An autonomous, multi-domain content engine

One brain. Many storefronts.

A single local program that researches, writes, evaluates, compiles, and runs a whole portfolio of niche book-review sites — each on its own five-dollar VPS. The writing and judgment run through Claude Code on a subscription, so the marginal cost of another article, or another entire site, is effectively zero.

Internally: BookEngine — the "customCMS" experiment, a content-management system with no CMS to log into.

The shape

The CMS lives on your machine. The sites are just output.

There is no admin panel on the internet. One local brain owns the database, does the thinking, compiles static HTML, and pushes it out. Each domain is a dumb, fast binary that only serves.

Local brain
Domain manager · pipeline engine · persona engine · analytics — one SQLite DB per domain, plus a global one
▼   ▼   ▼
scifi-reads.com
Sci-Fi & Fantasy
VPS · $5/mo
biz-books.co
Business & Investing
VPS · $5/mo
selfhelp-shelf.com
Self-Help & Psychology
VPS · $5/mo
The pipeline

Every article runs the same six steps.

Per domain, per book, on a schedule — the brain turns a title into a published, cross-linked review.

01
Research
Open Library & Google Books — facts, editions, metadata.
02
Synthesize
Draft the review against the domain's voice.
03
Review
Edit and fact-check the draft.
04
Eval
Score it; regenerate until it clears the bar.
05
Compile
Render to static HTML with internal links.
06
Push
Ship the files to the domain's VPS.
The two engines around it

Not just a writer — a librarian and a community member.

A book knowledge graph

Reviews aren't standalone pages. Every book cross-references the whole library, so a review can say "if you liked this, try…" and mean it. Tables carry a typed prefix — BK_books, AU_authors, BR_book_relationships — so the schema reads like a map.

A social persona engine

Personas that watch videos and read threads on TikTok, YouTube, Reddit, and X — engaging authentically to build credibility first, and only promoting once they've earned it. The analytics loop pulls data back from every VPS for cross-domain insight.

How it's built

One Go binary, two modes.

Language
Go with pure-Go SQLite (modernc.org/sqlite, no CGo) — cross-compiles to Linux trivially
Two modes
CLI (local): research, write, compile, push.  Server (VPS): serve static files + analytics endpoints
The LLM
Claude Code CLIclaude -p --output-format json. Subscription, not metered API — zero marginal cost
Data
One SQLite DB per domain plus a global one; the book knowledge graph with a TBL_ prefix convention
Research
API-only for v1 — Open Library, Google Books. Single domain first, then fan out
Hosting
Cheap VPS per storefront (~$5/mo) serving a static binary. The expensive thinking stays on one machine
Status

A local experiment that outgrew its language.

Local · not published

Filed under customCMS because that's the question it started as — what does a content management system look like with no CMS to log into? It began as a pure-Zig concept and became a working Go engine with a real pipeline, database layer, and compiler. It stays local: the interesting part isn't code anyone needs, it's the idea — move the whole editorial operation to one brain, and let the internet only ever see the output.

Experiments