Already exists01
Progress
Tasks · habits · goals · checkpoints
Existing OAuth + MCP
The system of intention. Keep plans and targets here—not raw sensor exhaust.
A PRIVATE INTELLIGENCE LAYER FOR ONE LIFE
The system that
notices.
Not a giant dashboard. Not an AI that watches everything. A quiet operating layer that connects what you planned, where your time went, how your body felt, and what actually changed.
01 · INTENTIONProgressGoals, tasks, habits
02 · ATTENTIONTimeHours, focus, context
03 · OPERATORKairoConversation, review, action
SLEEPFOCUSBODYPHONE
FRI · 31 JUL+0.62MOOD SIGNAL
ILLUSTRATIVE FORECAST
Your energy is likely to hold through 16:30.
Sleep regularity↑ 18%
Late phone use↓ 11%
Suggested: protect one 75-minute work block.
00 / THESIS
You already built the first
three organs.
Progress knows what matters. Time knows where attention went. Kairo knows how to talk with you. The missing piece is memory: a common event layer that can connect them to health, devices, work and lived experience—without forcing every signal into one app.
THE SIMPLE RULEIntegrate systems of record. Do not replace them.
01 / LIFE SURFACES
One life. Six signal families.
Every surface has a different honest path: existing system, direct API, native bridge, derived signal, or a two-second human log.
Already exists02
Time
Focus hours · categories · personal/work
Existing Convex + Apple apps
The system of attention. Completed timers become evidence against plans.
Native bridge03
Apple Health
Rings · sleep · workouts · calories
HealthKit iPhone companion
Activity summaries include Move, Exercise and Stand. Sleep stages and workouts are queryable.
Native bridge04
Mood
Valence · emotion · daily mood
HealthKit State of Mind + Kairo
Two small check-ins create the ground truth the prediction engine needs.
Native bridge05
Asthma & medication
Inhaler puffs · dose events · sick days
HealthKit + one-tap log
HealthKit includes inhaler usage and medication APIs. Sick days remain a deliberate annotation.
Native bridge06
iPhone attention
Screen time · categories · pickups
Device Activity extension
On-device Screen Time reports can expose duration, notifications and pickups with entitlement.
Direct API07
Mac activity
Apps · windows · browser · idle time
ActivityWatch local API
Keep detailed titles local; sync classified minutes rather than private window text.
Direct API08
Coding
Projects · languages · debugging · reviews
WakaTime + GitHub
Use editor heartbeats for time and GitHub events for shipped evidence. Never infer productivity from commits alone.
Direct API09
Work context
Calendar · Jira · deployments · incidents
Google Calendar + Jira + New Relic
Import metadata and outcomes, not message bodies or production payloads.
Direct API10
Music
Recently played · listening windows
MusicKit or Spotify
Useful as context. Treat genre or tempo interpretations as hypotheses, not mood labels.
Quick log11
Calls & connection
Meaningful calls · people time · social energy
Kairo check-in
iOS does not provide a clean historical Phone log API. Count meaningful contact, not who said what.
Quick log12
Books & learning
Books · piano · Hindi · chess · typing
Progress habits + quick evidence
Sessions come from Time; milestones and outcomes stay in Progress.
Quick log13
Movies & new things
Movies · recipes · travel · firsts
Kairo natural-language log
Subjective life events are easier to say than to integrate. Kairo turns one sentence into an event.
Computed14
Luck, mistakes, wins
Fortune · lessons · achievements
Nightly reflection
Keep the original note. Let the model suggest tags, but never rewrite the remembered event.
Direct API15
Blog & creation
Drafted · published · ideas captured
Git/RSS + Progress
Measure the creative loop—idea, draft, publish—not vanity traffic by default.
Quick log16
Private habit
Day status · urge · recovery
Encrypted local count
Red-tier data: discreet labels, no content, separate retention and an instant-delete control.
APPLE REALITY CHECK
HealthKit is strong. Screen Time is possible but privacy-sandboxed and entitlement-gated. Historical Phone call details are not a clean public API. The correct fallback is a meaningful-call count, not brittle scraping.
02 / MOOD INTELLIGENCE
Find your k. Earn the conclusion.
The goal is not to guess your mood from a universal formula. It is to learn which factors predict your own reported mood, at which lag, and with what confidence.
PERSONAL MODEL / V0
Signal contribution
Illustrative coefficients—not your data. Positive bars predict a more pleasant mood.
