What gets generated
From a workspace, CelinQ's documentation generator produces a complete static site: a landing page, full-text search, a page per package, and a page per diagram with the diagram itself rendered inline. Every page links back to the live model objects it describes, so a reader can navigate the architecture the way they would navigate the model itself — by following relationships, not by hunting through a document's table of contents.
Landing page & search
An entry point into the workspace, with full-text search across everything the profile includes.
Package & diagram pages
One page per package, one per diagram — diagrams rendered inline, not left as a static image export that goes stale silently.
Scoped publishing
Publish the whole workspace, or a profile scoped to selected packages or diagrams — exclusions win, so an excluded package never leaks into pages, search, or exported assets.
Where it can be published
A publish profile targets one of four real destinations: a local folder, a zip file, a SharePoint document library, or an SFTP server. Credentials for SharePoint and SFTP are stored the same way an AI provider's API key is — never in plaintext, never returned once set. A profile can run on demand, or on a schedule (fixed interval, daily, or weekly), checked once a minute by the server so a restart resumes the same cadence rather than firing immediately.
Continuous documentation: knowing when it's out of date
Every generated publication already records the exact workspace revision it was generated from. As of this year, that fact is surfaced directly: CelinQ compares a past publication's revision against the workspace's current one and reports whether it is stale, and by how many revisions. A publication generated an hour ago against a model that has not changed since is not stale. One generated last week, with twelve changes landed since, is — and CelinQ can now tell you that without regenerating it first just to find out.
Knowing something is stale is only half the question — the other half is what actually changed. A separate, deterministic Change Intelligence endpoint answers that directly: point it at the revision a stale publication was generated from, and it returns exactly how many elements were created, updated or deleted since, broken down by kind, plus the specific entities affected. No AI involved, no summarisation — the same real change records that already drive sync and Fusion, counted and listed.
What this is, and is not, today. Detecting staleness and listing exactly what changed since are both real and shipped. What is not yet built is a one-click regenerate wired to that answer, or automatic regeneration triggered by specific revisions — today, acting on a stale result is calling the same publish action again by hand, now with a precise reason to.
What generates the content
Page content — package summaries, element listings, diagram context — is produced deterministically from the model: the same workspace state always produces the same documentation, with no dependency on an AI provider being configured, reachable, or even enabled. Sovereign, AI-disabled deployments get the full documentation capability with nothing missing. Where an administrator has enabled AI assistance, it can enrich generated prose — but only ever on top of the deterministic facts already computed from the model, never inventing a fact the model itself does not contain.
What's not built yet
| PDF / DOCX export | Not built. Output today is HTML, inline SVG diagrams and JSON — no PDF or Word document generation. |
|---|---|
| Reusable documentation templates | Not built. A publish profile scopes what gets published (packages, diagrams); it does not yet let a team define a reusable report structure (executive summary, scope, risks, appendix) independent of the default page layout. |
| One-click regenerate on staleness | Detection and change listing are both real (see above); a UI action that regenerates a stale publication directly is not yet built — today it's the same publish action, run again. |
| Live, always-current viewer | The portal is generated — on demand or on a schedule — not continuously served from the live workspace. A viewer that never needs regenerating is a different, larger capability not built yet. |
Read next
Generate Architecture Documentation with CelinQ
Why documentation generated on demand beats a document someone forgot to update.
Turning a Repository into a Searchable Knowledge Base
How the same graph and revision history that power sync and Fusion double as the foundation for search.