Knowledge Bases

Knowledge bases are the foundation of Agent Assist. They store the documents, articles, and reference material that the RAG (Retrieval-Augmented Generation) pipeline searches to generate suggestions for agents.

How It Works

The assist pipeline follows this flow when a customer message arrives:

1. Classifier: The inbound message is evaluated to determine if it contains a question or request that needs an answer.
2. KB Search: The message is converted to an embedding and searched against the relevant knowledge base using vector similarity.
3. LLM Generation: The top matching chunks are sent to the configured LLM along with the conversation context. The LLM generates a suggested answer with source citations.
4. Delivery: The suggestion is pushed to the agent's widget in real time.

Sections

Section	Description
Creating Knowledge Bases	Step-by-step guide to creating a new knowledge base
Document Management	Upload documents and configure external sources
RAG Settings	Tune similarity thresholds, chunk sizes, and reranking

TIP

Each knowledge base can be mapped to specific CCaaS queues or conversation types. This ensures agents receive suggestions from the most relevant source material for the conversations they handle.