Kazeia
Project under development

Kazeia

Conversational agent for psychological support

A listening assistant for mental health teams and their patients, in text or voice.

Offline tablet · no data in the cloud
Designed for supervised clinical settings

Developed in Luxembourg.

The challenge

Psychological distress does not wait for the next appointment.

Risk of depression among 17–18 year-olds
34%

of girls

19%

of boys

Shortage

of mental health professionals, especially for children and young adults.

Sources: HBSC Luxembourg 2022 study (University of Luxembourg, Ministry of Health); National Mental Health Plan 2024–2028.

In Luxembourg, young people already confide in AI.

45%

of 17–30 year-olds use a chatbot for personal advice.

22%

discuss topics with it they would share with no one else.

Source: BEE SECURE Radar 2026 (SNJ, Kanner-Jugendtelefon).

Our response

Kazeia does not replace the clinician, it supports them.

Care teams cannot see every patient as often as they should. Used within the ward, Kazeia gives patients a non-judgemental listening space between sessions and eases the team's workload.

In text or voice, in the patient's language

Luxembourgish is coming soon.

A familiar voice, through voice cloning

Kazeia can speak with a voice the patient knows.

The clinician stays in control

PHQ-9-type questionnaires and results centralised on their workstation (in development).

How it works

Offline and secure, within the care unit.

Tablette hors ligne dans le service, synchronisée vers le poste central du praticien / Offline tablet in the unit, synchronised to the clinician's central workstation TABLET, WITHIN THE UNIT No internet needed Multilingual Luxembourgish coming soon 1Dialogue · in text or by voice 2Understanding · on-device AI 3Response · text or voice 4Assessment · PHQ-9 (coming) other tablets Local synchronisation over the internal network, no cloud, to the central workstation CLINICIAN'S CENTRAL WORKSTATION Collects each patient's results Tracks assessments over time Crisis and distress alerts, restricted to medical staff National health data sharing, from this workstation only (coming)

Everything runs on the tablet

Speech recognition, the language model and speech synthesis all run on the device. No internet connection, no cloud: no data travels over the internet.

The clinician centralises follow-up

Results and conversations are synchronised over the internal network, with no cloud, to the clinician's central workstation, where each patient's progress is tracked.

Designed for Luxembourg's framework

Only the central workstation, under the institution's control, will be able to share data within Luxembourg's national health framework (coming). The tablet never connects to the internet.

Patient safety

Kazeia is used only in supervised clinical settings and never replaces staff supervision. At-risk statements (distress, suicidal thoughts) are detected and raised as priority alerts, visible only to medical staff.

The interface

A calming presence, without a face.

During a conversation, a glowing orb breathes at rest and comes alive when Kazeia speaks. Each voice has its own hue.

  • IdleThe orb breathes slowly. Nothing is being recorded.
  • ListeningThe person speaks, and the orb follows the rhythm of their voice.
  • ThinkingSpeech is transcribed and the reply prepared, on the tablet.
  • SpeakingKazeia answers aloud in a few sentences, then gently prompts the patient.
Kazeia is speaking…
  1. You I can't sleep. It's been three nights. 21:47
  2. Kazeia Three nights is a long time, you must be exhausted. Has anything been on your mind lately that could be keeping you awake? 21:47
  3. You My head spins the moment I lie down. 21:48
  4. Kazeia Thoughts race at night. What keeps coming back? 21:48
Recreation of the patient screen. Kazeia's replies are deliberately short and keep the conversation going, without unsolicited advice.
Project status

Already running on tablets, under test conditions.

Kazeia is actively developed in Luxembourg.

Operational

  • Text or voice conversation on tablet, with no connection
  • Voice cloning: Kazeia speaks in a familiar voice
  • Detection of at-risk statements, alerts restricted to medical staff

In progress

  • Central workstation: result collection and patient follow-up
  • Luxembourgish language support

Next steps

  • Assessment questionnaires on the tablet (PHQ-9 and others)
  • Data sharing within the Luxembourg national framework
  • Clinical validation in Luxembourg, then CE marking

We are looking for care units in Luxembourg for a supervised pilot phase.

Pilot terms: please contact us.

info@kazeia.com www.kazeia.com