Conversational Knowledge Buddy

A user such as a CSR can use the Knowledge widget to interact with Knowledge Buddy directly from the interaction portal. However, when faced with a complex customer inquiry, it may be necessary for the CSR to ask multiple additional questions based on the initial request and response. This is where a conversational Buddy can provide the CSR with the tools needed to handle the customer issue and receive focused, relevant answers.

When you ask a conversational Buddy a question, Pega Knowledge Buddy performs the following steps:

1. Pega Knowledge Buddy fetches search results from the knowledge base by searching avector database for relevant content embeddings that match your question.
2. Pega Knowledge Buddy combines those results with the full conversation context and sends the combined prompt to a large language model (LLM) by using the Pega GenAI Gateway™.
3. The LLM processes the prompt and generates an answer. Because the conversational buddy includes the conversation history in every prompt, the LLM understands the context of your current question in relation to what has already been discussed.
4. The answer streams into the knowledge widget in real time. You can begin reading the response as the LLM generates it.
5. After you receive the answer, you can provide feedback on the response or share it directly from the conversational Buddy interface.

The conversational Pega Knowledge Buddy maintains the full history of your current conversation, so if you keep the same conversation window open, the Buddy retains what has been discussed. You can ask a follow-up question, for example, "Can you explain that in simpler terms?" or "What about for a premium account holder?" and the Buddy interprets the new question based on the previous exchange, instead of treating it as a new, standalone question.

If you close the conversation window and open a new one, the conversation history is reset.

A conversational Buddy can be coupled with streaming responses to provide a smoother user experience. Without streaming, you wait until the entire answer has been generated before seeing any content. However, with streaming enabled, the answer appears progressively as the LLM builds it. Becausevery large and complex LLMs can take time to generate a complete response, streaming reduces waiting time and improves efficiency for serving customers.

The following graphic illustrates the process:

A non-conversational Buddy does not retain context from previous questions, and is well suited for quick, standalone questions. By comparison, a conversational Buddy provides a chat‑like experience that supports natural, back‑and‑forth interactions. When you create a Buddy, choose the response style that best fits your use case. The following table compares between the two response styles: