← overview
Workers pulse in orange during inference · links flash during activity · IT sees the infra come alive
ADMIN DASHBOARD
Topology tab · lightweight reactive SPA
Multi-WebSocket (one per paired peer)
push events + federation pulse
CLUSTER
virtual pool
PRIMARY POOL
HQ site
admin WebSocket
GPU SATELLITE
compute datacenter
admin WebSocket
local admin WS
paired peer WS
worker-a
● busy
worker-b
● busy
worker-c
idle
worker-d
idle
worker-e
idle
job_start
tokens/s
Real-time flow:
1. Client calls the inference API
→
2. Pool assigns the right worker
→
3. Worker computes · animated pulse
4. Tokens streamed (live tps)
→
5. Worker emits job end + metrics
→
6. Pool propagates push · dashboard repaints
« See your models work in real time on your machines, across your sites,
without anything leaving your infrastructure — that's what the cloud will never give you. »
Worker busy (1.2s pulse)
Worker idle
Active link (flash)
WebSocket push (cross-pool)
• Broadcast admin WebSocket events
• Emitted on dispatch / job end / timeout
• Metadata: worker, tokens/s, model used
• Non-blocking async (zero added latency)
• Server-side filtering per admin session
• Pure CSS animations (pulse + flash keyframes)
• DOM targeting via data-attributes
• Busy workers + in-flight jobs state preserved
• Survives periodic re-render
• Unified animation regardless of pool
• One WebSocket per non-revoked paired peer
• Differential reconnect: only on change
• Auto-reconnect on drop
• Unified control-room view
• No intermediate proxy (zero cloud relay)