Open standard for machine-executable enterprise processes.

UAPF (Unified Algorithmic Process Format) is an open, machine-readable standard for expressing enterprise workflows, decision logic, roles, exceptions, and multi-agent execution semantics. It enables organizations to convert human workflows and undocumented operational knowledge into explicit, portable, AI-executable algorithms—the universal algorithmization layer for modern AI-driven enterprises.

UAPF was created in January 2026 while a group of industry practitioners was preparing a response to the Monetary Authority of Singapore's public consultation on AI Risk Management Guidelines. Drafting the answers exposed a gap existing methodologies and standards did not close: AI governance frameworks described the model well, but had no shared, machine-readable way to describe the process the model runs inside, the decisions it makes, and the cases it escalates to humans. Instead of inventing a new notation, we compiled the proven OMG cornerstones already trusted in regulated industries — BPMN 2.0, DMN 1.3, and CMMN 1.1 — into a single versioned, agent-readable package format. The first version of UAPF was published with the MAS submission.

Enterprises increasingly deploy AI agents to execute business processes, but:

• Most processes are described only in text.
• Many operational rules live only in employees’ heads.
• Process knowledge disappears when employees change roles.
• Every AI agent platform uses different internal structures.

UAPF solves this by providing:

• A universal import/export format.
• Alignment with established international standards (OMG BPMN/DMN/CMMN).
• AI-native capabilities (MCP, A2A).
• Version-controllable, machine-executable models.
• Interoperability across agents, orchestrators, and institutions.

• No new modelling language: Reuses existing ISO/OMG standards.
• Agent-native: Models include agent roles, capabilities, MCP tools, and A2A messaging schemas.
• Interoperable: Portable across modelling tools, automation engines, and AI runtimes.
• Composable: BPMN + DMN + CMMN + MCP + A2A → one algorithmic package.
• Governable: Supports metadata, provenance, auditability, and compliance.

A UAPF package is a zip-based container:

<process>.uapf/
  ├── manifest.json
  ├── process/
  │     ├── bpmn.xml
  │     ├── cmmn.xml
  ├── decisions/
  │     ├── *.dmn
  │     └── glossary.json
  ├── agents/
  │     ├── roles.json
  │     ├── capabilities.json
  │     └── bindings.json
  ├── integration/
  │     ├── mcp-tools.json
  │     ├── a2a-schemas.json
  │     └── api-endpoints.json
  └── metadata/
        ├── version.json
        ├── provenance.json
        └── compliance.json

Each file has a formal JSON Schema defined under /schemas.

UAPF aligns with globally adopted standards:

Schemas located in /schemas validate UAPF packages and ensure cross-platform compatibility, including:

• uapf-manifest.schema.json
• uapf-roles.schema.json
• uapf-capabilities.schema.json
• uapf-bindings.schema.json
• uapf-mcp-tools.schema.json
• uapf-a2a-schemas.schema.json

• Document intake → routing → approvals → archival.
• Inter-institution coordination.
• Automatic classification via DMN.
• Multi-agent distribution using A2A.

• Iteration workflows.
• Design → analysis → implementation flows.
• Agent-based code generation, testing, reporting.

• Procurement.
• Compliance workflows.
• Customer service logic.
• Risk evaluation.

An AI orchestrator (e.g., LangGraph, CrewAI, AutoGen) loads UAPF as:

1. Graph (from BPMN/CMMN).
2. Ruleset (from DMN).
3. Agent mapping (roles → tasks).
4. Tool interface via MCP.
5. Multi-agent coordination via A2A.

This enables full organizational operations to be executed by agents.

• Finalize schema versioning.
• Add canonical examples.
• Publish reference parser (Python/Node.js).
• Publish UAPF Playground (web viewer).
• Integrate Veritrust decentralized identity.

• Formal BPMN 3.0 agentic extensions.
• Simulation engine.
• Compliance and safety modules.
• EU AI Act alignment.

MIT License — open standard, free to use, free to implement.

Pull requests, issues, and proposals are welcome. Standardization direction is coordinated under the Neksus AI / UAPF working group.