From scattered documents to an approved process.
AI does the heavy lifting on the draft. Governance is built into the path — not bolted on after.
Process knowledge is scattered. Turning it into trustworthy BPMN is slow.
Knowledge is scattered
Process knowledge lives in PDFs, slide decks, screenshots and SOPs — never in one modellable place.
Modelling is slow and personal
Hand-modelling each process takes days, and the result depends on who drew it.
Notation drifts
Diagrams differ between teams, so they aren’t comparable — or reusable.
The approval trail is fuzzy
When an auditor asks who approved this, and when? — nobody can answer with confidence.
Four steps from raw document to approved, exportable process.
Ingest
Drop in the documents you already have — PDFs, slides, screenshots, SOPs. OpsCodex reads them in.
Draft
The AI proposes a BPMN draft with surgical, editable suggestions. You refine the model directly, with every element editable. Naming templates and shared notation keep every draft consistent across teams.
Review & approve
An assigned reviewer and approver sign off — separation of duties is enforced, with an audit log at every step.
Export
Approved processes export as standards-conformant BPMN XML, with the approval on record.
Around 70 percent less time to a model ready for review.
We modelled several processes once by hand and once with OpsCodex and measured the time, from the document to a model ready for review, corrections included.
The better evidence is your own process: in the demo we are happy to time it together with you.
Not one call to one big model.
OpsCodex breaks the draft into small steps you can check. Program code sets the order; the models read and propose. That is why an open-weight model on your own hardware gets you a draft you can check, too.
The six points as text
Read to fit the model
When a model is set up, OpsCodex detects how much it can take at once and whether it reads images. Every document is then split into portions that fit. A page that could not be read is reported as an open question instead of silently missing.
Diagrams and tables
A separate page reader looks at the pages as images, diagrams, tables and scans included, and passes on steps, rows and columns as structure. That reader can be a small open model; the drafting model needs no image input itself.
Inventory first
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.
Sourced, or an open question
Every element links to its passage in the document, and program code checks the quotes against the text. Whatever the AI assumes without a source becomes an open question, and approval stays locked while it is open. Where the material contradicts itself, that becomes an open question too.
Repeatable, as if by one hand
Who gets a lane, and how roles, departments and systems are named, is set by program code, and so are layout and BPMN XML. Two runs over the same material usually differ only in details, such as whether someone who only receives an email gets a lane of their own. Both results are clean, valid models. Two people model the same process with a different granularity, different terms and a different style. Only processes modelled alike can be compared and maintained together.
No frontier model needed
Every step gets a narrow, strong context: the right passage in the document, the inventory, the rules. Much is decided deterministically by program code. That is why OpsCodex needs no frontier model in any step, proven with open-weight models such as Gemma, gpt-oss and Mistral.
And after the draft: the AI compares processes, for example those of two sites, proposes a common version and points out gaps and overlaps between your processes.
The notation isn’t old. It’s settled.
BPMN 2.0 is maintained by the OMG and published as ISO/IEC 19510:2013 — a standard, not a vendor format. It stopped changing because it works. That stability is what lets your models outlive the tools that drew them. Far from fading, it has become the notation the agent era orchestrates on.
An OMG standard, ratified by ISO
Published by the OMG and standardised as ISO/IEC 19510:2013. The XML OpsCodex exports is readable by any conformant tool — no format lock-in, by design.
What agents run on
UiPath models processes in BPMN 2.0 in Maestro, its agentic orchestration product; Camunda, Flowable and Trisotech orchestrate their AI agents on it too.
Backed where it counts
Since October 2025, Fidelity, NatWest, Deutsche Bank and Capital One jointly back an open-source BPMN orchestration platform under Linux Foundation governance ↗ — regulated industries are betting on the standard, not away from it.
OpsCodex governs those models; execution engines run them. Same notation, different job.
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.