Inside Kay’s agentic loop

Observe → Reason → Act → Learn—within clear, safe boundaries you control.

Kay’s agentic loop continuously monitors behavioral inputs to detect patterns early. It reasons through these insights to determine appropriate interventions, acts by providing timely, compassionate nudges, and learns from outcomes to improve support. All actions occur within user-defined limits and include human-in-the-loop safeguards, ensuring control over data sharing and intervention levels.

People in a row, using phones. Diverse group in a casual setting, focused on their screens.
Magnifying glass.

Observe

01

Inputs

AM/PM check-ins track mood, stress, sleep, energy, social connection, environment, with optional free-text notes and engagement signals.

02

Optional Integrations

Apple Health data, such as sleep duration, can be integrated when enabled by the user to enhance monitoring accuracy.

03

Data Sharing

Users can opt-in to share data with providers via a secure code, maintaining control over their information.
Brain with light bulb, representing idea.

Reason

Cartoon woman with bun, glasses, striped shirt, jeans, sneakers sits on the floor, smiling.

Pattern detection across last 3–7 days for “drift.”

Pattern Detection

Signal ladder: Drift → Elevated → Moderate → High risk tiers.

Risk Tiers

Context awareness: mismatch between self-report and behavioral markers; sentiment from optional notes.

Context Awareness

A black outline of a lightning bolt on a white background.

Act

A clipboard with a strategy marked with arrows, circles, and "X"s.

Micro-interventions

Short, evidence-based coping strategies; “what’s doable now” prompts.
Black and white illustration of a vintage microphone.

Tone shift

Kay softens language when drift criteria are met.
Emergency siren icon.

Escalations

If risk crosses thresholds and you opted into provider linking, Kay prepares a concise snapshot for your provider.
Hand icon pressing finger, with arrows indicating movement.

User control

One-tap snooze, mute, or adjust frequency.
Outline of a graduation cap.

Learn

Young Black woman with glasses smiles at the camera while working on a laptop at a desk.

Providers can set clear guidance that shapes Kay’s day-to-day support. Define what to lean into, what to steer around, and how/when to surface guidance. Kay honors these settings in future recommendations—consistently, across every linked patient.


What you can configure

  • Strategy emphasis: highlight preferred approaches; de-emphasize others.
  • Boundaries & sensitivities: list topics, phrases, or contexts to avoid.
  • Tone & pacing: choose calmer vs. more directive tone; ultra-short vs. slightly longer prompts.
  • Escalation posture: when to suggest contacting the care team vs. offering self-serve steps first.
  • Language cues: plain-English terms you want Kay to use for your population (jargon off).
  • Cultural considerations: note community-specific norms Kay should respect.


Kay’s agentic loop—Observe, Reason, Act, Learn—within user-controlled boundaries and human-in-the-loop safeguards

Data stays yours. Share with a provider only when you choose.

Kay’s agentic loop processes inputs to detect behavioral patterns, enabling timely interventions while respecting user autonomy and data sharing preferences.