← System Design
SLW · Rivent platform

Where things run

The rooms are always on in cloud hosting. The work that happens while nobody is watching runs from one server. Nothing anywhere depends on a person being signed in.

1 · Always on — no machine of ours

The rooms — served from cloud hosting

Vercel · Railway · Cloudflare Pages. Nothing user-facing is served from a box we run.

Terminaldaily flow
LP FlowLPs
DealFlowcompanies
Quillwriting
Fordresearch
Slide Libraryslides
Portalrivent.dev
Up regardless of the server

The databases — Supabase

Where the work is asked for and where the answer lands.

App data Every room reads and writes its own hosted database.
Job queues Asking for work is just a row. Queueing a research job in Quill is fully hosted — no server involved yet.
a queued job sits waiting — the app's part is done

2 · The muscle — one server, automation only

slwdev

A single Hetzner box. Cron wakes every few minutes, each runner checks its queue, and exits immediately when there is nothing to do — an idle tick is free.

Quill — research Researches a queued firm or topic and writes back a brief with sourced claims. Every 10 min · daily full pass
Quill — voice Reads LP letters and decks, proposes house-voice elements with citations. Twice an hour
Quill — ingest Indexes documents as they land in the Library. Every 10 min
Ford — intake drain Pulls whatever arrived by email, upload or machine drop into the pipeline. Every 10 min
Slide Library — recompute Refreshes staleness across the deck. Plain code, no agent. Nightly
Platform housekeeping Portal sync, vault backup, watchdog. Plain code, no agent. Hourly · nightly · 5 min

Terminal, LP Flow, Podcast and DealFlow have no background work here at all — they are entirely self-contained in their hosting.

claims the job, writes the result back

3 · Whose credential does the work run on

Each app's own API key — never a person's login. A runner that cannot find its key stops, rather than quietly falling back to whoever happens to be signed in on the box. That matters because the alternative fails invisibly: the work keeps happening and looks healthy, while it is billed to a person and dies the moment they sign out, switch accounts or leave.

What breaks when something goes away

If this goes away
Anyone's laptop

No app and no job has ever depended on a personal machine.

Nothing stops
If this goes away
Anyone's Claude account

Agent sessions run on app keys, so no person is in the path.

Nothing stops
If this goes away
slwdev

Every room stays up and nobody is logged out. Quill stops researching, distilling and ingesting; Ford stops draining intake.

Automation stops
If this goes away
Hosting or Supabase

The affected room goes down, and any runner that writes into it has nowhere to put its result.

That room stops

slwdev is the fleet's single point of failure for automation only, never for availability. That is a deliberate trade — one box is cheap and simple — but worth knowing rather than discovering. Moving the runners into their apps' own hosting would remove the last machine dependency; not scoped, and the cost of the move is real.