n8n, Workflow Automation Without the Cloud Tax

I have a complicated relationship with n8n. I run it on a Coolify deployment, I have flows that move RZP webhooks into Telegram alerts, and I still mostly reach for a 40-line Python script when I need a new automation.

The reason is not that n8n is bad. It is that n8n is a visual graph, and a visual graph is great for non-coders and worse for me, because I already think in functions. Every n8n node feels like a bigger box than the operation it represents. When something breaks I want to grep, not zoom into a panel and click into a JSON sub-editor.

That said, n8n earns its place in my empire for two specific shapes of work. First, when someone non-technical needs to own a flow after I build it. I shipped a kaam-tracker invoice-followup flow in n8n exactly because the bookkeeper would eventually need to tweak the schedule and the recipient list. She can. She does. Second, when the flow has more than four conditional branches and the visual map actually helps me reason about edge cases. Slack-incoming, GitHub webhook, error-fork, retry-fork, escalation, notify, log; that lives better in n8n than in code.

The Indian-operator angle is cost. Zapier-Pro is roughly Rs 4,000 a month at the entry tier with five-figure run caps that you blow through fast on any real automation. n8n self-hosted on Coolify on a Rs 1,200 a month Oracle ARM box runs unlimited flows. The break-even is somewhere around 700 runs a month, and any business worth automating crosses that in week one.

The catch I learned late is the queue mode. Default n8n runs the executor and the workflow engine in the same process, and a single hot loop will starve everything else on the box. Switch to queue mode with Redis early. I did not, and a runaway Bluesky-poll flow on autokaam-related infra burned 92 percent CPU for three hours before I noticed.

For startup-scale work in India, I still recommend trying n8n once on a real workflow before committing. If you find yourself wishing for a print statement and a stack trace, switch to FastAPI plus a cron entry. If you find yourself building a flow with more than three diverging branches, n8n is the right home for it.

Zapier and Make pricing scales with run count, which murders any high-volume automation past hobby scale. n8n is fair-code (free for self-host, paid for SaaS) and runs comfortably on a Rs 1,200 a month Oracle ARM box via Coolify, removing the per-run tax entirely. The 2026 wave of agentic LLM workflows multiplies run counts by 10-100x compared to old form-fill automations, which makes the SaaS pricing math even more brutal and the self-host case stronger.