Personal AI OS

AI-READY PERSONAL INFRASTRUCTURE

Structure your digital life so AI can actually use it.

Personal AI OS is an open experiment in combining specialist apps, clean data flows and lightweight databases into an architecture that makes personal data understandable, portable and useful for AI.

See the architectureView real workflows

No AI guru claims. No fake automation. No one app pretending to do everything.

Garmin Fenix 8 used in everyday life
Real devices. Real data. Real friction.

THE CORE IDEA

AI is not the system. The architecture makes AI useful.

Today, much of the intelligence is still human: choosing the right tools, defining the source of truth, deciding what belongs where and creating reliable paths between data sources. AI becomes valuable only after that foundation exists.

ARCHITECTURE

Specialist tools stay specialized. The surrounding system makes the data AI-ready.

CaptureGarmin, Apple Health, specialist apps
→
OrganizeSources of truth, timestamps, clean ownership
→
Fill gapsAirtable and structured personal context
→
Use with AIAnalysis, interpretation and decisions

REAL-WORLD CASE STUDIES

Not demos. Working systems used in everyday life.

CASE 01

A personal data lake without another tracking app

Natural-language observations are stored as timestamped facts in Airtable. That preserves the raw information while keeping future calculations flexible.

  • One conversational input
  • No duplicate tracking workflow
  • Raw facts instead of frozen calculations
  • AI-ready history for later analysis
Airtable journal used as a personal data layer

CASE 02

Turning wearable signals into usable context

Wearables are already excellent at collecting data. The real challenge is preserving enough surrounding context so those signals can later be interpreted together with training, nutrition, routines and subjective observations.

Garmin health and activity overview

CASE 03 — UPCOMING

Making a specialist app part of a broader health-data workflow

A future case study on interoperability, Apple Health and how structured export can make a specialist app far more useful without turning it into an all-in-one platform.

No personal consumption history or identifiable health data will be published.

CASE STUDY — ONE QUESTION, MULTIPLE SYSTEMS

From fragmented data to one practical training decision.

The real benefit appears when several specialist systems can be queried in one conversation. A single request can combine recovery data, endurance history, strength load, nutrition context and real-life constraints into one actionable recommendation.

CAPTURE & SPECIALIST TOOLS

Garmin / FenixRunning, cycling, walking, sleep, heart rate, HRV and activities
LiftTrackStrength planning, completed sessions and muscle load
DrinkLogDrinks and estimated blood alcohol concentration
AirtableNutrition, subjective observations and everyday events

DATA & ACCESS LAYER

Apple HealthShared health-data route for compatible sources
FreddyGarmin Connect, Apple Health and Runalyze in one health context
MCP connectorsFreddy, LiftTrack and Airtable become directly queryable

CONVERSATIONAL DECISION LAYER

ChatGPT weekly contextGoal, schedule, symptoms, upcoming events and prior decisions
One usable answerHeart-rate targets, load ceiling, stop rules and practical execution

Some Garmin data reaches Freddy through more than one route. The practical system tolerates this overlap rather than pretending the personal data ecosystem is perfectly clean.

THE QUESTION

“I want to run 11 km today. How should I structure the session based on all currently available information?”
ChatGPT question querying Airtable, LiftTrack and Freddy

THE SYNTHESIS

Several systems contribute different parts of the answer.

FreddySleep, HRV, resting heart rate, recent endurance load and Garmin/Runalyze context
LiftTrackRecent strength session, loaded muscle groups and residual fatigue
AirtableNutrition, alcohol, symptoms and manually logged events
Chat contextTraining goal, injury history, family schedule and upcoming football

THE RESULT

A recommendation that is specific enough to use immediately.

  • 11 km as a controlled aerobic run rather than a tempo test
  • Target heart-rate range and explicit upper ceiling
  • Adjustment for recent strength work and uneven sleep
  • Stop or run-walk rules for pain, heavy legs or excessive heart rate
  • Practical guidance for shoes, warm-up and the rest of the day
Detailed training recommendation generated from multiple personal data sources
Freddy Apple Health source overview
Apple Health and its contributing sources inside Freddy.
Freddy Garmin Connect and Runalyze sources
Garmin Connect and Runalyze add further training context.

PRINCIPLES

The rules behind the project.

01

One source of truth

Each domain has one primary system. Avoid duplicate ownership.

02

Raw facts first

Preserve original observations. Recalculate derived values later.

03

Specialists over monoliths

Use focused tools that do one job well.

04

Human before automation

Automate only after the workflow and data model make sense.

05

Privacy by separation

Public documentation never shares the private personal data layer.

06

No hype

Document trade-offs, limitations and failures as clearly as successes.