As organizations accelerate the deployment of AI agents across critical business workflows, performance alone is no longer sufficient. Enterprises must also understand, validate, and govern how these systems generate their outputs—especially in regulated and high-stakes environments. To address this need, Dataiku, through its 575 Lab open source office, has introduced Kiji Inspector™, one of the first open-source explainability frameworks purpose-built for enterprise AI agents. The initial model family supported by the framework is NVIDIA’s Nemotron open models.
As companies move toward sovereign AI and develop more of their own infrastructure, the combination of NVIDIA Nemotron models and Dataiku’s Kiji Inspector enhances visibility into AI decision-making processes. Kiji Inspector directly tackles the “black-box” problem by providing built-in explainability for agent decisions. At its core, the framework uses a Sparse Autoencoder to analyze the moment an AI model selects a tool, identifying the signals behind that decision and translating them into clear, human-understandable explanations—without impacting system performance.
According to Hannes Hapke, Director of 575 Lab at Dataiku, many enterprises are already integrating AI agents into decisions that affect revenue, safety, compliance, and customer trust, yet lack structural visibility into how these systems reason. He emphasized that without explainability, scaling AI also scales uncertainty. Kiji Inspector aims to change this by allowing organizations to inspect and refine AI systems before risks materialize, enabling greater trust as agentic systems transition from experimentation to production-grade infrastructure.
This release builds on the growing collaboration between Dataiku and NVIDIA in delivering enterprise-ready generative and agentic AI solutions. NVIDIA Nemotron models provide high-performance, open-source capabilities designed for enterprise use, while Dataiku delivers orchestration, connecting data platforms, enterprise applications, and AI services within a governed environment.
Amanda Saunders, Director of Generative AI at NVIDIA, highlighted that scaling autonomous AI agents requires trust rooted in transparency and accountability. She noted that open models like Nemotron allow organizations to better understand, audit, and control their systems. By combining Nemotron with Kiji Inspector, enterprises gain deeper insight into the reasoning behind AI-driven decisions.
Dataiku’s focus on explainable AI has already resonated with more than 750 enterprise customers operating in complex and regulated industries. SLB, a global technology company in the energy sector, emphasized the importance of transparency in AI adoption. According to Sampath Reddy, Global Innovation Manager for Data & AI at SLB, AI delivers the most value when engineers can understand and rely on its outputs, enabling validated workflows and strong governance that support real-world deployment.
With the introduction of Kiji Inspector for NVIDIA Nemotron, Dataiku enables enterprises to leverage cutting-edge open-source AI capabilities while maintaining visibility, governance, and trust. As AI agents become more autonomous and deeply integrated into enterprise systems, explainability will play a critical role in ensuring regulatory compliance, operational reliability, and long-term business success.
Kiji Inspector for NVIDIA Nemotron is now available.
