# How it works > Mental model: sites, headlines, drafts, posts, runs, taxonomy. The pull-based delivery model. Why Mentionwell is headless, not a CMS. Mentionwell is a **headless content engine** with a public read-only API. Your destination site **pulls** content; Mentionwell never writes to your codebase. ## The data model Five top-level objects live in Mentionwell's database. Most operations move data from one to the next. - **Site** — one destination domain. Has a brand profile, taxonomy, delivery config, and an automation policy. - **Taxonomy** — a tree of content clusters. Headlines get attached to nodes for topical authority + internal linking. - **Headline** — a planned article. Has a target keyword, search intent, schedule, and lifecycle status (`proposed → approved → queued → generating → drafted → scheduled → published`). - **Draft** — a generated article waiting on review. Includes HTML, Markdown, FAQs, JSON-LD, sources, and a measured cost. - **Post** — a published article. Same shape as a draft plus `published: true` + `publishedAt`. Plus **Run** (one execution of the planner / writer / critic / publish flow, with cost + duration) and **Job** (one worker run on the queue: onboarding, draft, publish, autopilot, refresh). ## The pipeline ```text Site profile + taxonomy │ ▼ Headline ─────► Draft ─────► Post ─────► /api/public/.../posts (planner) (writer + (publish (your site fetches) critic) flow) ``` Every step writes to the database; the API serves directly off the latest published row. ## Pull, not push Older blog stacks **push** content into the destination (CMS → webhook → SSG rebuild → CDN). Mentionwell inverts that: your destination **pulls** on every revalidation. Three reasons: 1. **No deploy on every post.** You publish in Mentionwell and your site picks it up on its next ISR / SWR cycle. No CI run, no minutes burned. 2. **No coupling.** Your destination doesn't need a CMS adapter, plugin, or build-time content fetch. One API call replaces it. 3. **Caching is honest.** The API returns Cache-Control headers tuned for SWR. Your CDN takes care of the rest. The optional webhook (see [Webhooks](/docs/webhooks)) is for cases where you want zero-second publishing — it just nudges your destination to revalidate the relevant paths. ## What lives where | Concern | Where it lives | |---|---| | Article body (HTML, Markdown, JSON-LD, FAQ, TOC) | Mentionwell database, served via `/api/public` | | Site chrome (header, footer, layout) | Your destination repo | | URL structure (`/blog`, `/blog/[slug]`) | Your destination repo (typically) | | SEO metadata for the article | Mentionwell returns it; your destination renders it into `` | | RSS / JSON Feed / sitemap | Mentionwell serves them; you link to them from your destination | | Auth + roles | Mentionwell handles internal auth; your destination only needs the read API key | ## What Mentionwell is *not* - Not a CMS. There's no rich-text editor for non-technical users; it's an AI editorial pipeline. - Not a static site generator. Posts aren't built into your site at deploy time. - Not a CDN. The reader API has caching headers; your destination's CDN does the actual edge serving. > Once this model clicks, every weird symptom ("post isn't showing", "it's stale", "preview is fine but prod isn't") collapses to either: env var wrong, cache stale, or post not published yet. Treat those three as your first checks every time. --- Canonical URL: https://mentionwell.com/docs/concepts Live HTML version: https://mentionwell.com/docs/concepts Section: Get started Site index for AI ingestion: https://mentionwell.com/llms.txt Full reference: https://mentionwell.com/llms-full.txt