Self-hosting

Eureka Analytics runs on your own infrastructure as a set of containers. Every component is stateless except ClickHouse (analytics data), PostgreSQL (accounts and settings) and Redpanda/Kafka (the event log), which can be self-hosted or replaced by managed services.

Components

Service	Role	Port
collector	Receives tag and server hits (/v1/c, /v1/p.gif, /v1/s, /v1/config/{pid}.json)	8080
processor	Classifies actors, builds sessions, writes to ClickHouse in batches	—
api	Admin and reporting API, public API v1, prompt reporting	3000 (container 8080)
ui	Web app	5173 (container 80)
tag-cdn	Serves ah.js	8081
clickhouse, postgres, redpanda	Storage and event log	8123, —, 19092

Single server with Docker Compose

cp deploy/.env.example deploy/.env      # set every secret
docker compose -f deploy/docker-compose.yml --env-file deploy/.env up -d --build
# Collector http://localhost:8080 · Tag http://localhost:8081/ah.js · API http://localhost:3000 · App http://localhost:5173

The ClickHouse schema is created on first start. The API applies the PostgreSQL migrations on start.

With the demo data

docker compose -f deploy/docker-compose.yml --env-file deploy/.env --profile demo up -d --build
# then open http://localhost:5173/demo

The demo-data service generates 120 days of Summit Supply traffic (about 3.8 million events) the first time, then keeps generating live traffic.

Kubernetes

kubectl create namespace ah
kubectl -n ah apply -k deploy/k8s

The manifests deploy the collector (with a horizontal autoscaler, 3 to 60 pods), the processor (keep replicas at or below the number of Kafka partitions, 12 by default), the API, the app, the tag CDN, the demo traffic and an ingress for three hostnames (collect., cdn., app.). ClickHouse, Kafka and PostgreSQL are expected to be managed services (ClickHouse Cloud or the Altinity operator, Redpanda Cloud or MSK, Cloud SQL or RDS); set their URLs in the secret (deploy/k8s/secret.example.yaml, to replace with your secret manager).

Environment variables

Main settings in deploy/.env:

Variable	Purpose
CLICKHOUSE_PASSWORD, POSTGRES_PASSWORD	Database passwords
JWT_SECRET	Signs app sessions (32+ characters)
AH_IP_SALT_SECRET	Derives the daily salts used to hash IPs and cookieless visitor IDs. Required in production.
ANTHROPIC_API_KEY	Enables questions in plain English (Claude). AH_DISABLE_ASK=1 turns the feature off. AH_ANTHROPIC_MODEL and AH_ANTHROPIC_EFFORT tune it.
AH_INTERNAL_TOKEN	Protects internal endpoints (collection status)
STRIPE_SECRET_KEY, STRIPE_WEBHOOK_SECRET, STRIPE_PRICE_<PLAN>_<PERIOD>	Billing. Without them, plan changes apply immediately (development mode).
DEMO_ENABLED, DEMO_ASK_PER_HOUR, DEMO_ASK_DAILY_CAP	Public demo

Collector and processor (AH_ variables):

Variable	Default	Purpose
AH_MODE	standalone	collector, processor or standalone (both in one process, no Kafka: for small installs)
AH_PROPERTIES_FILE	properties.yaml	Properties, allowed hostnames, ingest keys, tag configuration. Reload with SIGHUP.
AH_KAFKA_BROKERS	localhost:9092	Kafka / Redpanda brokers
AH_CLICKHOUSE_DSN	clickhouse://default:@localhost:9000/ah	ClickHouse (native protocol)
AH_TRUST_PROXY_HEADERS	false	Read the client IP from CF-Connecting-IP / X-Forwarded-For behind a load balancer or CDN
AH_GEOIP_CITY_DB, AH_GEOIP_ASN_DB	—	MaxMind databases for geography and network
AH_FETCH_IP_RANGES	true	Download the bots' published IP ranges daily
AH_FLUSH_INTERVAL, AH_FLUSH_MAX_ROWS	1s, 50000	Batch size of database writes
AH_SESSION_TIMEOUT	30m	Server-side session split

Scaling

• The collector handles tens of thousands of requests per second per vCPU and scales horizontally.
• The processor scales with the Kafka partitions; events of a visitor stay in order within a partition.
• ClickHouse scales by shards (by property) and replicas; one cluster per region keeps data in its region.