An AI BPMN generator for when the diagram has to be approved.
Text, an SOP, a PDF, a Word file, a slide deck, an image or a chat description goes in. A BPMN 2.0 file with pools and lanes comes out, built for content, not just syntax: from a document, every role, step and decision is first listed with its quote, and the diagram is checked against that list. Then OpsCodex does what a generator does not: each step keeps that source, named people review and approve, and the record of who did what cannot be quietly edited afterwards. It runs in your own data centre.
Text · SOP · PDF · Word · Slides · Images · Chat → BPMN 2.0 XML
OpsCodex has an AI BPMN generator. The work starts after it.
An AI BPMN generator turns a process description into a BPMN 2.0 diagram: you paste text or upload a document, the AI proposes roles, tasks, decisions and events, and you download BPMN XML that other tools can open. OpsCodex does that, from several documents at once.
OpsCodex is an AI-assisted BPMN governance platform for industrial and regulated enterprises — AI drafts processes from existing documents, people review and approve, deployed on-premise.
Most generators end with the download. In a company the diagram then has to be checked by someone who knows the process, approved by someone accountable, and found later by the people who work from it. That is the part OpsCodex is built for.
What goes in, what comes out.
The points generator rankings compare, answered for OpsCodex.
Correct, not just valid.
Every ranking checks whether the export is valid BPMN 2.0. Valid is the easy part: a diagram can open in any modeler and still describe the process wrongly. Most of the work in OpsCodex goes into the content.
Inventory first, then the drawing
Before anything is drawn, four specialised passes list roles, steps, decisions and data, every line with its quote. A second reader without AI looks for everything that can act. Only then is the process modelled, and the model is checked against the inventory, so no step in the documents goes missing quietly.
Diagrams, tables and scans too
A separate page reader looks at every page as an image, diagrams, tables and scans included, and passes on steps, rows and columns as structure. A page that could not be read is reported as an open question instead of silently missing.
Gaps and contradictions on the table
Where the documents contradict each other, or the AI would have to assume something they do not say, it does not guess quietly: that becomes an open question, and approval stays locked until it is settled.
The same rules in every process
Who gets a lane, and how roles, departments and systems are named, is set by program code, and so are layout and BPMN XML. The Process Advisor also checks the model against modelling rules. So the review is about the process, not the drawing.
None of this replaces the review; it makes the review easier: the reviewer sees where every step comes from and what is still open. We publish no accuracy figure; a pilot on your own documents settles that. How the AI drafts →
The rankings check the file, rarely the approval.
Published rankings of the best AI BPMN generators compare valid BPMN 2.0, XML export, pools and lanes, the editor, the free plan and the entry price. Fair questions when one person needs one diagram.
Whether each step comes from the source or was assumed by the AI. Whether the author can approve their own diagram. Whether anyone can change the record later. Where the documents and prompts go. Of the four rankings we read, one names approvals as a criterion at all. In a regulated company, these questions decide whether the diagram counts.
Rankings read on 5 November 2026: justflow.it · bpmnstudio.com · bpmnai.com · lucidflow.ai
Five questions the rankings leave open.
A source for every step
Every element of the draft carries its passage in the document, one click shows it, and program code checks the quote against the text. Whatever the AI assumes without a source becomes an open question. While it is open, approval stays locked.
Author, reviewer, approver: three people
The author cannot approve their own diagram, and that is enforced, not a setting someone can switch off quietly. An exception takes an administrator and a written justification, carries its own status, and every exception can be listed.
An audit trail nobody can quietly edit
Every state change and every sign-off goes into an append-only log. The database itself rejects changes and deletions, and the database role the application runs under has those rights revoked.
In your own data centre
OpsCodex runs on-premise as an instance of its own, with Docker Compose or on your Kubernetes platform, without internet access if you need it.
A data path to the AI you choose
Your own model server on your own network, so no document leaves it, or a provider such as Anthropic, OpenAI, Google or Mistral under your own contract. By default the text of prompts and answers is not stored.
And the day after sign-off, the concierge answers employees’ questions only from approved processes, also as a browser extension next to ERP or MES. In daily work →
A valid diagram can still be wrong.
Andreas Hense of Hochschule Bonn-Rhein-Sieg, who publishes a video series on business process modelling, gave a specialised AI BPMN generator a pharmacy process: prescriptions are collected every hour and made up into packages, customers pick them up later. The diagram opened in bpmn.io and was syntactically correct. In content it was not: it put the hourly collection run in one chain with the pickup, although that run, in his words, “is a separate control flow, it’s a separate business process”. Only someone who knows the process finds that.
That applies to every AI draft, ours included. It is the argument for a review that cannot be skipped, not against AI: in OpsCodex the person who checks is never the author, and every assumption is an open question on the table before anyone signs.
Source: A. Hense, “BA Copilot AI for BPMN Modeling – Can It Replace Traditional Methods?”, YAWL User Group, 2025-10-26, 03:29 · doi:10.5281/zenodo.18364350. The tool tested, BA Copilot, has since ceased operations. His correct model of the same process: “BPMN Business Process Modeling for Beginners – Avoid Common Mistakes with a Pharmacy Example”, 2025-09-24, 04:17.
What the generators document.
