Vibe Modeling (AI Assistant)

Vibe modeling is the conversational way to build models in the BESSER Web Modeling Editor: instead of dragging every element onto the canvas by hand, you describe what you want in plain language and the modeling agent creates and edits the diagram for you. You stay in control – the agent proposes changes, applies them to the live canvas, and you keep refining by chatting.

The assistant is powered by the modeling agent, a separate BESSER service the editor connects to over a WebSocket. On the public editor at editor.besser-pearl.org the agent is hosted for you; for a local deployment you point the editor at your own instance (see Running it locally).

Opening the assistant

The assistant appears as a floating chat button in the bottom-right corner of any editor page. Click it to open the chat panel. A colored dot next to the title shows the connection status:

  • green – connected and ready

  • amber – connecting

  • red – disconnected (the modeling-agent service is unreachable)

Type a request, press Enter, and watch the agent apply the result to the canvas. Suggested quick actions (e.g. create, generate code, export) appear as one-click chips to speed up common follow-ups.

What you can ask it to do

The modeling agent understands natural-language modeling requests such as:

  • Create a model from scratch – “Model a library with books, authors and members, where a member can borrow many books.”

  • Modify the current diagram – “Add an email attribute to Member and make it the identifier.”, “Rename Book to Publication.”, “Delete the Loan class.”

  • Add OCL constraints in plain language – “Add a constraint that a member cannot borrow more than 5 books.” The agent authors the B-OCL and attaches a constraint box to the target class. (OCL is only emitted when you explicitly ask for it.)

  • Trigger code generation – ask it to generate, and it launches the same generation pipeline as the Generate menu.

Supported diagram types

The modeling agent can create and edit these diagram types:

  • Class diagrams (including OCL constraints)

  • Object diagrams

  • State machine diagrams

  • Agent diagrams

  • GUI (no-code) diagrams

  • Quantum circuit diagrams

  • BPMN diagrams

Diagrams are laid out by a deterministic layout engine so results are stable and readable, not scattered across the canvas.

Voice and file input

Beyond typing, you can:

  • Speak your request – record a voice message in the browser and the agent transcribes and acts on it.

  • Attach a file (up to 10 MB) – for example a screenshot, a hand-drawn sketch, or an existing model to convert into a B-UML diagram.

Running it locally

The assistant is optional: the editor works without it, but the chat widget only connects when a modeling-agent service is reachable. When running the editor from source, point it at your agent instance with the UML_BOT_WS_URL environment variable (default ws://localhost:8765). See Environment Variables for the full variable list and the modeling agent repository for how to run the service.

Note

Code generation and validation triggered from the assistant still rely on the BESSER backend at http://localhost:9000/besser_api. Start it alongside the editor and the modeling agent for the full experience.