Consult-prep note · live demo
Carry what you told an AI to a human professional.
Tsunagi turns a person’s ChatGPT conversations about a personal struggle into a one-page consult-prep note, in their own words, that they can hand to a counselor or coach they choose. No diagnosis. No account. Nothing stored on our servers.
Responses are generated by AI, not by a professional. For adults (18+). A fictional sample is built in, so you can try it without personal data.
Problem
- People now talk through personal struggles with general-purpose chatbots.
- Chatbots are not professionals, and they tend to agree. In Cheng et al. (Science, 2026), 11 AI models affirmed users’ actions 49% more often than humans did[1]. OpenAI has said that, in some cases, user memory made sycophancy worse[2].
- When someone does decide to see a counselor or coach, what they worked through stays in a chat log they can’t easily bring along. The first session starts from scratch.
Product
-
Bring the conversation
Paste a chat or load a text file. You see, edit and delete exactly what will be sent, and you consent before anything leaves your browser.
-
Answer a few short questions
One question at a time. You don’t have to answer everything; anything left out stays marked “not yet filled”. You can make the note whenever you’re ready.
-
Take the note to a person
What you most want help with; background, with events kept apart from your interpretation; what you’ve tried; what you want and want to avoid; whether you see a doctor; practical preferences; and questions to explore together. Copy, download or print it. You carry it and you hand it over.
Words an AI used about you earlier are not recorded as facts. They become questions for the professional (“an AI mentioned X — does it apply to me?”). If there are signs of crisis, the demo puts safety first and shows Japan’s public crisis lines.
Next: the same protocol as a Claude skill and as custom-GPT instructions, so the note can be made where people already talk.
Why Japan
- Japan ranked 61st of 147 in the World Happiness Report 2026, down from 55th the year before[3].
- About 4 in 10 people say they feel lonely (Cabinet Office survey, published April 2026)[4].
- Help exists but is split. There are 76,547 registered Certified Public Psychologists (Kōnin Shinrishi) as of June 2026[5], while AI self-care apps and human-counseling marketplaces run separately. In our desk research we did not find a consumer service that lets people carry their context between the two. That search was not exhaustive.
- Regulation shapes the design. In Japan, software that judges severity or triages people can be a medical device. Tsunagi is designed to stay outside that line: no diagnosis, no severity scores, no routing decisions, and the person carries the note. We plan to confirm this with the regulator’s consultation desks.
Principles
Don’t hold context hostage
Making and taking the note stays free. The note belongs to the person, in a plain format they can take anywhere.
No diagnosis, no triage
No condition names, no severity, no deciding where someone “should” go. General information is the same for everyone; the person chooses.
Nothing stored
Our server keeps no copy and writes no logs of what you send. Text goes to a US AI provider (Anthropic or OpenAI, named on the consent screen) only after you consent.
Humans decide; claims stay measured
We are not clinicians and we are recruiting professionals as advisors. We only claim what we have measured about how the AI behaves.
Who is building it
Kazuya Hibara, based in Japan. He builds AI agents and automation for businesses. He is studying psychology on his own and is not a clinician, so he is asking counselors, coaches and Certified Public Psychologists to advise on the questions, the wording and the crisis steps.
The goal is to raise Japan’s well-being, not engagement. No fear-based or insecurity-based marketing. It starts as a business so it can reach people, and a nonprofit form is on the table as more people join.
Contact: contact@kazuyahibara.com (please don’t send personal health information).
Sources
- Cheng et al. “Sycophantic AI decreases prosocial intentions and promotes dependence.” Science 391(6792), 2026. doi:10.1126/science.aec8352
- OpenAI. “Expanding on what we missed with sycophancy,” May 2, 2025 (“in some cases, user memory contributes to exacerbating the effects of sycophancy”; it also says there is no evidence memory broadly increases it). openai.com
- World Happiness Report 2026, as reported by Kyodo News (March 20, 2026). sakigake.jp
- Cabinet Office of Japan, minister’s press conference on the loneliness and isolation survey, April 14, 2026. cao.go.jp
- Science Council of Japan, report summary (August 2026): 76,547 registered as of the end of June 2026. scj.go.jp (PDF)