Skip to content
Theme
OnDevice Tools

GamingGame Releases

Kult gaming companion uses Gemini to answer players by voice

London startup Kult is building an always-on game helper that combines voice replies with a second-screen feed. The system uses Gemini models and automated game knowledge to support over 100 titles.

Illustration of a Kult gaming companion showing voice answers and a second-screen guide

Key takeaways

  • Kult is testing a voice and second-screen game helper with alpha players.
  • The system uses Gemini Flash for routine replies and larger Gemini models for harder reasoning.
  • Answers are based on automatically maintained game knowledge rather than model memory.
  • Kult says it supports over 100 games and plans to add computer vision.
In this article
  1. What the companion does
  2. How Gemini models are used
  3. Knowledge and personalisation
  4. Availability and next steps

What the companion does

The post describes a common frustration. Players often leave a game to search for help, then struggle to return to the same task. Kult wants guidance to arrive without that break.

The companion works out which game is running and how far the player has progressed. It can reply through speech while the player keeps their hands on a keyboard or controller. A second display can show maps, video, images, builds or links when more detail is useful.

Kult says the product adapts over time. It notices what a player asks about and what they skip. That helps it match answers to a player's style, such as rushing through a game or exploring every corner.

How Gemini models are used

Illustration: AI, Gemini logo, Google logo, gaming and the labels Gemini, Gemini Flash, Google Cloud

Three demands shape the design: delay, correctness and price. A game helper must answer quickly, give reliable information and stay cheap enough for many questions during one session.

Kult uses the lighter Gemini Flash tier for routine conversational turns. This model handles many small requests and produces answers based on Kult's game knowledge. Larger Gemini models are reserved for rarer tasks that need deeper reasoning, such as work done once for each game.

Google Cloud Blog says Gemini models offer a wide range of performance and cost options. Through Model Garden, part of the Gemini platform, Kult can also use a large catalogue of non-Gemini models. Kult says the latest Gemini Flash models sit at a useful balance of price, speed and quality.

The company reaches models through the Gemini platform and the Gemini Developer API, an interface for calling models. A routing layer chooses the backend, meaning the service that runs the model. That backend runs on Google Kubernetes Engine, a Google Cloud service for managing applications across many machines.

Kult also uses Jev, a model that only makes decisions and returns a confidence score. Routine choices can be settled quickly and cheaply. Less certain cases move to a larger Gemini model.

Knowledge and personalisation

Kult does not rely on a model's memory for game facts. For each supported game, it builds a structured knowledge base and keeps that information current. This is important because games can change with patches.

The model's role is to reason over that knowledge. It also tracks the current session, including the game, recent actions and earlier questions. That allows a short question to make sense without extra explanation.

Answers arrive in two forms. The spoken reply is brief. At the same time, a visual feed appears on a second screen with richer material.

The feed uses an open standard for interfaces generated by AI agents. The model does not write code. It selects from a catalogue of components designed by Kult, and the app renders them in Kult's style.

Each model is kept in a registry. When a new generation ships, Kult tests it against human-written benchmark answers. It swaps in a model only if it wins.

Availability and next steps

Kult is in alpha with its first players. The company plans to open access to more users over the coming months. It says adding a game is automated rather than a manual project.

That automation matters for a small team. It lets Kult target new releases and large catalogues without a large editorial operation. Google Cloud Blog says most playing happens in the first weeks after a game launches.

Kult also says it has close working ties with Google Cloud and direct access to engineers. The next planned step is computer vision. In the coming months, Kult intends the companion to become aware of what is happening inside the game.

The questions players ask also guide development. They show where knowledge is thin and what to build next. Kult says each session can improve the system for everyone.

For players, the main idea is simple. Help should arrive in the moment, without leaving the game. For developers, the example shows how cloud AI services can support a consumer product that needs many small, fast model calls.

Sources

  1. The Blueprint: How Kult built a gaming companion that actually knows your gameGoogle Cloud Blog

Tags

Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.