// the take
AlchemyCMS is a Rails engine that turns an existing Rails app into a full CMS — admin dashboard, versioned pages, multi-language/multi-domain support, and a REST API for headless use. It's for Rails teams who want a CMS they can mount into their own app and extend with their own models, rather than a hosted SaaS or a JS-stack headless CMS.
The admin UI is built with Rails ViewComponents (ingredient editors, dashboard widgets, link dialogs), not a pile of legacy partials, which makes it realistic to override or extend individual pieces. It's a true Rails engine — you bring your own User model and auth, and it integrates into an app you already have instead of demanding a dedicated one. The project maintains a real upgrade path: a stable branch per supported Rails version going back to Rails 3, plus an `alchemy:upgrade` rake task that automates most of the migration work between major versions. CI runs Brakeman security scanning, lint, and coverage on every push, and there's a dedicated script (and CI check) that catches migrations that would silently destroy data.
That migration checker exists because of a real landmine: on SQLite, common schema changes (rename_column, remove_column, add_timestamps, etc.) make Rails rebuild the table, and with FK cascades on, that silently deletes every child record — page versions, elements, ingredients. Rails' own safeguard against this doesn't work inside a transaction, so it's not just a theoretical footnote, it's something a contributor or self-hoster can hit for real. Image handling (Dragonfly) renders to local `public/pictures` by default; anything beyond a single server needs a separate extension or manual S3 config, it's not solved out of the box. Because Alchemy ships a branch per Rails version, bumping Rails in your app means waiting on or coordinating with Alchemy's own release cycle rather than just updating a gem. And despite the "headless" pitch, it's still a server-rendered Rails monolith underneath with hard dependencies on ImageMagick and a SQL database — no lightweight or edge deployment story.
We read the README, the directory tree and a key source file in November 2026. Nobody pays to be listed. How we judge →