docling-pipelines-slim
Overview
docling-pipelines-slim is a lightweight variant of docling-pipelines that excludes certain operator dependencies.
When to Use Slim
Use this variant if:
- Your codebase doesn't need the built-in extract_operator
- You want a lighter installation and will add extraction support later via pip install "docling-pipelines-slim[extract]"
Installation
pip install docling-pipelines-slim
To enable extraction operator support:
pip install "docling-pipelines-slim[extract]"
Usage
All imports and usage are identical to the standard docling-pipelines package:
from docpipe.cli.docpipe_cli import main
from docpipe.lib.docpipe_flow_manager import DocpipeFlowManager
# Run a flow
manager = DocpipeFlowManager(flow_file="my_flow.json")
result = manager.execute()
Building from Source
Build slim wheel
Note: The command below is for local builds only. In CI, the slim wheel is built and published to PyPI by the GitHub Actions release workflow alongside the main wheel.
cd slim/
uv build --wheel
Result: slim/dist/docling_pipelines_slim-<VERSION>-py3-none-any.whl (version derived from Git tag via hatch-vcs).
Install from wheel
# Standard wheel installation
pip install ./dist/docling_pipelines_slim-*.whl
# With support for extraction operator
pip install "./dist/docling_pipelines_slim-*.whl[extract]"
Operator Availability
Core operators are available in slim. Optional operator support requires installing the corresponding optional group.
| Operator | Default | Optional Group |
|---|---|---|
ingest_source |
✓ | — |
extract_operator |
✗ | extract |
chunker (fixed-size / sentence modes) |
✓ | — |
chunker (hybrid / docling-split modes) |
✗ | extract |
embeddings (litellm, watsonx, ollama providers) |
✓ | — |
embeddings (HuggingFace local provider) |
✗ | extract |
lang_detect, redaction |
✓ | — |
vectordb, acl_operator |
✓ | — |
| All quality operators | ✓ | — |
See Also
- Parent README
- docs/reference/OPERATORS.md — Operator reference
- docs/guides/FLOW_CONFIGURATION_GUIDE.md — Flow authoring guide