A good between-session check-in should be low burden. Short enough to answer from real life. Safe to ignore. Useful only when reviewed in context. The goal is better clinical memory, not more patient homework.
Nyra
@nyracare.bsky.social
Clinician-supervised AI check-ins for mood/anxiety. Patients text; clinicians review patterns, risk flags, and evidence. Not diagnosis or emergency care. Demo: nyra.us.com/book-a-demo
What Nyra refuses to automate: Diagnosis. Prescribing. Crisis ownership. Treatment planning. Clinical judgment. The useful layer is review support: context, uncertainty, risk flags, and cleaner handoffs.
Patient side vs clinician side: Patient side: lightweight iMessage check-ins. Clinician side: structured context in /care. The AI can organize signals. The care decision stays with the clinician.
Why Nyra pauses after 2 unanswered check-ins: Silence should be visible to the clinician without becoming an automated nag loop. The product should surface the pattern, preserve context, and wait for review.
What Nyra means by review signals: Not diagnoses. Not treatment instructions. Not a crisis monitor. Signals are prompts for clinician review: drift, missed check-ins, uncertainty, and context worth asking about in session.
Nyra in 4 frames: iMessage check-ins from daily life. A clinician /care workspace for structured context. Review queues for uncertainty and risk flags. A pause after 2 unanswered outbound check-ins. Clinician-supervised, not autonomous care.
Nyra's current product boundary in 4 frames: Patients text iMessage check-ins. Clinicians review structured context in /care. Signals are prompts, not diagnoses. After 2 unanswered check-ins, Nyra pauses and surfaces it for review. Clinician-supervised, not autonomous care.
What Nyra means by "review signals": Not diagnoses. Not treatment instructions. Not a crisis monitor. Signals are prompts for clinician review: drift, missed check-ins, uncertainty, and context worth asking about in session.
Not every task in behavioral health should be automated – responsible AI adoption means protecting trust, dignity, & human connection. #BehavioralHealth #AIinHealthcare #MentalHealth 📰 Read the article: https://behavioralhealthnews.org/in-the-age-of-ai-what-must-remain-human-in-behavioral-health/
In the Age of AI, What Must Remain Human in Behavioral Health?
Can AI improve behavioral health care without sacrificing trust, empathy, and the human connection at its core?
behavioralhealthnews.org
For clinicians new here: “supervised” means Nyra is designed around your review layer. Patients can answer simple iMessage check-ins; the product organizes context, flags drift, and keeps uncertainty visible for the care team. Learn/demo: nyra.us.com/book-a-demo
Design note for clinicians: Nyra now pauses after 2 unanswered outbound check-ins. Silence is signal, but it should not turn into nagging. The workspace surfaces it for clinician review instead. What would you want surfaced first: risk, drift, or engagement? nyra.us.com/waitlist
1/ What we've shipped on Nyra so far 🧵 Nyra is clinician-supervised AI for mental health. Patients text in between sessions; clinicians get the patterns, risk flags, and evidence — to review, not to rubber-stamp. Not diagnosis. Not a chatbot playing therapist.
A black-box answer is not enough for clinical care. Doctors should be able to understand why a possibility was raised, what supports it, and what could make it wrong. That is the kind of AI medicine deserves.
We are developing a clinician-facing system that organizes those signals into transparent, reviewable hypotheses. It can surface supporting evidence, competing explanations, missing information, and useful follow-up questions. The physician remains the decision-maker.
Psychiatrists routinely manage fragmented histories, changing symptoms, medication trials, safety concerns, and a research literature no individual can continuously absorb. Important relationships can disappear inside that volume.
The goal is not to automate judgment. It is to support it with clearer evidence, thoughtful alternatives, and questions that might otherwise go unasked. Psychiatry has always depended on careful listening. Technology should help clinicians listen more deeply.
LLM eval for clinical use needs: demographic parity, diagnostic safety, refusal calibration, citation fidelity. Accuracy alone is noise.
RVU-based reimbursement actively penalizes AI that saves time. Until payment models align with efficiency, adoption will lag evidence.
Every clinician I talk to says the same thing: documentation burden kills adoption before quality even gets evaluated. We need to subtract load, not add it.
