PagerMind investigates every Prometheus alert with your own metrics, logs and dashboards, then tells you in Slack or Teams whether it matters. Most nights, the answer is that it doesn't.
5,000 free credits. Work email required. Read-only access to your tools.
PagerMindHostHighCpuLoad on web-03pg_backup climbed#ops-alertsLoad peaked at 14.7 during the 03:00 database backup and was back to 2.1 by 03:12. CPU stayed under 70% and no service logged errors.
No action needed. The backup window explains it.
PagerMind, triaged in 87s
PagerMind is an AI SRE: it does the first ten minutes of an on-call investigation so you start from an answer instead of a raw alert.
Point an Alertmanager receiver or a Grafana alerting contact point at your PagerMind webhook. Nothing to install in your cluster.
Your filters, maintenance windows, a repeat cooldown and a storm limit catch the duplicates. Those alerts are recorded on your dashboard and cost nothing.
Each workspace gets its own agent. It reads the trend in Prometheus, breaks it down by container, searches Loki, opens the Grafana dashboards your team already uses and checks what it has seen before on that host.
About 90 seconds after the alert, a short note says what happened, what it means and whether anyone needs to act, as a reply on the original alert thread.
Mark a verdict wrong and write what really happened. The correction becomes guidance for that alert the next time it fires.
Every note starts with the verdict, names the host, explains the evidence in plain sentences and ends with one action.
The alert fired but nothing is wrong: a backup, a deploy, a spike that already cleared. You stay asleep, and the reasoning is there in the morning.
Something is broken and still broken. The note says what, where, since when, and what the evidence points to, so whoever is paged starts in the right place.
Not an outage yet, but trending the wrong way: a disk filling, error rates creeping up. Worth a look in working hours.
PagerMind queries your tools live during an investigation. Your metrics and logs stay where they are.
An AI agent with write access to production is a second incident waiting to happen. PagerMind can look at everything you connect and change nothing. The host agent runs only commands on a fixed read-only list, and anything else is refused before it runs.
Every new workspace gets 5,000 free credits, enough to see PagerMind on your real alerts. A typical investigation uses about 100 credits, and filtered alerts use none.
When the credits run out, add your own OpenAI, OpenRouter or Anthropic API key in settings and PagerMind keeps going on your key. No card needed to start.
PagerMind is an AI SRE for teams that run Prometheus and Alertmanager. When an alert fires, it queries your own metrics, logs and dashboards, works out whether the alert matters, and posts a short verdict to Slack or Microsoft Teams in about 90 seconds: false alarm, real issue or needs attention.
An AI SRE is a software agent that does the first part of an on-call engineer's job: it reads an alert, looks at the systems involved, gathers evidence and writes up what is going on. It does not replace the on-call engineer; it does the first ten minutes of investigation so the engineer starts from an answer instead of a raw alert.
PagerDuty and Opsgenie decide who gets paged and when. PagerMind decides what the alert means. It investigates the alert and posts a verdict, and it works alongside your paging tool rather than replacing it.
No. PagerMind only reads. It runs queries against your Prometheus, Loki and Grafana, reads Kubernetes pod state and logs, and, if you install the optional host agent, runs a fixed list of read-only commands on your servers. It never restarts, deletes or edits anything.
A Prometheus server PagerMind can reach over the internet with authentication (basic auth or Cloudflare Access), an Alertmanager or Grafana alerting webhook pointed at PagerMind, and a work email address. Slack, Microsoft Teams, Loki, Grafana and the host agent are optional.
About 90 seconds. A typical investigation makes around 20 read-only queries before it writes the verdict.
New workspaces get 5,000 free credits, and a typical investigation uses about 100 credits. When the credits run out, add your own OpenAI, OpenRouter or Anthropic API key and PagerMind keeps investigating on your key.
No. Before any investigation, PagerMind checks your filters, maintenance windows, a repeat cooldown and a storm limit. Alerts they catch are still recorded on your dashboard, but no AI runs and no credits are used.
Yes. Mark a verdict wrong and write what really happened. PagerMind turns the correction into guidance for that alert and applies it the next time the alert fires.
Anyone with a work email address. Each company gets one PagerMind organization; when a colleague signs up later, they ask the organization's owner for access.
Connect Prometheus, point Alertmanager at PagerMind and send a test alert. The first verdict usually lands within 15 minutes of signing up.