India's 1st sovereign AI product launched at AI Impact Summit
What's the story
At the India AI Impact Summit, Arinox AI and KOGO have launched what they call India's first sovereign artificial intelligence (AI) product. The innovative system is based on the concept of "AI in a box," and is designed to run locally without internet connectivity. This revolutionary technology aims to address concerns about data security, integrity, and costs associated with AI adoption in enterprises.
Product launch
CommandCORE is powered by NVIDIA hardware
The latest iteration of this innovative system, dubbed CommandCORE, is powered by NVIDIA hardware. Arinox AI and KOGO have partnered with NVIDIA and Qualcomm for its agentic stack. "The future of AI is private, on an enterprise level too," said Raj K Gopalakrishnan, CEO and Co-Founder of KOGO AI. He emphasized the importance of keeping AI private for organizations to exponentially increase their intelligence and learning capabilities.
Security concerns
'AI in a box' proposition
The "AI in a box" proposition is not just technical, but also ideological. Organizations using public foundational models expose operational insight, which can be risky for sensitive industries. Gopalakrishnan warned that when such industries share data with foundational models and cloud-based AI services, they are also sharing intelligence. This highlights the security and privacy concerns associated with agentic AI deployments.
Threat analysis
Security concerns in enterprise AI deployments
An analysis by security platform HiddenLayer, AI Threat Landscape 2025, found that 88% of enterprises are worried about vulnerabilities from third-party AI integrations. This includes popular tools like OpenAI's ChatGPT, Microsoft Copilot, and Google Gemini. The findings underscore the need for robust security measures in enterprise-level AI deployments to mitigate potential risks associated with third-party integrations.
Scalability
CommandCORE can handle focused workloads and repeatable tasks
CommandCORE is designed to simplify the complexities of hardware, software, and application layers that an enterprise would have to manage independently. It can handle focused workloads and repeatable tasks, scaling up to large clusters for an entire workflow. Enterprises can connect multiple units together for scalability, and choose from three model configurations at a starting price of ₹10 lakh.
Applications
Small and medium model variants for specific use cases
The small variant of CommandCORE can run a model with 1-7 billion parameters, making it ideal for batch processing or HR onboarding processes. The medium model handles around 20-30 billion parameters for complex agents with inference. As AI adoption grows in regulated and sensitive environments, firms will need accelerated computing platforms that can operate entirely on-premise under strict security controls.