Resources · Article

Grafana OnCall OSS is archived. Here are your three paths

Jose Escrich

Founder, WHAWIT · September 4, 2026

If you self-host Grafana OnCall, the code went read-only on 2026-03-24 and, unless you had wired your own Twilio account, the phone stopped ringing the same day. Here is what changed, what still runs, and three realistic ways forward.

What happened, and when

OnCall OSS entered maintenance mode on 2025-03-11 (Grafana Labs published the announcement the next day), with full archival scheduled for 2026-03-24. The repository now lives at grafana-cold-storage/oncall, read-only, AGPL-3.0, with 3,887 stars and 424 forks at the time of writing. Its README opens with: “On 2025-03-11 Grafana OnCall (OSS) entered maintenance mode and was archived on 2026-03-24.”

The same day Grafana switched off the Cloud Connection, the link to Grafana Cloud that carried SMS, calls and mobile push for installs that had not configured their own Twilio credentials.

What stopped working on 2026-03-24

According to Grafana’s OSS set-up documentation:

• SMS and phone calls that relied on the Cloud Connection are no longer supported.
• Mobile app push notifications for OSS users are no longer supported. The Grafana IRM mobile app worked with self-hosted stacks only until 2026-03-24.
• New releases: a read-only repository does not receive fixes.

What should still run, per the repository’s own feature list: schedules, escalation chains, integrations, Slack and Telegram notifications — for as long as your containers do. What is gone, unless you run your own Twilio, is the part that wakes people up.

Grafana’s own suggestions are to bring your own Twilio, Exotel or Zvonok account for SMS and calls, to route push through Pushover, Gotify or ntfy, or to move to Grafana Cloud IRM.

The three paths

1. Grafana Cloud IRM. The supported successor. Per the Grafana Cloud pricing page, the Free tier is “limited to 3 active IRM users per month”; the Pro plan has a “$19 per month” platform fee that includes 3 active IRM users, then “$20.00 / active IRM user” beyond that. Migration is Terraform or the OnCall API; Grafana notes that its IRM migration tools “don’t currently support Grafana OnCall OSS as a source platform”, and every alert source has to be repointed at the new integration URLs. You get the mobile app back, and Grafana’s own AI-powered investigations. You also move your on-call off the infrastructure you chose OSS to keep it on.

2. Do it yourself. Keep the archived OnCall running, wire your own Twilio credentials for SMS and voice, and a Pushover, Gotify or ntfy webhook for push. Cheapest in licence cost. The trade: you are now the vendor of an unmaintained codebase that pages people at night.

3. WHAWIT. Native on-call (schedules, layered rotations, overrides, escalation policies) that pages over WhatsApp, SMS, phone call, Slack, Microsoft Teams, Discord and email, with an AI investigation that runs before the page and travels inside it. Your Grafana, Prometheus and Loki stay where they are; WHAWIT reads them through your own mcp-grafana server. We do not publish list pricing; engagements start as a pilot on your own incidents.

Connecting Grafana to WHAWIT today

What exists right now:

1. Run mcp-grafana next to your Grafana and mint a service account token. The endpoint must be reachable over HTTPS from outside your network: the SSRF guard rejects localhost and private ranges.
2. Add it from the connector catalog. Pick the Grafana card, paste the URL and token; the token is stored in Secret Manager, never in the app database.
3. Discover tools, then allow the ones you want. Everything arrives denied. Allow dashboards, Prometheus and Loki queries and alert state; the on-call agent teams and Radar use them as evidence, and every call lands on the incident timeline.
4. Optional: your own operations server. If you expose runbooks or restarts as an MCP server, WHAWIT can run them from the incident with approval gates: the approver sees the exact payload before anything reaches your infrastructure.

Logs and metrics today come from the connected providers (Datadog, New Relic, CloudWatch, GCP, Azure, Sentry, BetterStack, Honeycomb) or from our Node OpenTelemetry SDK. A direct Loki connector exists in our codebase but is not exposed to customers yet.

What is not there yet

Two things a Grafana shop will ask for:

• A webhook endpoint for Grafana Alerting and Alertmanager. Today WHAWIT ingests alerts from Opsgenie (and from the connected log/metric providers), not from a generic contact point.
• An importer for OnCall OSS (schedules, escalation chains, integrations, routes), the way we import from Opsgenie. Not available today.

If either stands between you and a decision, tell us: it changes what we build next.

Choosing

Least change: Cloud IRM, at the cost of moving on-call to Grafana’s cloud and paying per active user. Lowest licence cost: DIY, until the first CVE you patch yourself. If you want the page to arrive with the investigation attached and your Grafana stack untouched, that is what the pilot tests, against your own incidents.

And if you also run Opsgenie: Atlassian ends Opsgenie support on 2027-04-05 — the REST APIs stop working that day, and data that was not migrated is permanently deleted. The same three paths apply.