Unified observability plane

Global Command Centre

See metrics, events, logs and traces for the agent fleet in one place — then dig into evidence when something breaks. Read-only; no autonomous remediation.

Open active P1 Evidence / logs

Read-only Agent OS

No shell execution
No cluster-admin
No secret reads
No database writes
No firewall changes
No autonomous remediation

Observability signals

Metrics · events · log pipelines · traces — unified fleet view

Nodes
840
infra
Clusters
31
infra
Agents
30
APM
Error events
4
events
P2
2
events
SLA risks
3
traces
Log drains
4
logs
Approvals
5
gates

Live telemetry

Fleet metrics & monitors

Streaming timeseries widgets and threshold monitors — Datadog-style dashboard layout, vendor-neutral simulated feed (1.5s ticks).

LIVEupdated now

CPU

41.7%

fleet avg

Latency p95

229ms

model gateway

Error rate

0.46%

production

Throughput

879rps

agent invoke

Agents busy

51%

active workers

CPU utilization

Agent hosts · last ~60s

Request latency

Gateway p95 · ms

Live event stream

Rolling ingest from collectors & agents

Collector heartbeat · eu-west-1
23:23:26
Latency probe elevated · model gateway
23:23:31
Collector heartbeat · eu-west-1
23:23:34
Integrity check · evidence artefact
23:23:30

Error rate

Failed invokes / total

Throughput

Requests per second

Active monitors

Threshold checks on live series — monitors-as-code pattern (no vendor lock-in)

Fleet CPU anomaly
ok
avg(last_5m):cpu.utilization{scope:agents}
41.7% · thr > 85%
Gateway latency p95
ok
avg(last_5m):gateway.latency.p95
229ms · thr > 400ms
Error rate spike
ok
sum(last_5m):errors.rate{env:production}
0.46% · thr > 2.5%
Log pipeline lag
ok
avg(last_5m):pipeline.lag.p95
1.7s · thr > 3s

Telemetry pipeline

Collectors forward structured container logs from agent sidecars — ops pattern analogous to Fluent Bit DaemonSets tailing /var/log/containers/*.log.

Log collectors (DaemonSet)
4/4 nodes
healthy
Structured JSON parse
cri-o · containerd
healthy
Export endpoint
EU residency
healthy
Pipeline lag p95
1.4s · live
healthy

Open evidence / log viewer

Application latency

Model gateway request performance — live p95 overlay on seeded providers

OpenAI

195 ms live · US / EU routing

healthy

Anthropic

225 ms live · US / EU routing

healthy

Google Gemini

255 ms live · US

degraded

Azure OpenAI

285 ms live · EU (Sweden Central)

healthy

AWS Bedrock

316 ms live · EU (Frankfurt)

healthy

Ollama (on-prem)

346 ms live · On-premise

healthy

vLLM Cluster

376 ms live · On-premise (sovereign)

healthy

Open Model Gateway

Infrastructure health heatmap

Composite health by customer and environment — see everything in one place

Customer	production	staging	dev	dr
FS Core Banking Platform	96	91	86	70
Nordic Payments Rail	99	99	84	79
Card Issuing Services	99	93	88	83
Grid Telemetry Fabric	88	83	78	73
SCADA Edge Estate	85	80	75	70
Clinical Data Platform	93	88	83	78
Imaging AI Workloads	99	98	82	77
National Registry Services	96	91	86	70

Error & incident timeline

Historical event volume (seeded) — live series above for last-minute fleet health

Token spend and retry waste

USD per day across all tenants

Active incidents

Ordered by severity and SLA exposure

Why is fs-prod-cs-tool2 NotReady?

inc-4821 · FS Core Banking Platform · production · SLA at risk

Payment rail latency above 850ms p95

inc-4818 · Nordic Payments Rail · production · SLA at risk

SCADA edge cluster losing telemetry batches

inc-4809 · SCADA Edge Estate · production

Clinical DB replica lag exceeding 90s

inc-4802 · Clinical Data Platform · production · SLA at risk

Recurring incident patterns

Signature clustering across 30 days

Registry egress reset during image pull

3 tenant(s) · last seen 2026-08-02

Kafka consumer rebalance storm

2 tenant(s) · last seen 2026-07-30

Replica lag during nightly ETL

1 tenant(s) · last seen 2026-08-01

Edge collector telemetry drops

1 tenant(s) · last seen 2026-08-01

Gateway config rollout latency regression

2 tenant(s) · last seen 2026-08-02