Where do the credentials live, and why is that the first question?
Because an automation platform is the single account that can read everything. To move a lead from a form into your CRM and raise an invoice, it holds API keys for the form, the CRM and the accounting system at once — a concentration of access that no individual employee has.
With a hosted platform those credentials sit with the vendor. Make is explicit about the mitigation available: you choose an EU or US data centre at signup, and choosing EU keeps the processing and the stored credentials inside European jurisdiction. That choice is made once and matters permanently.
With a self-hosted platform there is no vendor holding anything. n8n self-hosted, Kestra, Node-RED and Automatisch all run on infrastructure you control, so the credentials never leave it and you remain the sole controller. For an organisation whose automations touch payroll, patient records or financial systems, that is frequently the only acceptable arrangement.
The practical test: list the systems a workflow will touch and ask whether you would give one contractor standing access to all of them simultaneously. If the answer is no, the automation platform needs the same scrutiny you would give that contractor.
App integration, data orchestration or embedded — which are you buying?
App-to-app integration moves records between SaaS products when something happens: a form submission creates a CRM record, a paid invoice triggers a fulfilment task. n8n, Make and Automatisch do this, and it is what most people mean by automation.
Data orchestration runs pipelines on a schedule with dependencies, retries and backfills — extract, transform, load, then rebuild yesterday when the source system was wrong. Kestra does this, replacing Apache Airflow, Prefect and Dagster rather than Zapier. Buying an app-integration tool for a data pipeline means discovering it has no concept of a backfill.
Hardware and protocol automation wires physical things together: sensors, controllers, MQTT and Modbus. Node-RED owns this, running on anything down to a Raspberry Pi, and no SaaS automation platform competes.
Embedded integration means shipping integrations inside your own product, so your customers connect their tools without you building each connector. elastic.io and Locoia both offer this, white-labelled, which is a completely different purchase made by a software vendor rather than by an operations team.
How does automation pricing turn expensive?
Per-operation billing, and it surprises people because the unit is not what they think. Make prices by operations, where every step in a scenario counts — a workflow with a router, three branches and an aggregator consumes several operations per run, not one. Make's free tier gives 1,000 operations a month and Core starts around $9, which is far cheaper per operation than Zapier, but the arithmetic still scales with workflow complexity rather than with business value.
Self-hosting removes the meter entirely. n8n self-hosted has no per-task pricing at all, which is the argument that wins for high-volume automations: a workflow running every minute costs the same as one running daily. Kestra, Node-RED and Automatisch are the same.
What self-hosting costs instead is operational attention, and n8n says so directly — a server, a queue, upgrades and someone responsible when a workflow silently stops. For a handful of low-volume automations that person's time exceeds any subscription.
Locoia sits at the other end at about €200 a month, priced for mid-market operations rather than small teams, which is a deliberate position rather than an oversight.
Does "open source" mean what you think it means here?
Not uniformly, and the distinction has consequences. Kestra is Apache 2.0 and Node-RED is Apache 2.0 — genuinely open, permissive, no restrictions on use. Automatisch is AGPLv3, which is genuinely open and strongly copyleft.
n8n is source-available under the Sustainable Use Licence, which is not classic open source. The code is readable and self-hosting is free and unrestricted for internal business use, but the licence limits offering n8n as a service to others. For almost every buyer that changes nothing; for anyone planning to build a product on top of it, it changes everything, and it is better learned now than at launch.
elastic.io splits the difference: the connector SDK is open source while the platform is proprietary, so you can write and own your connectors without owning the engine.
What open source actually buys in this category is exit and inspection — a workflow engine holding every credential you have is a reasonable thing to want to read the source of.
Where do automation projects usually fail?
Silent failure. A workflow that stops running does not announce itself; the records simply stop arriving, and somebody notices three weeks later when a report looks wrong. Error handling and alerting are the difference between automation and a liability, which is why Make ships error handlers, Locoia builds alerting in, and Kestra offers replays and backfills to repair the gap once found.
The second failure is automating a process nobody agreed on. If two departments disagree about when a lead becomes an opportunity, encoding one version in a workflow makes the disagreement permanent and invisible. Automate the process after it is settled, not to settle it.
The third is the single point of knowledge. One person builds forty workflows in a visual editor with no versioning, then leaves. Kestra answers this by making workflows declarative YAML that lives in version control and can be reviewed like code; Node-RED is explicit that it has no built-in team collaboration or versioning, which is a real constraint at team scale.
The fourth is credential sprawl. Every workflow that stores its own copy of an API key is a key nobody rotates. Centralised credential management and a documented owner for each connection is unglamorous and is what keeps this working past year one.
