Abstract sculpture of interwoven metallic filaments and particles

AI on-call engineer — 30 nights free

Kwozi takesthe night shift.

Kwozi answers your alerts at night, runs the runbooks you signed, and wakes you only when customers feel something it can’t fix.

by guglielmogiannattasio.it

How it works

One alert, at 03:12.

Kwozi reads it, finds what changed, and runs the runbook you signed. The same alert, three times.

  1. Arrives

    search-api03:12

    p99 latency 1.9 s, over the 800 ms threshold for 5 minutes

    Datadog monitor “Search p99 latency”

  2. Read

    search-api38 alerts grouped03:13
    Started
    14 min after deploy #4812 “Boost recent listings”
    Customers
    1.4% of searches take over 2 s. Your line is 1%.
    Runbook
    Roll back the last deploy, signed by Dana on 12 Aug

    Rolling back #4812 in 0:30

    Rolled back #4812 at 03:14. p99 back to 310 ms by 03:19. Nobody woken.

    Called Dana at 03:13. The rollback waits for her answer.

  3. Handled

    search-api03:19

    Rolled back #4812 at 03:14. p99 back to 310 ms by 03:19.

    Nobody woken. Added to the morning note.

The morning note

What you wake up to.

At 08:00 Kwozi posts this to Slack and email. Set the rules and read what each sample night would have said.

Morning noteTue 14 Oct, 22:00 to 08:00

Quiet night. Nobody was woken.

Alerts
214
Causes
5
Fixed
4
Wake-ups
0

What happened

  • 23:41logs-2 disk at 91%Compressed 38 GB of old logs, disk now at 52%, in 4 min.Fixed
  • 01:12checkout-api 3 pods out of memoryRestarted 3 pods, in 3 min.Fixed
  • 03:12search-api p99 latency 1.9 s after deploy #4812Rolled back #4812, p99 back to 310 ms, in 7 min.Fixed
  • 04:05email-workers CPU at 92%, 18k emails queuedScaled from 4 to 6 workers, queue empty by 04:31, in 26 min.Fixed
  • 05:30status-page TLS certificate expires in 9 daysLeft for you. No customers affected.For you

For you this morning

  • Raise the memory limit on checkout-api. 3 pods ran out at 01:12.
  • Fix deploy #4812 before shipping it again.
  • Renew the status page TLS certificate before 23 Oct.
1%
Restart crashed pods
Roll back a bad deploy
Scale up under load
Clean up full disks
Fail over a database
Always wake for payments

Sample nights from a team of 12 engineers running 40 services. Yours starts after your first night.

Pricing

Priced per pager.

You pay for the engineers on the rotation, never for alerts. The first 30 nights are free.

The lights of Chicago and the surrounding Midwest at night, seen from the International Space Station
Chicago at night from the International Space Station, 8 September 2026.

Questions

Before the first night.

What teams ask before they hand Kwozi the pager.

What does Kwozi connect to? Alerts from Datadog, Grafana, Prometheus Alertmanager, Sentry, CloudWatch or any webhook. Deploy history from GitHub, GitLab or Argo CD. It reaches people through Slack, push, text and phone.

Does it replace our pager? It can. Or it sits in front of the pager you have: every alert reaches Kwozi first, and only the ones that need a person go through.

Can it make things worse? It runs only the runbooks you signed, inside their limits: restart, scale up to 2×, roll back the last deploy. It never deletes data, touches secrets or pushes code.

What if it decides wrong about waking me? When it isn’t sure whether customers feel something, it wakes you. You set the line per service and can change it any time.

Where do runbooks run? On a small runner inside your own cloud account, so Kwozi never holds your production credentials.

How long does setup take? About an hour to connect alerts, deploys and Slack. Kwozi then reads in read-only mode for a few nights before you hand it anything.

Start free

Sleep through tonight.

Connect your alerts this evening and read your first morning note at eight.