Saga, the hyperscaling platform for AI, and Etermax, the studio behind Trivia Crack, the No. 1 trivia game on mobile with 150 million annual users, launched today an AI agent for Willy – the official character of the iconic game.
In collaboration with Etermax, Saga has transformed Willy into an interactive AI agent designed to engage players across social platforms, with a clear objective: deepen community engagement and increase game installs with new fans.
“We’re very excited about this. It’s the first agent that we’re doing that is across multiple social channels, so this will be on X and Instagram to start with,” said Rebecca Liao, CEO of Saga, in an interview with GamesBeat.
Users will immediately recognize Willy’s fun, smart, and caring personality, just like the character they have come to love from the game. Willy also opens the door for character-based AI interactions to meaningfully influence user intent before a download ever occurs, offering a new approach to user acquisition for games.
Trivia Crack is a good candidate as the first such game to get an agent because it has such a recognizable brand identity. But Liao said the firm would also like to move to a completely self-service model, where a developer or creator and create their own character agent and deploy it across social channels.
The core point of Willy is to improve user acquisition, or to get more people to sign up to play the game. At some point, he will recommend that a player download the game. User acquisition has become so difficult and costly, since players have very little time for discovering new games.
“It’s getting new users in the door or reactivating old users as well. And the goal is for Willy to deliver more game downloads and to help Trivia Crack sell in-game assets — any sort of merch,” Liao said. “So Trivia Crack hopes to see revenue increase from the use of the agent, and then we get a cut of that revenue.”
From game character to companion and user acquisition engine
While Saga started as a Web3 gaming company, Etermax is not using any Web3 features. And Saga has pivoted toward AI agents in a way that allows it to serve non-blockchain customers.
Rather than acting as a traditional in-game feature, Willy’s AI agent is designed to live where players already spend time – on social platforms, including Instagram and X to start. Willy can chat with users about any topic, not just game details, become a companion and post on his own and Etermax’s social accounts. He can guide any users interested in playing the game or purchasing Trivia Crack in-game assets or merch.
“Character AI agents are a thrilling new form of entertainment for consumers, but they become truly valuable to studios and creators when they’re tied to real products, real IP, and real outcomes,” said Liao. “This collaboration is about leveraging IP that the Trivia Crack team has worked hard to create for a far more organic and effective distribution and user acquisition engine.”
Why this matters for games
User acquisition has become increasingly competitive for mobile studios, with rising costs and declining returns across traditional ad channels. By turning a familiar character into a persistent, interactive presence, Etermax aims to reduce friction between discovery and download.
“Our thesis is simple: character-driven AI can meaningfully move the needle on installs for a game that already has massive global recognition,” said Fernando Vasconez, Head of Gaming Investments and Partnerships at Saga. “If it works, this becomes a repeatable model for how traditional studios can extend their IP into social-native environments to solve the user acquisition problem in the game industry, and eventually expand to other non-gaming industries.”
Creating the AI agent
The work was done quickly in a couple of months, partly because Saga has already created the training model for the AI.
She added, “What we needed from them was the core material on which to train the agent. So all their historical files around gameplay and then visual files as well. We had to know how Willy had been drawn in the past or animated in the past, and we took all of that and put it into the training model.”
It didn’t take long after that to come up with a first draft.
Willy will launch tomorrow for audiences to interact with in a fun, positive and upbeat way, Liao said.
“He can talk to five-year-olds. He can talk to 85-year-olds. And we also wanted him to be very encouraging when playing trivia and doing different trivia questions with people. I think a lot of people will try to actually play with Willy just because he comes from Trivia Crack,” Liao said.
Liao said the team created the AI agent character so he would respond even if you didn’t talk about the game or trivia at all.
“He’s this cute thing you want to hang out with. He’s very empathetic,” Liao said. “He isn’t overly sentimental at all. He doesn’t do any sort of toxic positivity, but he tries very hard to be there for you, in case you tell him that stuff is going on.”
Safety guard rails
As for safety guard rails, Liao said the team tried to break him and he held up. He will stay away from religion or politics. He won’t swear. Liao said LLM-based agents like Willy don’t do well on open-ended tasks.
“The LLM is really just there to provide the largest phase of training data, but also to divvy up the tasks for responding to sub agents. So the sub agents in our system do the majority of the work, and these are all small language models that we have developed,” Liao said. “Some of them, like the brand safety agents, are capable of removing certain words from the vocabulary.”
In avoiding certain topics, Willy will try to give some answer that steers the conversation in another direction.
Keeping costs down
The reality of AI today is that AI tokens are required for every question and answer. That means there is some cost associated with each query a user sends. That can be tough for the company if users send out too many queries. But Liao said the price of AI tokens has come down with improvements in processing power at data centers. The wager is that this cost of AI tokens will be a lot lower than advertising a game to get more users.
Also, if a particular game requires more tokens than expected, then Saga will move that game to a highe tier in pricing.
Built on Saga’s AI infrastructure
Willy’s AI agent is powered by Saga’s hyperscaling infrastructure, which uses a proprietary training model combining LLMs and SLMs to produce fine-tuned character AI Agents. In addition, the Saga platform enables programmable AI agent interactions and future expansion into rewards, payments, and monetized engagement.
While the initial focus is on user acquisition, the system is designed to support deeper integrations over time.
“We’re always looking for new ways to meet players where they are and bring them into the Trivia Crack universe,” said Joaquín Toro, Chief Planning Officer at Etermax, in a statement. “This collaboration allows us to deploy AI-driven engagement as a new acquisition channel, while staying true to the characters and experiences our players already love.”
Saga’s goal is to come up with AI agents for lots of games — each of them with their own personalities. There are more such games on the way, Liao said.