APMN — AI Process Model and Notation¶

APMN is an open specification for modelling AI-native business workflows. It extends BPMN 2.0 with first-class constructs for agents, RAG pipelines, confidence routing, MCP tool calls, and human-in-the-loop review — everything you need to describe a modern AI workflow in a vendor-neutral, executable way.

Why APMN?¶

Problem	APMN solution
BPMN has no concept of an AI agent	`agentTask`, `ragTask`, `memoryTask` node types
No standard way to express confidence-based routing	`confidenceGate` gateway
Human review patterns are ad-hoc	`humanInLoopTask` with timeout and escalation
MCP tool calls are invisible in diagrams	`mcpToolTask` with server + tool attributes
Process diagrams can't be compiled to real infrastructure	TwinTrack compiles APMN → Orkes Conductor, Google ADK

Quick Start¶

Open the APMN Modeller to design your workflow visually.
Export as APMN YAML.
Use TwinTrack to compile to Orkes Conductor JSON or Google ADK Python.

Specification¶

→ APMN v0.1 Specification

Schema¶

YAML Schema — JSON Schema for APMN YAML files
BPMN Extension XSD — XML extension for BPMN 2.0 tools

Examples¶

Patient Preadmission — multi-agent clinical intake workflow

Resources¶

Resource	Link
Specification	apmn.kshetra.studio/spec/v0.1
GitHub	kshetra-studio/apmn
APMN Modeller	apmn-modeler.kshetra.studio
TwinTrack (compiler)	bpmn2ai.kshetra.studio
License	Apache 2.0