Evidence-based nudges that meet employees where they already work.
Open Source — github.com/wayneww2016-sketch/keepwell-ai
Cultural stigma keeps usage at 2-5%. Employees won't self-refer until crisis. We need a low-barrier digital front door.
Lunch-and-learns, wellness fairs, and office yoga skip DC operators, factory workers, and night-shift staff entirely.
Programs designed at HQ don't account for 10+ countries, different work rhythms, languages, or cultural norms.
KeepWell AI delivers micro-interventions inside the tools employees already use, personalized to schedule, weather, and language.
Detects time of day using circadian + ultradian rhythm science
Checks live weather (temperature, rain, UV) for context
Selects from 8 wellness dimensions, avoiding repetition
Delivers actionable nudge + 1-line science backing
Monitors patterns; surfaces EAP resources when needed
26°C, Partly Cloudy
Physical
"Step outside for a 5-min walk. Let your eyes focus on something far away."
Ref: Outdoor walking combines Attention Restoration Theory with physical movement (Kaplan, 1995)
11:30 PM Detected
Emotional
"Sleep debt cannot be repaid on weekends. Close the laptop."
Ref: Walker, 2017 — prefrontal cortex impairment at 50%
Not just "drink water." A holistic model covering every aspect of human well-being, each with 5-8 evidence-based nudges.
Physical
Emotional
Occupational
Social
Intellectual
Environmental
Financial
Spiritual
Ultradian Rhythm (Rossi, 1991) — 90-min work/rest cycles
Circadian Chronobiology (Walker, 2017) — time-of-day optimization
Nudge Theory (Thaler & Sunstein, 2008) — choice architecture
Transtheoretical Model (Prochaska, 1997) — stages of change
Self-Determination Theory (Deci & Ryan, 2000) — autonomy, competence, relatedness
JD-R Model (Bakker & Demerouti, 2007) — resource vs. demand balance
Affect Labeling (Lieberman et al., 2007) — naming emotions reduces amygdala activity by 43%
Attention Restoration Theory (Kaplan, 1995) — nature micro-breaks restore directed attention
Two-Process Model (Borbely, 1982) — sleep hygiene timing for shift workers
Never commands. Always suggests. Explains the "why." Respects autonomy. Max 1 nudge per 90 min.
Night-shift presets correctly identify 2 PM as "sleep time" for DC operators. No other tool does this.
Live weather via Open-Meteo (free, no API key). Rain = indoor suggestion. UV ≥ 6 = sunscreen warning.
Schedule nudges to team channels via cron. Also supports Telegram and generic webhooks.
English + Traditional Chinese. Ready for APAC. Adding Japanese and Korean is straightforward.
If emotional scores stay low for 3+ days, gently surfaces EAP recommendation. Non-diagnostic, one-time.
Visual HTML dashboard with streaks, dimension bars, and heatmaps. Shareable with managers (anonymous).
Multiple channels ensure no one is excluded — IDE users, Slack teams, mobile workers, or frontline staff.
MCP Server for Kiro, Cursor, Claude Code, Codex
Scheduled pushes to team or DM channels
Personal mobile notifications via bot
Corporate deployment via webhook
Cron (Linux/Mac) or Windows Task Scheduler. Respects quiet hours — no messages during sleep.
5-question onboarding auto-generates config. Or use presets: Early Bird, Night Owl, Shift Worker, Remote Flexible.
MCP server in their IDE. Gets nudges during 90-min ultradian breaks. Slack channel for team challenges.
Telegram bot with shift-worker preset. Weather-aware nudges. EAP gateway catches sustained stress early.
Slack DMs in local language. Flexible schedule preset. Weekly reflection prompts on Fridays.
Anonymous weekly dashboard showing dimension trends across regions. No individual data. Informs program design.
When threshold triggers, surfaces the correct local EAP hotline number and booking link. Tracks referral-to-use rate.
Track: nudge engagement rate, EAP referral conversion, dimension score trends, absenteeism correlation.
Digital-first. Evidence-based. Scalable across APAC.
The EAP front door employees will actually walk through.
Open Source — MIT License
github.com/wayneww2016-sketch/keepwell-ai