Automating for real with n8n

n8n has become my Swiss Army knife for automation: open source, self-hosted, and flexible enough to stay out of my way when a flow grows.

Why self-host

Keeping n8n on my own infrastructure means the data never leaves the machine until I decide it should. For workflows touching sensitive information, that's the difference between "doable" and "not a chance".

• Full control over versions and dependencies
• No artificial cap on the number of executions
• Freedom to put a reverse proxy in front and handle auth my way

The webhook → validate → act pattern

Most useful flows start with a webhook. My rule is simple: always validate the input before doing anything.

// Pseudo-logic of a Function node
if (!body.email || !body.email.includes('@')) {
  throw new Error('Invalid email')
}
return { json: { ok: true } }

A webhook without validation is just an elegant way to fill your database with garbage.

Integrating LLMs without blowing up costs

When I add an LLM node to a flow, I treat it as an expensive, fallible call:

1. Aggressive caching on repeated inputs
2. Explicit timeouts
3. A deterministic fallback when the model doesn't answer

Automation has to stay predictable even when the "smart" part throws a tantrum.