Changelog

What changed.

Two things ship here: the hosted control plane, and the open-source agent in tagged releases. Dashboard changes need nothing from you. Agent changes arrive when you pull the image.

DockTail Cloud DockTail agent All agent releases on GitHub →

DockTail Cloud

cloud.docktail.org

  1. A public vantage, and receipts for alerts

    • –Funnel-exposed services are now probed by Cloud itself, from the open internet. "Up locally but nobody outside can reach it" is a verdict, not a guess.
    • –Every alert shows whether it was actually delivered.
    • –Docker down/up events go through the same flap detection as checks, so a restart loop opens one incident instead of one per restart.
    • –Incident log capture is on by default. Workspaces that had switched it off stay off.

  2. Onboarding, redone

    • –Notifications are set up before you pick a plan. Email to the workspace owner is on automatically, so no workspace can end up with alerts and nowhere to send them.
    • –A workspace without a plan sees the pricing table right on the overview instead of a banner.

  3. Homelab annual at $30

    • –Pay Homelab yearly and it is $30, half of what twelve months at $5 would cost. Monthly stays at $5.

  4. Fleet ribbon, Business plan, free trials

    • –The overview leads with a fleet health ribbon: up, down and empty per time slice, with a 365-day view. Host-down paints every service on the host as down instead of leaving a gap.
    • –Business plan: 30 hosts included, expandable to 100, 365-day health history, 15 alert channels, advanced alert routing, priority email support.
    • –Homelab and Team open with a 30-day trial that carries the full plan.
    • –Health history is stored as transitions instead of raw check results, with a retention window per plan.
    • –An unhealthy Docker healthcheck is treated as critical, not a warning.

  5. Tailnet vantage from the control plane

    • –The tailnet vantage now reads the Tailscale control plane, so "awaiting approval", "not advertised by this host" and "definition gone" are distinct states. One credentialed agent per tailnet answers for all hosts; credentials never leave it.
    • –A host whose agent goes silent is shown as host offline, and one outage opens one incident rather than several.

  6. Clearer failure reasons

    • –Incidents say what actually happened instead of a generic down. An OOM kill is only called an OOM kill when Docker confirms it; a plain exit 137, a restart loop and a failing healthcheck each get their own wording.

  7. Enrollment windows for workspace keys

    • –A key accepts new hosts only during a window you choose, from 15 minutes to 24 hours, and can be reopened from Settings → Keys. Each key shows the hosts bound to it.
    • –The DockTail agent is marked as such in the host's workload list, and reports its own release version.

  8. The rest of the box

    • –Containers without docktail.* labels are listed per host with image, state, ports, CPU and memory, and can be watched for Docker-event incidents. Services and other containers share one Workloads table.
    • –Host vitals behave on Proxmox LXC guests, and the services table filters by compose project.

DockTail agent

ghcr.io/marvinvr/docktail · from GitHub releases

Agent releases are tagged on GitHub with their notes. This list loads from there.

Releases on GitHub

Pull ghcr.io/marvinvr/docktail:latest to update. Older agents keep working with Cloud and simply report fewer signals.