Skip to main content

Summarizer Agent

The Summarizer Agent example highlights the following features:

  • Input Filtering
  • html function
  • extractUrl function

Summarizer Agent

The Summarizer Agent summarizes web pages. To achieve this, we augment the input message with the content of the web page that the user requests. This provides the Large Language Model with all the data required to summarize the web page.

In the filterInput block, we use the extractUrl function to extract any web page url from the inputMessage. If a valid url is contained with the input, its content is added to the inputMessage.

In addition, the html function is used to filter out the html tags to help reduce the overall size of the web page document.

agent {
    name = "summarizer-agent"
    description = "Agent that summarizes web pages."
    prompt {
        """
       You are a helpful agent. 
       You help customers by summarizing webpages. 
       Keep your answer short and concise.
     """
    }
    filterInput {
        val url = extractUrl(inputMessage).firstOrNull()
        if (url != null) {
            debug("Loading url: $url")
            val html = html(url).getOrThrow()
            inputMessage = inputMessage.update("""
                User question: ${inputMessage.content}
                The webpage $url contains the following text:
                $html
             """
            )
        }
    }
}

Example question

Please summarize the following article: https://www.theregister.com

After the web page has been loaded, follow up questions about the web page can be asked. This is because the content of the web page remains in the transcript that is provided to the Agent.

The Agent can also be extended to use the pdf function to summarize pdf documents.