Pourquoi Archyno

Archyno is a modelling tool that runs in a browser. It draws UML 2.5.1 and ArchiMate 4, plus ER models and BPMN, and it exports to the formats an architecture team is actually asked to hand over.

The thing that makes it different from a drawing tool is one design decision: there is a metamodel, and everything reads it. The palette is generated from it, the connection rules are checked against it, the shape a given element gets is looked up in it, and the AI is constrained by it. A drawing tool with a UML stencil will happily let you draw a composition from an actor to a use case. Archyno refuses, because the metamodel says that relationship does not exist.

A diagram in Archyno is a view of a model, not a picture. The same element can appear on four diagrams, and it is one element in all four - rename it once and it is renamed everywhere it appears, because there was only ever one of it.

That is what a modelling tool is for, and it is the thing people discover they needed about six months into maintaining a folder of unrelated PNGs. A change to a component name in a drawing tool is a change in however many files happen to show it, found by memory.

The project tree holds the model; diagrams hang off it. Everything in a project shares one element pool, so the layered view, the sequence diagram and the deployment diagram are three windows onto the same system rather than three drawings that agree by accident.

Most tools export a picture. The question that decides whether a tool survives contact with an enterprise is whether it can hand the model to whatever the rest of the business runs - and in European enterprise architecture, that is very often Sparx Enterprise Architect.

The .qeaexport is the one worth understanding, because it is the one that is hard. A .qea is a SQLite database carrying Enterprise Architect's own repository schema - dozens of tables, with EA's internal identifiers and conventions. Archyno produces one by copying a blank repository EA itself created and writing the model into it, which is why the result opens as a repository rather than as something that needs importing.

Ask most AI diagram features for a class diagram and you get something diagram-shaped: boxes with plausible names, connected by whatever line the model felt like emitting. It looks right until an architect reads it.

Archyno's generation is bounded by the same metamodel the palette is. The model proposes elements and relationships; anything the notation forbids is rejected before it reaches the canvas. You get a diagram that is valid UML, which is a much smaller claim than "a correct diagram" and a much more useful one - it means the output is a starting point you edit rather than one you check line by line.

Describe a system in a paragraph, and the elements are drawn one at a time while you watch. The Free tier includes 150 credits a month, which is enough to find out whether this is useful to you before deciding anything.

Open a diagram somebody else has open and you see them: their cursor, moving live, and a marker on the element they are editing. Cursors ride a realtime channel and never touch the database, so they cost nothing and stay smooth.

The element someone is editing carries a soft lock. It is soft on purpose - it keeps two people out of the same property panel without turning the diagram into a queue.

Access is per project. A project can be shared with named people, and every read and write is checked in the database itself rather than in the application - so a link to a diagram you were never given is a 404 rather than a page that loads and hopes the client hides the right buttons.

There is no save button. Geometry is flushed after 400ms of stillness, text after 600ms of not typing, and the status bar tells you which state you are in.

The part worth stating plainly is the failure case, because that is where autosave usually lies to you. If a save fails - the connection dropped, a deploy moved the endpoint - the change is not discarded and the editor does not claim to be saved. The edited elements stay marked as unsaved, the status bar offers a Retry, and the next successful flush carries them. What you cannot do is keep working productively with no connection at all: the model lives in the database, so an extended outage means an extended Retry rather than an offline session.

Undo is 100 steps deep on every tier, including Free.

The case for one person is mostly about friction. A consultant or an engineer who needs a diagram this afternoon is choosing between a drawing tool that will let them draw something wrong, and a desktop modelling suite that needs a licence, an install and an afternoon of its own before it draws anything.

• Nothing to install, nothing to pay. The Free tier is one project, unlimited diagrams, 300 elements and the complete notation, with no card. You find out whether the tool suits you before it costs anything.
• The notation checks itself.Nobody reviews a solo modeller's work before a client sees it, so the tool being unable to draw an invalid relationship is doing the job the reviewer would have.
• Exports that end an argument. A client on Sparx gets a .qea, a developer gets Mermaid in the repository, a slide gets a PNG with no watermark on any tier. One model, whichever form the person asking happens to need.
• AI for the blank page. 150 credits a month, and what comes back is valid UML you edit rather than a picture you check.

The paid Individual tier adds projects, unlimited elements, the interop formats and code generation. It is not open yet - see the note at the end.

For a team the calculation is different, because the expensive problem is not drawing a diagram. It is that six people have drawn six diagrams of the same system, in four tools, and nobody can say which one is current.

• One model, not one file each. A project holds the elements and the diagrams are views of them. When the component is renamed, every diagram in the project is renamed - which is the difference between a model a team maintains and a folder a team stops trusting.
• Access checked in the database. A project is shared with named people, and every read and write is authorized by row-level security rather than by the application. A link to a diagram somebody was never given is a 404, not a page that loads and hopes the UI hides the right buttons.
• People can see each other work. Live cursors and a soft lock on the element being edited, so two architects in the same view are coordinating rather than overwriting - within the limits set out above.
• The Sparx handoff survives review. If the architecture board runs Enterprise Architect, the model goes to them as a repository it opens natively, which is usually the constraint that decides whether a browser tool is allowed at all.
• Where the data sits is written down. Hosting and database are in the EU, every sub-processor is named on the privacy page, and the security page states what is and is not claimed - including the certifications we do not hold.

Every tool has an edge, and the ones that do not say where it is waste your evaluation week. Ours, as of today:

The paid tiers are not open yet, so the Free tier is what there is to judge today: one project, unlimited diagrams, 300 elements, the complete notation, PNG and SVG without a watermark, and 150 AI credits a month. No card.