The Anders Loc Platform

Your AI. Your hardware. Your rules.
Architecture Document v2 — March 14, 2026 — Sterngold Family

The Anders Loc Platform

Your AI. Your hardware. Your rules.

A family platform for sovereign, personal AI — running on your own machines, private by default, connected when you choose. Built by the Sterngold brothers. Open to anyone who wants to join.

Version 2 — March 14, 2026


What Is This?

Imagine every person in the family has their own personal AI assistant. It knows your work, your goals, your documents. It runs on your laptop — not in someone else's cloud. Nobody can see your data. But when you want to share something with family, you can.

That's what we're building.

For Alex: A system that ingests your trading data from Interactive Brokers and Fidelity, tracks your portfolio across accounts, quizzes you on CFA Level II topics, automates your monthly budget reports, and surfaces market insights from the sources you already follow (Trading Economics, FXStreet, Fed data, S&P Global) — all private, all yours. Your Chief of Staff, but for finance.

For Nina: A personal workspace that helps you build your business — writing, client management, website — with AI that understands your methodology and speaks your language.

For Vlad: The command center that already exists (Mission Control, coaching tools, writing assistant, health tracking) — now made portable, shareable, and upgradable.

For everyone: The family infrastructure that grows with us.


The Principles (Plain Language)

We borrowed ideas from companies that built platforms serving millions — Uber, Booking.com, Tesla, Amazon. But we're three people, not three thousand. So we took the principles and left the complexity.

1. Your data is yours

Everything runs on hardware you own. Cloud services are connectors, not requirements. If the internet goes down, your AI still works.

2. Each person owns their space

Alex can change anything in his setup without breaking Vlad's. Nina can build her business tools without asking permission. Like apartments in a building — shared foundation, private rooms.

3. Set up once, update forever

We set up the hardware once. Every improvement after that is a software update — pushed remotely, no physical access needed. Vlad can fix Alex's machine from Amsterdam while Alex is in Atlanta.

4. No back doors

Services talk to each other through proper channels, not by reaching into each other's files. This is what makes it safe to share infrastructure without sharing secrets.

5. Keep it simple

We use the simplest tool that works. No enterprise software for three people. When something breaks, anyone should be able to understand what went wrong.

6. Rebuild from zero

If a machine dies, we buy a new one, run one command, and everything comes back exactly as it was. Nothing lives only in someone's head.

7. Privacy has tiers, not walls

Vlad and Alex share openly (brothers). Nina has her own boundary — she sees what she's invited to. Future clients get complete isolation. This is built into the structure, not just a promise.

8. The best for everyone

Nina gets the same quality of tools as Vlad — just with a different interface. The less technical person gets a browser dashboard. The technical person gets a terminal. Same power, different door.


What We Built So Far (Honest Assessment)

We've been building since February 6, 2026 — about five weeks. In that time, we created a lot. We also broke a lot. Both matter.

What Works Well

System What it does Status
Mission Control Personal dashboard — tasks, calendar, boards, writing pipeline, health, portfolio Running daily, localhost
13 websites sterngold.nl, werkanders.com, Sterngold's Cooking, EventRadar, FoodLog dashboard All live on Cloudflare
FoodLog Voice-based food diary with AI calorie estimation, health data tracking Live, 30+ days of data
Writing system 15 essays, gallery + workshop model, interview-edit process, editorial pipeline Active, first LinkedIn post (11K impressions)
Anders Loc Local AI on methodology corpus — ask questions, get sourced answers Prototype working (Ollama + ChromaDB)
22 skills Daily rituals (/hi, /wrap, /gn), writing (/vladvoice, /editor), coaching (/solutions, /resettl), finance (/cfo) 15 active, 7 parked
9 automated jobs Stock prices, Obsidian sync, weekly reports, monthly reviews Running via launchd
Security Air-gapped health/finance folders, pre-commit hooks, API key rotation, sensitivity gates Layered, tested

What Broke (Lessons That Shape the Platform)

