Questions covered

The shortest useful model

Classify the system by its primary role:

  1. advises — coaching;
  2. inhabits — in-game;
  3. acts — agentic;
  4. relates — social;
  5. preserves — continuity.

Then examine its inputs, permissions, time horizon, and player controls.

Frequently asked questions

What is an AI game companion?

An AI system that supports a player through guidance, interaction, action, social presence, memory, or continuity related to video games.

What are the five main types?

Coaching, in-game, agentic, social, and continuity companions.

Does it need screen access?

No. Screen vision is one possible input, mainly useful for current-moment coaching.

Is it always a chatbot?

No. It may be an in-game character, voice system, agent, overlay, or background continuity service.

Can it play the game for me?

Some agentic companions can perform permitted game actions. Many other companions only advise, converse, or preserve context.

Can it remember progress?

Yes, when it has authorized storage. Reliable systems should separate games, characters, and runs.

Is a static wiki an AI game companion?

No. A wiki can be a source used by an AI companion, but a static page is not adaptive AI.

Is every AI NPC a companion?

No. The NPC must hold a meaningful supporting role. Opponents and background characters do not automatically qualify.

Does companion mean romantic companion?

No. The term can describe coaching, collaboration, embodiment, or continuity without relationship simulation.

Is an AI game companion the same as a gaming assistant?

They overlap. Assistant emphasizes tasks; companion can describe a broader and more continuous role.

What is a continuity companion?

A system that preserves and restores meaningful gaming context across sessions, chats, devices, characters, or runs.

What is Companion Play?

A disclosed continuity-companion implementation created by Raynor Eissens. It creates gaming restore points inside existing LLM conversations.

Is this taxonomy an industry standard?

No. It is an editorial model designed to describe observable architectures.

Can one product have several types?

Yes. A system may coach, act, and preserve continuity at the same time.

How should I evaluate a product?

Check its context sources, permissions, action authority, storage, correction controls, deletion, and current official documentation.