Introduction

Docling is a powerful document parsing, conversion, and chunking solution that supports OCR, layout analysis, and table extraction. While the docling-serve Docker image provides a convenient REST API, it does not automatically download required models on first use—unlike the standalone version.

This article explains how to preload models into a custom Docker image to ensure smooth deployment without runtime delays.

Why Preload Models?

By default, Docling downloads models (e.g., layout, tableformer) when first needed. However, in containerized environments:

• No persistent storage → Models vanish when the container restarts.
• Network dependency → First-time usage requires internet access.
• Performance overhead → Downloads slow down initial requests.

Preloading models into the image avoids these issues.

Step-by-Step: Building a Custom Image

Create a Dockerfile

Extend the official docling-serve image and download models to a known path (models):

FROM quay.io/docling-project/docling-serve
RUN docling-tools models download layout tableformer -o /models

Note: Replace layout and tableformer with your required models.

Build the Image

docker build -t my-docling-serve .

Configure Deployment

.env File

Set the artifacts path to match the preloaded models:

DOCLING_SERVE_ENABLE_UI=true
DOCLING_ARTIFACTS_PATH=models

docker-compose.yml

services:
  docling:
    image: my-docling-serve:latest
    container_name: docling-serve
    restart: unless-stopped
    ports:
      - "5001:5001"
    env_file:
      - .env

Start the Container

docker compose up -d

Accessing the API

Once running, the service exposes:

• Base URL: http://127.0.0.1:5001/
• OpenAPI Docs: http://127.0.0.1:5001/docs
• Test UI (if enabled): http://127.0.0.1:5001/ui/