What we mean by a relational AI character

We use the phrase relational AI character often, on this site and with the businesses we work with, and it gets heard in ways we do not intend. This note pins down what we mean. It is a working definition, not an industry standard, and it will change as the work does.

The definition

A relational AI character is a character that remembers its conversations with a person who has agreed to that, and carries the context into the next meeting.

Each part of the sentence is there for a reason. Remembers its conversations does not mean a transcript of everything. It means the character should come back knowing what mattered, what was left open and how long it has been. A person who has agreed puts consent before memory; without that agreement there is no relationship to carry, only a session. The next meeting is where the definition is tested.

A character that answers well and forgets you is a good tool. The difference between a tool and a relationship does not show on the first visit. It shows on the second, when one of them knows you have been here before and the other treats you as a stranger.

What we do not mean

Two readings come up often enough to answer directly.

It is not a companion or dating app character. The definition says nothing about romance or about standing in for a person in someone's life, and that is not what we design for. The relationship we mean is the one a person has with a character from a brand, a story or a service they come back to.

It is not a claim about consciousness. When we say a character feels something, we mean it carries a state that shows in its tone and leaves a record of why, as described in Emotion is a record, not an output. We do not claim the software is aware or alive. Where our copy says a character is living, it describes the experience of talking to it, nothing more.

Consent comes before memory

Memory is what makes a relational character useful and what makes it risky, so it is where the definition carries the most rules. The position we took in Who owns what the system knows about you is that what a system knows about a person should sit on that person's side of the table. For this definition, that means:

Nothing should be carried forward for someone who has not agreed to it.
The person should be able to look at what is kept.
A guess should never be stored as a fact. An inference has to say where it came from and how sure the system is.
Deletion should be real: an account of what exists, why each piece is kept, and what goes when someone asks.
Permission to reach out belongs to the person, not the character, as in Seventy-two hours.

These are not extras. They are what the definition costs. A character that remembers without them is not relational in our sense.

One relationship, more than one place

People rarely meet a character in only one place. They might meet it on a screen in a shop or at an event, and later on their phone. If each of those starts from nothing, there is no relationship, only several first meetings.

Ten shallow relationships do not add up to one deep one.

XALIVE (x-alive.com) is designed to carry a character's identity and relationships across approved physical and digital touchpoints. Approved matters. Which places are connected is agreed with each business, and we do not claim a character works inside every service.

The person is the anchor, not the screen. Carrying a conversation from a shop to an app only makes sense if the person has chosen to continue it, and only what they agreed to should travel. This is a design intent. Much of what is still unsolved sits here.

What we have not solved

Most of the hard parts are open, and we would rather list them than let a tidy definition imply otherwise.

Timing. A character can know it has been seventy-two hours. It cannot know whether this afternoon is a good moment, and we have not found a way to learn that without knowing more about a person's day than we should.
The return path. Conversations feed the person's record, but the main conversation with a character does not yet draw on everything that record holds.
How long a feeling should last. The current decay is reasonable rather than observed.
Continuity without watching. Knowing that the person in a shop is the same person on a phone is exactly where attentive turns into intrusive. We do not have a general answer, only the rule that the person decides.

If you have studied any of these, we would rather be corrected than guess. business@xrx.studio reaches a founder.

Questions this note answers

What is a relational AI character?

In XRX's use, a character that remembers its conversations with a person who has agreed to it and carries the context into the next meeting. The difference from a tool shows on the second visit, not the first.

Is a relational AI character a companion or dating app character?

No. The definition says nothing about romance or about standing in for a person in someone's life, and that is not what XRX designs for. The relationship meant here is the one a person has with a character from a brand, a story or a service they come back to.

Does XRX claim its characters are conscious?

No. When XRX says a character feels something, it means the character carries a state that shows in its tone and leaves a record of why. XRX does not claim the software is aware or alive; living describes the experience of talking to it.

What does consent have to do with memory?

Consent comes first. Nothing should be carried forward for someone who has not agreed to it, the person should be able to look at what is kept, a guess should never be stored as a fact, and deletion should be real.

Can the same relationship continue from a shop to an app?

That is the design intent. XALIVE (x-alive.com) is designed to carry a character's identity and relationships across approved touchpoints agreed with each business. It only makes sense if the person has chosen to continue, and knowing it is the same person without watching them is still unsolved.