Real-time service monitoring with Chernoff faces. Spot outages at a glance — happy faces mean healthy services, angry faces mean something's wrong.
ChernoffOT queries any Prometheus-compatible endpoint and turns your services into expressive cartoon faces. Point it at your existing Prometheus, Amazon Managed Prometheus, Google Managed Prometheus, Grafana Cloud, or any other PromQL-compatible backend.
All services healthy
Some services degraded
Quick Start
Docker
docker build -t chernoffot .
docker run \
-e PROMETHEUS_URL=https://your-prometheus-endpoint \
-p 3000:3000 \
chernoffotOpen http://localhost:3000 — you'll see faces for every discovered service.
Docker Compose
Copy the provided docker-compose.yml, set PROMETHEUS_URL in a .env file, then:
echo 'PROMETHEUS_URL=https://your-prometheus-endpoint' > .env docker compose up
From Source
bun install PROMETHEUS_URL=https://your-prometheus-endpoint bun run dev
Prerequisite: Bun runtime (
curl -fsSL https://bun.sh/install | bash)
How It Works
Each service in your infrastructure becomes a face. By default, facial features encode these four outage signals:
| Feature | Default Metric | Good → Bad |
|---|---|---|
| Mouth | Error rate | 😊 Smiling → 😟 Frowning |
| Eye Size | Latency (p95) | 😑 Calm → 😳 Wide open |
| Brow Angle | Traffic anomaly | 😌 Relaxed → 😠 Angry |
| Head Color | Health score | 🟢 Green → 🔴 Red |
Sweat drops appear when a service is critically unhealthy (< 30% health).
You can remap which normalized metric drives each facial feature from the in-app mapping modal (? button in the header).
Architecture
Your Services → Prometheus-compatible backend → ChernoffOT → 🎭 Faces
ChernoffOT auto-discovers your services, classifies them by type (app, database, cache, queue, worker), and fetches type-appropriate metrics via PromQL. No manual service registration required.
ChernoffOT works with any PromQL-compatible backend. Just point PROMETHEUS_URL at it:
| Provider | PROMETHEUS_URL |
|---|---|
| Amazon Managed Prometheus | https://aps-workspaces.<region>.amazonaws.com/workspaces/<id> |
| Google Managed Prometheus | https://monitoring.googleapis.com/v1/projects/<project>/location/global/prometheus |
| Azure Monitor Prometheus | https://<workspace>.prometheus.monitor.azure.com |
| Grafana Cloud | https://prometheus-prod-XX.grafana.net/api/prom |
🚧 Coming soon: zero-config bundled Prometheus. Join waitlist to get notified→
ChernoffOT currently issues direct requests to PROMETHEUS_URL and does not add auth headers or cloud request signing. For secured endpoints, route through an authenticated proxy/gateway and point PROMETHEUS_URL to that proxy.
Configuration
All configuration is via environment variables with sensible defaults:
Core
| Variable | Default | Description |
|---|---|---|
PROMETHEUS_URL |
http://localhost:9090 |
PromQL-compatible endpoint |
PORT |
3000 |
Port the ChernoffOT server listens on |
REFRESH_INTERVAL |
15 |
Dashboard refresh interval in seconds |
Service Discovery & Labels
| Variable | Default | Description |
|---|---|---|
LABEL_SERVICE |
service_name |
Primary label identifying services |
SERVICE_LABEL_ALIASES |
service_name,service |
Comma-separated fallback label names |
LABEL_STATUS |
http_response_status_code |
Label used for HTTP status codes |
METRIC_DURATION |
http_server_request_duration_seconds |
Primary request duration metric name |
SERVICE_TYPE_OVERRIDES |
— | Force service types: svc1=db,svc2=cache |
Tuning
| Variable | Default | Description |
|---|---|---|
DISCOVERY_CACHE_MS |
60000 |
How often to re-discover services (ms) |
CLASSIFIER_CACHE_MS |
600000 |
How often to re-classify service types (ms) |
METRICS_CACHE_MS |
15000 |
How often to re-fetch metrics (ms) |
ACTIVE_WINDOW_SEC |
300 |
Seconds before a service is considered missing |
MISSING_RECENTLY_SEC |
3600 |
Seconds before a missing service is considered stale |
CATALOG_TTL_SEC |
604800 |
Seconds before a stale service is pruned (7 days) |
Service Classification
ChernoffOT automatically classifies services by inspecting which metrics they expose:
| Type | Detection Signals |
|---|---|
| Application | http_server_*, http_requests_*, grpc_server_* |
| Database | postgres_*, mysql_*, mongodb_*, lock/connection metrics |
| Cache | redis_*, memcached_*, hit/miss, eviction metrics |
| Queue | kafka_*, rabbitmq_*, nats_*, lag/depth metrics |
| Worker | job_*, task_*, queue consumer metrics |
Each type uses different PromQL queries optimized for that workload. Override with SERVICE_TYPE_OVERRIDES=my-redis=cache,my-pg=db.
Compatible Backends
Any backend that speaks PromQL:
- ✅ Prometheus / VictoriaMetrics / Thanos / Cortex / Grafana Mimir
- ✅ Amazon Managed Prometheus / Google Managed Prometheus / Azure Monitor
- ✅ Grafana Cloud / Coralogix / New Relic (Prometheus endpoint)
- ❌ Datadog (no PromQL support)
Tech Stack
- Runtime: Bun — fast JavaScript/TypeScript runtime
- Frontend: Vanilla JS + SVG (no framework, no build step)
- Backend: Single TypeScript file, ~2200 lines
- Dependencies: Zero runtime dependencies
- Container: Alpine-based, < 100MB image
Development
bun install PROMETHEUS_URL=http://localhost:9090 bun run dev
License
MIT