How chat refinement works
The AI maintains full awareness of the current state of your diagram. When you type an instruction, it reads the existing elements, connections, pools, and lanes, then applies the minimum change needed to satisfy your request. Unrelated parts of the diagram stay exactly as they are.Open the chat panel
With a diagram open in the editor, locate the Chat panel on the right side of the screen. If it’s collapsed, click the chat icon in the right toolbar to expand it.
Type your instruction
Describe the change you want in plain English. Reference element names or positions to help the AI identify exactly what to modify (see tips below).
Press Enter or click Send
The AI processes your instruction and updates the diagram immediately. The changed elements are briefly highlighted so you can spot what moved.
The AI edits the existing diagram rather than regenerating it from scratch. Every element you’ve already added or adjusted — whether by AI or manually — is preserved unless your instruction explicitly asks for it to change.
Example instructions
The following examples show the range of changes you can request in a single chat message. Copy and adapt them to match the element names in your own diagram.More example instructions to try
More example instructions to try
Tips for effective instructions
Following a few simple guidelines helps the AI understand your intent and produce accurate results on the first try.Be specific about element names
Reference elements by their exact label when possible. “Add a gateway after ‘Check Credit Score’” is clearer than “add a gateway in the middle.” If you’re not sure of the exact name, a close approximation still works well.
Describe what you want, not how to implement it
Say “add an exception path for rejected documents” rather than “insert an exclusive gateway with a condition expression set to rejected equals true.” The AI handles the BPMN mechanics — you focus on the business logic.
One meaningful change per message
For complex restructuring, break your changes into a series of focused instructions. This makes it easier to review each result and course-correct if needed, rather than untangling a large change all at once.
Use follow-up messages to correct
If the AI’s interpretation isn’t what you expected, don’t undo and start over — just describe the correction in your next message. The AI will adjust only what needs to change.