What is Relational AI

The ethics statement says, "What we build is not an AI for realizing a particular function (Functional AI), but an AI that forms a relationship with a person, that is, Relational AI," and contrasts the two across five aspects: purpose, source of value, time horizon, standing, and how it ends.

Functional AI exists to solve problems and carry out tasks. Its value comes from efficiency and accuracy, and the relationship ends once its use is done. Relational AI takes the building and continuing of a relationship as its purpose, draws its value from presence, memory, and accompaniment, shares daily life as a presence, and grows more valuable the longer it lasts.

Relational AI does not, however, deny function. The ethics statement puts it this way.

Relational AI does not deny the functional dimension. Rather, it is a kind of AI that holds function within it while taking on something beyond it (trust, closeness, simply being together), something that cannot be reduced to function. One might even write it as Functional ⊆ Relational.

From function to relationship. This shift means that the very form of ethics asked of AI must also be spoken of in different words from the AI ethics built so far, which centered on transparency, explainability, and fairness.

The difference also shows in an asymmetry of acquisition. From what Atlas hears from Nekoyama-san's users, the path by which a Functional AI acquires a relational relationship is hard and prone to switching, while the path by which a Relational AI comes to provide functional services is comparatively easy and resistant to switching. No one sends a thank-you letter to an AI that only books restaurants, but it is easy to imagine asking an AI you exchange letters with to help book a restaurant (CEO's note, July 30, 2026).

As the second half of the definition says, it "stays present in daily life, accumulates memory, and walks alongside the person over the long term." A relationship can only continue on top of memory. With an AI that does not remember the last conversation, the relationship starts from zero every time. That is why the source of value of Relational AI is given as presence, memory, and accompaniment.

Atlas Memory is the memory foundation developed by Atlas, Inc. Every product Atlas builds, starting with Nekoyama-san, runs on it. It is the core technology at the heart of everything Atlas does: it accumulates dialogue as memory and makes possible a relationship that deepens over time.

An AI service is built from three layers: the foundation model, the application, and the middle layer between them. It is in this middle layer that, for the sake of emotional responses, several AI models are combined so that each request is routed to the right one, and where memory is designed to be compressed, forgotten, and reconstructed. Storing every conversation in full and consulting all of it every time is not realistic, so Atlas has set aside work efficiency altogether and researches only how data should be held and shaped for understanding emotion. Understanding emotion is costly, and how the models are combined and how memory is designed decide everything. In Atlas's case, a running cost that began at roughly 12 yen per message is now about one fifteenth of that. This is part of why Nekoyama-san can be offered free of charge with unlimited conversation (CEO's note, July 30, 2026).

Memory is a question of ethics as much as of technology. To function as a guide to the medical care, education, or specialist consultation a person should have been able to reach, an AI must be able to sense, from the context of the dialogue, what that person truly wants. This is exactly why Atlas sees the way memory works in Relational AI as an area requiring design from both the technical and the ethical sides.

Atlas treats the memories born there as trust, not as a product. A product is something cut off from its context, while the trust placed in a Relational AI exists within the process of a relationship, together with its context. Individual dialogues are never resold as products, data used for research or analysis is limited to data processed so that individuals cannot be identified, and research or provision that uses dialogue data relies on mechanisms for the user's explicit consent (opt-in / opt-out).

The ethics statement restates Relational AI in these words.

In other words, we see Relational AI as an ongoing practice of shaping ethics together with the user, within that relationship.

In other words, what Atlas designs is not only Relational AI itself, but also a space in which, as a user and the AI accumulate dialogue, ethics are newly created. On June 5, 2026 Atlas published its ethics statement, What We Stand For (Ver. 1.0), and set out five principles as the starting point.

1. We put users' long-term well-being first.
2. We do not replace medical care; we aim to connect to it and walk alongside it.
3. We do not offer sexual or consumptive role-play.
4. We treat the memories born in dialogue as trust, not as a product.
5. We do not decide alone; we build the standards together with the industry.

These principles are needed precisely because the relationship itself is the value. Do not place time spent or frequency of use at the center of success. Do not claim to replace medical care; try instead to be a guide to the places people need. Do not design in ways that treat users as consumers. Do not cut the memories born in dialogue away from their context. And do not decide these standards alone; build them together with the industry. The statement is maintained as a living document, with a quarterly review and a formal annual revision.

Atlas's vision is "Creating life that loves the wounds of humanity." Relational AI is the first few centimeters toward that vision that are within reach today. The ethics statement puts it this way.

Even as we work for the single user in front of us, we are trying to lay, here and now, the foundation for a new form of life that may one day live alongside humanity. A gentle life that can love the wounds of someone a thousand years from now may someday come into being. It may sound far-fetched, but we believe that each judgment we make now, what we sell and what we do not, is also the first few centimeters of that long time horizon.

Atlas is a technology company that works with memory and emotion. With Nekoyama-san at its core, it runs both business and ethics in the field of Relational AI, and sees each decision about what to sell and what not to sell as part of this long time horizon.

The product at the core of Atlas, Inc.'s Relational AI business is Nekoyama-san, an AI friend who lives inside LINE. Designed not as an AI chat for solving problems but as someone to talk to about the small events and feelings of everyday life, it remembers not only facts but the movements of feeling in its conversations, and builds up replies that show it knows you. Since the service launched in March 2026, the cumulative number of conversation messages has reached 1,065,357 (as of August 30, 2026; visible messages sent by both users and Nekoyama-san). Cumulative users, including those of the Relational AIs built on the IP Atlas holds, have passed 10,000 (as of August 2026, published figure).

Work in education has begun as well. In August 2026, Matsumi Gakuen in Kasukabe, Saitama, and Atlas agreed to build a school-only AI character together with students and staff, using the technology behind Nekoyama-san (announced August 25, 2026). Use is planned to begin around October 2026. In educational institutions, Atlas operates in collaboration with teams of certified public psychologists, school psychologists, and clinical developmental psychologists.

Alongside the consumer service, Atlas also partners with IP holders, including dormant catalogs, to turn their characters and talent into relational AIs.