Validate
Detect and validate ODPS products, ODPC catalogs, ODPG graphs, ODPV vocabulary, and Data Contract references before they move into production workflows.
Agent-Ready Data Product Tooling
v0.3.4
The SDK for agent-ready data products
A practical Python SDK and MCP server for loading, validating, explaining, generating, searching, traversing, importing, exporting, and publishing Open Data Product standards, external OKF context bundles, Data Contracts, ODPR workflow recipes, LLM workflows, and portfolio workspaces.
$ pip install open-data-products
Open source and useful? Give the SDK a GitHub star to help others find it.
Built to give AI agents the context to operate
The SDK turns data product standards into executable context. Teams get validation, developers get Python functions, and AI agents get structured outputs, summaries, workflow recipes, bundled resources, graph reasoning, Data Contract reports, and safe tools for real workflows.
Explain and summarize
Turn YAML and JSON into readable explanations, structured validation results, references, small summaries, and compact context hints that avoid flooding an agent context window.
Generate
Use local runtimes or hosted providers to create standards-aware ODPS, ODPC, and ODPG artifacts from business source material.
Run recipes
Use ODPR recipes to describe repeatable SDK workflows, including the steps, providers, inputs, outputs, validation checks, and review policy for each run.
Discover resources
Expose bundled schemas, prompt templates, vocabulary records, catalog object records, and graph object records through Python, CLI, and MCP surfaces.
Reason over graphs
Summarize, traverse, analyze, and extract trusted focus-node context from ODPG relationships, ODPC objects, and ODPV terms.
MCP Server Included
Give AI agents the context layer for data product work
The SDK includes Model Context Protocol support so agent hosts can operate on Open Data Products files and OKF context bundles through safe tools, structured resources, standards-aware context, graph relationships, and LLM-assisted generation.
Agent-ready tools
Expose validation, explanation, reference discovery, summaries, compact context artifact discovery, OKF bundle validation, OKF concept listing, search, traversal, graph reasoning, and product context actions through MCP.
ARWS manifest
Generate an agent manifest with
open-data-products manifest --json for hosts that understand ARWS-style tool discovery.
Local and service workflows
Use the SDK with local LLM runtimes including Ollama, LM Studio, vLLM, llama.cpp server, embedded llama.cpp through
llama-cpp-python, and LocalAI, or hosted providers such as Claude, OpenAI, OpenRouter, and Groq.
Designed for local and provider LLM support
The SDK keeps generation workflows provider-aware while staying grounded in the Open Data Products standards family. Choose a local runtime for private or offline work, run a GGUF model directly with the optional embedded llama.cpp extra, or connect a hosted provider when managed model access fits better. Generation configuration can be extended with additional local models, hosted providers, and OpenAI-compatible endpoints.
Run models on your machine
Use Ollama, LM Studio, vLLM, llama.cpp server, LocalAI, or another OpenAI-compatible local chat server for local generation workflows.
Run GGUF models without a server
Install open-data-products[llama-cpp] and select llamacpp-embedded when a workflow should call llama-cpp-python directly instead of a separate local API server.
Bring the local model you trust
Work with model families such as Llama, DeepSeek, Qwen, Mistral, Mixtral, Phi, Gemma, Code Llama, StarCoder, Yi, Command R, Falcon, Granite, Nemotron, Vicuna, and WizardLM when exposed by your runtime.
Use managed LLM APIs
Connect service providers such as Claude, OpenAI, OpenRouter, and Groq through generation configuration, then validate generated ODPS, ODPC, ODPG, ODPV, and data contract artifacts before use.
Default provider suggestions by SDK job
These provider profile names are suggested by the default SDK generation configuration and can be selected with --provider <name> in CLI commands. Add your own profiles when another model or provider fits the job better. The LLM selection guide expands the decision path for validation, generation, graph, portfolio, localization, CI, and release-review jobs.
| SDK job | Suggested local profile | Suggested stronger or production profile |
|---|---|---|
| Validate, build, render, or create TOON/GCF sidecars | No LLM | No LLM; keep these deterministic in CLI and CI. |
| Fast draft generation | ollama-gemma3n, ollama-llama, ollama-mistral, or ollama-phi |
claude or openai when the draft becomes release-facing. |
| Structured ODPS, ODPC, or ODPG YAML generation | ollama-qwen3, ollama-qwen3-14b, or llamacpp-embedded with a local GGUF model |
lmstudio-gemma4-12b, claude, or openai. |
| Graph inference and portfolio review | lmstudio-gemma4-12b or ollama-qwen3-14b |
claude or openai when consistency matters more than local-only execution. |
| Localization or release review | lmstudio-gemma4-12b |
claude or openai for production polish and review confidence. |
| ODPR workflow recipe runs | Recipe-selected provider profile | Let the workflow contract decide which steps are deterministic, local, hosted, or hybrid. |
Example SDK commands
Use the suggested provider profile names from the default configuration when drafting artifacts, refreshing portfolios, or reviewing localized outputs.