What happened Root cause Platform lesson
FoodLog stopped working for 2 days Redeploying the Edge Function silently re-enabled JWT verification, breaking the webhook. Nobody noticed. Every service needs a health check. Redeploys need post-deploy verification — automated, not manual.
Health data stopped flowing (Mar 12) iOS killed the background app (Health Auto Export). The app reports "exported" but the data never arrives. Don't depend on third-party iOS background apps. Use Apple-native automation (Shortcuts) or pull-based sync instead.
Context compaction lost work Claude Code hit context limits mid-task. Background agents disappeared. Edits failed because prior file reads were forgotten. Save progress early and often. Don't trust that a long session will complete. Checkpoint before it's too late.
Test data polluted real data Test webhook entries merged with real food entries (20-min session window). No way to tell them apart. Test environments must be separate from production. Even for one user.
3000-line monolith Edge Function FoodLog grew from 200 lines to 3000 lines in one file. Hard to debug, hard to update, risky to deploy. Services should be small and independent. Split when a file has two responsibilities.
Scripts depend on exact file paths Move a file, break a script. Launchd jobs, symlinks, Python scripts all have hardcoded paths. Migration means updating 50+ paths. Use relative paths and environment variables. The platform should work regardless of where it's installed.
Manual deployment Deploying a website means: copy files to temp folder, git commit, push, verify. Different steps for each site. One deploy command per service. Automate or document. Never rely on memory.
No backups of local-only data VersGil and Health were only on one machine. If it broke, the data was gone. Now fixed (local git vaults, today). But need off-site backup too.
Secrets scattered across 7 locations .env files in ~/.config/eventradar, ~/.config/versgil, ~/.config/foodlog, plus Supabase secrets, SSH keys, OAuth tokens. Central secret management. One place to look, one process to rotate.

These aren't failures — they're the foundation for building it right.


The Five Domains

Each domain is a self-contained area with its own purpose. Today these are folders on a laptop. Tomorrow they become independent services that can run on any machine.

Today (March 2026)

Domain What Who uses it
Health Food logging, weight tracking, Apple Watch data, blood work, mental health Vlad
Finance Portfolio tracking, spending, tax, pension, property, FIRE planning Vlad, Alex
Content Websites, writing, LinkedIn, coaching materials Vlad
AI Loc Local AI engine — Ollama models, RAG knowledge bases, coaching AI Vlad (Alex soon)
Shared Mission Control, task management, family docs, infrastructure Everyone

In Two Years (March 2028) — Projection

Domain What changed Who uses it
Health Unified family health dashboards. Each person tracks their own data privately. Shared anonymized trends if desired. Vlad, Alex, Nina
Finance Shared investment tracking (brothers). Individual spending stays private. Tax season automation. Vlad, Alex
Content Vlad's writing + Nina's business content + shared family media. Each person has their own publishing pipeline. Vlad, Nina
AI Loc Multiple knowledge bases: Vlad's coaching methodology, Alex's CFA corpus, Nina's business framework. Each sovereign. Everyone + clients
Shared Family dashboard. Shared calendar. Trip planning. Recipe sharing. Home automation. Everyone
Coaching (new) Client-facing services: intake forms, session prep, progress tracking. Air-gapped per client. Vlad's coaching clients
Education (new) Learning tracks, study companions, skill tracking. Alex's CFA → potentially a product for others. Alex, Nina, clients

Domains grow organically. We don't build "Education" until someone needs it. But we design the platform so adding a new domain is trivial — one folder, one config, one deploy.


The People

Three roles, equal value

Role Who Interface What they can do
Builder Vlad Terminal, Claude Code, Docker, Git Full admin. Builds and maintains the platform.
Power User Alex Claude Code, browser, Git basics Uses AI tools, manages his own space, pulls updates. Can code if he wants to.
User Nina Browser, simple commands, guided setup Full access to her services through friendly interfaces. Equally powerful — different entry point.

The rule: The User role is not "less than" — it's "different interface to the same power." Nina's browser dashboard can do everything Vlad's terminal can do. If it can't, that's a bug, not a feature.

