3. Add knowledge to AI Agents

This is part of a series of articles that I am writing to better understand and explain how AI agents will transform how we create and consume content in the near future. If you want to start from the first article, please start here.

In the previous article, 2: What makes an AI agent unique?, I discussed how AI agents become unique through their specific instructions, domain knowledge, and actions.

Let’s delve into the concept of knowledge. Simply put, you visit a dentist for teeth cleaning, an accountant for financial management, and a primary care physician for health check-ups. In each scenario, you seek their services because they possess the domain expertise that you require.

Consider another instance. Suppose you use a Software as a Service (SaaS) platform for payments and encounter a question. Your first instinct might be to initiate a Google search, which often results in a barrage of ads and irrelevant site links. After some scrolling, you find the correct website, click on it, and are presented with FAQs and support articles. Now, you must sift through this information to determine if it answers your question.

Imagine, however, if there was an AI agent equipped with FAQs, blogs, articles, and other useful information. All you would need to do is pose your question to this AI agent, and it would provide the appropriate information. This approach not only saves time and effort for the user but also enhances the overall user experience.

The importance of domain-specific knowledge in AI agents cannot be overstated. These agents, armed with specialized knowledge, can offer more accurate and insightful responses, thereby revolutionizing the way users seek information and interact with platforms. This is the future of AI - niche applications that are tailored, efficient, and truly transformative.

To see how this works, we created this functionality on PlebAI, where anyone can create AI agents. When creating them, you can add knowledge by attaching a PDF document, text file, or even a URL from a website. Behind the scenes, these data are retrieved, transformed into vectors, and stored in a vector store. This data can be both public and private as it is stored securely and only shared with LLM (Large Language Model) in the form of embeddings.

With this knowledge, the LLM can easily answer the user’s questions correctly.

Here’s how to add knowledge:

1. Go to PlebAI
2. Click ‘Create Text AI agents’
3. Fill in all the necessary information
4. Add knowledge in the form of documents or website URLs
5. Start using them either privately or publicly

Once you add all the necessary information, the AI agent is now equipped to answer any user question related to the stored knowledge. You can also enable user web browsing so that it can retrieve any additional knowledge available on their website. Here’s the response from the Zaprite Help AI agent answering a question from its FAQ.

Marc Andreessen, a renowned entrepreneur and investor, has been quoted as saying, “The content of each new medium is the old medium.” This quote encapsulates the idea that each new form of media tends to repurpose the content of its predecessor.

With this view, we can start to push website and app content into AI agents and make them available inside any chat interface.

Let me know what you think and if you have any feedback or comments.