Dataiku, the enterprise AI platform, has introduced Kiji Inspector™, an open-source tool designed to make AI agent decisions understandable and auditable. The release, through Dataiku’s 575 Lab, is aimed at organizations using AI in high-stakes or regulated environments where transparency is critical.
Kiji Inspector currently supports NVIDIA Nemotron open models, helping enterprises see how AI agents arrive at their decisions. At its core is a Sparse Autoencoder that analyzes model behavior in real time, translating complex signals into clear explanations teams can trace, validate, and trust—without slowing operations.
“Embedding AI agents into decisions that affect revenue, safety, and compliance requires visibility into how these systems reason,” said Hannes Hapke, director of 575 Lab at Dataiku. “Kiji Inspector lets organizations inspect AI decision-making before risk becomes reality.”
The tool reflects a broader collaboration between Dataiku and NVIDIA. While NVIDIA Nemotron delivers high-performance open-source AI models, Dataiku provides scalable orchestration, connecting data, applications, and AI systems under one governed framework. Amanda Saunders, NVIDIA director of generative AI, noted that combining Nemotron with Kiji Inspector enhances transparency, auditability, and control for enterprise AI deployments.
Enterprises are already recognizing the value. SLB, a global energy technology company, highlighted the importance of explainability in operational AI systems. “AI delivers real value when engineers can understand and rely on its decisions,” said Sampath Reddy, SLB’s global innovation manager for data & AI.
By enabling explainable reasoning in AI agents, Kiji Inspector aims to support long-term AI success, regulatory compliance, and enterprise confidence, even as organizations scale autonomous systems. Kiji Inspector for NVIDIA Nemotron is available now, with more information on deployment at Dataiku’s website.
