Observability

System status

Models

error rate · picker shards · nonce / 1,000,000

Incoming vs served

per model · last 24 hours · 30-minute buckets

Tokens per second

served completions · last 6 hours · 1-minute samples

Error rate

per model · last 24 hours · 30-minute buckets

Error rate (5m)

Error rate (1h)

Error rate (24h)

Error breakdown

last 24 hours · capacity shown under By reason

Stacked by model so you can see which model is driving failures over time.

Capacity & load

per pool node · live

Inference pool

Loading nodes…

Loading pool capacity...

Each table is one pool node. Charts and model error rates above are the combined pool. Capacity figures are that node's view of network compute weight — not escrow ids, wallets, or host addresses. Usable shards are active runtimes still under the chain nonce cap (1,000,000).