AI BPMN generators that rankings and AI assistants name, as their vendors describe them on their own public pages. “Not documented” means we found nothing about it there, not that the product cannot do it.
| Tool | Input | BPMN 2.0 file | Approval workflow | Author ≠ approver | Audit trail | Hosting and AI |
|---|---|---|---|---|---|---|
| OpsCodex | Text, chat, PDF, Word, PowerPoint, Excel, OpenDocument, images | Yes, with pools and lanes | Author → reviewer → approver | Enforced; an exception needs an administrator and a written reason, and is logged | Append-only, protected by the database | Your data centre; any model you choose, also on your own hardware |
| Procevia | Text, voice, PDF, DOCX, CSV, images; AI interviews on the Team plan | From the Pro plan | Draft → review → approved, with a before/after view, from the Pro plan | Another person by default; the organisation can switch this off | Version history; an audit log is not documented | SaaS on AWS in Paris; AI by Mistral |
| bpmnai.com | Chat, pasted text, document upload | Paid plans, or for credits | “Review workflows” on Team and Enterprise | Not documented | Audit log on Enterprise | SaaS in the USA; AI by OpenAI and Anthropic |
| Just Flow It | Text, voice; PDF and photo upload on Pro | All plans, with pools and lanes | Not documented | Not documented | Version history on Pro | SaaS, data stored in the EU; AI by Anthropic and Google |
| BPMN Studio | Chat, pasted notes | All plans; one pool with lanes | Not documented | Not documented | Version history | SaaS, storage in the EU; AI by OpenAI |
| LucidFlow | PDF, Word, Excel, PowerPoint, images, text | Paid plans | Not documented | Not documented | Not documented | SaaS; AI by Google Gemini, Anthropic as fallback |
Swipe sideways to see all columns.
Sources, read on 5 November 2026. procevia.de/pricing, docs.procevia.de/…/prozess-beschreiben, docs.procevia.de/…/prozessakte, docs.procevia.de/…/team-verwalten, docs.procevia.de/…/versionsverlauf, procevia.de/trust · bpmnai.com/text-to-bpmn, bpmnai.com/release-notes, bpmnai.com/pricing, bpmnai.com/privacy · justflow.it, justflow.it/pricing, justflow.it/privacy · bpmnstudio.com, bpmnstudio.com/pricing, bpmnstudio.com/privacy · lucidflow.ai, lucidflow.ai/privacy
Something outdated or unfair? Write to [email protected]; we would rather correct this page than defend it.
One diagram, for yourself, today.
If you need a quick diagram for a workshop or your own notes, and nobody has to approve it, a self-service generator with a free tier is faster and cheaper. OpsCodex is worth it when the process has to hold up in an audit, process documents may not go to a public cloud, and the people doing the work should find the approved version, not the latest download.
Questions about AI BPMN generators.
What is the best AI BPMN generator for regulated industries?
It depends on what happens to the diagram. If it has to be approved, look for four things: every step traceable to its source, author, reviewer and approver enforced as different people, an audit trail nobody can quietly edit, and operation in your own data centre with a model you choose. OpsCodex is built for that case. For a single diagram that nobody has to approve, a self-service generator is enough.
Can AI generate BPMN from a PDF or Word document?
Yes. OpsCodex reads PDF, Word, PowerPoint, Excel and OpenDocument files, plain text, images and scans, several at once, and drafts a BPMN 2.0 model with pools and lanes. Every element links to its passage in the document, and whatever the documents do not say becomes an open question.
How do I turn text into BPMN 2.0?
Paste the description into a new process, or drop it in as a text or Word file: OpsCodex first lists roles, steps, decisions and data, each with its quote, then models the process and lays it out. Or describe the process in the AI chat and build the model step by step. Either way you get BPMN 2.0 XML that any conformant BPMN tool can open, and SVG, PNG, PDF or draw.io if you need them.
Is AI-generated BPMN reliable?
Not without a review. A model can be valid BPMN and still describe the process wrongly. OpsCodex makes the review checkable: every element drafted from a document shows its source, assumptions become open questions that lock approval, and a named reviewer who is not the author signs off. We publish no accuracy figure for the draft; a pilot on your own documents settles that.
Can an AI BPMN generator run on-premise?
OpsCodex does. It runs as an instance of its own in your data centre, with Docker Compose or on Kubernetes, and works with an open-weight model on your own hardware, proven with models such as Gemma, gpt-oss and Mistral. Then no document leaves your network.
Is there a free version of OpsCodex?
There is no public free tier. You can see OpsCodex on one of your own documents in a one-hour demo, and then run a pilot in your own environment. Write to [email protected].
See OpsCodex on one of your own processes.
Bring a PDF, a slide deck or a messy SOP, or use our sample data. In one hour we run the process end-to-end — draft, approval, export — and the concierge answers questions on it. Afterwards you get the process and an analysis report.
“In our consulting projects we spent weeks working through interviews, and in the workshop everyone still described the same process differently. Gaps and overlaps surfaced late, and the approved version vanished into a repository. We built OpsCodex for the whole road: from the draft through optimisation to the work that follows.”
Tillmann Schatz · Product owner OpsCodex, Senior Manager at Mews PartnersOpsCodex is built by a team at Mews Partners.