AI Control Hub
From prompt to a running, tested process
A database-driven application is built in dialogue today: you describe what you need and receive a clean relational model you can afterwards inspect in any tool. For data applications, vibe coding has long worked. For business processes it does not – BPMN 2.0 stands in between, a standard from another era, without a successor, with lanes, gateways and symbols that require training courses.
The AI Control Hub closes that gap. It is the MCP server of our platform – and it produces not just a diagram, but a process that actually runs afterwards. Vibe coding, but for executable business processes. And it does not stop at building: the AI then plays the process through itself – as every participating role, with deadlines actually fired and a send log – before handing it over.
Other AI draws the process.
Ours builds it – and tests it.
What a prompt produces
Tap or hover the components
- Prompt
-
- Process model
- Forms
- Logic
- Actions
- Ready to run
The process model
Fully modelled in Workflow Studio – with steps, branches and responsibilities. Not a draft to be reworked, but a model the engine can execute.
You watch it take shape live
Most AI integrations stop at the backend: the model talks to the system via MCP, and at some point a result appears. Ours goes one step further – into the interface you are working in anyway.
Through a WebSocket connection into Workflow Studio the diagram appears while you are prompting. You do not review a result afterwards – you follow it being built and step in when something goes the wrong way.
The AI tests what it has built
Producing a diagram is one thing. Knowing whether the process actually runs is another. That is why the AI Control Hub does not stop at building: the AI starts the generated process, logs in as every participating role in turn, works through it via the normal user interface and then reads back what really happened – not on screen, but in the stored runtime state.
- As every role Test accounts for every group are provisioned automatically. The applicant sees no processing action, the department sees its approval button – and if it is missing even though inbox and notification are correct, that is caught.
- Deadlines actually fired Reminders, escalations and expiry are really triggered and measured in the test run, not skipped. You see when the reminder went out and to whom.
- Every notification on record Every e-mail sent is recorded in the send log of the case: recipient group, time, template. Evidence instead of eyeballing.
- Errors no user interface shows A case can sit in the right inbox, notify the right group and still have skipped a required stage. Because the AI reads the state rather than just the display, it finds exactly this class of error – in its own work.
- Assumptions stay visible Whatever the source material does not settle – group boundaries, responsibilities, deadline lengths – is marked as an assumption in the process, with a warning sign in the node name. Reconciling it remains coordination work, not development work.
A person initiates, follows along and decides on the assumptions. What the AI delivers is a process that has been played through before you see it – with a runtime transcript instead of a screenshot.
We turned core components of our platform into MCP servers back in 2025 – at a time when the Model Context Protocol was only just emerging. MCP is widely established today, and other vendors expose their diagram editors through it as well, mostly at the BPMN level. Our difference sits one layer deeper: not just having something modelled, but producing a process that can be executed – and playing it through before it is handed over.
And because it can be put differently
We once packed the same thought into 78 seconds – about kickoff workshops, 200-page specifications and wall-sized BPMN diagrams.
Key Features
Standardized AI Integration via MCP
By implementing as an MCP server, the AI Control Hub offers a future-proof, open interface. Connect your BPM environment seamlessly with a variety of Large Language Models (LLMs) such as ChatGPT (OpenAI), Claude (Anthropic), Gemini (Google), or internal company models - securely and standardized.
Create processes by chat
Describe the desired workflow in dialogue and a complete process emerges: a diagram with actions, low-code logic and matching forms – either visually via the Visual Workflow Builder or in BPMN 2.0 via the Advanced Process Designer. The result is not a draft to be retyped, but directly executable.
Played through before you see it
The AI tests the generated process itself: with test accounts for every role, deadlines actually fired and a send log. Permission errors and skipped stages surface there, not in production. What comes back from the test run is a runtime transcript – not a screenshot.
AI-assisted Form Creation
Define required data fields and structures for your process steps simply in dialogue. The AI Control Hub translates your requirements via AI directly into fully configured forms within the Form Builder. This saves time and ensures consistency.
Full Control and Traceability
Although you use the power of AI, you always maintain control. The AI Control Hub acts as a controllable interface. All AI-initiated changes or suggestions are traceable within your BPM environment and subject to your governance policies.
Flexibility and Independence
Thanks to the open MCP standard, you are not tied to a specific AI provider. Choose the model that best suits your requirements and switch if necessary without having to change your core integration.