n8n-nodes-stillrunning

Heartbeat + AI-workflow monitoring for n8n. Reports when your workflows succeed, fail, or go quiet to StillRunning , so you find out the moment one fails silently, instead of when a client does.

This is a community node. Installing community nodes.

Why

n8n tells you a workflow errored while it's running. It doesn't tell you when a scheduled workflow never fired, froze, or got stuck in the queue , the failure modes that don't write an execution record. StillRunning watches for the ping that never arrives and alerts you (email / Slack / Discord / webhook), with the AI-spend fields parsed when you send them.

Install

npm i n8n-nodes-stillrunning

Or in n8n: Settings → Community Nodes → Install → n8n-nodes-stillrunning.

Setup (Tier 1 , no credentials beyond your ping token, ~30s)

1. Create a workflow at stillrunning.ai and copy its ping token (Integration tab).
2. In n8n, add a StillRunning credential and paste the token. (Self-hosted StillRunning? set the Base URL.)
3. Success path: drop a StillRunning node at the end of your workflow, operation Report Success. Drag a direct LLM node's output into the optional token/cost/model fields to track AI spend.
4. Failure path: in your workflow's Error Workflow, add a StillRunning node, operation Report Failure. Leave Read Error From Input on , it reads the failed node + error message straight from the Error Trigger payload. No key needed.

Operations

| Operation | Use it for | Sends | |---|---|---| | Report Success | end of the happy path | event=success + optional duration, tokens, cost, model, tool calls | | Report Failure | inside the Error Workflow | event=fail + failed node + error message (+ execution id) | | Heartbeat | long-running / step-based jobs | a bare "still alive" ping |

Every ping carries source: "n8n" and the n8n executionId.

Cost for AI Agent workflows (Tier 2)

n8n's AI Agent node drops token usage from its output wire (a documented n8n limitation), so the wired token/cost fields only flow from direct LLM nodes (OpenAI, Anthropic, …). To capture cost for Agent workflows, use the Tier-2 Error Workflow enrichment template (coming with the docs): it calls GET /executions/{id} inside your own n8n using a local credential , the key never leaves your instance , and pushes the per-node timing + token usage to StillRunning.

Design note

Monitoring must never break the workflow it monitors. A failed ping is reported on the node's output item (stillRunning.ok = false), never thrown.

License

MIT