Slurm, Kubernetes, Docker & Linux collectors — read-only infrastructure monitoring. See the engine →
◎ LightOS FabricSign upLaunch console
Infrastructure & AI observability

The Operating System for AI Fabrics

Monitor your workloads and measured resources across Slurm, Kubernetes, Docker and Linux. Connect AI traces and explore fabric decisions separately.

CONTROL PLANEGPU RACKS · RDMATELEMETRY ↑OCS · MEMS12 PORTS

4

infrastructure collectors

OTLP

Python & Node traces

Read-only

infrastructure access

120

recent snapshots per environment

One control plane

Measure workloads. Explore decisions.

LOCK-FREE RING BUFFERZERO-COPY · 100µs BINSB0B1B2B3B4B5B6B7NIC 0NIC 1CONSUMERREPLAY LOG

Module 01 · Ingest

Measurements from your infrastructure

Read-only collectors report available resource measurements and workload state from Slurm, Kubernetes, Docker and Linux hosts.

  • Outbound HTTPS collection
  • Revocable environment credentials
  • Missing hardware measurements remain unavailable
GPU 07 · NCCLENGINECLOSED-LOOPQ3 · DEPTH 1,480 / TRIP 1,200CONGESTED · 18%HEALTHY TRUNKREROUTED · 82%

Module 02 · Decide

Reroute congestion before packets drop

Explore congestion and GPU-skew rules in the fabric simulator. Decisions are suggestions; this app does not execute hardware commands.

  • Queue depth trip at 1,200 packets
  • Straggler detection at 15% skew
  • Every action logged and replayable
OPTICAL WARDENOCS · MEMS ARRAYPATH A · LOSS 2.2 dB · BER 1e-9MAKE-BEFORE-BREAKPATH B · LOSS 0.4 dB · ACTIVEGPU STACKZERO-DROP SWITCH

Module 03 · Protect

Explore optical health decisions

Simulated optical loss and bit-error measurements demonstrate suggested make-before-break actions. Real optical monitoring requires a vendor-specific source.

  • Pre-fail at 2.2 dB loss or BER 1e-9
  • Clearly labeled simulated feed
  • No hardware dispatch or zero-drop guarantee

Watch your fabric think

Open the built-in simulator, inject a fault, and inspect suggested actions. Hardware commands are not executed.

Get started