What the fleet is doing · live
–
news items / 24h
–
live sources
–
cross-bloc corroborations
–
entities tracked
–
vectors indexed
a 24/7 news-corroboration engine — synthesized by a local multi-model LLM stack
–
AI stack · live
–
status
–
models resident
–
VRAM in models (GB)
–
tokens generated*
–
throughput (tok/s)
–
requests active
one iGPU serving the IDE, the news engine and the story world — OpenAI-compatible, on-prem. *lifetime tokens, accumulated across model restarts (best-effort).
Home network · live
–
devices online
–
devices seen
–
subnets
–
mesh peers (up/total)
–
containers
–
hosts reaching internet
–
security risk / 100
live device discovery + service fingerprinting across the LAN & NetBird mesh — aggregate counts only, no device detail leaves the network.
Automation health · live
–
–
–
–
the scheduled jobs with no UI of their own — nightly backups, hera-pi's dynamic-DNS + port-forward refresh, and the collector behind this page. Each is checked by its output — did it run, succeed, and recently? — not by a ping, so a silent failure surfaces here instead of staying invisible.
What my services use
–
RAM · used / fleet (GB)
–
VRAM models · used / fleet (GB)
–
CPU · used / fleet (cores)
–
Services
Fleet capacity (used / total)
Fleet compute · hardware (fixed specs)
–
CPU threads
–
RAM (GB)
–
VRAM (GB)
–
AI compute (TOPS)*
–
Raw compute · FP32 (TFLOPS)†
–
Storage (TB)
*AI TOPS = vendor peak, mixed precision (AMD NPU INT8 + NVIDIA FP4). †Raw = estimated peak FP32 (GPU+CPU). Ballpark figures, not a benchmark.