YAAD GPU LAB · OPERATORKill switches
Public execution (broker may claim approved jobs)
New submissions
Auto-approve new requests and teams (skip the approval queue)
Team workspaces (new teams + key changes; deletes always run)
Compute pool (teams and their agents submit batch jobs; the broker places them by capacity)
The CSI broker also has its own local switch (execution_enabled) — either one off stops dispatch.
Sparks (team capacity)
| Spark | Health | Teams | Disk free | Reported |
|---|
Help desk
Every request from students and their AI agents. The automatic host answers most tickets within minutes. Waiting for operator approval means the fix needs you. Open the ticket, decide, reply, and set the status.
| Updated | Status | Team / Spark | From | Problem |
|---|
Teams
| Team | Status | Spark | Members | Disk / procs / GPU |
|---|
End of event: deletes every team account and all its data on every Spark. Verify afterwards with python -m src.main teams wipe-all --confirm on the broker.
Awaiting approval
| Requested | By | Capability |
|---|
Recent jobs
| Requested | By | Status | Node | Result |
|---|
Who can use the lab
Anyone on the list below, plus any confirmed account at an allowed domain.
Allowed email domains
Currently allowed: loading…. Anyone with a confirmed email at these domains can sign in.