting every possible response, you define the companion’s personality, knowledge, and behavioural guidelines through prompt engineering and system prompts. The ACE SLM uses a system prompt that you define, covering the character’s backstory, personality traits, knowledge boundaries, and behavioural rules. For example, PUBG’s Ally companion uses a system prompt that defines it as “a helpful but cautious squadmate who prioritises survival, communicates tactical information clearly, and follows the player’s lead.” The companion’s knowledge is further shaped by a Retrieval-Augmented Generation (RAG) system that can reference game-specific information—map layouts, weapon stats, enemy types, and mission objectives. Key design considerations include: defining clear behavioural boundaries to prevent the AI from acting outside intended parameters, implementing “safety rails” that override AI decisions in critical gameplay moments, designing fallback behaviours for when the model is uncertain or VRAM is constrained, and creating distinct companion personalities that players can bond with over time. The most effective ACE companions balance dynamic interaction with predictable gameplay reliability.</p>
Real-World Implementation: Lessons from PUBG Ally Duo Mode
<p>PUBG’s Ally Duo Mode, running from June 16-30, 2026, provides the first real-world test of ACE in a shipped game. The mode pairs players with an AI-controlled squadmate that communicates via voice, makes tactical decisions, and interacts with the environment. Early analysis reveals several important lessons. First, player engagement is high—players spend 40% more time in matches with the ACE companion compared to solo play, and session retention is up significantly. Second, voice interaction quality is critical—players who use voice chat with the ACE companion rate the experience 2.5x higher than those who rely on text communication. Third, companion reliability matters more than personality—players prefer a competent AI squadmate that follows orders correctly over one with more personality but worse tactical decisions. Fourth, context awareness is the biggest technical challenge—the ACE companion sometimes fails to understand complex multi-step instructions or misunderstands game state. KRAFTON is using player feedback to refine the companion’s model and system prompts. The beta mode has generated enormous interest, with over 1 million matches played in the first 48 hours, and NVIDIA and KRAFTON are expected to extend the trial beyond June 30.</p>
Optimization, Debugging and Performance Tuning
<p>ACE Game Agent performance optimization requires careful management of GPU resources. The combined model footprint of 4-5GB VRAM means ACE companions compete with the game itself for GPU memory. Key optimization strategies include: lower the TTS model quality setting for background companions (reserving full quality for main story characters), use streaming model inference that shares VRAM across frames rather than loading the full model at once, implement model LODs that reduce model complexity based on distance from the player, and cache common responses to avoid re-running the SLM for frequently asked questions. The SDK provides profiling tools under the NVIDIA ACE Developer Menu that show VRAM usage, model inference times, and audio latency. Target metrics: ASR latency under 200ms, SLM response under 500ms, TTS synthesis under 300ms, and total end-to-end latency under 1.5 seconds for natural conversation flow. Debugging ACE NPCs requires a different approach from traditional AI—the SDK includes a Conversation Log that records all player inputs, model responses, and internal reasoning, enabling developers to audit and improve the AI’s behaviour over time. Prompt iteration, not code changes, is the primary debugging technique for ACE companions.</p>
Frequently Asked Questions
Does the ACE Game Agent SDK work on consoles?
Currently, the SDK is designed for PC games running on NVIDIA GeForce RTX GPUs. Console support is under investigation but faces challenges due to fixed VRAM allocations on PS5 and Xbox Series. Cloud-streamed ACE on NVIDIA GeForce NOW is a potential path for console and mobile games.
Can ACE companions use custom AI models?
The SDK supports custom SLM models through ONNX Runtime and TensorRT. Developers can fine-tune the provided Qwen 3.5 4B model on their own game data or replace it entirely with a custom model, provided it meets the SDK’s API requirements.
How much does the ACE SDK cost?
The ACE Game Agent SDK is free for developers through NVIDIA’s Developer Program. There are no runtime licensing fees per player. Developers only pay for their own development costs and any cloud services they choose to use alongside the SDK.
What happens if a player doesn’t have an RTX GPU?
ACE companions require an NVIDIA RTX GPU with 8GB+ VRAM. For players without compatible hardware, developers can implement fallback behaviour that uses simpler, non-ACE NPC AI or cloud-based ACE inference through NVIDIA GeForce NOW.
Technology Team
Expert reviewer at Verdict — testing AI productivity tools since 2023.
More Guides
How to Use ChatGPT for Work: A Complete Productivity Guide
Master ChatGPT for workplace productivity with practical workflows for email, research, analysis, and content creation. Includes real-world prompts and strategies used by professionals.
ProductivityBest AI Tools for Freelancers in 2026: Complete Toolkit
A curated guide to the best AI tools that help freelancers work faster, produce better results, and earn more. From writing to design to automation, build your AI-powered freelance business.
Get the AI Tool Brief
Weekly picks, productivity tips, and early access to new reviews — straight to your inbox.