Future people

Person When Role What they get
Michelle When she's ready User Her own space — connected to Vlad's for shared trips, cooking, planning
Erik (Michelle's son) Later User Learning companion, age-appropriate AI tools, gaming perhaps
Coaching clients 2027 Isolated User Their own House — completely air-gapped from family data

Each new person is the same process: create their space, configure their AI, connect to shared services they're invited to. No custom engineering per person.


The Technology Stack (What's Under the Hood)

This section is technical. Alex — this is for you. Nina — skip to "How You'll Use It" below.

Hardware Plan

Phase Machine Role When
Now Vlad's MacBook Air M3 8GB Start building the platform here This week
Phase 1 Vlad's MacBook Air M5 15" 32GB Main development + Docker services ~Mar 25
Phase 2 Alex's MacBook Air M5 13" 16GB Alex's own House Same time
Phase 3 Mac Mini M4 Pro 64GB Always-on family server (24/7) Later 2026

When the Mac Mini arrives, all services move there. The laptops become pure clients — thin, fast, no server duties.

Networking

All machines connect through Tailscale — a secure, zero-configuration network. Think of it as a private tunnel between all family devices, no matter where they are in the world.

┌──────────┐     ┌──────────┐     ┌──────────┐
│ Vlad Air │ ←──→│ Mac Mini │←──→ │ Alex Air │
│(Amsterdam)│     │ (24/7)  │     │(Atlanta) │
└──────────┘     └────┬─────┘     └──────────┘
                      │
                ┌─────┴──────┐
                │ Nina's Mac │
                │(Amsterdam) │
                └────────────┘

Vlad can fix Alex's machine from Amsterdam. Nina can access shared dashboards from anywhere. All encrypted, all private.

Services (Docker)

What is Docker? Think of it as a shipping container for software. Each service runs in its own container — isolated, portable, reproducible. If the Mac Mini breaks, you buy a new one, run one command, and all the containers come back exactly as they were.

The entire platform is defined in one file (docker-compose.yml):

Service What it does Who uses it
Mission Control Task dashboard, calendar, boards Everyone
Ollama Local AI models (runs on your hardware, not cloud) Everyone
Anders Loc Knowledge base + RAG (ask questions about your own documents) Everyone (each with their own data)
HAE Server Collects Apple Watch health data automatically Vlad
Grafana Beautiful health dashboards Vlad (eventually everyone)
Cloudflare Tunnel Secure access from outside the home network Shared

Starting everything: one command. Updating everything: pull + restart. That's it.

Version Control (Git + GitHub)

What is Git? "Track Changes" for your entire project — but smarter. Every change is recorded with who made it and when. You can go back to any previous version. Nothing is ever lost.

What is GitHub? A website that stores Git projects in the cloud. Private — only the family can access it. It's the single source of truth for all platform code and configuration.

One private repository: house-of-anders

house-of-anders/
├── docker-compose.yml      ← one file, whole platform
├── domains/
│   ├── health/             ← FoodLog, Apple Watch, Withings
│   ├── finance/            ← portfolio, spending, tax
│   ├── ai-loc/             ← Ollama config, knowledge bases
│   └── content/            ← website deploy scripts
├── users/
│   ├── vlad/               ← Vlad's AI config + skills
│   ├── alex/               ← Alex's AI config + skills
│   ├── nina/               ← Nina's AI config + skills
│   └── shared/             ← shared scripts + configs
├── docs/
│   ├── what-is-docker.md   ← guide for non-technical users
│   ├── what-is-git.md      ← guide for non-technical users
│   ├── runbook.md          ← how to maintain
│   └── onboarding.md       ← how to add a new person
└── infrastructure/
    ├── tailscale/           ← network config
    ├── backups/             ← backup scripts
    └── cloudflare/          ← tunnel config

For Alex: You'll use git pull to get updates. That's one command. Everything else is automated. For Nina: Updates happen automatically. You don't need to know Git exists.


