Integrations — On-call and alerting
NerveStax and PagerDuty
Every alert is triaged first. Only real incidents reach your pager.
What we do
What NerveStax does with PagerDuty
Route your Airflow, dbt and webhook alerts to NerveStax. An agent investigates each one with read-only tools, resolves the noise with its reasoning written down, and escalates the rest to PagerDuty with the likely cause and the evidence it found. You keep PagerDuty for rotations, schedules and who gets woken up.
-
Triage before anyone is paged
Alerts land in NerveStax first. The agent reads the failed task's logs, the lineage around it and what changed recently, then reaches a verdict: resolved, or escalate.
-
Close noise with a written reason
Retries that already self-healed, transient infrastructure blips and fan-out from one upstream failure end as resolved, with the full transcript of what was checked kept in the product.
-
Page with the cause attached
A real incident escalates to PagerDuty carrying the likely cause, what is downstream of it and the evidence behind that reading, so the person woken up starts from a diagnosis.
-
Unsure still pages you
A crash, a timeout, a missing verdict or an inconclusive one all escalate. Nothing is silently dropped, and an uncertain agent escalates rather than inventing a cause.
-
Written down before delivery
The escalation is recorded in NerveStax before it is sent, so a paging outage can delay the page but cannot lose the incident. Every triage stays readable afterwards.
Background
About PagerDuty
PagerDuty is an incident response service that decides who is on call, pages them, and tracks the incident until it is closed. Most data platforms already send their failures there.
How it connects
What connecting it looks like
-
01
Create a PagerDuty integration
Add an Events API integration on the service that should receive data platform incidents, and copy its integration key.
-
02
Add the key in NerveStax
Paste it in as the escalation target for the workspace. It is stored encrypted and no screen shows it back.
-
03
Point your alerts at NerveStax
Use the per-project webhook URL for Airflow failure callbacks, dbt run and test results, or anything else that can POST JSON.
-
04
Watch the first night
Read the triage transcripts in the morning: what was resolved, what was paged, and why. Adjust what you route from there.
Scope
What this does not cover
- Escalation is one-way: NerveStax pages PagerDuty. It does not sync incident state back, write incident timelines, or manage your schedules.
- Triage is read-only. It never fixes an alert by changing something; code fixes still arrive as a pull request a person approves.
- Every escalation also lives in NerveStax with its full diagnosis, whether or not the page is delivered.
Questions
PagerDuty and NerveStax
Q01Do you replace PagerDuty?+
No, we page it. PagerDuty still owns rotations, schedules, escalation policies and who gets woken up at 3am. What changes is what arrives: instead of a task name and a retry count, the page carries a diagnosis, because the alert was already investigated before it became a page.
Q02What happens to the alerts that do not page?+
They end as resolved with the reasoning recorded, and stay readable in NerveStax. Nothing is deleted or silently dropped. If you disagree with a verdict, the transcript shows every log it read and every check it ran, so a wrong call is visible in seconds.
Q03What if the triage agent itself fails?+
It escalates. Fail-open is the design: a crash, a timeout, a missing verdict or an inconclusive one all page you. Anything short of a proven resolution reaches a human, which is also why an alert storm short-circuits straight to escalation.
Q04Can it change something in production to fix an alert?+
No. During triage the agent holds read-only tools, with no repository or write access while a raw alert payload is in its context, and payloads are treated as untrusted input rather than instructions. When a code change is the fix, it ships as a tested pull request you approve.
Q05What can send alerts into it?+
Anything that can POST JSON. Airflow task and DAG failure callbacks and dbt run and test results are understood natively; everything else arrives through a generic webhook with a per-workspace token, including checks in your warehouse or your observability tool.