Handle: docling
URL: http://localhost:34491

Docling simplifies document processing, parsing diverse formats — including advanced PDF understanding — and providing seamless integrations with the gen AI ecosystem.

# [Optional] Pre-pull the image
harbor pull docling

# Start the service
# --open is optional, to open in browser
harbor up docling --open

• Harbor runs official docling-serve service
• When running with nvidia GPU, set one of the CUDA-enabled images in the harbor config:

  harbor config set docling.image ghcr.io/docling-project/docling-serve-cu124

• As of July 2025, Cuda 12.6 and Cuda 12.8 images are broken, only Cuda 12.4 works
• You can use Docling with CPU only, but it will be much slower
• Main use-case is to pre-process documents from the internet or local files with the Docling API

Open built-in docs at http://localhost:34491/docs or use the official documentation

Following options can be set via harbor config:

# The port on the host where Docling endpoint will be available
HARBOR_DOCLING_HOST_PORT       34491

# Docker tag to use for Docling image
HARBOR_DOCLING_VERSION         latest

# Docker image to use for Docling (CPU by default)
HARBOR_DOCLING_IMAGE           ghcr.io/docling-project/docling-serve-cpu

# URL to open with `harbor open` command
HARBOR_DOCLING_OPEN_URL        http://localhost:34491/ui

You can set extra environment variables for Docling via harbor env. See the official configuration reference for more options.