I built an external watcher for n8n — because the engine reporting on itself is zero information when the engine is wrong

Built with n8n
1

A few weeks ago there was a great thread here about silent failures — workflows that stop triggering, or run green and do nothing. I’ve been running automations for clients for years, and that exact pain made me build something for myself. After the discussion here (thanks pirateprentice, ASHIM, Aghassi, dima — that thread was the best design review this tool ever got), I cleaned it up properly and I’m opening it up.

FlowPulse — read-only monitoring for n8n, from the outside:

  • Connects via the n8n API (URL + API key, ~2 minutes). Zero changes to your workflows — no heartbeat nodes, no error-workflow wiring, no per-workflow setup tax.
  • Learns each workflow’s normal cadence and alerts when one goes quiet — including the case an in-workflow heartbeat can’t catch: the trigger that dies, so nothing runs at all.
  • Classifies failures into real categories — expired OAuth ≠ API down ≠ rate limit ≠ your bug. On my own instance it turned a year of “465 errors, all unknown” into 16 distinct categories with the failing node named.
  • One alert per incident, not one per failed execution — cooldowns, recovery detection, and re-alert on reoccurrence built in. Alert fatigue is how monitoring gets muted.
  • Telegram + email. Dashboard with health per workflow, top failing nodes, error clusters.

What it doesn’t do (yet): it won’t catch “ran green but produced garbage” — output-baseline detection (“this slot normally produces 12, today 0”) is what I’m building next, and destination reconciliation (the fourth layer from the thread) is on the long-term map. If you need those today, the DIY patterns in that thread are the honest answer.

I’m looking for a handful of people who run n8n in production — especially for clients — to try it and tell me what’s wrong with it. Free, no card. I’d rather hear brutal feedback from people who live this than build in the dark.

:backhand_index_pointing_right: https://flowpulse.us

The silent failure: when your n8n workflow succeeds but does nothing
2

Sorry for slow reply. A week is too long for someone who opened a tool up and said break it.

The credit is generous. Half of what is in that thread came back to me sharper than I put it in.

One thing to break, and it is the same bug we keep circling. FlowPulse is a thing on a schedule that reports on other things on a schedule. If your poller dies it goes quiet, and quiet is exactly what your product means by everything is fine. No alerts and no watcher look the same from outside. Whatever covers that cannot be FlowPulse.

The cold start you already named yourself. Workflow with no cadence learned yet is the window where a condition is most likely to be wrong, and that is the same hole my counts have.