Nightwatch · AI SRE

The engineer on call while you sleep.

When the page fires at 3 AM, Nightwatch acks it, reads the logs, the dashboards and the last deploy, then drafts the fix PR and the incident summary. It wakes a human only when judgment is genuinely required.

Join the rotation How it works

Private beta · read-only by default · SOC 2 in progress

Fig. 01 · a four-minute incident, replayed verbatimruntime 04:38

02 · The 3 AM tax

3:07 AM.

The hour your roadmap pays for your reliability

Pager fatigue isn’t a tooling gap. It’s a staffing model where your best engineers do their worst work. Woken at random. Dropped into a maze of dashboards. Re-deriving context a machine already had. Fixing the same class of regression they fixed last month.

The on-call rotation was designed for a world where only a human could read the logs. That world ended.

median time spent just gathering context before any fix begins: 38 min
of night pages end in a known, mechanical remediation: 63%
good engineering decisions made half-asleep: 0

03 · Operating loop

What happens while you don’t wake up.

Triage
Acks in seconds, not snoozes.
Nightwatch acknowledges the page, dedupes the alert storm, pulls the runbook and the service map, and states a working hypothesis before a human would have found their glasses.
Investigate
Reads everything. Forgets nothing.
Logs, dashboards, traces, the last ten deploys, that config change from Tuesday. Every claim cites its evidence: line numbers, chart links, commit hashes. The morning review takes minutes, not a meeting.
Act
Drafts the fix, under your rules.
Rollbacks, reverts, scale-ups and feature-flag kills, proposed as PRs and runbook actions, executed only inside policies your team wrote. Nothing touches production that you didn’t pre-authorise.
Escalate
Wakes you only for judgment.
Some calls are genuinely human: data-loss risk, customer comms, an unknown unknown. For those, Nightwatch pages with the full story attached. You wake up briefed, not blind.

04 · Patched into

Reads what your team reads. Speaks where your team speaks.

at launch in development

nightwatch core

PagerDutypages in · resolves outavailable at launch
Datadogmetrics · logs · tracesavailable at launch
Grafanadashboards · alert rulesavailable at launch
GitHubdeploys · diffs · fix PRsavailable at launch
Slacksummaries · approvalsavailable at launch
Opsgeniepages in · resolves outin development
Prometheusmetrics · alert rulesin development
Sentryerrors · releasesin development
CloudWatchmetrics · logsin development
Kubernetesevents · rolloutsin development

Connect via scoped, revocable API credentials, read-only by default. Something custom in your stack? Nightwatch speaks anything with an API and a log line.

05 · Rules of engagement

An agent in production earns trust in writing.

Read-only by default: Nightwatch ships with zero write scopes. You grant actions one policy line at a time, and revoke them just as fast.
Every action reviewable: Every query, every conclusion, every proposed change lands in an immutable audit log, replayable step by step like the incident above.
Your telemetry stays yours: Customer logs and metrics are never used to train models. Data is processed in-region, encrypted in transit and at rest.
SOC 2 Type II in progress: Audit underway, report expected this year. SSO/SCIM, custom DPAs and on-prem runners available on the Fleet plan.

nightwatch.policy.ymlENFORCED

# least privilege, written by your team
mode: read_only

allow:
  - logs.read
  - metrics.read
  - deploys.read

actions:
  pr.draft: auto
  rollback: requires_approval
  db.write: never

escalate:
  sev1: page_human_immediately
  uncertain: page_human_with_brief

Fig. 02 · guardrails are config, not promises

06 · Pricing

Less than one engineer’s lost morning.

Priced per on-call rotation, not per seat. The whole team reads the summaries; nobody pays to sleep.

Night Shift

$390/mo per rotation

Triage and investigation. Nightwatch observes, concludes and briefs. It never touches production.

Acks, triages & investigates every page
Evidence-cited incident summaries in Slack
Up to 10 services
30-day audit log

Most teams

Full Rotation

$1,180/mo per rotation

The whole loop. Guarded actions under your policies: rollbacks, fix PRs, runbook execution.

Everything in Night Shift
Drafted fix PRs & policy-guarded rollbacks
Runbook execution with approvals
Unlimited services · 1-year audit log
Postmortem drafts for every incident

Fleet

Customannual

For platform orgs running many rotations, with the compliance team in the room.

SSO / SCIM provisioning
On-prem runners & in-region processing
Custom DPA · audit log export
Dedicated environment & support engineer

※ Waitlist teams lock founding pricing: 40% off the first year, forever-grandfathered.

07 · Early access

Hand the pager to something that never sleeps.

We’re onboarding a small number of engineering teams into the private beta, pager-heavy stacks first. Founding pricing locked for everyone on the list.

A Lunvex Labs venture