Repository · thirty-one canvases · AI agents
Here, a wrong model does not get published.
Even when an AI wrote it.
Cartano reviews every model as you draw: 191 good-practice rules that advise as you go, and only hold back publishing for a real error. Our AI agents go through the same review: they draft, correct themselves, and hand you a clean draft — with the count of rounds.
Free trial from 1 November, no credit card. Plans from €12 excl. VAT a year, available to subscribe now.
Who it's for
One tool, four ways to use it. Start with yours.
“Half our days go into redrawing diagrams that are wrong before the final presentation.”
“Our map exists in four versions, and nobody knows which one is right.”
“Nothing links a table to the process that fills it.”
“Thirty sets of diagrams, and one evening to mark them.”
The agent workshop
A conversation produces a plausible diagram. A Cartano agent produces a model that passes the rules — and when it cannot, it tells you which ones.
It asks its questions and restates the need. You validate the thesis before it produces anything.
It searches your repository, reuses your existing objects and writes a draft.
The 191 rules run on the draft, as on any model in the space.
It starts again until zero errors — or stops and says what it could not satisfy.
Accept, rework or reject. Nothing enters the repository without you.
A single truth
No architecture suite goes down to the data contract; no data tool goes up to the business capability. Cartano holds the whole chain, on one repository.
An application, a database, an actor exist once. Canvases are views of them: you place an existing object, you do not copy it.
Every object is described by a finer diagram — a process by its sub-processes, a system by its containers. The breadcrumb brings you back.
Activities × data, activities × applications: the matrices flag data read but never produced, an activity with no tool.
Thirty-one canvases
Not a free drawing tool: each canvas knows its standard, and its rules guide you through it.
Entities and attributes, Merise or Codd, generated DDL, published data contract.
Lanes, activities, data read and produced, applications used, XML export.
Three-layer map, IS urbanism plan, data flows between blocks.
Motivation, capabilities, value stream, technology, organisation, roadmap, free canvas.
Classes, use cases, components, packages, sequence, states — and the generated code.
Each level zooms into the previous one; Structurizr DSL and C4-PlantUML export.
Nine canvases that draw nothing: they decide. A computed residual risk, an expired proof, a deprecation with no date, a missed notice period — each sets a verdict the organisation must answer, and blocks publishing until it does.
The product, not mock-ups
Screenshots of Cartano as it runs online, on insurance and fleet management examples (the demo data is in French).
Open format
Your current drawings do not go in the bin, and you leave whenever you want, with everything.
Each page of a draw.io file becomes a document, positions, sizes and colours kept to within half a pixel. The report lists everything that did not make it.
A Git mirror of every document, the whole space export, and the formats your teams already read. Export stays free, on every plan.
Governance and hosting
A Cartano agent inherits the safeguards of the interface. And your data stays where your CISO wants it.
Every action undoes step by step, cascades included. An undo erases nothing: it writes a compensating action.
Who changed what, when, with the before and after. The log is append-only — nobody rewrites it, us included.
An agent publishes nothing and creates no type. Not an instruction in a prompt: it is refused by the database.
Each space is isolated at the database level itself. Single sign-on, roles enforced on every write.
Shared SaaS, dedicated instance, or your own Postgres database. Daily backup, restore replayed every week.
A switch truly turns the agent workshop off in a space: no call goes out. And never any training on your data.
What we do not do
No connector to a CMDB or a cloud provider: the repository fills through import or by hand. Our job is a correct model you can stand behind, not an inventory that updates itself.
No shallow industry library. You start from your taxonomy, which the metamodel lets you extend without development.
No simultaneous editing. An architecture model is decided by three people; what matters next is that the repository tells the truth and the log says who changed it.
Where Cartano stands
Everything described on this page works online. We use it to map our own system.
Subscriptions open now, free 30-day trial from 1 November. Online, backed up, monitored: Thirty-one canvases, checks, agents, full export, SSO.
Subscribe online, with no order form. Readers stay free, and so does export.
A written service commitment: availability, recovery time, published maintenance windows.
A wrong model that refuses to publish, an agent that corrects itself, the full export. No slide deck.
CARTANO