Agentic AI is the next big thing in large language models, but most Agentic AI systems are cloud-based. While these are broad language models trained to handle a wide range of tasks and queries, there are some huge drawbacks to the cloud-based LLM setup, including energy costs, environmental impact, and data security. Cloud-based AI agents aren't always a good solution.

Enter Intel's AI Assistant Builder. While designed more for Intel's enterprise customers, the AI Assistant Builder (formerly "AI Superbuilder") is designed to help you curate and train your own AI Assistant. And it's all local, on-device AI. No Cloud is involved. To find out what makes AI Assistant Builder so unique, I sat down with Intel's Dr. Olena Zhu to discuss the software's current functionality, future, and how it stacks up against the competition.

How does Intel AI Assistant Builder work?

All you need is a couple of software downloads.

The hardware requirements for running the Assistant builder aren't too strict, assuming you have an Intel Core Ultra 100 series or Intel Core Ultra 200 series system handy. Currently, the best option is Intel's Core Ultra 200V series of processors, as they have a more powerful NPU. You also need at least 16GB of system memory, at least 4GB of storage available, and integrated Intel Graphics or better.

Once you've downloaded and installed the AI Assistant Builder software, you can start feeding it data and training it by asking questions and submitting corrections when the agent is wrong. You don't need to know much about coding or programming to start customizing your own agent. You just need to have data.

A custom, secure AI Agent

You can build whatever AI Assistant you need.

Before you download the AI Assistant Builder beta, you'll be asked to choose an agent type from one of five use cases: Sales, HR and Talent, Finance, Medical, and API Services. But Intel's custom assistant builder isn't so limited. These five choices were designed to showcase potential uses for the agent rather than hard-locking you into a specific set of features.

As Dr. Zhu explained, "These are just examples of what we build," as Intel has tailored several AI Assistants for its various partners.

"So you can build, you can build anything you want", Dr. Zhu confirmed. "We have those as an example, but on our GitHub and there's the API guide, and we just did a big training for a large enterprise company. With over 30 Fortune 500 companies, and we show them, 'Hey, you can build whatever you like, an agent or assistant that will be helpful for your industry, for your usage ... if you have something specific and if our API service is still not enough, then talk to us, and we can help."

And because the Assistant Builder is developed entirely on-device from your own data, your data remains secure. So you don't need to worry about a Cloud-based AI service having access to sensitive business data, medical records, legal records, or personal data in order to train an AI assistant for your needs. Dr. Zhu confirms the heightened data security of the Assistant Builder is a key factor for many of Intel's clients who need to protect sensitive data, and that gives Intel's AI Assistant Builder a leg up over other Agentic AI solutions like OpenAI's Operator, Cognition Labs' Devin AI, Anthropic's Claude, or Amazon's Nova Act.

However, you can also take advantage of Intel’s hybrid solution (demoed at Computex) if you need to share the custom trained AI Agent across multiple devices or even give it cloud access for certain tasks that require a larger language model. As Dr. Zhu explained, "For sensitive information ... you don't need to send it to the cloud because that can be well handled by a small language model on an AI PC. However, once you are done with that, you can send the high-level or masked information to the cloud, and then the cloud can perform all the rendering and picture generation... So it's hand-to-hand. So that's also another big trend that we are embracing, and the Superbuilder (AI Assistant Builder) is evolving with those vendors."

And since the AI Assistant Builder is still in beta, the software isn't yet in its final format. New updates are expected to roll out. Intel confirmed at Computex that Model Contex Protocols will be added to the AI Builder system in the near future.

An AI Assistant Marketplace

Dr. Zhu has an open-source vision of the future.

But other than Model Context Protocols, where will AI Builder go next? After all, the AI industry is constantly evolving and trying new technologies and optimizations. While Dr. Zhu couldn't get into specifics, she was able to share her vision for the future of Intel's AI Assistant Builder system.

"So my vision is Superbuilder (AI Assistant Builder) should eventually grow to a marketplace or it's like an ecosystem," Dr. Zhu enthused. "From handling different types of local files to supporting various workflows or automatic computer uses So we'll make this very simple. It's just like picking an outfit type of thing. I like that, and I like this, and I want to hook it up so it's mine, and I want to use it for doing something."

Dr. Zhu's vision is not unlike the Adobe plugin marketplace or even Intel's existing GitHub for AI software solutions. So I could easily see the Intel AI Assistant Builder system taking on a more modular style of customization. I wouldn't be surprised if Dr. Zhu's vision for an AI solution marketplace becomes a reality in the very near future.