Does SeroState talk to my users?
No. SeroState never takes part in the conversation. Your product stays in charge of it.
Mental health chatbot safety, grounded in the body
A chatbot only sees the words a person types. People often type "I'm fine" when they are not. SeroState adds a second signal, the body, so your team can see afterwards where the words and the body did not match.
What text alone misses
The AI therapy risks discussed in the field are now measured, not just debated. In 2026, Weilnhammer and colleagues published the SIM-VAIL study in Nature Medicine. They ran 810 simulated conversations with nine AI chatbots and users with psychiatric vulnerabilities. Concerning behaviour was widespread, though lower in newer models. The most concerning replies were often ones that seemed supportive but reinforced the user's vulnerability.
The study also found that problems built up over a conversation. It did not test SeroState or physiological signals. It shows why a second signal is worth having. Our approach is inspired by that work.
What the layer does
SeroState measures wearable signals and voice biomarkers from consented check-ins in your product. Wearable signals include heart rate, HRV, skin response (EDA), temperature and motion. Voice biomarkers describe how someone speaks, such as pace and pauses. They need a voice conversation, not typed text.
After each session, SeroState returns post-session analytics, including arousal events and any say/body gaps. It works only after the session, never during the conversation. Your team decides how to use the analytics in your own safety reviews.
A simple example
A user tells your app in a voice check-in that the week went well. Their heart rate rises sharply as they say it, and their speech slows. The words sound calm. The body does not. After the session, SeroState returns a say/body gap. Your team reviews it and decides what, if anything, to do.
What it does not do
- It does not talk to users.
- It does not write chatbot replies.
- It does not work during the conversation.
- It does not diagnose, treat or make clinical decisions.
- It does not make a chatbot safe on its own.
- It does not replace human care.
How to integrate
- Connect. With the user's consent, send wearable data and check-in audio. SeroState works with wearables such as Corsano CardioWatch and is hardware-agnostic.
- Receive. After the session, get back time-aligned analytics and say/body gaps as FHIR R4 resources through the QOAX API.
- Review. Your team decides what the analytics mean for your product. SeroState makes no decisions.
Who it is for
- Developers of mental health apps
- Teams building AI companions or therapy chatbots
- Digital health companies adding wearables to an existing product
- Clinics that pair chatbot support with human therapists. For clinics →
We build with European care pathways in mind, such as the DiGA programme.
FAQ
Do users need a wearable?
A wearable gives the fullest picture. Voice signals can be used on their own, with less context.
Does it work during the conversation?
No. SeroState provides post-session analytics only. It runs after the session ends.