Feiten gathers air quality data from AirGradient ONE indoor air monitors and shows it on a self-hosted dashboard.

Sensors publish their readings over MQTT to a VerneMQ broker. An Elixir/Phoenix backend subscribes to the broker, persists each reading into a TimescaleDB hypertable, and serves a JSON API. A React frontend renders the readings as time-series charts.

Currently deployed in production on a homelab, behind Cloudflare Access. Code hosted on self-hosted Gitea instance, using Gitea CI and Docker container registry. Deployed using Portainer.

docker compose up -d        # Postgres (TimescaleDB) + VerneMQ
cd backend
mix setup                   # install deps, create + migrate DB, run seeds
mix phx.server              # http://localhost:4000

The backend reads its database and MQTT settings from environment variables (see backend/config/dev.exs for defaults). The MQTT subscriber starts with the application; set config :feiten, start_mqtt?: false to disable it.

cd frontend
npm install
npm run dev

docker-compose.prod.yml runs the full stack — TimescaleDB, VerneMQ, the backend release, and the Caddy-served frontend — using prebuilt images from the project registry.

Copy .env.prod.example to .env.prod and fill in the required values.

Optional settings (PHX_HOST, HTTP_PORT, MQTT_INTERNAL_PORT, MQTT_PUBLIC_SCHEME, MQTT_PUBLIC_PORT, POOL_SIZE) have sensible defaults documented in .env.prod.example.

docker compose -f docker-compose.prod.yml --env-file .env.prod up -d

VerneMQ exposes a plaintext MQTT listener on MQTT_INTERNAL_PORT (default 1883). This is appropriate for a trusted LAN; MQTT credentials traverse the network in plaintext. For internet-facing deployments, put a TLS terminator in front of the broker.