Studio Atelico has opened Early Access for an AI engine designed to run on the player's own hardware. GamesIndustry.biz reported the launch on October 30. The Atelico AI Engine supports language, image, speech and character systems, with integrations for Unity, Unreal Engine and Godot.

That's an interesting proposition for developers who want AI to participate in the game itself. If a character's response or a newly generated creature can be handled locally, the design has fewer reasons to depend on a remote service answering every request. The appeal becomes especially clear when the feature is something players might use repeatedly.

Local Generation Changes What a Designer Can Try

Atelico's pitch is that its models ship with the game and run on the device, avoiding cloud inference charges for those operations. That changes one part of the calculation for a developer considering a feature built around player input.

Imagine a creature-creation system that players enjoy experimenting with. Repeating the process is part of its attraction: change a description, see what happens, then try something stranger. If every attempt requires a paid remote request, popularity could become an awkward design constraint. Local processing offers another way to approach that problem, although the work still has to fit the player's machine.

There is room for enjoyable experimentation here. A developer could make generation something players return to throughout an adventure, provided the results arrive quickly enough and remain worth using. The useful question is what that interaction adds to the game. Producing more content isn't much of an achievement if players have no reason to care about it.

The Dialogue Can Still Be Written by Writers

One of the more appealing features on Atelico's engine page is its approach to authored dialogue. Writers can supply lines, and the system can retrieve an appropriate response based on meaning and context. Character states can control which responses are available.

That offers an interesting middle ground for conversations. A player could phrase a question naturally while the game responds with material its writers have chosen. The flexibility sits in recognising what the player is asking and selecting a suitable line, while the character's voice and story remain deliberately written.

For a mystery or role-playing game, that could be useful. A suspect's willingness to talk might change as the conversation progresses, or a companion might respond differently after a particular event. Those possibilities still require careful design, but they give developers a specific reason to explore the technology beyond making every character endlessly chatty.

Bobium Brawlers Gives the Technology a Clear Job

Bobium Brawlers provides a concrete example of what the studio is pursuing. Atelico describes a card-and-dice brawler in which a player describes a creature and the game generates its artwork and deck on the phone.

The attraction is easy to understand. Describing a ridiculous creature and then finding out how it works in a battle could create a stronger connection than simply receiving another random card. Whether that holds up over time depends on the rules around the generation: how distinct the creatures feel, how their abilities interact and whether an unusual description leads to an interesting decision.

There is also relevant context in GamesIndustry.biz's July interview with the studio. Chief product officer Chris Olson said its artists were paid and discussed a draft Artist Rights Contract covering permitted training uses and participation in the upside. The studio said its creature-generation models were trained on its own artists' work.

Those are the studio's descriptions of its arrangements, rather than something we've independently audited. They do identify questions worth asking of any production pipeline: whose work is involved, what permission covers its use and how the people making it benefit.

The Player's Hardware Still Has to Run the Game

Atelico's GPU-sharing documentation addresses the obvious practical challenge. The graphics hardware needs to draw the game while also processing AI. The studio describes a scheduler with different priorities for rendering and inference, and publishes results from its own Windows 11 and RTX 5080 demonstration.

That's the kind of detail developers need to investigate. Performance in a published demonstration doesn't establish how a particular game will behave across its target devices. A quiet conversation screen and a crowded combat encounter place different demands on the hardware. Teams need to decide when processing should happen, what can wait and how the experience behaves when resources are tight.

Developers can register for early SDK access through the engine's waitlist. For those interested in AI as a gameplay system, this looks worth following: a defined set of tools, a concrete game example and an approach that gives local processing a central role. The next useful evidence will come from how developers make those pieces work inside games people want to keep playing.

Sourcing note: this article uses Atelico's official engine and GPU-sharing material, its Bobium Brawlers description, and GamesIndustry.biz's October 30 report and July interview. We haven't tested the SDK; performance and production claims remain attributed to their sources. Featured image: Studio Atelico, via GamesIndustry.biz.

Would you be more interested in AI characters or generated creatures if the processing happened on your own device? Let us know your thoughts in the comments. And if you've got a game you'd like us to review, send it over to editor@gamevibe.io.