Mission Control: one view for a whole operation
I built this for my own operations before I'd ever bring the approach to yours. Here's how it works — and what it proves.
Most small operations don't fail for lack of tools — they drown in them. Mail in one place, calendar in another, contacts somewhere else, tasks, bills, content, system health each in their own tab, each with its own login, and no single place that tells the truth. Mission Control is my answer to that for my own work: one private command center that pulls it all into a single view. This is how I built it, and why it's the clearest proof of the approach I bring to a business.
01The problem
The cost was never any one tool — it was the switching and the scatter. Information lived in ten places, nothing was reconciled, and there was no command view of the whole operation. Every "quick check" meant opening five tabs. The day was death by a thousand context switches.
02The constraints
- One person building it. No team — so it had to stay simple enough for one person to own and maintain for years.
- Privacy non-negotiable. It holds personal and financial data, so it had to be private and tailnet-only — never publicly exposed.
- Modest hardware. A VPS with no GPU: no heavy frameworks, no local model inference.
- Keep running while being built. A live daily tool, not a lab project — every change had to be safe and incremental.
03The solution architecture
A single web app that pulls each domain into one view. A stdlib-only Python backend (deliberately no web framework) exposes the routes and proxies the underlying services; each domain has its own small store; an AI assistant sits on top over a retrieval layer; and the whole thing is served privately over an encrypted tailnet.
04The implementation
- Frontend: a single-page shell plus one JavaScript module per tab — vanilla JS, no framework.
- Backend: stdlib-only Python — routes, API handlers, and service proxies; isolated mail and calendar micro-services; a small store per domain; notifications and scheduled jobs.
- AI where it fits: the assistant answers and triages over a retrieval layer on top of the unified data — AI on top of deterministic plumbing, not sprinkled everywhere.
- Built live, in small steps: 255 commits over about five weeks, each a safe increment to a system already in daily use — never a big-bang rewrite.
05The results
- One dashboard replaced a dozen scattered tools — mail, calendar, contacts, tasks, bills and finances, content, and system health now live behind a single command view.
- Live, self-hosted, and in daily use — not a prototype.
- 255 commits in ~5 weeks, delivered without ever taking the tool down — sustained, safe, incremental delivery.
- No framework debt: stdlib-only means nothing breaks on a dependency upgrade, and one person can own it — on a modest, GPU-less server.
An honest note on proof. Mission Control is private and holds real financial and personal data, so there are no screenshots here — showing them would defeat the point. I'm describing the architecture and the capability, not the contents. I also don't claim a specific "hours saved" number, because I didn't measure one; the numbers above are the ones I can actually stand behind.
06Lessons learned
- Stdlib-only was the right call for a one-person, long-lived system — fewer moving parts, nothing to maintain but the code you wrote. Durability came from subtraction.
- Build it live, in small commits. Two hundred small, safe increments beat any big rewrite for a tool you can't afford to have down.
- Unify the view; keep the plumbing deterministic. Put AI where it genuinely helps — assist and triage — not everywhere. The data layer stays plain, predictable code.
- Know when not to consolidate. Not every tool belongs in one view — only the ones you touch daily. Pulling in the rest would have been effort without payoff.
07The reusable pattern: One View
The move generalizes to any small business drowning in disconnected tools:
- Inventory the tools and data you touch daily.
- Find the disconnections — where data doesn't flow and you re-enter it by hand.
- Define the single view — the one place that should tell the truth.
- Connect and consolidate — eliminate the duplicate entry.
- Keep one canonical place, and let AI assist on top of it.
That's One View — and I proved it on my own operations before bringing it to anyone else's. If your business runs across a dozen disconnected tools, this is the approach I'd bring to it.