Cassava AI adds NVIDIA Nemotron to its multi-model exchange

Expanding the AI Factory

Cassava Technologies has announced a significant expansion of its AI Factory capabilities by incorporating NVIDIA Nemotron open models. This move is designed to help enterprises and service providers innovate with artificial intelligence in a more effective and affordable manner. By integrating these specific models, the company aims to provide the necessary capacity and economics for businesses to scale their AI solutions.

The deployment happens through the Cassava AI Multi-Model Exchange, also known as CAIMEx. This platform provides a managed inference service, which means companies can access the models via APIs without the burden of building or operating their own model-serving infrastructure. This setup gives organizations the flexibility to select the best mix of models for their specific needs while maintaining a consistent technical framework.

According to Ahmed El Beheiry, Group Chief Technology & AI Officer at Cassava Technologies, the goal is to improve the efficiency of every GPU cycle. He noted that the move provides enterprises and service providers with more choice in how they access and deliver these capabilities to their customers. By focusing on the underlying economics of the hardware, Cassava is attempting to make high-level AI more accessible to a broader range of industries.

Powering Agentic AI

A primary focus of this expansion is the support for agentic AI. Unlike standard chatbots that simply provide information, agentic AI can work through complex tasks and interact directly with various business tools. NVIDIA Nemotron provides the specific reasoning and tool-use capabilities required to make these types of autonomous actions possible for corporate environments.

Because the models are open, developers have access to the weights, training data, and recipes used to create them. This transparency allows for a customizable foundation where developers can fine-tune the models to create specialized agents. These agents can be tailored to perform very specific functions that are unique to a company's internal workflows or niche applications.

As these agents begin to take on longer-running tasks and gain access to sensitive business systems, the complexity of the software increases. Cassava is addressing this by providing a framework where these agents can operate reliably. The focus is on moving beyond simple prompts toward systems that can navigate a series of steps to complete a goal.

Security and Governance

Security and guardrails are becoming critical requirements as businesses mature in their use of AI agents. When an AI system is granted the power to access internal systems or perform actions on behalf of a user, the infrastructure must be able to enforce strict rules. Cassava is addressing this by utilizing NVIDIA OpenShell to manage these workloads.

The use of NVIDIA OpenShell allows the company to bring together model compute, governed agent execution, and independent security enforcement. This means that the security layer is built into the infrastructure itself rather than being an afterthought. It provides a way to ensure that the AI agents stay within their intended boundaries while still having the freedom to perform their assigned tasks.

This integrated approach is designed to foster trust and confidence in data security and control. By providing a managed environment, Cassava helps enterprises ensure that their data remains protected while they explore the benefits of agentic workflows. It creates a controlled space where the risks of autonomous AI are mitigated by technical guardrails.

Optimizing AI Economics

To make these advanced capabilities more affordable, Cassava has introduced the CAIMEx Token Optimiser. This tool is specifically designed to improve the economics of both generative and agentic AI by identifying information that has already been processed. Instead of processing the same content multiple times, the system reuses the existing information where appropriate.

This optimization strategy reduces the overall computing requirements for the user. By minimizing unnecessary processing, the Token Optimiser helps customers use their AI tokens more efficiently as their usage scales. This is particularly important for agentic AI, which may involve many repetitive steps or long-running tasks that could otherwise become very expensive.

The benefits of this optimization apply to both the models hosted by Cassava and third-party AI models. This broad compatibility ensures that enterprises can reduce costs across their entire AI ecosystem. It represents a move toward more sustainable and predictable spending for companies that are moving away from experimental AI toward production-level deployment.