Autonomous observability · WebSocket streaming

Autonomous full-stack observability with self-healing intelligence.

Observability that closes its own incidents.

Multi-cluster Kubernetes, container fleets, synthetic probes and chaos experiments in one real-time command center — with incident intelligence that groups the noise, verifies recovery and writes the post-mortem.

Join early access

PRIVATE EARLY ACCESS · OFFICIAL RELEASE Q2 2027

devoptic / command-center

STREAMING LIVE {{ wsLatency }}ms · {{ msgs }} msg/s {{ envClusterCount }} clusters · {{ envOpen }} open

Event waterfall {{ envLabel }}

Cluster fabric

{{ c.name }} {{ c.pods }} pods

{{ c.cpu }}%

Anomaly triage

Acknowledged

Container saturation

orders-worker-3 · cpu 94% · 6m · {{ firstCluster }}

14:41:52 WARN queue depth 4,812 · consumer lag 38s

14:42:03 WARN gc pause 412ms · heap 91%

acked by @mira · 2m ago · fingerprint svc:orders-worker

Investigate

Tail logs

6.5M+ metrics per node, minute-bucketed and indexed

<50ms dashboard query budget at fleet scale

90%+ alert noise removed by semantic fingerprinting

0 lost alerts — transactional outbox delivery

The story of an incident

34 minutes, one page, zero manual cleanup.

eu-west-k8s-prod · GKE

Open

T+00:00pod/payments-7f9c CrashLoopBackOff · restart 1

T+00:09docker orders-worker-3 cpu 97% · saturation

T+00:21pod/payments-a81b CrashLoopBackOff · restart 2

T+00:34pod/payments-c3d9 OOMKilled · exit 137

14alerts a traditional stack would page on in the next minute

VIGIL

silence_detected

payments-api stopped emitting logs for 90s. The surrounding buffer is frozen before the pods are rescheduled and the evidence disappears.

14:41:52 WARN pool=pg-payments wait 1840ms

14:41:58 WARN heap 241Mi / 256Mi

14:42:03 ERROR alloc failed · runtime: out of memory

14:42:03 ·· no output 90s ··

14:43:33 VIGIL snapshot sealed · 50 lines

50-line snapshot · MinIO

s3://forensics/payments-api/14-42.log

Fingerprint engine

Investigating

14→1persistent incident, one page to on-call

compose_servicepayments-api

container ids7f9c · a81b · c3d9 · …

contributorsk8s.pod.restart · docker.cpu_spike

Stability window

Resolved

6/6 promoter checks · every 5 min30m stable

# PIR — INC-2291 payments-api
duration      34m · auto-resolved
root cause    OOMKilled · limit 256Mi
evidence      s3://forensics/payments-api/14-42.log
pages sent    1 (would have been 14)

PIR delivered to #inc-payments

01Incident intelligence

Fourteen restarts. One incident.

Fingerprinting strips ephemeral container IDs and groups alerts by service identity, so a crash-looping replica set pages once. A six-stage state machine then tracks it to recovery and closes it for you.

fingerprintcompose_service → container_name → service_name → host + image. UUID and hex IDs discarded.

relapse memoryRecurs within 7 days: the original investigation reopens. After that, it is linked as a relapse.

auto-resolutionA promoter daemon checks every 5 minutes and closes after 30 minutes of confirmed stability.

post-incident reviewMarkdown PIR with timeline, blast radius, contributor keys and resolution notes.

INC-2291payments-api

RAW ALERTS · 14 payments-api-7f9c2d1e payments-api-a81be04f payments-api-c3d90b77 + 11 more

→

FINGERPRINT svc:payments-api 1 incident · 1 page

RESTARTS / MIN30 MIN WINDOW

Replay lifecycle

# PIR — INC-2291 payments-api
duration       41m · auto-closed
blast radius   3 replicas · {{ firstCluster }}
root cause     OOMKilled after config v2.14 (limit 256Mi)
contributors   k8s.pod.restart · docker.oom · gitlab.pipeline#4918
resolution     limit raised to 512Mi · 30m confirmed stable

Multi-cluster health · {{ envLabel }}

ok pressure critical

{{ c.name }} {{ c.prov }} {{ c.pods }} pods · {{ c.deploys }} deploy · {{ c.sts }} sts · {{ c.ds }} ds

CPU

MEM

k8s.pod.cpu_spike14:02:11 ⇄ k8s.pod.cpu_normal14:06:40 severity P2 → P4

02Kubernetes fabric

Every cluster, watched at once.

