NVIDIA ACE

NVIDIA ACE is a suite of RTX accelerated digital human technologies, and its 2026 direction is the part worth paying attention to: characters that run locally on the player's GPU rather than calling a cloud service. The ACE Game Agent SDK, released as an open source beta at Unreal Fest 2026 alongside Unreal Engine 5 plugins, exposes three API layers. Agent APIs are stateful and own conversation history, driving the inference loop and tool call chains autonomously with minimal application code. Chat APIs are stateless and hand developers direct control over inference. RAG APIs enable semantic, lexical and hybrid retrieval from developer built databases so responses stay grounded in current game knowledge. Underneath sit small language models capable of planning at the frequency real time decision making demands, plus multimodal SLMs that let characters hear audio cues and perceive their surroundings. The In-Game Inferencing SDK manages plugin based inference alongside rendering using compute in graphics. Because everything runs on device, there is no network latency, no per interaction server bill and no internet requirement, and character memory persists across an entire campaign on the player's machine. It has shipped in real titles including inZOI's Smart Zoi NPCs and NARAKA: BLADEPOINT. The trade off is honest: the inference cost moves onto the player's GPU.