Skip to main content

Notifications (Messages)

EcoLink raises in-console notifications for the events you care about — low balance, instance termination, team invites, model deployments. They show up on the bell icon in the top bar and in the Messages page.

What triggers a notification​

Balance warnings​

As your balance approaches zero with running resources, notifications fire at 60 / 30 / 20 / 10 / 5 / 3 / 1 minutes of remaining runtime (based on your current burn rate — sum of hourly rates across all active GPU instances, clusters, and inference instances).

Severity escalates as the clock gets shorter:

  • warning at 60 / 30 / 20 min
  • critical at 10 / 5 / 3 / 1 min

Instance / cluster / inference lifecycle​

  • Instance terminated — duration reached — a GPU instance or cluster auto-stopped at its estimated duration
  • Instance terminated — credit depleted — balance hit $0 with the resource still running
  • Instance terminated — failed — a pod crashed and couldn't recover
  • Instance preempted — a GPU instance was preempted by the platform (temporarily stopped); can recover when capacity frees up
  • Instance recovered from preemption — a preempted instance is running again

Team​

  • Invite accepted — someone accepted your team invite
  • You were invited — someone invited you to their account

Platform​

  • Maintenance notice (rare) — scheduled maintenance windows
  • Service disruption (rare) — incidents affecting your resources

Where to find notifications​

Bell icon​

The bell icon in the top-right of the console shows a red dot + count when you have unread messages. Click to see the 5 most recent; "View all" links to the full Messages page.

Messages page​

Console → Messages (under Account) — full list, most recent first. Each row shows:

  • Icon — severity (info / warning / critical) with matching color
  • Title — one-line summary
  • Message — full detail (resource ID, amounts, relevant links)
  • Time — relative timestamp

Click Mark all read to dismiss the red dot.

What NOT to expect notifications for​

We intentionally don't raise notifications for:

  • Successful API calls (every one would be spam)
  • Auto-scaling or replica changes (platform models)
  • Pod restarts inside a running inference instance (the instance stays available)
  • Routine storage billing (daily, predictable)