The on-call OSS space has a problem. Of the three self-hostable PagerDuty alternatives our directory recommends, zero are currently alive.

Last week's first repo-health snapshot flagged ~11% of all listed self-host alternatives as not-alive. This week we're zooming in on the worst-affected single category: on-call rotation and incident paging. It's not 11%. It's 100%.

If you're paying $21–41/user/month for PagerDuty and were planning the self-host escape hatch, the directories you're staring at — including ours — are largely out of date. Here's what May 2026 actually looks like, and what to do instead.

The three we listed (and what their health pill says today)

Three projects, three different ways of being not-alive. Worth taking each one in turn — the failure modes generalize to any OSS-buying decision in 2026.

Grafana OnCall — archived two months ago

The most painful entry. Grafana archived their own on-call project on 2026-03-24. The repo's README points at "Grafana IRM" — Grafana's roadmap-rebranded successor. As of May 2026, IRM is a paid Grafana Cloud feature, not a self-host package.

For a self-hoster, that means: the codebase is frozen, no PRs accepted, the next features live behind a SaaS paywall. What makes this trap harder to spot is that the repo still has recent commits — the archive happened after a scramble of farewell fixes. A tool that surfaces only last_commit_at would tell you OnCall is fine. The signal that catches it is archived: true.

Lesson: when you're picking a self-host target, last-commit-recency is half the answer. The other half is the archive flag, and roughly a third of the "dead" repos in our snapshot are dead specifically because of archived: true, not because their commits dried up.

LinkedIn Oncall — nine months silent

LinkedIn's Oncall was always a partial story. It's the rotation/calendar layer; you paired it with linkedin/iris for actual paging. Two processes, MySQL underneath, a 2018-vintage UI. Acceptable trade-off if it shipped.

It hasn't. Nine months without a commit on a tool that needs to keep up with Twilio API versions, Slack Block Kit revisions, and Prometheus AlertManager schema is a real risk. Open-issue count is 78. The iris half (linkedin/iris) shows similar stagnation.

This is a textbook "acquihire-grade abandonment": a single open-source project that an internal team kept alive while a single maintainer was around, then stopped touching when that person rotated off. Not malicious, not formally dead — just no one driving.

Cabot — repo no longer exists

cabotapp/cabot returns 404 from the GitHub API today. Renamed, deleted, or transferred to a private namespace — we genuinely don't know which.

Cabot was the smallest install of the three: a Pingdom-plus-paging combo with just Postgres + Cabot. The minimal surface area made it attractive for tiny teams. But a 404 from api.github.com/repos/cabotapp/cabot means no clone, no fork, no historical issue tracker. Whatever knowledge was in there is — for any practical self-host purpose — gone.

We hold it as unknown in our snapshot rather than promoting it to dead, because someone might restore the namespace next month. But the 404 is itself the strongest "do not put this on a 2026 roadmap" signal we have.

Why the on-call vertical decays harder than most

A few observations from staring at this category specifically:

1. On-call tools live or die on integrations that age fast. Twilio outbound API versions, Slack Block Kit, Prometheus AlertManager schema, PagerDuty's own webhook format for receiving migrations. Six months of maintainer silence and you've got at least one of those broken. A static site generator can ship a single commit a year and still be useful. An on-call tool cannot.

2. The vendor money in this space is real. PagerDuty, Splunk On-Call, Opsgenie, FireHydrant — all well-funded SaaS with vested interest in killing OSS competition. Grafana OnCall arcing into Grafana IRM is the textbook end state: get acquired-by-roadmap, archive the public repo, sell the same code as a hosted feature.

3. It's a high-trust install. People deploy a status page on a whim. People do not deploy "the thing that decides who gets woken up at 3am" without serious vetting. So the user base of OSS on-call tools skews smaller and more demanding — which inverts the casual-contributor loop healthier OSS projects depend on.

The combination is brutal. You need ongoing integration maintenance, the vendors will hire away your one maintainer, and your user base is too small and too cautious to cushion the gap with PRs.

What you should actually self-host for paging in 2026

Two real options, neither of which is on our /pagerduty/ page yet — this is a directory bug we'll fix in the next sweep:

1. Keep — keephq/keep

• 11,805 stars, last commit yesterday (2026-05-09), AGPL-3.0
• Bills itself as the open-source AIOps and alert management platform. Modern stack (Python + Next.js), active Discord, regular releases.
• The clearest current self-host alternative to PagerDuty. Native integrations for Prometheus, Datadog, Sentry, Slack, Twilio. Workflow engine for routing rules.

The fact that we hadn't surfaced this on our /pagerduty/ page when we wrote it is the editorial bug this whole post is paying off. The June 2026 sweep will add it.

2. Hand-rolled: AlertManager + a calendar + Twilio

Deeply boring. Prometheus AlertManager handles routing. Google Calendar (or any calendar with an iCal feed) holds the rotation. A tiny webhook script translates "AlertManager fired → who's on call right now per the calendar → Twilio outbound call/SMS".

Not a product. Not a deploy. A weekend script. But it's what a lot of small teams have actually converged on while waiting for the OSS on-call market to settle, and it has zero supply-chain dependency on anyone else's roadmap.

What this means if you trusted our /pagerduty/ page last week

Honestly: the page is misleading right now. Our pill colors will tell you Grafana OnCall is dead, but the directory still lists it as the lead alternative. That's an editorial state we have to fix, not a "feature".

This is exactly the failure mode a freshness-first directory is supposed to catch quickly — and the reason we built the graveyard page (every dead/stale repo across our 100 SaaS pages, with backlinks to the SaaS each is listed under) and the public JSON API (so anyone aggregating us downstream gets the freshness signal too).

If you were planning a 2026 PagerDuty migration and our /pagerduty/ page was your starting point — thank you for reading carefully. Your fallback today is Keep, or hand-rolled AlertManager + calendar + Twilio. Don't put grafana/oncall, linkedin/oncall, or cabotapp/cabot on a 2026 self-host roadmap.

What's next

This is the second post in the monthly snapshot series. The May post established the overall ~11% not-alive headline. This post zooms in on the worst category. Future posts in this series will cover:

• The first month-over-month diff (June 2026 — what flipped state)
• Linktree-style "link in bio" tools (BioDrop dead, littlelink stale — same pattern, smaller stakes)
• Form builders post-Typeform (ohmyform abandoned — what's actually shipping)

Want the diff post when it lands? Subscribe to the monthly digest from any SaaS page on the directory.

We'd rather you find out from us than from a git clone followed by a confused stare at the archive banner.