Latest updateInvestigating22:38 UTC ()
Deploys are waiting up to 15 minutes for a runner. Running deploys are not affected.
- Identified22:56
Two runners lost their cache volume. They are being replaced; queued deploys start as they come back.
SolutionsPlatform teams
Platform, infrastructure, IT and SRE teams get a page their colleagues check before they ask. Each one stays inside: behind Google or Microsoft sign-in, a password, your office network or your app's sign-in.
Current status
Latest updateInvestigating22:38 UTC ()
Deploys are waiting up to 15 minutes for a runner. Running deploys are not affected.
Two runners lost their cache volume. They are being replaced; queued deploys start as they come back.
At a glance
Uptime, 30 days
99.58%Services operational
3of 4Active incidents
1Access
One access mode per page. It is checked before anything renders, and the badge, widget, JSON and feeds follow it, so nothing leaks through an embed.
Who gets in
Visitors from your office or VPN. Nobody has to sign in.
Starter and up
Everyone else sees
Downgrades
If a plan lapses, the page keeps refusing visitors it would have refused before. Switching it to public is always your call.
Who it's for
The systems everyone inside relies on rarely get a status page. Give each team one, and let colleagues subscribe from it.
Platform teams
Deploys, the internal gateway, queues. Engineers look before they ask, and an incident posted there emails everyone who subscribed.
IT
Mail, VPN, payroll. Staff check the page before they file, and the all clear reaches them the same way.
SRE leadership
Put each team's SLOs on one page with the error budget left, judged by the same rules everywhere.
You run it
The team that runs everything else sets Observer up. It reads what you already have, only dials out, and lives in your repository.
Sources
Prometheus, OpenTelemetry, CloudWatch, Loki, Elasticsearch and your databases, plus HTTP, TCP, DNS, TLS, gRPC and WebSocket checks, all from one agent inside your network.
Network
It reads inside your network and sends one number and its verdict per metric. Credentials stay on the host, and nothing connects in.
Install
Run it next to your data with Docker, as a single binary, or on Kubernetes with the Helm chart.
As code
Keep checks, thresholds and pages in observer.yaml, review them in pull requests, and apply them with the CLI or its GitHub Action. A dry run shows the diff first.
Coding agents
Connect the MCP server from Claude Code, Cursor, Codex or another MCP client, and your agent can add the checks and the pages for you.
Verdicts
Comparisons are strict, so a value equal to a threshold never matches, and a status change has to hold through a short confirmation window before a page flips.
Private metrics
Mark a metric private and it stays off every status page, internal ones included. Your team still hears when it breaks.
Off every page
A private metric is left out of the page, its uptime calendar, widget and Reliability Center, and it can never change a page's headline.
Still watched
Its status changes still reach your alert channels, and the console keeps it in the metrics list, marked private.
Scope
Same setup
Same agent, same metrics, same incidents. An internal page is a page with its access mode set, and it can go public again at any time.
Pages per plan
Starter runs 1 status page, Pro 3, Business 10, Enterprise as many as you need. Internal pages count like any other.
Where Observer stops
Rotations and escalation stay in PagerDuty or Grafana IRM, and finding the cause stays with your team and your AI SRE. Observer feeds them, and proves the impact.