OpenWALDO
A community project building a shared, licensed and auditable pool of AI training data.
OpenWALDO is an open source project launched in August 2026 by Gregory Kurtzer, sponsored by his company CIQ. The name stands for Weights, Artifacts, Licenses, Data and Origins, and the goal is to open up the one layer of AI that almost nobody opens: the training data.
The argument is that open weights only get you halfway. You can download and run an open model, but you still cannot see what went into it. OpenWALDO records where every piece of training material came from, what rights were claimed over it and who contributed it, using ordinary open source machinery: the index lives in Git so changes are reviewable, and the data is verified by content hashes. Models trained from it can ship with an AI Bill of Materials. At launch the corpus held 124 billion tokens across 75.1 million documents, which is a starting point rather than a rival to what the big labs train on.