Fast local draft
open-data-products generate \ --input source_docs/products/ \ --kind product-reference \ --provider ollama-gemma3n \ --output generated/
Portfolio refresh
open-data-products portfolio refresh \ generated/portfolio/ \ --provider claude
Localization review
open-data-products portfolio localize \ generated/portfolio/ \ --languages "fi,sv" \ --provider lmstudio-gemma4-12b
Context bundles and sidecars for agent workflows
OKF support lets teams exchange external Markdown/frontmatter knowledge bundles, while TOON and GCF give agents smaller sidecar files for repeated catalog rows and graph relationships. ODPR recipes can make those context choices part of a workflow contract, so each repeatable agent run knows which bundles, sidecars, validation checks, providers, and review steps to use while ODPS, ODPC, ODPG, and ODPV remain the canonical structured artifacts.
Import and export knowledge context
Validate Open Knowledge Format bundles, summarize concepts without full Markdown bodies, import concepts as generation-ready source docs, and export ODPC catalog or portfolio artifacts as portable OKF context.
Compact repeated objects
Generate TOON views for ODPC catalogs and ODPG graphs when repeated structures need to fit more cleanly into model context windows.
Experimental graph context
Emit GCF files for ODPC and ODPG workflows, including graph nodes and edges that are packed for agent prompts and review automation.
Make context repeatable
Use workflow contracts to decide when agents should use OKF bundles, TOON, GCF, canonical YAML, provider profiles, and validation gates for each repeatable run.
Keep source files authoritative
Use OKF, --toon, and --gcf as context exchange or build outputs; keep standards YAML as the reviewed, validated, and exchanged source of truth.
One SDK. Three execution paths.
Use the same standards-aware capabilities from Python, from the terminal, or through an agent host.
1
Python API
Call top-level helpers such as load, validate, explain, summarize, resolve references, and list resources.
2
Unified CLI
Run human-readable commands by default and add --json for scripts, agents, and CI jobs.
3
MCP server
Launch open-data-products serve so compatible agent hosts can call safe SDK tools over stdio.
4
ARWS manifest
Render open-data-products manifest --json for agent systems that need a compact tool manifest.
5
Guide paths
Move from install and validation into LLM generation, graph reasoning, catalogs, and portfolio workspaces.
Guided paths into the SDK
The GitHub repository includes course-style guides. The website highlights the most useful entry routes instead of listing every lesson at once.
Install, validate, explain
Begin with Python setup, install the package, and validate your first standards file.
Use local or hosted LLMs
Configure generation, keep prompts project-owned, and create reviewable fragments from business notes.
Connect vocabulary, graphs, and catalogs
Search shared terms, build graph context, and assemble ODPC catalogs that humans and agents can review.
Build the complete workspace
Move from separate commands into a connected portfolio with HTML review pages and agent-ready YAML.
Now on Udemy
Data Product SDK Masterclass
The Data Product SDK Masterclass is now available on Udemy. It turns the current SDK updates into a guided, practical learning path for building scalable, agent-ready data product value management workflows. Use the course page to enroll, or reach out on LinkedIn for voucher access.
- 3.5 hours of on-demand video
- 19 downloadable resources
- 2 executable Colab workbooks
- Full GitHub repository with all materials
- Practical SDK workflows for validation, generation, catalogs, and graphs
Portfolio Builder
Turn source lanes into a connected data product portfolio
The portfolio workflow packages generation, ODPC catalogs, ODPS products, ODPG graphs, review pages, localization, and version history into one practical path for business-facing work.
# Build and review the workspace
open-data-products portfolio build --output generated/portfolio/
open-data-products portfolio localize generated/portfolio/
open-data-products portfolio explain generated/portfolio/
Designed for the Open Data Products standards family
The SDK gives developers and AI agents one practical interface across product descriptions, ODPC catalogs, ODPG graphs, ODPV vocabulary, ODPR workflow recipes, and data contract workflows. OKF is supported as an external context bundle format, not as a fifth Open Data Products standard.
Product
Defines the data product, its access, quality, SLA, pricing, support, license, and strategy.
Catalog
Organizes products, objectives, use cases, KPIs, and demand signals at portfolio level.
Graph
Connects products to business value through nodes, edges, and relationship paths.
Vocabulary
Creates shared meaning for humans, platforms, and AI agents working with the standards.
Recipes
Defines workflow contracts for repeatable SDK runs, provider choices, validation gates, and review steps.
Data Contracts
Supports contract-aware workflows for structure, expectations, quality, access, and product handover.
Start with a clean environment
Install the SDK, check the CLI, then run a human-readable validation command. Add --json when an agent, script, or CI job needs stable machine-readable output.
# Create a portfolio project folder mkdir data-products-portfolio cd data-products-portfolio # Create and activate a virtual environment python3 -m venv .venv-sdk source .venv-sdk/bin/activate # Install the SDK pip install open-data-products # Check the command and validate a file open-data-products --help open-data-products validate examples/product.yaml open-data-products explain examples/product.yaml open-data-products summary examples/product.yaml
Reference docs for deeper work
Use the landing page and guides for orientation, then jump into the focused references when you need exact command behavior, API details, or agent-host setup.
API and commands
Use the Python API and CLI references when integrating the SDK into automation, products, or validation pipelines.
LLM, ODPR, and Data Contract workflows
Configure generation providers, extend provider profiles, use ODPR recipes as workflow contracts, validate contracts, and produce product-level reports.
MCP, ARWS, OKF, and handoff
Connect the SDK to agent hosts, inspect safe tools, give agents a compact routing file, and use OKF bundles or compact context sidecars when available.
FAQs
What is the Open Data Products SDK?
It is the Python SDK, CLI, and MCP server for working with the Open Data Products standards family. It helps teams and AI agents validate, explain, generate, search, traverse, and publish ODPS, ODPC, ODPG, ODPV, Data Contract, and portfolio artifacts.
Why do Data Contracts matter here?
Many industries treat contracts as the operational boundary between data producers and consumers. The SDK supports contract-aware workflows so product descriptions, schemas, expectations, quality rules, access details, and governance context can be reviewed together.
What are workflow contracts?
Workflow contracts are ODPR recipes that describe how a repeatable SDK run should work. They are similar to reusable skills or playbooks for AI agents, but with explicit steps, inputs, outputs, provider profiles, validation gates, context formats, and review policy so humans, CI jobs, and agents can run the same workflow consistently.
Is this only for AI agents?
No. Developers can use the Python API and CLI directly, while AI agents can use the same capabilities through MCP, ARWS manifest metadata, and agent-facing context files. The goal is one standards-aware toolkit across human and agent workflows.
What are TOON and GCF used for?
They are optional compact context sidecars for ODPC catalogs and ODPG graphs. TOON and experimental GCF help pack repeated rows and graph relationships for prompts, summaries, and review automation while YAML remains the canonical standards file.
What is Open Knowledge Format?
Open Knowledge Format, or OKF, is an external Markdown/frontmatter context bundle format for portable human and agent review. The SDK can validate OKF bundles, list concepts safely for agents, import concepts as generation source documents, and export ODPC catalog or portfolio artifacts as OKF context. OKF is not a fifth Open Data Products standard; ODPS, ODPC, ODPG, and ODPV remain the canonical structured standards.
Can it work with local and hosted LLMs?
Yes. Generation workflows can use local runtimes such as Ollama, LM Studio, vLLM, llama.cpp server, embedded llama.cpp through llama-cpp-python, and LocalAI, or hosted providers such as Claude, OpenAI, OpenRouter, and Groq. Generated artifacts should still be validated before use.
What does the portfolio builder create?
It turns source lanes such as objectives, use cases, signals, and product notes into connected portfolio outputs: ODPC catalog YAML, ODPS product files, ODPG graph context, reviewable HTML pages, localized pages, and version history.
Is there a Udemy course for learning the SDK?
Yes. The Data Product SDK Masterclass on Udemy includes 3.5 hours of on-demand video, 19 downloadable resources, 2 executable Colab workbooks, a full GitHub repository with all materials, and practical SDK workflows for validation, generation, catalogs, and graphs.
Build data product workflows that AI agents understand
Use the SDK as the context and execution layer for AI agents working with data product standards, OKF context bundles, ODPC catalogs, ODPR workflow contracts, data contracts, validation, graph traversal, vocabulary helpers, compact sidecars, and agent-ready workflows.