GPU Watchdog — Admin
alerts, bot connection, history
admin
Dashboard
Log out
Sign in
Control token
Sign in
GPU controls
two clicks — arm, then confirm
Primary
—
Holding the card
—
Resident in VRAM
—
Open connections to Ollama
—
Stop request
Free the GPU
Force stop
Secondary
—
Resident in VRAM
—
Callers from this box
—
Free the GPU
force stop not available
Telegram bot
token is write-only — never displayed back
Bot token
Chat ID
Detect
Save bot
Send test message
Alert rules
edge-triggered, with cooldowns — never one message per request
Alerts enabled
GPU active / idle
Temperature above
°C, repeat every
min while hot
Thermal throttling
GPU stops reporting
Unmetered GPU load
Save rules
Alert history
Refresh
When
GPU
Event
Message
Delivery
no alerts sent yet