Knowledge Buddy behavior when no vector results found – LLM invocation and extensibility question
Hi-We are evaluating Pega GenAI Knowledge Buddy and observing the following behavior:
When a user asks a question and no relevant content is found in the vector store, Knowledge Buddy responds with “I don’t know based on the available information”. From our observation, in this scenario the LLM is not invoked, and the response is returned purely based on retrieval outcome.
Our understanding of a typical RAG pattern is that retrieved content augments the prompt, but the LLM is still invoked, even when retrieval returns no or low-confidence results (allowing the model to reason, respond generically, or use alternate sources).
Hence, we would like clarification on the following:
Hi-We are evaluating Pega GenAI Knowledge Buddy and observing the following behavior:
When a user asks a question and no relevant content is found in the vector store, Knowledge Buddy responds with “I don’t know based on the available information”. From our observation, in this scenario the LLM is not invoked, and the response is returned purely based on retrieval outcome.
Our understanding of a typical RAG pattern is that retrieved content augments the prompt, but the LLM is still invoked, even when retrieval returns no or low-confidence results (allowing the model to reason, respond generically, or use alternate sources).
Hence, we would like clarification on the following:
- Is Knowledge Buddy intentionally designed to skip LLM invocation when no vector results are found?
- Is there any supported configuration, extension point, or roadmap capability that allows:
- Always invoking the LLM regardless of retrieval outcome, and/or
- Falling back to external or broader knowledge sources (for example, internet or non-ingested sources), similar to tools like Microsoft 365 Copilot?
- If not configurable, can this behavior be considered a hard product design constraint, specifically to enforce grounded, enterprise-only answers and prevent hallucination?
This clarification will help us determine whether Knowledge Buddy is suitable for:
- Strict enterprise knowledge use cases only, or
- Broader “assistant-style” use cases that require exploratory or open-domain responses.
Product & Version Details
- Pega Platform: 25.1.1
- Knowledge Buddy: 25.1.1‑104
- Knowledge Management: 25.1.1‑22
Looking forward to your guidance on the intended behavior and supported options.
Regards-Rajarshi