Training happens once, when a model is built — this open source graph is verified reference data to train on. Grounding happens at inference, every time a question is asked — stable IDs anchor answers in source truth.
Training happens once, when a model is built. The graph gives model builders verified relationships and consistent IDs to learn from — reference data with provenance instead of scraped guesses.
Pretraining, fine-tuning, and context extension of open-weight models, all on open source data that is free to download for everyone.
Grounding happens at inference, every time a question is asked. Resolving a name to a canonical ID with evidence is the difference between an answer and a guess: match rate by join key — stacking fields helps, joining on a canonical ID resolves it.
Ask an agent for a competitive analysis on Helios and name-matching returns five different companies. Joining on a canonical ID returns one verified entity with its full graph: master record, locations, patents, news, people.
IDC forecasts 1.3 billion AI agents in operation by 2028. Every new agent that rebuilds the same entity map is redundant infrastructure; every agent that guesses identity from a name is a hallucination waiting to happen. The time to standardize is before the next 10x wave.
Open, shared reference data is "the essential bedrock for the next generation of agentic AI."
Jim Zemlin, CEO, Linux Foundation — The Value of Open Data on Global Entities
The Agentic AI Foundation, the Linux Foundation home of open agentic standards including MCP and A2A, defines how agents interact. OpenData.org identifies the participants: verified counterparties, persistent identity, interoperable identifiers.
OpenData.org enables the vision of Open Weights and American AI Leadership and helps fulfill America's AI Action Plan: open, verified entity data as shared AI infrastructure.