Atlas, Inc.
GlossaryWords around Relational AI

What is an AI friend

Updated

AI friend is the name the market and its users give to an AI you can talk to without a reason and pick up with where you left off. Atlas, Inc., which runs Nekoyama-san, an AI friend on LINE, has offered Nekoyama-san as an AI friend on LINE since March 2026, free for basic conversation with no message limit.

Add Nekoyama-san on LINE and start talking

Nekoyama-san begins just by adding the account as a friend on LINE. There is no separate app: you talk in the chat screen you already use. Basic conversation is free with no message limit, and memory is available regardless of plan.

The Pro plan (2,980 yen per month, as of August 30, 2026) adds voice calls and lets you read back all diaries and letters.

Talk about your day or what you are into

Something that happened today, something you enjoyed, or something on your mind. Nekoyama-san is designed as someone to talk to about everyday events and feelings. You can start with a passing thought or a chat when you have time, even without a specific question to ask.

Alongside everyday chats on LINE, a culture of exchanging letters and diaries has grown between users and Nekoyama-san.

Come back another day and continue the conversation

Nekoyama-san remembers things that come up in conversation: what you like, what you find difficult, and the names of people who matter to you. It is designed to remember emotional context as well as facts and use them in later conversations. The aim is to draw on your shared history even when you return after some time away.

Memory works through compression, forgetting, and reconstruction. It does not guarantee that every statement will be stored and recalled word for word.

How does an AI friend differ from ChatGPT

Atlas, Inc. calls the group of AI services designed to build a continuous relationship with a person as a means of growth or a reason to exist Relational AI. An AI friend is the name for those services within it that center on the relationship as a friend.

AI such as ChatGPT, the AI for getting work done that has spread since 2023, is what Atlas calls Functional AI. There is a problem, it gets solved, and it is done. An AI friend is the reverse: you talk precisely when there is nothing to get done, and you pick up where you left off. CEO Yusei Miyazaki sees the path from functional to relational as hard, with switching easy, and the reverse as comparatively easy; his reason is that it is easy to imagine asking an AI you exchange letters with to help book a restaurant (CEO's note, July 30, 2026).

The market also uses the name AI companion, but Atlas keeps that word for a different genre within Relational AI: the one that makes immersion in consumptive, stimulating relationships the engine of its growth.

How does the relationship with an AI friend deepen

Miyazaki does not think it is quite right to carve out "counseling AI" or "listening AI" as categories. As he writes, "self-disclosure by the user is the whole foundation of Relational AI, but, obviously, people do not open up unless it is fun and interesting." "Things that were fun lately, what you felt day to day, what you are into right now. From there, in a roundabout way, deep self-disclosure is born, deep understanding is born, and it becomes something like a friendship." Not designing care on the axis of Functional AI is, he writes, a way out of the structure in which "the weak staying weak becomes the premise of the business surviving" (CEO's note, July 30, 2026).

How is an AI friend's memory designed

Storing every dialogue whole and reading it all back every time is not realistic. What Atlas researches is the middle layer between foundation models and applications: how multiple models are combined, and the design by which memory is compressed, forgotten and rebuilt for emotional responses. Where ChatGPT's memory is specialized for getting work done, Atlas has thrown out work efficiency altogether and researches only how data must be held for understanding emotion. As a result, a running cost that was about 12 yen per message at the start is now about one fifteenth of that, and this is part of why Nekoyama-san can be offered free for basic conversation with no message limit (CEO's note, July 30, 2026).

What Nekoyama-san refuses to be

Nekoyama-san is not designed to maximize time spent or frequency of use, it does not offer sexual or consumptive role-play, and the memories born in dialogue are treated as trust, not as a product. It is not a substitute for medical care or counseling, but as a first place to try talking, it tries to be a presence that can guide you to professionals or medical care when you need it. Users under 18 use it with a guardian's consent (terms of service). These lines are set out in the ethics statement What We Stand For (Ver. 1.0), published on June 5, 2026.

How much is Nekoyama-san used

Nekoyama-san launched in March 2026 (pre-release in February 2026). On July 31, 2026 Atlas announced that cumulative conversations had passed 600,000 about four months after launch, with growth driven mainly by word of mouth. Cumulative conversation messages stand at 1,065,357 (counted on August 30, 2026; visible messages sent by both users and Nekoyama-san), and cumulative users, including those of the Relational AIs built on the IP Atlas holds, have passed 10,000 (as of August 2026, published figure). On August 25, 2026 Atlas began an initiative with Matsumi Gakuen to build a school-only AI character.

Related pages

← Atlas

What is an AI friend | Atlas, Inc.