Knowledge Base Overview

The knowledge base is how your AI agent learns about your business. Instead of relying only on its general training, your agent searches your uploaded content in real time to answer callers with accurate, company-specific information.

Under the hood, Zentrix Agent Studio uses a Retrieval-Augmented Generation (RAG) pipeline -- a technique that combines vector similarity search with large language model generation to produce grounded, factual responses.

How It Works

The knowledge base pipeline has two phases:

1. Ingestion (when you add content)

1. You upload or paste text content (FAQs, product docs, policies, etc.)
2. The system splits your content into chunks of approximately 2,000 characters each, with 200-character overlap between chunks to preserve context across boundaries
3. Each chunk is converted into a numerical vector (embedding) using OpenAI's text-embedding-3-small model
4. The chunk text and its embedding are stored in Supabase PostgreSQL using the pgvector extension

2. Retrieval (during a live call)

1. A caller asks a question -- for example, "What is your refund policy?"
2. The question is converted into an embedding using the same model
3. A search_knowledge() database function performs cosine similarity search against all chunks belonging to that agent
4. The top 5 most relevant chunks (above a 0.7 similarity threshold) are returned
5. These chunks are injected into the agent's context, and the LLM generates a response grounded in your actual content

Three Ways to Add Content

Zentrix Agent Studio provides three methods for building your knowledge base, available in Step 4 (Knowledge Base) of the setup wizard:

Method	Best For	Details
Upload / Paste	Existing documentation	Paste text directly or upload content. See Uploading Content
Prompt Builder	Structured agent instructions	Drag-and-drop section builder with templates. See Prompt Builder
AI Generate	Quick starting point	Auto-generates FAQ from your company details. See AI Generate

You can use all three methods together. Each method creates a named source in your knowledge base, making it easy to organize and manage content over time.

Multiple Sources Per Agent

Each agent can have multiple knowledge sources -- for example:

• "Company FAQ" -- general questions about your business
• "Refund Policy" -- detailed refund and return procedures
• "Product Catalog" -- descriptions and pricing for your products
• "Troubleshooting Guide" -- step-by-step solutions for common issues

During a call, the agent searches across all sources simultaneously. The most relevant chunks are selected regardless of which source they belong to.

Sharing Knowledge Between Agents

If you run multiple agents (e.g., a Receptionist and a Customer Support agent), you can copy knowledge from one agent instance to another. This saves time when both agents need access to the same company information. See Managing Sources for details.

Knowledge Base in Your Dashboard

Your customer portal displays the total number of knowledge chunks for each agent instance. You can view all sources, see chunk counts, add new content, and remove outdated sources from the My Agents page.

---

A Product by BRTNeura Technology LLP