Home / Engineering / How Kolaboro is built
How Kolaboro is built
One edge worker, static pages, published data files and one database, with every moving part checked from outside. A map of where a request goes and where data lives.
Updated 18 November 2026
Request path
A page request reaches the nearest Cloudflare location. Static pages come from the asset store. Project, place and ranking pages are built by the worker from the published data files and kept in the edge cache, so most answers need no database call.
Data
Research runs on a separate branch: scripts read developer pages and official registers and write one project file, then a rebuild publishes the data files the site reads. Every fact keeps its source and the date it was read. A hash of each daily dataset goes into a public chain.
Database
Accounts, cases, follows and counters live in a Postgres database. The worker calls a fixed list of server functions; no page reads tables directly, and personal data never goes into the published files.
Deploys
Every change is a git commit on the site branch, which Cloudflare Pages deploys in one to three minutes. Before each push the worker bundle is built and its size compared with the last build. The deploy guard checks the live site every 30 minutes and restores the last good files if its checks fail twice.
Model spend
The few site features that call a language model stop by themselves at a daily spend cap of 0.80 USD. Daily spend is public, and the checks page raises an alert above 1 USD a day.
Maintenance mode
One switch (an environment setting, off today) makes every page answer with a short notice in the visitor’s language and a retry time, while the data API and the status page keep working.
See the maintenance notice · /api/open/ai-spend · /data/snapshots/