One dashboard for your scheduled workflows, Claude Code runs, and the agents that run when your users show up. We notify you the moment a heartbeat goes quiet, a duration doubles, or a cost spikes.
Free for up to 25 monitors · No credit card required
The most expensive failures are the quiet ones.
“Two of its four agents had locked into a recursive loop, exchanging clarification requests and verification instructions back and forth, thousands of times, around the clock… Nobody on the team knew until the cloud bill arrived.”
“The cron job didn’t run… and there was no error. No logs. No warnings. No trace. Just silence.”
“the 7:10 AM trigger didn’t fire. There were no errors and no workflow runs triggered at the expected time.”
“my clients are losing orders and leads, and incomplete executions are piling up.”
One ping URL. We catch all of it, the minute it happens.
The most common failure isn’t an error. It’s silence, a dropped schedule, a frozen container, a queue stalled overnight. On anything scheduled, we expect a ping every interval. Miss it past the grace window and we tell you, on the channel you actually read.
We baseline what “normal” looks like for every workflow. The moment a run takes meaningfully longer than its own history, you get pinged.
Send tokens, cost, and model with each ping. We catch the run that costs ten times what it should, before the monthly invoice does.
The process stayed alive, the calls succeeded, the output came back well-formed and completely wrong. Hand us a schema for a good result, and we flag the run that doesn’t match.
A run isn’t one ping, it’s a chain. We group every step by trace and show you what each one did, what it cost, and which model ran it. Observability for your agents, not just an up-or-down light.
One agent run. Every step, token, and dollar, in one place.
node <(curl -fsSL https://stillrunning.ai/cli/wrap.mjs) \
"<your-ping-url>" -- claude -p "your prompt"Wraps any non-interactive claude -p run. Auto-captures tokens, cost, model, tool calls, and duration, and groups the run into a StillRunning outcome. Your terminal output stays exactly as a bare claude -p.
All SDKs & docsEvery ping lands in a typed event log. We track success, failure, start, log. We compute duration server-side when you correlate start and end. We hold 14 days of history on free, 90 on Pro.
Heartbeat goes quiet, duration doubles, or cost spikes, we email you within minutes. Maintenance windows when you deploy. Phone and Slack on Pro.
Free forever with commercial use. Upgrade only when you outgrow it.
Run a real business on the free tier. Commercial use included.
For production workloads. Faster intervals, deeper history.
Cancel in one click. No calls, no emails.
For agencies and on-call rotations. SSO included, no per-seat add-on.
Cancel in one click. No calls, no emails.
Enterprise (custom retention, dedicated support): hello@stillrunning.ai
A GET or POST to a URL we give you. The simplest version is one line of curl at the end of your script. The richest version is a JSON body with tokensIn, costUsd, model, and any metadata you want to attach.
Two ways. Install the n8n-nodes-stillrunning community node, it gives you Report Success, Report Failure, and Heartbeat operations, and reads the failed node straight from your Error Workflow. Or drop in a plain HTTP Request node pointing at your ping URL, connected to both the success and error branches.
Yes. Wrap any non-interactive claude -p run: node <(curl -fsSL https://stillrunning.ai/cli/wrap.mjs) "<url>" -- claude -p "..." and it auto-captures tokens, cost, model, tool calls, and duration from Claude's own output, with no flags or code changes, and groups the run into a StillRunning outcome. Your terminal output is unchanged. The Claude Agent SDK and Vercel AI SDK have native one-line wraps too.
Cost spikes per run, yes, send us costUsd and we’ll alert when a run jumps multiples above its own baseline. A cumulative spend circuit-breaker that can kill a running agent is on the roadmap. Email hello@stillrunning.ai if you want a heads-up when it ships.
Yes, the failure mode pure heartbeats miss. Give us a JSON Schemafor what a good result looks like, send your output on the ping, and we flag the run that finished clean but returned the wrong shape, the agent that “failed while continuing to work.”
We watch for overdue pings and dispatch the moment one goes late. Most email alerts land in your inbox within minutes of the missed run. Phone alerts on Pro arrive faster.
Workflows imported from Cronhub (or created with an email tied to them) get a one-click magic-link claim. Same workflows, same ping URLs, same history, now bound to your login.
Probably. 25 monitors covers most freelance setups. The forcing function to upgrade is usually interval (you want 30-second resolution) or retention (you want more than 14 days of run history).
Free for up to 25 monitors · No credit card required
Heartbeat monitoring for AI workflows and agent scripts.