Back to Dashboard · Project Hub

AI is a source of data.
You are the clinician.

Every AI feature in astrasera — emotion detection, body language signals, session summaries, clinical inference cards — exists to give you more to notice. None of it tells you what to conclude, what to diagnose, or what to do. That judgment is yours alone.

The AI surfaces observations. You decide what they mean.
Last reviewed March 2026
The governing principle of all AI in astrasera: AI features provide supplementary observational data to licensed clinicians. They do not generate clinical conclusions, diagnoses, treatment recommendations, or risk assessments. Every signal, summary, or inference the AI produces must be independently reviewed, interpreted, and validated by the treating clinician before any clinical decision is made.

What the AI provides

  • Observational signals from facial landmarks, gaze, and head pose — measured, not interpreted
  • Pattern detection across sessions — flagging what appears to co-occur repeatedly
  • Transcription of spoken words with timestamps
  • Summaries of prior session content to support recall
  • Structured draft notes from session transcripts — for the clinician to edit, verify, and approve
  • Named inference cards that describe possible signal combinations — framed as hypotheses for the clinician to explore, not findings

What the AI does not do

  • Diagnose any mental health condition, disorder, or clinical presentation
  • Direct, recommend, or advise on treatment, medication, or therapeutic approach
  • Assess risk — including suicide risk, self-harm, danger to others, or crisis states
  • Replace clinical assessment, the therapeutic relationship, or your professional judgment
  • Produce documentation that is legally or clinically valid without your review and approval
  • Make any determination about a client's mental state, capacity, or wellbeing

Your responsibility as the clinician

  • You are the licensed professional. All clinical decisions are yours alone.
  • Any AI-generated note, summary, or signal must be reviewed and independently verified before being used in clinical documentation.
  • AI observations that appear clinically relevant should be explored through direct clinical inquiry — not treated as confirmed findings.
  • You remain fully responsible for the care you provide, including documentation, diagnosis, and treatment planning.
  • Consult your licensing board's guidance on AI-assisted documentation before using AI-generated notes in official records.

Scientific basis and limitations

  • Behavioral signals (gaze, blink rate, facial movement) are based on a body of research — some well-replicated, some actively contested between research groups.
  • The mapping of facial expressions to discrete emotions (Ekman's basic emotions framework) is a useful model but has been significantly challenged by Barrett's Constructed Emotion theory and others. These signals are probabilistic, not deterministic.
  • Signal accuracy is affected by lighting, camera quality, the client's angle to camera, glasses, face coverings, and internet latency.
  • Signal norms are derived from research populations that may not represent your client's cultural, ethnic, neurological, or developmental background.
  • Composite clinical inferences (e.g., "pre-cry cluster," "shame posture") are clinical heuristics, not validated diagnostic criteria.
Key point: Individual signals mean very little in isolation. A client looking away from the camera while discussing a topic could reflect avoidance, concentration, cultural norms, or a distraction in the room. The AI cannot distinguish these. You can.

Emergencies and crisis situations

  • AI features are not designed to detect, flag, or respond to crisis situations, suicidal ideation, self-harm, or acute psychiatric emergencies.
  • Do not rely on the absence of an AI alert as indication that a client is safe.
  • Follow your standard clinical protocols and professional obligations for all risk assessment and crisis intervention — AI plays no role in this process.
  • If a client appears to be in distress during a session, respond as you would in any clinical encounter — your direct clinical judgment takes full precedence.

AI data is for clinician use only

  • All behavioral signals, inference cards, and emotion data visible in the Analysis panel are shown to the therapist only — they are not visible to the client.
  • Do not share raw AI signal data with clients as if it were clinical fact. If you choose to discuss observations with a client, frame them as your clinical observations, not as AI findings.
  • Session transcripts and AI-generated notes are covered by the same confidentiality obligations as all other session content.
  • Inform clients, consistent with your informed consent process, that session AI tools are in use.

Regulatory and legal status

  • Clinical Decision Support software: astrasera AI features are designed to function as Clinical Decision Support (CDS) tools under the 21st Century Cures Act (2016). They are intended to support, not replace, the decisions of licensed clinicians, and the basis for all AI outputs is transparent and independently reviewable by the clinician. Under this framing, these features are not regulated as medical devices by the FDA.
  • Not a medical device: astrasera AI does not acquire or analyze medical images, signals, or physiological measurements for the purpose of diagnosis or treatment. Facial landmark tracking is provided as observational behavioral data, not as medical assessment.
  • HIPAA: AI features process session content that may constitute Protected Health Information (PHI). Refer to the Notice of Privacy Practices for full details on how PHI is handled, stored, and protected. AI Edge Functions are planned to migrate to AWS Bedrock to achieve full HIPAA BAA coverage.
  • Licensing board compliance: The use of AI tools in clinical practice is subject to guidance from your state licensing board and professional associations (APA, NASW, NBCC, etc.). Requirements around informed consent, documentation, and scope of AI use vary by jurisdiction. You are responsible for compliance with all applicable rules in your jurisdiction.
  • No warranty of accuracy: astrasera makes no warranty, express or implied, regarding the accuracy, completeness, or fitness for purpose of any AI-generated signal, summary, or output. All outputs are provided as-is for informational purposes and must be independently verified by the treating clinician.
CDS — Not FDA-regulated as a medical device HIPAA-covered session data Clinician responsibility model Not for emergency or crisis use

By using AI features in astrasera, you acknowledge

That all AI-generated signals, summaries, inference cards, and clinical notes are supplementary observational data — not clinical findings, diagnoses, or treatment recommendations.

That you, as the licensed treating clinician, bear sole responsibility for all clinical decisions, documentation, and the care of your clients. No AI output changes or reduces this responsibility.

That AI signals are based on research that is in some areas well-replicated and in other areas actively debated. You will apply your clinical judgment to determine whether and how any AI observation is relevant to a given client in a given moment.

That AI features are not appropriate for emergency or crisis assessment, and that you will follow your standard clinical protocols for all risk-related situations regardless of AI output.

astrasera is a tool that gives you more to notice — and trusts you to know what to do with it. Every signal the AI surfaces is the beginning of a clinical question, not the end of one.