AI diagram generation

Generate a BPMN diagram from a natural-language description, and use the modify and analyze actions to refine an existing diagram with AI.

With AI generation you describe a business process in plain English and Bonita Process Designer builds a complete BPMN diagram for you on the canvas. Two companion actions let you keep working with AI once a diagram exists: Modify applies a natural-language change to the current diagram, and Analyze returns a written analysis without touching the canvas.

AI generation requires the Creator or Admin role. Readers cannot generate, modify, or analyze diagrams.

Prerequisites

Before the AI features are available, the installation must be set up by an Admin:

  • The installation is registered (company name supplied during first-time setup).

  • AI generation is enabled and an AI provider is configured.

If either condition is not met, the AI affordances in the editor are disabled. When you have access but generation is unavailable for the installation, contact your Admin. See First-time setup and Application settings.

Generate a diagram from text

When you create a new diagram, a first-time choice appears on the empty canvas:

  • Start from scratch — edit the diagram manually.

  • Generate with AI — open the AI generation dialog.

Choose Generate with AI, then:

  1. Type a description of the process you want. A live character counter shows how many characters remain. The maximum is configurable by an Admin (default 500 characters; range 200–2000). English prompts only.

  2. Optionally pick an AI provider for this generation (see Choosing an AI provider).

  3. Select Generate diagram.

The AI panel with a description and the character counter

A clear, structured description gives the best result. For example:

A simple approval process: the user submits a request, the manager reviews it.
If approved, it goes to execution. If rejected, it returns to the user for revision.

How generation runs (async job + progress)

Generation runs as a background job, so it survives a closed tab, a network drop, or a proxy that cuts the connection. The diagram is still produced and delivered to you when you return.

After you submit, a progress overlay opens and the bar advances through honest, phase-based steps:

Phase What is happening

Queued

Submitted, waiting for a worker slot.

Planning

The structure of the process is being planned.

Generating

The diagram is being produced (the longest phase).

Validating

The generated BPMN is being checked.

Finalizing

Layout is applied and the result is finished.

When an estimated time is known, the bar fills smoothly toward the next phase; otherwise it shows an indeterminate pulse rather than a fabricated percentage. Select Continue working to dismiss the overlay without cancelling — the job keeps running and the result is imported when it completes.

When the job succeeds, the diagram is imported onto the canvas, auto-zoomed to fit, and saved automatically. Unsupported BPMN elements are converted or removed on import, with a migration log if any changes were made — see Supported BPMN elements.

If you close or reload the editor while a job is in flight, Bonita Process Designer reattaches to it on the next load: a running job resumes its progress overlay, a job that finished while you were away is imported immediately, and an expired result shows an honest warning.

Modify and analyze an existing diagram

The AI dialog has Generate, Modify, and Analyze tabs. Unlike generation, Modify and Analyze run inline — while they work, an elapsed-seconds counter is shown.

  • Modify — describe a change (for example, "Add a manager approval step after the first task"). The returned BPMN replaces the current canvas. Modify counts against the AI quota, like generation.

  • Analyze — ask a question about the current diagram (for example, "What are the bottleneck risks?"), or leave it blank for a general description. The answer is returned as text only; the canvas is untouched. Analyze is read-only and does not count against the quota.

Choosing an AI provider

When more than one AI provider is configured for the installation, the generation dialog offers a provider picker. Selecting a provider overrides the Admin default for that one generation only — it does not change the installation default and is not remembered for the next generation. If only one provider is configured, no picker is shown.

AI generation quota

Each group can have a cap on the number of AI generations it may run over a rolling 30-day window. The cap is set per group by an Admin and is shared by everyone in the group, regardless of role — an Admin generating on a group’s diagram is counted and capped like anyone else.

  • Generate and Modify count against the quota.

  • Analyze does not (it is read-only).

The window is rolling: there is no scheduled reset. Older generations simply age out of the 30-day window, freeing capacity again.

When a group is at its limit, the request is rejected immediately — before any AI work starts, so it never consumes provider tokens. You see a warning toast on the Generate flow, or an inline error on the Modify tab, naming the group and its usage (for example, 100/100 in the last 30 days). Wait for usage to age out of the window, or ask an Admin to raise the limit. A group with no cap set is unlimited; a cap of 0 blocks AI generation for that group entirely.

For setting and monitoring quotas, see Manage AI quota.

Limitations

  • English-language prompts only.

  • Prompt length is limited (200–2000 characters, default 500).

  • Generation typically takes 20–40 seconds; a job that runs too long is stopped and can be retried.

  • A succeeded result is retained for a limited time; if you return long after it finished, you may need to generate again.

  • Very complex processes may not generate accurately — review and refine the result.

  • The AI service must be reachable from the installation.

Related pages