Whetstone is the first product Cyberelf Labs is commercializing, and it has changed shape over time.
Earlier public work focused on evaluating and controlling the promotion of AI artifacts—models, adapters, prompts, agents, and workflows. That work explored sealed evaluations, contamination checks, regression detection, baseline comparisons, realistic multi-step evaluation, verifier-backed receipts, item-level evidence, and PASS / HOLD / BLOCK release decisions.
Broader Cyberelf and Whetstone architecture later explored identity-bound admission, formal obligations, signed runtime permissions, scoped and reversible control, and private deployment boundaries.
Whetstone now sits inside a wider investigation of persistent, tool-using agents and trustworthy agent infrastructure. These stages are related, but they are not interchangeable. This page keeps their scopes and claims separate.