Skip to content

Workflow engines run business processes (approvals, cases, onboarding, orders) from a diagram or definition, keeping state, waiting for people and calling services. This is process orchestration, a cousin of container orchestration (Orchestration Platforms). Camunda use cases are in Common Camunda Use Cases.

Tools at a glance

Tool Type Notes
Camunda 8 BPMN engine (Zeebe) + tools Modeler, Operate, Tasklist, Optimize, Connectors; SaaS or self-managed
SpiffWorkflow / SpiffArena Python BPMN engine and low-code platform Citizen-developer friendly, open source
n8n Visual automation (nodes and triggers) Great for integrations and webhooks
Temporal Code-first durable workflows Workflows written as code, strong retries
Apache Airflow / Prefect Data pipeline schedulers Jobs and DAGs rather than human processes
AWS Step Functions Managed state machines See 6. AWS Developer Services and Serverless
Azure Logic Apps / Durable Functions Managed workflows See 7. Azure Application Development
Django Viewflow Workflow library for Django Used in the Django tests

BPMN in a Nutshell

BPMN 2.0 (Business Process Model and Notation) is a standard diagram language that is also executable.

flowchart LR
    S((Start)) --> A[User task: submit application]
    A --> B{Gateway: complete?}
    B -->|No| C[User task: request more info]
    C --> A
    B -->|Yes| D[Service task: call verification API]
    D --> E[User task: officer approves]
    E --> F((End))
Element Meaning
Event (circle) Start, end, timer, message, error
Task (rounded box) User task (a person), service task (code or API), script, business rule
Gateway (diamond) Decision (exclusive), parallel split / join, event-based
Sequence flow Order of steps
Pool / lane Participants or roles
Subprocess A process inside a process
Boundary event Timer or error attached to a task (e.g. escalate after 3 days)

Process variables carry data; DMN tables hold decision rules readable by business users; FEEL is the expression language in Camunda 8.

Camunda 8

flowchart LR
    M[Web Modeler / Desktop Modeler] -->|deploy BPMN| Z[Zeebe engine]
    Z --> W[Job workers: your code]
    Z --> T[Tasklist: human tasks]
    Z --> O[Operate: monitor instances]
    Z --> C[Connectors: REST, Slack, email...]
  • Zeebe is the cloud-native engine; it schedules work and job workers (your services) poll for jobs, do the work and complete them
  • Operate shows running and failed process instances; Tasklist is the inbox for human tasks
  • Connectors call REST APIs and SaaS without writing a worker
  • Try it with the Camunda 8 Code Studio (Camunda 8 Code Studio Tutorial)

SpiffWorkflow and SpiffArena

  • SpiffWorkflow is the Python library that executes BPMN
  • SpiffArena is the platform around it: a BPMN.js web editor, process groups, forms, message correlation, service connectors and monitoring
  • Runs from Docker or locally (Python with uv, MySQL / PostgreSQL / SQLite for development)
  • Aimed at citizen developers: flow-chart and spreadsheet-style tools to capture rules that everyone can read and execute

n8n

  • Triggers (webhook, schedule, app event) start a workflow of nodes (HTTP request, database, email, code)
  • Self-host with npm install -g n8n or Docker, UI on port 5678
  • Has a REST API, so another app (e.g. Django) can create and trigger workflows programmatically
  • Good for integration glue and automations; for long-lived human processes with audit and versioning, a BPMN engine fits better

Choosing

Need Good fit
Long-running, human-heavy business processes with diagrams the business can read BPMN engine (Camunda, Spiff)
Quick integrations between SaaS apps and webhooks n8n
Workflows defined in application code with strong reliability Temporal, Durable Functions, Step Functions
Data pipelines on a schedule Airflow / Prefect
Workflows inside a Django app Viewflow or a BPMN engine called over its API

Case-management example

A case moves through submit → triage → investigate → decide → close. Django stores the case and permissions; the engine (n8n or BPMN) owns the process state, timers (SLAs) and notifications. Keep one source of truth for status and let the other react to events.

Related: AI Agents and Workflows (agent pipelines are workflows with human gates too), 2. Docker Compose (engines usually run beside a database in Compose).