services:
  anythingllm:
    image: mintplexlabs/anythingllm:latest
    container_name: anythingllm
    restart: unless-stopped
    ports:
      - "127.0.0.1:3035:3001"
    cap_add:
      - SYS_ADMIN
    environment:
      STORAGE_DIR: /app/server/storage
      DISABLE_TELEMETRY: "true"
      AUTH_TOKEN: ${AUTH_TOKEN}
      JWT_SECRET: ${AUTH_TOKEN}

      # LLM = LiteLLM gateway (Claude Sonnet 4)
      LLM_PROVIDER: generic-openai
      GENERIC_OPEN_AI_BASE_PATH: http://host.docker.internal:4000/v1
      GENERIC_OPEN_AI_MODEL_PREF: anthropic/claude-sonnet-4-20250514
      GENERIC_OPEN_AI_API_KEY: ${LITELLM_API_KEY}
      GENERIC_OPEN_AI_MODEL_TOKEN_LIMIT: "200000"
      GENERIC_OPEN_AI_MAX_TOKENS: "4096"

      # Embeddings = local Ollama on host (bge-m3, multilingual incl. Czech)
      EMBEDDING_ENGINE: ollama
      EMBEDDING_BASE_PATH: http://host.docker.internal:11434
      EMBEDDING_MODEL_PREF: bge-m3
      EMBEDDING_MODEL_MAX_CHUNK_LENGTH: "8192"

      # Vector DB built-in
      VECTOR_DB: lancedb

      # Chunking — slightly larger than default for Czech doc style
      TEXT_SPLIT_CHUNK_SIZE: "1500"
      TEXT_SPLIT_CHUNK_OVERLAP: "250"

    extra_hosts:
      - "host.docker.internal:host-gateway"
    networks:
      - localai
    volumes:
      - anythingllm-storage:/app/server/storage
      - anythingllm-hotdir:/app/collector/hotdir
      - anythingllm-outputs:/app/collector/outputs
      # Intranet docs available read-only for the ingestion script
      - ./intranet:/intranet:ro

volumes:
  anythingllm-storage:
  anythingllm-hotdir:
  anythingllm-outputs:

networks:
  localai:
    external: true