How You'll Use It

Alex's Day

Wake up. Open laptop. Your dashboard shows: IBKR portfolio up 1.2% overnight, 3 options expiring this Friday, CFA study streak at 12 days. The AI pulled overnight Fed balance sheet data and flagged a repo rate anomaly — "review before market open." Check your 44-column trade tracker — Claude already reconciled yesterday's fills from IB. Quick CFA drill: "Explain pension accounting under IFRS vs US GAAP" — the AI generates a concise explanation with a practice question, sourced from your Mark Meldrum materials. Budget update: the monthly report you used to spend 3 hours preparing was auto-generated at midnight — spending, positions, NAVs, all consolidated from IB, Fidelity, and NNEK. Vlad pushed an update to the shared investment research — it's already on your machine.

Nina's Day

Wake up. Open laptop. Your dashboard shows today's clients, pending proposals, and a draft blog post the AI started based on your notes from yesterday. Click the blog post — review it, adjust the tone, publish it to your website. A client sent documents — ask your AI to summarize them (they stay on YOUR machine, not in any cloud). Need to write a proposal in Dutch? Your AI knows your style and your methodology. Check the family shared folder — Vlad added a recipe for Saturday dinner.

Vlad's Day

/hi — Mission Control opens. Health card shows weight, HRV, steps. Calendar shows Resettl session at 10. Focus card: Preply prep is priority. Write for 90 minutes — the AI holds your voice, your references, your editorial process. Coach a client at 2 — the session prep was generated from their intake form, running on local AI, no cloud involved. /wrap at 5. Everything saved, versioned, backed up. Push an update to Alex's config — his machine gets it automatically.


Budget

Item Cost Notes
Docker Desktop Free Personal use
Tailscale Free Up to 3 users, 100 devices
GitHub Private Free Unlimited private repos
Cloudflare Tunnel Free Secure external access
Backblaze B2 backup ~€6/month 10GB free, then €0.006/GB/month
Domain renewals ~€30/year werkanders.com etc.
Total ~€6/month Hardware already planned separately

Principle: free first, pay for reliability when free breaks. Frugality is the mother of creativity — but we don't compromise on backups.


Backup Architecture

Layer What How often Where
Time Machine Full machine backup Continuous External drive
Git repositories All code and config Every commit GitHub (encrypted at rest)
Local vault git VersGil + Health (sensitive) Daily at 23:00 Local only — never pushed
Off-site backup Critical data Daily Backblaze B2 (encrypted)
Recovery drill Can we rebuild from zero? Quarterly Test on spare machine

The rule: If the Mac Mini catches fire, we buy a new one, clone the repo, restore the backup, and docker compose up. Same platform. Zero manual steps.


What's Next

This week

When machines arrive (~Mar 25)

Later


For Alex — What We Learned From Your Answers

You answered the questionnaire (thanks!) — here's what we're building for you based on what you said:

What you need What we'll build Priority
Monthly budget report takes 3+ hours Auto-generated monthly report from IB, Fidelity, NNEK data. Spending, positions, NAVs consolidated. High — automate first
Trade journaling is manual Auto-ingest daily IB trade reports → structured journal with entry/exit, P&L, thesis notes High
CFA Level II (Aug 2026) RAG on your Mark Meldrum + CFAI materials. Quiz mode, concept explainer, weak-area drilling, Ethics infographics High
Portfolio tracking across IB + Fidelity + NNEK Unified dashboard: positions, P&L, margin, 52-week high/low, option vol skew High
Market data from 6+ sources Automated weekly digest from Trading Economics, FXStreet, Treasury TIC, Fed, S&P Global EPS Medium
Dr. Mark Meldrum Discord/YouTube transcripts Record + transcribe his daily market reviews for when you can't dial in Medium
Password manager (currently Excel!) Move to 1Password or iCloud Keychain — Excel passwords are a security risk Quick win
Want to see code + thinking process Claude Code shows everything by default — code, reasoning, file changes. You'll love this. Built-in
SSH maintenance must be logged Tailscale logs all connections. We'll add a maintenance log file. Trust + transparency. Built-in

Your Excel Files (Structure Analysis)

We looked at the structure (not the numbers!) of your sample files:

INVESTMENTS - APPLIED.xlsx — 5 sheets: - Trades & Positions (256 rows × 44 columns — this is your main tracker) - IB Transfers, Margin tracking, Prompts (your own AI prompts!), Trades archive - This is a sophisticated workbook. The AI will need to understand your column structure to automate reporting.

