Studio Atelico

Two years ago we asked a simple question: what if the AI in a game lived in the game? Not in a data center, not behind an API key, not paid by the token, but on the device in the player's hands. Today that's a product: we are announcing the @StudioAtelico AI Engine. It's a library that runs inside your game and brings together the AI a game needs: language models for dialogue and narrative, image generation, voices and speech recognition, lip-sync, memory and search, guardrails, and a behavior engine for characters with needs, emotions and relationships. Why on-device? Because cloud AI and games want different things. A game has about 16 milliseconds to draw a frame, and a round trip to a server takes longer than that. A cloud bill grows with every player, so the more people love your game, the more it costs you. And players' conversations with your characters are their business, not ours. What surprised us most was that running the model wasn't the hard part. Sharing the machine was. Our first try ran a language model next to a game on the same GPU, and the game dropped from 123 FPS to 3. So we built the engine to work with the renderer instead of fighting it. The same scene now holds 121 FPS while a model generates text. The other surprise was how much small models can do when the engine keeps their output in the shape the game expects. That's how Bobium Brawlers builds a creature's art and deck on a phone, and how GARP runs a village of autonomous characters on one GPU. Next, we want other developers building with it. The engine plugs into Godot, Unity and Unreal, has a C ABI, and runs on Windows, macOS, Linux, iOS and Android. If you're making a game where the AI should be part of the play and not a service you rent, I'd love to talk. Blogpost: atelico.studio/blog/the-atel… Atelico AI Engine overview: atelico.studio/ai-engine/