docker-compose.yml

# ── Quick start with GGUF (recommended, 3-5x faster than ONNX) ──
# docker compose up -d
#
# Model is auto-provisioned on first start by scripts/provision.sh
# (mount a volume to persist models across restarts).
#
# For ONNX instead: set LIBRAVDB_EMBEDDING_BACKEND=onnx-local
# and volume-mount your ONNX model + runtime directories.

version: "3.8"

services:
  libravdbd:
    image: ghcr.io/xdarkicex/libravdbd:latest
    container_name: libravdbd
    ports:
      - "50051:50051"
    volumes:
      - libravdb_data:/var/lib/libravdbd/data
      - libravdb_models:/var/lib/libravdbd/models
    environment:
      - LIBRAVDB_GRPC_ENDPOINT=tcp:0.0.0.0:50051
      - LIBRAVDB_DB_PATH=/var/lib/libravdbd/data/data.libravdb
      - LIBRAVDB_EMBEDDING_BACKEND=gguf
      - LIBRAVDB_EMBEDDING_PROFILE=nomic-embed-text-v1.5
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "libravdbd", "health"]
      interval: 30s
      timeout: 5s
      retries: 3

volumes:
  libravdb_data:
  libravdb_models: