Open Data Products Python SDK

open-data-products is a Python SDK and CLI for the OpenDataProducts.org standards family: ODPS, ODPC, ODPG, ODPV, and ODPR.

Use it to validate data product documents, build catalogs and graphs, inspect portfolio source intake, run LLM-assisted generation workflows, expose a local MCP server, and give AI agents a consistent standards-aware API.

Install

pip install open-data-products

Optional extras:

# Outlook .msg intake support
pip install "open-data-products[email]"

# Data Contract CLI integration
pip install "open-data-products[contracts]"

# Embedded llama.cpp generation support
pip install "open-data-products[llama-cpp]"

# Development tools
pip install "open-data-products[dev]"

Python 3.8 or newer is required.

What It Provides

Quick CLI Examples

Most commands print human-readable output by default. Add --json for CI, scripts, MCP clients, and agents.

Run the SDK through Python:

# Validate and inspect standards documents
python3 -m open_data_products.cli validate examples/product.yaml
python3 -m open_data_products.cli explain examples/product.yaml --json
python3 -m open_data_products.cli refs examples/product.yaml --json
python3 -m open_data_products.cli summary examples/product.yaml

# Discover bundled schemas, prompts, vocabulary records, and guidance
python3 -m open_data_products.cli resources --json
python3 -m open_data_products.cli resources --id generation.prompt.system --json

# Agent surfaces
python3 -m open_data_products.cli manifest --json
python3 -m open_data_products.cli serve

After installation, the console script provides the same commands:

open-data-products validate examples/product.yaml
open-data-products explain examples/product.yaml --json

Portfolio source intake can be inspected without calling an LLM:

python3 -m open_data_products.cli portfolio intake \
  --objectives sources/objectives/ \
  --use-cases sources/use-cases/ \
  --signals sources/signals/ \
  --products sources/products/ \
  --config generation.config.yaml \
  --json

Portfolio build uses the same source lanes and prompt budget controls:

python3 -m open_data_products.cli portfolio build \
  --objectives sources/objectives/ \
  --use-cases sources/use-cases/ \
  --signals sources/signals/ \
  --products sources/products/ \
  --output generated/portfolio/

LLM Generation

Generation defaults to local Ollama-compatible settings, and can also use embedded llama.cpp, OpenAI-compatible local servers, NVIDIA NIM, Claude, and hosted providers configured in generation.config.yaml.

open-data-products config generation --copy-to generation.config.yaml
open-data-products config generation --config generation.config.yaml --check

open-data-products generate \
  --config generation.config.yaml \
  --input source_docs/products/ \
  --kind product-reference \
  --output generated/

Documentation

The full SDK guide that used to live in this README is now here:

• Full SDK guide

Focused user guides:

• Command guide
• Portfolio intake guide
• LLM generation
• Agent surface
• Recipe workflows
• Data Contract workflows
• API reference

Project references:

• Documentation index
• Examples
• Agent handoff

Development

git clone https://github.com/Open-Data-Product-Initiative/odps-python
cd odps-python
pip install -e ".[dev]"
pytest -q

Before publishing a package, verify the PyPI description renders:

python3 -m build
python3 -m twine check dist/*

Acknowledgments

Thanks to the Open Data Product Initiative community, Chris Howard / Kitard for the original odps-python foundation, devlouie for the MCP layer and agent surface, and the Data Contract CLI project for the optional contract execution engine.

License

Apache License 2.0. See LICENSE for details.