On-call tooling is the least glamorous thing in an observability stack and the only part that has to work while everything else is broken. It survives people quitting, phones on silent, and escalation chains nobody rehearses until 3am on a holiday weekend.
Two of the defaults for that job died within a year of each other. Atlassian's migration page states that Opsgenie runs "without interruption until April 5th, 2027," and after that "the product will no longer be accessible; any customer data in Opsgenie that has not been moved will also be deleted." Grafana OnCall OSS, the answer everyone gave when the reply was "self-host it," went into maintenance mode in March 2025 and was archived in March 2026. I put the full comparison, with per-vendor screenshots, on DevToolLab.
This Is a Deletion Date, Not a Deprecation
Worth being precise about the difference, because it changes your timeline. Plenty of products get "deprecated" and then quietly run for another five years. This is not that. Atlassian published a calendar date and attached data destruction to it. Rotations, escalation policies, integrations and incident history all have to be somewhere else by April 5, 2027 or they stop existing.
Atlassian's own destination is Jira Service Management, which absorbed the alerting and on-call features, and they ship automated migration tooling into it.
The Free Self-Host Option Closed Too
Grafana OnCall OSS carried the self-hosted recommendation for years. Its docs now open with a caution notice: archived as of March 24, 2026, development continuing in Grafana Cloud IRM instead. The repository was relocated to grafana-cold-storage/oncall and GitHub flipped it read-only on June 5, 2026.
The last release still runs. Nobody is shipping patches for it, which for something holding your paging path is a different risk calculation than for a CLI tool.
Six Replacements, Priced
Every figure here came off the vendor's own pricing page, checked September 18, 2026. Annual rates are shown where the vendor publishes a discount.
| Tool | Entry | Mid | Free tier | Open source |
|---|---|---|---|---|
| PagerDuty | $25 Professional | $49 Business | 5 users | No |
| incident.io | $19 Team | $25 Pro | Basic, forever | No |
| Rootly | $20 Essentials | Enterprise on request | No | No |
| FireHydrant | $25 Pro, annual | Enterprise on request | 10 responders | No |
| Jira Service Management | $20 per agent | $51.42 Premium | 3 agents | No |
| Keep | Self-hosted, free | Cloud on request | All of it | MIT outside ee/
|
PagerDuty is the incumbent and charges accordingly: $25 per user monthly, $21 annually, with Business at $49 or $41. Its real product is the integration catalog, so if your alert sources are a mess it is the shortest path. The bill stings as soon as you are paying full seats for engineers who take the pager twice a year.
incident.io treats the Slack or Teams channel as the incident surface instead of a separate console. The Basic tier is genuinely free forever and covers single-team on-call and a status page. Team is $19 monthly or $15 annually, Pro is $25.
Rootly opens at $20 per user and markets an AI layer: similar-incident lookup, a scribe for meeting notes, in-incident chat help. Judge the scheduling and escalation primitives first, since those are the parts you are actually replacing.
FireHydrant is the one with a genuinely different pricing shape, and I think it is the most under-discussed option here.
It charges per responder, not per seat. If 25 people out of 200 engineers carry the pager, that gap against seat-based pricing is enormous. Its free tier covers 10 responders, which is the most generous starting point of any commercial option in the table.
Jira Service Management is the low-risk migration and the awkward one. Same vendor, first-party tooling, no export-reimport of history. It is also the vendor that just killed the product you are migrating off, and per-agent pricing is a different shape from what Opsgenie charged, so model the bill before committing.
Keep is the surviving open-source answer: MIT licensed outside its ee/ directory, 12,334 stars, v0.54.3 shipped September 9, 2026, commits landing continuously. It bills itself as AIOps and alert management rather than an on-call scheduler, so it is strong at correlating and deduplicating alerts across sources and thinner if what you want is a rotation calendar with reliable phone delivery. LinkedIn's oncall also still exists, BSD-2-Clause and roughly 1,257 stars, but its last commit was August 2025.
The Costs That Are Not on the Pricing Page
Three things decide the real number, and the original post works through each in more detail.
Seat versus responder pricing can differ by a multiple on identical headcount, so count who can actually be paged before you compare quotes. Voice and international SMS are metered almost everywhere and they are the feature that has to work; PagerDuty's free tier, for instance, includes 100 international phone and SMS notifications monthly, which a US-only rotation will never notice and a distributed one will hit fast.
The one that eats the schedule is integration rebuild. Rotations and escalation policies port fine. The long tail of systems pointed at your current webhook endpoints does not, and that inventory is always longer than anyone guesses. A webhook signature verifier is worth having open while you repoint them, and an uptime SLA calculator helps settle which alerts deserve a phone call at all.
Picking One
Already committed to Atlassian: Jira Service Management, with the bill modeled first. Running incidents in Slack: incident.io, cheapest commercial entry and the Slack model is the product. Small pager rotation inside a big org: FireHydrant, on responder pricing alone. Messy alert sources and no appetite for risk: PagerDuty. Set on self-hosting: Keep, with engineering time budgeted, because you are shaping an alert platform into an on-call tool rather than unboxing one.
Wrapping Up
A deletion date with data destruction attached is not a roadmap item to revisit next quarter. The commercial field sits between $19 and $25 at entry, FireHydrant has the free tier worth taking seriously, and Keep is the one maintained open-source option left standing.
Plan against the calendar instead of the feature matrix: inventory what pages you today, count your real responders, and run a live rotation on a trial while you still have the option to change your mind.
Top comments (0)