Install with Docker

Stache is a single container. The embedded SQLite database, photos, and AI model cache all live in one data volume — there’s no database service to run alongside it.

Requirements

• Docker with the Compose plugin (docker compose version)

Compose file

services:
  app:
    image: ghcr.io/graphichealer/stache:latest
    restart: unless-stopped
    environment:
      PORT: 8080
      DATA_DIR: /data
      TZ: ${TZ:-UTC}   # digest "time" setting uses this timezone
    volumes:
      - data:/data   # SQLite db + photos + model cache all live here
    ports:
      - "8080:8080"
    # Lets the AI provider URL reach services on the Docker host
    # (e.g. Ollama at http://host.docker.internal:11434/v1).
    extra_hosts:
      - "host.docker.internal:host-gateway"

volumes:
  data:

Start

docker compose pull
docker compose up -d

Open http://<host>:8080, create the first household account, then head to Labels to print your first batch.

Verify it’s healthy:

curl localhost:8080/api/health

Memory

The app idles around 200 MB. Photo-similarity search (CLIP) loads lazily on the first photo upload or search — once warm the app sits around 450–550 MB. On very small hosts, set STACHE_NO_ML=1 to disable embeddings entirely; everything else keeps working.

One caveat on plain HTTP

The live camera viewfinder, clipboard copy buttons, and offline PWA features need a secure context — localhost counts, http://192.168.x.x does not. Over http://<lan-ip>:8080 the app still works (photo capture falls back to the file picker), but for the full experience front it with HTTPS: a reverse proxy like Caddy or Nginx Proxy Manager, a Tailscale cert, or your Home Assistant ingress URL.

If you do proxy it, also set STACHE_BASE_URL (and STACHE_TRUST_PROXY=1 so client IPs survive the hop) — see Configuration.

Updating

docker compose pull && docker compose up -d

Migrations apply automatically at boot.

Backups

Everything lives in the data volume — backing up Stache is backing up that one folder. See Backups for the built-in scheduler.