Index · 5 open files

01QuantPodPROTOTYPE02DB.aiLIVE DEMO03NighthawkBUILDING04FloatBUILDING05HearthBUILDING

Nighthawk · overnight on-call

You don’t have an on-call rotation. Now you do.

Nighthawk watches production overnight. It fixes what it is sure of, and it scores the blast radius before it touches anything. On payments, auth, and data-loss paths, it does not guess.

When it is not certain, it escalates, and the escalation is the product. It pages you with a complete case file: the incident, a reproduced failing test, every fix it tried and why each was accepted or rejected, and ranked root causes.

A live, read-only demo. No login, no signup, no sales call.

In early access: we’re taking on a handful of pilot teams and talking to backers who want in early.

Already running Nighthawk? Sign in to your dashboard

Read-only by default · protected paths never shipped without sign-off

risk gate · tonightINC-2231

guard retry in auth/middleware

blast radius · every authenticated session

REJECTED

cap retries on payments/retry/backoff.ts

isolated path · 41 tests green · reversible

PR #2231
root-cause confidence 0.58 < 0.80 gate→ escalated
Two patches, one verdict eachOpen the live demo

01 · The escalation

When it cannot be sure, it does not guess. It briefs you.

At 03:12 a SEV-2 fires for checkout-api 5xx errors. Nighthawk acknowledges it without paging anyone, reproduces the failure as a red test, and traces it to a deploy that added a retry on timeout. Its first fix would guard the retry inside auth middleware; the risk gate rejects that because auth is a protected path with a blast radius of every session. A second, contained fix on an isolated payments retry path passes the tests and opens PR 2231 behind a flag. The durable root cause sits in the protected payment-capture path, where Nighthawk’s confidence is 0.58, below the 0.80 gate, so it escalates: it compiles a case file with the incident, the failing test, both fixes and why each was accepted or rejected, and three ranked root-cause hypotheses, then pages the on-call engineer.

INC-2231 · checkout-api
REPLAY
03:12:04ALERTcheckout-api · 5xx rate 0.2% → 6.4% · SEV-2
03:12:06ACKnighthawk on it · on-call @sam not paged
03:14:18REPROisolated the request, wrote a failing test
test/checkout/capture.spec.ts · expected 200, got 502 · RED
03:16:40TRACEpayment-capture p99 8.1s · gateway timeouts ×318
deploy 4b1e0c landed 03:01 · added retry-on-timeout
03:19:32FIX 1drafted patch · guard the retry in auth middleware
03:19:33GATEblast radius scored · every authenticated session
REJECTED · auth is a protected path · will not auto-apply
03:24:50FIX 2narrower patch · cap retries on payments/retry/backoff.ts
isolated path · 41 tests green · capture.spec.ts now PASS
03:25:12PRopened PR #2231 · contained, reversible, behind a flag
03:26:03GATEsymptom held · root-cause confidence 0.58 < 0.80
03:26:05ESCALATEcompiling case file · paging @sam with the story
Fig. 01 · the escalation is the product

02 · The risk gate

Judgment is the feature. The fix is the easy part.

Plenty of tools can write a patch at 3 AM. The hard part is knowing which patch is safe to ship and which one should wake a human. Nighthawk decides with three numbers, every time.

Confidence score
Every action carries a number. Nighthawk acts on its own only above the line your team sets. Below it, it escalates instead of guessing. The gate ships at 0.80.
Blast radius
Before it touches anything, it scores the reach. One row, one service, or every session. A small, reversible change clears the gate. A wide one never auto-applies.
Protected paths
Payments, auth, and anything that can lose data are off limits to automation by default. Nighthawk reads them, reasons about them, and writes up a fix. It cannot ship one without you.
nighthawk.policy.ymlENFORCED
# the gate, written by your team
gate:
  min_confidence: 0.80
  max_blast_radius: service

protected: # never auto-written
  - payments
  - auth
  - data_migrations

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

escalate:
  below_confidence: page_with_case_file
  protected_root_cause: page_with_case_file
Fig. 02 · guardrails are config, not promises

03 · Pricing

One price per workspace. Pay for pagers, not seats.

A responder is anyone Nighthawk can page. Each tier includes a few; extra responders are a flat $30/mo.

Lookout

Free

Your first pager. No card.

  • 1 responder · 1 service
  • Email + Telegram channels, blast-and-repeat escalation
  • Webhook + manual incident composer
  • 7-day incident history
  • Autonomy locked to NEVER TOUCH / SUGGEST ONLY (levels 0–1)

Night Shift

$400/mo

Detect, reproduce, brief. It never touches production.

  • 5 responders included · +$30/mo per extra responder
  • All channels (Email, Discord, Slack, Telegram), full escalation engine
  • Up to 10 services · 30-day audit log
  • Risk profile up to FIX WITH REVIEW (autonomy levels 0–2)
  • Reproduced failing tests on every incidentRolling out
  • Ranked root-cause hypotheses on every incidentRolling out

Fleet

Customannual

For platform orgs with the compliance team in the room.

  • SSO / SCIM
  • On-prem runners & in-region processing
  • Custom DPA · audit log export
  • Dedicated environment & support engineer
  • Org-wide risk policy templates
What’s a responder?
Anyone Nighthawk can page and who can acknowledge. People, not rotations or seats.
What counts as a service?
One app or system you connect, by its ingest token or a monitor webhook.
What happens when nobody acks?
It keeps paging on your interval until the cap, then logs the incident as exhausted.

Waitlist teams lock founding pricing: 40% off year one, grandfathered.

04 · Get involved

Give the 3 AM call to something with judgment.

We’re onboarding a small number of pilot teams now, the ones without an on-call rotation first. You get founding pricing and a direct line to the people building it.

I agree to Lunvex Labs storing my email to contact me about early access, per the Privacy Policy.

Backing early infrastructure? Talk to the founders.

Already on the team? Sign in to your dashboard

A Lunvex Labs venture