Case Study

The Minimal Stack

Minimal infrastructure. Maximum output.

Zero Cloud Compute Zero Monthly SaaS Zero Vendor Lock-In Zero Token-Ware

6 enterprises · 43 automations · 90K+ lines of production code

Local-first. Deterministic. No cloud dependency.

Philosophy

The Thesis

Every dependency is a liability. Every SaaS subscription is a monthly vote of no-confidence in your own capability.

Zero Token-Ware

Token-ware is the new bloatware. SaaS products that charge per API call, per automation run, per AI inference — turning your operational costs into an unpredictable meter that scales against you.

The VindexAI approach: use AI to build the system, not to run it. The system executes on deterministic logic. AI is the architect, not the engine. Once built, it costs nothing to operate.

Deterministic by Default

90% of email triage is deterministic rules — keyword matching, sender classification, label routing. Only truly unknown senders hit Claude. The same principle applies to every pipeline: invoicing, exception detection, CRM sync, calendar coordination.

If a process follows a repeatable rule, it runs on code. Period. AI inference is reserved for genuine ambiguity — the 10% that actually requires judgment.

“Every dependency is a liability. Every SaaS subscription is a monthly vote of no-confidence in your own capability.”

Local-first means data sovereignty by default. No GDPR concerns. No vendor can shut you down. No pricing changes at renewal. No "we're sunsetting this feature" emails.

Cost Comparison

The Numbers Don't Lie

Monthly infrastructure cost: industry typical vs. the VindexAI stack.

Category Industry Typical VindexAI
Cloud compute $500–$5,000/mo $0
SaaS tools $200–$2,000/mo $0
Database hosting $50–$500/mo $0 (SQLite)
Workflow automation $150–$600/mo $0 (self-hosted n8n)
AI inference $100–$1,000/mo ~$32/mo (Claude Max)
TOTAL $1,000–$9,100/mo ~$32/mo
ANNUAL $12K–$109K Minimal

Claude Max ($32/mo) is the only recurring cost — and it’s used for building and iteration, not runtime operations. The production system runs on deterministic code.

Architecture

The Stack

Six components. Two machines. Everything else is code.

Compute

Mac mini “Olympus”

M-series. Always-on. Runs the entire back office.

01
Builder

MacBook “LT”

Interactive sessions. Code. Deploys. Mobile command.

02
Data

SQLite (WAL mode)

Single-file databases. No server process. Copy to backup.

03
Automation

n8n (self-hosted)

43 workflows. $0/month. Unlimited executions.

04
Web Server

Caddy

Automatic HTTPS. Single Go binary. Zero config drama.

05
Networking

Tailscale

Zero-config mesh VPN. Machine-to-machine anywhere.

06
6 Enterprises
43 Automations
60K+ Lines of Code
2 Machines
Decision Log

Why These Tools

Every technology choice was made for the same reason: eliminate dependencies.

SQLite (WAL mode) vs Postgres / managed DB

Single-file databases. No server process. Concurrent reads via WAL. Backup = copy the file. Zero admin overhead.

n8n (self-hosted) vs Zapier / Make

$0/month vs $150–$600/month. Unlimited workflows. Full code access. No execution caps.

Caddy vs nginx / Apache

Automatic HTTPS with zero config. Single Go binary. Simpler Caddyfile vs nginx.conf.

Tailscale vs OpenVPN / WireGuard manual

Zero-config mesh networking. No port forwarding. No firewall rules. Just works.

Local Macs vs AWS / GCP / Azure

One-time hardware cost. No monthly bills. Data sovereignty by default. No vendor can shut you down.

Doctrine

Local-First Principles

The rules that keep the stack minimal and the operator sovereign.

Data Sovereignty

All data stays on your hardware. No third-party cloud stores your operational data. No GDPR compliance concerns because the data never leaves the building.

No Vendor Lock-In

SQLite files are portable. n8n workflows are exportable JSON. Caddy configs are plain text. Nothing requires a proprietary runtime. You can migrate in hours, not months.

Trivial Backups

SQLite backup = copy the file. n8n backup = export JSON. The entire operational state of six enterprises fits on a thumb drive. Disaster recovery is measured in minutes.

Deterministic First

If it can run as a rule, it runs as a rule. AI inference is expensive and non-deterministic. Use it to build systems, not to run them. 90% of operations are pure logic.

One-Time Hardware

Two Macs. Bought once. No monthly compute bills. No auto-scaling invoices. No surprise overages. The marginal cost of running one more automation is zero.

Mesh, Not Hub

Tailscale creates a zero-config mesh between machines. No central server. No single point of failure. Each machine is independently useful. Together, they multiply.

ROI

The Math

What the industry spends vs. what actually works.

Industry Typical

  • × $12K–$109K/year in recurring infrastructure
  • × Vendor lock-in across 5–15 SaaS platforms
  • × Per-execution billing that scales against you
  • × Data spread across third-party clouds
  • × Migration takes months when a vendor sunsets

VindexAI Stack

  • Minimal annual cost — Claude Max is the only line item. See our methodology for how we design these systems.
  • Zero vendor lock-in — portable formats only
  • Unlimited executions — marginal cost is zero
  • All data on local hardware — sovereign by default
  • Full stack migrates in hours — it’s just files

At the low end, the industry stack costs 31x more. At the high end, it costs 283x more. And it still requires more maintenance, more integration work, and more vendor management.

Build Lean. Run Free.

Minimal infrastructure, maximum output isn't a compromise. It's what happens when you stop paying rent on your own infrastructure and start owning it.