Budget template.xlsx — 9 sheets: - Budget (296 rows × 83 columns — full year, very detailed) - Gas/Electric regression analysis (you're modeling utility costs!) - Coca Cola shares + Shareworks (equity compensation tracking) - Investment modeling (6.7% return scenario) - Loan schedules (550K, 20yr, 1.79%) - This is CFO-level personal finance. The automation potential is enormous.

What's Next for You

  1. Pick a color for your MacBook Air 13"
  2. Share your CFA study PDFs (we'll build the study companion from them)
  3. Export one month of IB activity statements (CSV or PDF — we'll build the auto-ingest from this)
  4. Think about: which monthly task do you want automated FIRST?

The Vision

Today it's three people and a few laptops. Tomorrow it could be a family platform that grows with us — each person sovereign, each connected, each upgradable. The pattern works for three people. The same pattern works for thirty.

But we start with three. And we start now.

Your AI. Your hardware. Your rules.


Appendix: The Philosophical Foundation

Why we build this way — principles borrowed from the best, filtered through our own experience.

From Uber — Domain Gateways

"Group related services behind a single gateway interface so consumers see one door per domain, not dozens of internal services."

Each project (EventRadar, VersGil, CFA Study) becomes a "domain" with one clear entry point — one API, one config, one deploy. New projects plug into the platform without touching existing ones.

From Booking.com — Autonomy at the Edges

"Push decision-making to the edges by giving every unit the tools to test and ship independently, without central approval."

Vlad saw this firsthand — from 35 developers to a platform serving millions. The principle that made it work wasn't microservices, it was autonomy. Each person (Vlad, Alex, Nina) can use and update their slice without breaking anyone else's. Independence is structural, not just a promise.

From Tesla — Ship Hardware, Update Software Forever

"Design the platform once for maximum capability, then improve everything through software updates pushed to the same unchanged base."

The Mac Mini is our "shipped hardware." Set it up once with maximum headroom (64GB, containerized services), then every improvement is a software deploy — new AI model, new service, new family member's app. Never re-architect the base; only push updates to it.

From Amazon — The API Mandate

"All functionality must be exposed through service interfaces. No back doors, no direct data reads, no shared-memory shortcuts. Design every interface as if it will be consumed by an external party." — Jeff Bezos, 2002

Even for three people, never let one project reach into another's files. Alex's portfolio tracker doesn't read Vlad's health data. This is what makes the jump from "my scripts" to "family platform" possible — clean contracts between components.

From the Self-Hosting Community — Declarative Everything

"Define your entire infrastructure in version-controlled config files so you can rebuild from zero with one command."

Docker Compose, environment configs, deploy scripts — all in Git. If the Mac Mini dies, you buy a new one, clone the repo, and docker compose up. Same platform, zero manual steps.

From Günther Anders — The Organ Metaphor

"The machine doesn't replace you. It's an organ you grew."

Every AI component should feel like an extension of the user, not a separate tool they visit. Anders doesn't live in a browser tab — it lives in the terminal, in the daily ritual, in the morning briefing. Alex's CFA companion shouldn't be "an app" — it should be woven into how he studies.

From Steve de Shazer — Minimal Elegance

"The simplest intervention that works is the best one."

Docker Compose, not Kubernetes. Mono-repo, not 10 repos. Shell script before CI/CD pipeline. Three lines of code is better than a premature abstraction. Only add complexity when the current solution genuinely breaks. But when reliability matters (backups, security), invest in the Miele, not the LG — technology serves a purpose.

Our Own — Document the Build

"The process of building AI Loc IS the content for AI Loc."

Every architecture decision, every Docker config, every migration script is potential educational material. Write it as if a practitioner will read it. This document isn't afterthought documentation — it's the product's first demo.


Built by the Sterngold brothers. One family, one platform. March 2026.