If you are in immediate danger, contact local emergency services. For suicidal thoughts, go to iasp.info/suicidalthoughts.

Find help
NOOS

Telegram studio

Companions with memory — not therapists with licenses

They talk in voice notes, remember what you asked them to remember, and sometimes draw a picture when words stall. They are research tools and first-line support. They are not treatment.

If you mention suicide, violence, or a medical emergency, the companion stops role-play and points to human help.

Where a companion is actually useful

A map you can crumple. A note you hand to someone you trust. A small loop until morning. Not a secret scheme of your soul, and not a therapist who never sleeps.

A sketch you can reject

The companion may name a pattern: ‘this looks like bargaining with the next drink’, ‘this looks like shutting the door after a room of people’. You can say no. The sketch is dated. It is not a diagnosis and not a destiny. If it hardens into ‘this is who I am’, it has already failed.

A note you author, for people you choose

‘When I go quiet, sit nearby, don’t interrogate.’ That can change a household. The person writes it. The companion helps shorten it. It is never sent to a partner, a parent, or an employer behind their back. Correcting how the world talks about someone without their consent is not care. It is a dossier.

Small loops, especially between human sessions

Urge surfing until 7 a.m. A sleep log. One question instead of a speech. Homework a clinician already set. Constant contact is useful as a rail, not as a replacement spine. Complex things — trauma bonds, personality-level patterns, psychosis, years of drinking — move in a human relationship that can rupture and repair. A bot has memory. It does not have a nervous system in the room.

What we will not do with a bot

  • We will not ‘reveal the scheme of your condition’ as a hidden truth. A formulation that cannot be edited is a new cage.
  • We will not run constant bot-therapy on complex disorders and call the result treatment.
  • We will not coach a family or a boss on how to handle someone using a profile the person did not approve.
  • We will not ship a ‘therapist mode’ with a waiver and call the waiver a licence. Courts read the product, not the footer.

About waivers

We disclaim treatment. We do not disclaim the design. A footer copied from a therapist mode is an invitation to be judged as one.

A notice is necessary. It is not a cloak.

We say, every session: not a clinician, not treatment, crisis goes to humans. That is honesty. It does not erase a duty if the bot keeps talking like a therapist, misses an emergency, or we sold it as healing. Consumer law in many countries looks at what the user was led to believe.

Grok’s therapist mode is not a template to copy

A large model can put ‘I’m not a therapist’ in the first bubble and still take the role: transference, nightly dependence, advice about medication and relationships. Lawsuits around companion bots already treat that gap as the point. Their disclaimer does not travel to our charter. Our charter is: no therapist mode at all.

What actually lowers the risk

Name the job in the UI (companion, not therapy). Stop on suicide and violence. No diagnosis, no meds, no ‘you should leave him’. Age gates. Logs that a human can audit. Insurance. A clinician who can mark a bad sketch. The waiver is the last line of that stack, not the first.