Independent watchers boot for each connected cluster — EKS, GKE, AKS or bare metal. No context switching, no cluster left unobserved. Crash loops, scheduling stalls and node pressure surface before users feel latency.

per-cluster workersWorkloadReadinessWatcher, K8sMetricWatcher and an alerting engine for every cluster.

workload coveragePods, Deployments, StatefulSets and DaemonSets.

tri-pillarPrometheus metrics, Loki logs and Grafana panels in one console.

contributor keysA spike paired with its matching normal event de-escalates severity automatically.

03API testing & chaos

Break it on purpose. Watch it recover.

Inject latency, error bursts and packet drops against a live service and assert that fallbacks hold. Time-aware baselines learn daily and weekly traffic curves, so real regressions stand out from normal peaks.

chaos engineControlled latency, HTTP error codes and network drops.

time-aware baselinesOutliers flagged by deviation from the learned curve, not static thresholds.

contract guardPayload drift and schema regressions caught continuously.

live presenceSee who else is editing a test suite, in real time.

Chaos playground

Injected latency{{ latency }}ms

Error burst

▌

Run experiment

Drag past 750ms or pick 25% to trip an incident.

VIGILbilling-svc · log volume

2 anomalies

ERROR SPIKE +340%

SILENCE

14:0014:1514:3014:4515:00

Forensics snapshot captured · MinIO

s3://forensics/billing-svc/14-42.log · TTL 30d

Download .log

04Vigil log intelligence

The log lines you would have grepped for, already saved.

Vigil tails every container and pod, flags volume spikes and error-rate escalations, and notices when a service goes quiet. On anomaly it freezes the surrounding buffer to object storage — the evidence is there before anyone opens a terminal.

EMA detectionStreaming anomaly scoring per container and pod.

silence sentinelsilence_detected fires when a worker stops emitting logs entirely.

compliance hookscompliance_violation events on policy-matching lines.

forensics freezePre- and post-incident buffers written to MinIO or S3 with retention TTL.

The rest of the command center.

Six more domains · one console

Docker fleet

Sub-50ms overviews over multi-million-row metric tables — four round-trips, maximum. Saturation and failing health checks surface within 30 seconds.

{{ c.cpu }}%

Acknowledge anomaly

Acknowledged

PATCH /acknowledge-anomaly · audited

Synthetic endpoints

Dual-path probing: downtime is declared only when the central scheduler and an edge gateway agree. p50/p95/p99, status codes and certificate expiry.

/v1/checkoutp99 42ms

/v1/auth/tokenp99 88ms

/v1/catalog/searchp99 412ms

SLA 30D99.994%

SSL EXPIRY61 days

PATHS2/2 agree

Topology & service catalog

A service map drawn from observed traffic, not a diagram someone forgot to update. Tier and owning team come from the catalog.

payment-gatewayT1 · team-payments

p95 18ms · 2.1k/min · 0.0% err

auth-serviceT1 · team-identity

p95 214ms · 640/min · 0.4% err

stripe-apiexternal

GitLab deploy attribution

Pipelines, commits and merge requests overlaid on production metrics. When an incident fires, you know whether the last deploy did it.

#4921 · @devops · +12ms

checkout-api · p95 latencylast 2h

Real-time streaming fabric

A WebSocket hub on a decoupled event bus. Incidents, state changes and telemetry are pushed to the browser — nothing polls, nothing refreshes.

PUSH LATENCY{{ wsLatency }}ms

THROUGHPUT{{ msgs }}/s

DELIVERED{{ totalMsgs }}

Environments & access

Pick an environment once and every metric, incident and cluster view follows. RBAC, tenant isolation, TOTP and an immutable audit log.

Try it — the console and cluster matrix above follow your pick.

Architecture

Event-driven from agent to browser.

01Agents & edge gatewaysDocker, Kubernetes, synthetic probes

02Ingest · GoMinute-bucketed, indexed time series

03EventBusDistributed or in-memory pub/sub

04WebSocket hubMultiplexed, targeted broadcast

05Console · React 19Transactional outbox for notifications

GoGorilla WebSocketReact 19PrometheusLokiGrafanaMinIO / S3GitLab CIEKS · GKE · AKS

Point it at a cluster. Watch the noise drop.

Early access opens in controlled waves ahead of the Q2 2027 release. Connect a Kubernetes context or a Docker host and see your first fingerprinted incident in minutes.