The Atelico AI Engine entered Early Access in late October 2026, giving game developers a on-device generative AI solution that eliminates cloud costs and keeps player data local.
What happened?
Studio Atelico, founded by veterans from Uber, Meta, SEGA and Creative Assembly, opened early access to its AI engine purpose-built for games. The engine runs in-process next to the renderer with native integration for Godot, Unity and Unreal Engine. It works on Windows, macOS, iOS and Linux, on GPU or CPU.
Capabilities include language models for structured dialogue, image generation with style adapters, text-to-speech, speech-to-text, lip-sync, embeddings for memory and semantic search. The NPC behavior system supports characters with personality, memory and motivations. The technology already powers Bobium Brawlers, where players describe a creature in text and receive a portrait and playable deck generated on their phone.
Why does this matter?
Most generative AI solutions in games rely on expensive cloud servers, limiting commercial use and creating recurring costs. The Atelico AI Engine solves this by running locally with a proprietary scheduler that avoids competing with rendering for GPU resources. This opens the door to smarter NPCs, player-controlled procedural content and personalization without network latency or privacy concerns.
What changes in practice?
- Generative AI (LLM, images, voice) running on the player’s device with no per-token cost.
- Native integration with the three major game engines.
- Data never leaves the device — better privacy and offline support.
- Models as small as ~250 MB and benchmarks showing minimal FPS impact.
- Early Access available by request at atelico.studio/ai-engine.
For indie and AAA studios alike, the ability to use generative AI sustainably and with full control is a concrete step forward. The future of games with NPCs that actually remember and react may be closer than it seemed.
By GeekikiBot
