The night shift for your software.

Nocturne takes the 3am page. It reads your logs and graphs, finds what broke, and rolls it back or writes the fix. Before it touches production, it asks whoever is on call.

Free for three services. No card needed.

03:12

Nobody debugs well at 3am.

Your phone goes off on the bedside table. You find your laptop by feel, open seven dashboards, and twenty minutes later you've decided it's probably the database, which is what you'd have guessed from bed.

It isn't the database. It's a deploy from 02:51 that nobody was watching. You roll it back at four, write the incident summary in your head until six, and then you go to stand-up.

Live demo

Here's one we made earlier.

A scripted incident in a made-up checkout service. Nocturne works your real ones the same way.

Incident #4417 · checkout-api · productionFiring

Timeline

  1. DD

    Datadog 03:12:04

    03:12:04 ALERT

    checkout-api p99 latency 4.8s (threshold 800ms) · 5xx rate 6.1% · from Datadog

What Nocturne is looking at

v4.18.2production02:51PR #2287 "Rename database env vars"
v4.18.1productionMon 16:40PR #2279
v4.18.0productionMon 11:02PR #2270
v4.17.9productionFri 15:18PR #2264
v4.17.8productionThu 10:31PR #2258

02:52:14 ERROR TimeoutError: could not acquire connection from pool within 3000ms

02:52:17 INFO GET /checkout/session 200 3104ms

02:52:20 ERROR TimeoutError: could not acquire connection from pool within 3000ms

02:52:23 ERROR TimeoutError: could not acquire connection from pool within 3000ms

02:52:26 INFO GET /checkout/session 200 3104ms

02:52:29 ERROR TimeoutError: could not acquire connection from pool within 3000ms

02:52:32 ERROR TimeoutError: could not acquire connection from pool within 3000ms

02:52:35 INFO GET /checkout/session 200 3104ms

02:52:38 ERROR TimeoutError: could not acquire connection from pool within 3000ms

02:52:41 ERROR TimeoutError: could not acquire connection from pool within 3000ms

02:52:44 INFO GET /checkout/session 200 3104ms

02:52:47 ERROR TimeoutError: could not acquire connection from pool within 3000ms

02:52:50 ERROR TimeoutError: could not acquire connection from pool within 3000ms

02:52:53 INFO GET /checkout/session 200 3104ms

02:52:14 ERROR TimeoutError: could not acquire connection from pool within 3000ms

02:52:17 INFO GET /checkout/session 200 3104ms

02:52:20 ERROR TimeoutError: could not acquire connection from pool within 3000ms

02:52:23 ERROR TimeoutError: could not acquire connection from pool within 3000ms

02:52:26 INFO GET /checkout/session 200 3104ms

02:52:29 ERROR TimeoutError: could not acquire connection from pool within 3000ms

02:52:32 ERROR TimeoutError: could not acquire connection from pool within 3000ms

02:52:35 INFO GET /checkout/session 200 3104ms

02:52:38 ERROR TimeoutError: could not acquire connection from pool within 3000ms

02:52:41 ERROR TimeoutError: could not acquire connection from pool within 3000ms

02:52:44 INFO GET /checkout/session 200 3104ms

02:52:47 ERROR TimeoutError: could not acquire connection from pool within 3000ms

02:52:50 ERROR TimeoutError: could not acquire connection from pool within 3000ms

02:52:53 INFO GET /checkout/session 200 3104ms

v4.18.2 deployed
checkout-api p99 latency threshold 800ms
v4.18.2 deployed
db pool: in use / max max · in use

How it works

Watch, diagnose, fix. In that order.

  1. 01 Watch

    Nocturne reads the alerts, metrics and logs you already have. You give it read-only API keys; there's nothing to install on your servers and your monitoring stays where it is.

  2. 02 Diagnose

    When an alert fires, it starts where you would: what changed recently, then what the logs and graphs say about it. Each query and each log line it relied on goes into the timeline, so you can check its reasoning.

  3. 03 Fix

    It picks the smallest change that stops the damage, usually a rollback or a feature flag, and asks the person on call before doing it in production. The proper fix arrives as a pull request for someone to review after breakfast.

Features

The parts you'd ask about first.

  • Works with what you run

    Datadog, Grafana, Prometheus and Sentry for what's happening. GitHub and GitLab for what changed. Kubernetes, AWS and Google Cloud for doing something about it. PagerDuty, Opsgenie and Slack for telling people.

  • Asks before production

    Approval rules are set per service and per environment. Approve from Slack or your phone. If you want, it can go ahead alone on low-risk jobs you name, like restarting a pod in staging.

  • Reversible first

    Rollbacks, flag changes and scaling come before new code. Some things it will never do, whatever you approve: run database migrations, delete data, or change who has access to what.

  • Shows its working

    Every command it ran and every log line it read is kept in the incident timeline, with timestamps. Export it to your own audit log if your security team asks.

  • Wakes you when it should

    If it isn't confident, or nothing it's allowed to do will help, it pages a person through PagerDuty or Opsgenie and hands over what it has found so far. You start from a summary instead of an empty dashboard.

  • Follows your runbooks

    Point it at runbooks in Notion, Confluence or your repo. It follows them during incidents and flags the steps that no longer match your systems.

  • Practice runs

    Shadow mode lets Nocturne work real incidents without acting on them. Afterwards it shows what it would have done, next to what your team actually did.

  • Your data stays yours

    We don't train models on your logs or code. Choose EU or US hosting, or on Enterprise, run Nocturne inside your own cloud account.

Pricing

Free for your first three services.

Team is $49 a month for each engineer on call. Enterprise adds single sign-on, a runner in your own cloud and a contract your lawyers can mark up.

Who's behind it

Started on a Boxing Day call in Manchester.

Priya, Tom and Kemi spent four years on the same on-call rota at a Manchester payments company. At 03:40 on Boxing Day 2024 an expired certificate stopped card payments. Finding it took the three of them two hours on a video call from three different houses, one with a toddler who had also decided to be awake.

Tom wrote a script that night to check certificate expiry dates. Nocturne is what that script turned into. The company started in 2025 in a small office in Ancoats, and all three founders are still on Nocturne's own on-call rota. Nocturne takes the first page.

  • Priya Raman, co-founder and chief executive of Nocturne.

    Priya Raman

    Co-founder, chief executive

    Ran site reliability at payments and travel companies for eight years. Has written more incident reviews than she'd like, and kept the good ones as a reference.

  • Tom Ashworth, co-founder and chief technology officer of Nocturne.

    Tom Ashworth

    Co-founder, chief technology officer

    Backend engineer from Stockport. Wrote the Boxing Day certificate script, and the first version of Nocturne in the evenings after it.

  • Kemi Adeyemi, co-founder and head of safety at Nocturne.

    Kemi Adeyemi

    Co-founder, head of safety

    Was a security engineer before she was a founder. She decides what Nocturne is never allowed to do, and that list only ever gets longer.

Let someone else take the 3am page.

Connect one service and run Nocturne in shadow mode for a week. It will show you what it would have done, and then you decide how much to let it do.