Frequently asked questions.

OpsCodex is an AI-assisted BPMN governance platform for industrial and regulated enterprises. AI drafts BPMN 2.0 processes from the documents you already have — PDFs, slide decks, SOPs — and named reviewers and approvers sign off before anything reaches your teams. It runs on-premise, inside your own environment.

Yes — on-premise is the primary deployment model. OpsCodex runs inside your perimeter, and your documents and models stay in your tenant. AI inference goes where you configure it: an external provider under your own contract, or a private endpoint inside your perimeter. There is no multi-tenant cloud dependency.

OpsCodex models processes in BPMN 2.0 and exports standards-conformant BPMN XML as specified by the OMG. Approved processes stay portable: the exported XML is readable by any conformant tool, and layout fidelity is worth testing against your own target tools during a pilot.

Classic BPM suites focus on modelling repositories or workflow automation. OpsCodex focuses on the governance gap before and after: AI drafts the model from your existing documents, an enforced author–reviewer–approver split decides what becomes official, and approved processes feed daily work as searchable answers and generated work instructions. It complements execution engines rather than replacing them.

No. AI output is always a proposal. Every process passes a mandatory six-eyes workflow — author, reviewer and approver are different people — and nothing is published or exported without a human sign-off, recorded in an append-only audit log.

Process owners, quality managers and operational-excellence teams in industrial and regulated enterprises — organisations that need audit-proof process documentation: who approved what, when, and in which version.

OpsCodex is built by a team at Mews Partners, a consultancy advising industrial and regulated enterprises on PLM and process governance since 1992, with around 350 consultants across seven offices. The product grew out of that practice, which is why its governance model mirrors how approvals actually run in industrial organisations rather than imposing a generic workflow. Product ownership sits with Tillmann Schatz, Senior Manager at Mews Partners. Demo requests are answered directly by the OpsCodex team rather than routed through a sales process.

It is the default, and it cannot be bypassed silently — but it is not a straitjacket. Administrators can enable an exception path for cases that genuinely need one. Even then the review still has to be completed on the current version, the administrator must enter a written justification, the exception is recorded as its own decision in the audit trail, and the process carries a distinct “exception approved” status rather than being indistinguishable from a normally approved one. Auditors can therefore list every exception ever granted and read the reason for each.

No. BPMN 2.0 stopped changing because it was finished, not because it was abandoned. BPMN 2.0 is maintained by the OMG and published as ISO/IEC 19510:2013, and has stayed deliberately stable, so a model drawn today is still readable by any conformant tool years later. It is also what the current generation of AI orchestration runs on: UiPath models processes in BPMN 2.0 in Maestro, its agentic orchestration product; Camunda and Flowable orchestrate AI agents on it; and in October 2025 Fidelity, NatWest, Deutsche Bank and Capital One launched an open-source BPMN orchestration platform under Linux Foundation governance. A notation that stops changing is what makes a process base portable.

Yes. That is the point of keeping the approved version in one place. OpsCodex publishes each approved process as a frozen release snapshot, and that snapshot is the single source everything else reads: the employee portal, and a read-only interface for AI assistants over the Model Context Protocol. Process logic therefore lives in a governed layer of its own instead of being locked inside the modelling tool, the same argument data teams make for a semantic layer, applied to processes. Consumers get more than text: every answer carries its governance envelope (revision, approver, release date, review status and a link back to the process), so an assistant can cite rather than claim. Access is read-only by construction: the interface has no write operations at all, and drafts and work-in-progress revisions are not in the published snapshot in the first place, so there is nothing to filter out. It is off by default, opened per token, and every access is logged.

Not as-is — by design. OpsCodex exports the approved business-level process (descriptive BPMN in the sense of the BPMN 2.0 conformance classes); an engine imports it as the starting structure, and your automation team adds what execution needs: service-task implementations, gateway conditions as evaluable expressions, process variables and data mappings, forms and task assignments, message, timer and error definitions, and engine-specific extensions. Those details live in the engine project — OpsCodex makes no round-trip promise and never overwrites them. The division of labour is deliberate: the approved process in OpsCodex stays the single source of truth for what should happen, the engine model implements how, and a newly approved version reaches the automation team as a governed change — revision, approver and release date on record — not as a file that silently replaces their work. What you gain is that automation starts from a reviewed, signed-off process instead of an outdated diagram — and AI agents can read the approved version directly over the read-only MCP interface.