These new services are designed to enable the next generation of the VAST AI OS to meet the growing demands of organizations scaling mission-critical AI initiatives.
At VAST Forward 2026, VAST Data unveiled two major additions to its AI Operating System: the VAST Data PolicyEngine and VAST Data TuningEngine. These new services are designed to enable the next generation of the VAST AI OS to meet the growing demands of organizations scaling mission-critical AI initiatives. Together, the engines ensure AI systems are trusted, explainable, and continuously learning, with PolicyEngine governing agent activity and TuningEngine managing model tuning to maintain alignment with organizational expectations.
The PolicyEngine addresses critical concerns around AI access and governance. As AI workflows increasingly interact with sensitive organizational data, the risk of data leakage, uncontrolled agent communication, and untracked actions grows. PolicyEngine enforces inline policies that control how agents access shared memory, tools, knowledge bases, and other agents. By requiring permissions and context-aware checks before actions execute, and maintaining tamper-proof logs of every interaction, the system ensures a zero-trust framework where all AI activity remains observable, explainable, and auditable.
The TuningEngine complements this by managing continuous learning and model optimization. Leveraging outcomes from agentic pipelines and curated feedback, TuningEngine supports methods such as LoRA fine-tuning, supervised fine-tuning, and reinforcement learning. Candidate models generated from these processes can be evaluated, benchmarked, and deployed either manually or automatically, creating closed-loop learning systems. Integration with NVIDIA’s NeMo Data Designer further enhances the platform’s ability to train and fine-tune Nemotron models. Together, these engines advance VAST Data’s vision of a “thinking machine”, capable of observing, reasoning, acting, evaluating, and improving while safeguarding security and explainability. The PolicyEngine and TuningEngine are expected to be released by the end of 2026.
