services:
  docmost:
    image: ghcr.io/vvzvlad/gitmost:latest
    depends_on:
      - db
      - redis
      - embeddings
    environment:
      APP_URL: 'http://localhost:3000'
      APP_SECRET: 'REPLACE_WITH_LONG_SECRET'
      DATABASE_URL: 'postgresql://docmost:STRONG_DB_PASSWORD@db:5432/docmost'
      REDIS_URL: 'redis://redis:6379'
      # #530 semantic search: the GLOBAL embedding provider (the `embeddings` TEI
      # sidecar below). A workspace that configures its own embedding provider
      # overrides this. TEI is OpenAI-compatible, so the endpoint is /v1.
      EMBEDDING_ENDPOINT: 'http://embeddings:80/v1'
      EMBEDDING_MODEL: 'intfloat/multilingual-e5-small'
      EMBEDDING_API_KEY: 'unused'
      EMBEDDING_DIMENSIONS: '384'
      # MUST match the --revision the sidecar pins (see EMBEDDING_REVISION in
      # .env). A revision bump changes the embedding fingerprint (PR-2 swaps
      # generations); keep the two in lockstep.
      EMBEDDING_REVISION: '${EMBEDDING_REVISION}'
      # e5 models require these input prefixes; empty them for a non-e5 model.
      EMBEDDING_QUERY_PREFIX: 'query: '
      EMBEDDING_DOC_PREFIX: 'passage: '
    ports:
      - "3000:3000"
    restart: unless-stopped
    # The app already serves precompressed (brotli/gzip) static assets with
    # long-lived cache headers and gzips dynamic API responses. For the best
    # cold-load latency you can OPTIONALLY put a reverse proxy (caddy / nginx /
    # traefik) in front with HTTP/2 (or HTTP/3) and brotli enabled — none is
    # required for compression to work.
    volumes:
      - docmost:/app/data/storage

  db:
    # pgvector image (same Postgres major as postgres:18) so `CREATE EXTENSION
    # vector` succeeds for the page_embeddings RAG table.
    image: pgvector/pgvector:pg18
    environment:
      POSTGRES_DB: docmost
      POSTGRES_USER: docmost
      POSTGRES_PASSWORD: STRONG_DB_PASSWORD
    restart: unless-stopped
    volumes:
      - db_data:/var/lib/postgresql

  redis:
    image: redis:8
    command: ["redis-server", "--appendonly", "yes", "--maxmemory-policy", "noeviction"]
    restart: unless-stopped
    volumes:
      - redis_data:/data

  # #530 Text Embeddings Inference (TEI) sidecar — the GLOBAL embedding provider
  # for semantic search. OpenAI-compatible, reached only on the internal network
  # (no published port). The model + revision are PINNED so the embedding
  # fingerprint is stable; set EMBEDDING_REVISION to a real commit sha in .env.
  embeddings:
    image: ghcr.io/huggingface/text-embeddings-inference:cpu-1.9
    command:
      - "--model-id"
      - "intfloat/multilingual-e5-small"
      - "--revision"
      - "${EMBEDDING_REVISION}"
    restart: unless-stopped
    # Cache downloaded model weights so a restart does not re-download them.
    volumes:
      - tei-models:/data
    # GET /health is TEI's readiness probe. (Drop this block if your TEI image
    # variant ships without curl.)
    healthcheck:
      test: ["CMD-SHELL", "curl -fsS http://localhost:80/health || exit 1"]
      interval: 30s
      timeout: 5s
      retries: 5
      # Model download on first boot can be slow; don't flap as unhealthy meanwhile.
      start_period: 120s

volumes:
  docmost:
  db_data:
  redis_data:
  tei-models: