Back to AI game news
AI game news

Atelico Opens Early Access for On-Device Game AI Engine and NPC Tools

By GameSlash

Atelico opened limited Early Access on October 29, 2026, to an AI engine that runs NPC dialogue, images and voice locally in Godot, Unity and Unreal games, with company-reported GPU tests.

On October 29, 2026, Studio Atelico opened limited Early Access to its AI Engine for games. It brings generative AI onto the player's own device, with integrations for Godot, Unity and Unreal. This is an early SDK programme, not a tool anyone can already download and ship.

The company announced the launch in its official October 29 post and accepts developer interest through its Early Access site.

Watch Studio Atelico's AI Engine Debut video

What can a game creator actually build?

The engine combines local language models, character memory and behavior controls, image generation, speech recognition, speech synthesis and audio-driven facial animation. It also offers structured output and moderation safeguards. Developers still need to author gameplay rules and integrate their own scenes and characters; this is not a prompt-to-complete-game button.

Can AI run without freezing the game?

A central claim is GPU sharing. In Atelico's own test of a 4K scene with 10 million triangles on an RTX 5080, the game ran at 123 FPS with no AI, 121 FPS while a 3-billion-parameter language model generated text using its balanced scheduler, and 3 FPS when the model ran in a separate CUDA context. These are vendor measurements under particular settings, not independent or universal frame-rate guarantees.

Original GameSlash chart of Atelico-reported frame rates:123FPS without AI,121FPS with CGS Balance and3FPS with a separate CUDA context
Graphic: GameSlash, based on Atelico-reported measurements. Not an independent benchmark or gameplay image. Source: Studio Atelico

The company also reports faster text decoding than llama.cpp on 22 of 24 combinations across an RTX 3090 and an Apple M1 Max, tying on the other two. Exact results depend on the model and hardware.

Original GameSlash diagram summarizing Atelico-reported decode results versus llama.cpp:22faster and2tied out of24tests
Graphic: GameSlash, summarizing the counts of Atelico-reported test outcomes, not percentage speed gains. Source: Studio Atelico

What developers should check before trying it

The practical value is building responsive NPC conversations, AI-generated character art and locally processed voice without paying a cloud API for each request. The trade-off is that player devices must have the necessary compute and memory, while developer integration and performance testing remain necessary. The engine is not generally available yet.

Developers can inspect supported features and join the waitlist on the official Atelico AI Engine page.

View source