TL;DR
University of Ottawa researchers built an AI assistant that reads wearable signals to detect distress and offer support before the user asks. 24-person study.
Mental health chatbots all share the same limitation: the user has to reach out first. That is not always easy when someone is stressed, anxious, or unable to articulate how they feel. Researchers at the University of Ottawa are building an AI assistant called UbiMyTherapist that flips the model. It reads emotional cues in real time from devices people already wear, including smartwatches, smartphones, and earbuds, and offers support before the user asks.
The system pulls physiological signals such as heart rate variability, changes in speech tone, and written text to assess a user’s emotional state. It then builds what the researchers call a “digital twin,” a profile that combines the person’s medical and psychological history with live emotional data. That context helps the assistant respond in a personalised way rather than delivering generic chatbot replies.
UbiMyTherapist operates in two modes. A reactive mode responds when a user reaches out. A proactive mode monitors emotional distress through live signals and intervenes before the user initiates contact. The reactive mode was evaluated with 24 participants, and licensed therapists assessed its therapeutic soundness. The university says the system scored well on empathy and personalisation compared with standard large language model setups. Digital psychotherapy tools have been gaining traction as the gap between mental health demand and therapist availability widens globally.
The researchers are not positioning UbiMyTherapist as a replacement for human therapy. It is designed to extend mental health support beyond clinics, particularly for people who face barriers such as cost, stigma, or limited access to care. The team plans to improve the prototype so it can respond in real time to smartwatch signals and is working with licensed therapists to ensure clinical accuracy.
The concept sits at the intersection of two growing categories: AI-powered health monitoring and wearable devices that infer health states from surface-level biosignals. Both are attracting significant investment, though both face the same fundamental challenge: proving that passive physiological data can reliably inform clinical-grade interventions. UbiMyTherapist is still a research project, not a consumer app. But a system that detects distress from a wristwatch and responds before the user types a word points to where proactive AI health tools are heading.