Sleep regularity+0.71
Late phone use-0.63
Exercise minutes+0.46
Deep work+0.39
Sick-day signal-0.52
FACTOR SCORE
kᵢ = direction × ΔCV error × stability × coverage1
Label reality
Morning energy + evening mood, ideally backed by Apple State of Mind valence.
2
Build lagged features
Today, yesterday, 3-day and 7-day baselines. Standardize against your own normal.
3
Validate through time
Train on the past, test on the future. Never random-shuffle a life timeline.
4
Beat simple baselines
The model must outperform “same as yesterday” and your rolling weekly average.
5
Explain cautiously
Start with Elastic Net. Report stable predictive drivers, not invented causes.
What the model may say:“Late phone use predicted lower next-morning energy in 7 of 9 held-out weeks.”What it must not say:“Your phone caused your mood.”
03 / ARCHITECTURE
A small spine, not a data empire.
At personal scale, Postgres, scheduled jobs and three thin collectors are enough. No Kafka. No universal rewrite. No LLM sitting in front of raw health data.
SOURCEiPhone + Watch
SOURCEMac + Devbox
SOURCECloud services
SOURCEHuman moments
MINIMUM NECESSARY DATA
EDGE COLLECTORHealth bridgeHealthKit
EDGE COLLECTORAttention bridgeScreen Time + Mac
EDGE COLLECTORConnector jobsAPIs + webhooks
NORMALIZED EVENTS
THE NEW PIECE
Life Event Ledger
Immutable source events, daily aggregates, consent, retention and provenance.
{
type: "health.sleep",
start: "2026-07-30T23:41+05:30",
duration_min: 392,
value: 0.81,
privacy: "amber",
source: "healthkit",
confidence: 1.0
} QUERY · LEARN · ACT
ProgressIntent vs reality
TimeAttention evidence
Insight workerFeatures + k
KairoBriefs + experiments
What you intend to change.
Goals, tasks, habits, deadlines, Bingo checkpoints and explicit commitments.
Where attention actually went.
Timers, categories, projects, focus sessions and personal/work boundaries.
What happened.
Source events, aggregates, mood labels, evidence links and retention rules.
What happens next.
Questions, briefings, one useful nudge, weekly diagnosis and small experiments.
04 / TRUST MODEL
The most personal data
gets the least exposure.
“Jarvis” fails if it becomes surveillance with nicer typography. Raw data stays close to the device; the shared system receives aggregates whenever detail is unnecessary.
GREENOperational
Steps, focus totals, task completion. Cloud sync allowed.
AMBERSensitive
Mood, sleep, illness, medication. Encrypted; explicit source consent.
REDIntimate
Private habit, notes, people. Local/default aggregate; separate delete control.
01
No message or call content
02
No raw window titles in cloud
03
No health data in general LLM context
04
Every insight links to provenance
05
Delete by source, category or date
06
Model opt-out leaves tracking intact
05 / BUILD ORDER
Useful in a week. Intelligent in a season.
Data collection is not the milestone. A useful daily loop is. Each phase must produce a visible benefit before another collector is added.
00
Week 1Prove the loop
Create the event contract, connect read-only Progress and Time, and collect morning/evening mood. Produce one nightly brief.
✓Canonical event ledger
✓2× daily mood
✓Progress + Time summaries
✓Kairo nightly note
08:00How are you starting?
→Energy, mood, sleep confidence. Ten seconds.
DAYSystems observe quietly.
→Health, attention, progress and context.
21:30What mattered today?
→Mood, win, mistake, luck, new thing.
SUNDAYKairo explains the week.
Patterns, uncertainty and one experiment.
06 / RESTRAINT
What we deliberately do not build.
01
A universal new app
Progress and Time remain independent products with clear jobs.
02
A giant streaming stack
Personal scale needs boring jobs and a reliable database.
03
A productivity score
Mood, health and focus are different signals—not one moral grade.
04
Invisible mood diagnosis
Self-report remains ground truth; prediction is assistive and uncertain.
05
An LLM memory dump
Kairo receives summaries and provenance, never every raw event by default.
06
An alert for everything
One useful nudge beats a life full of notifications.
07 / THE DESTINATION
A life you can query—
without reducing it to data.
KAIRO / WEEKLY REVIEW
“Why did last week feel better?”
Three factors were unusually aligned: sleep timing varied by only 34 minutes, phone use after midnight fell on five nights, and you completed four focused work blocks.
This is predictive, not causal. Confidence: moderate. Best next test: preserve the sleep window while allowing normal phone use for one week.