How to build a Chat with Knowledge Base AI Agent

TL;DR

• Lets users ask natural‑language questions and get precise, cited answers from your docs, wikis, or tickets.

• Built with a workflow: Input → KB retrieval → LLM generation → Output  .

• Keeps answers grounded—reduces search time, confusion, and reliance on experts. 

• Ideal for teams struggling with fragmented documentation and inconsistent search. 

• Fast to launch using Stack AI’s Workflow Builder. 

Common Pain Points of Searching for Information

• Information scattered across tools, hard to locate. 

• Traditional search returns links—not actionable answers. 

• Keeping answers consistent as content scales is tough. 

• Slow workflows, repeated questions, and outdated info. 

• Low trust in answer quality and source reliability. 

What the Agent Delivers

• Natural‑language interface pulling from your docs with citations. 

• Clear, contextual answers—not just links. 

• Reduces dependency on SMEs due to consistent quality. 

• Easy to configure via Workflow Builder. 

• Keeps knowledge retrieval fast, reliable, and scalable. 

Workflow Overview

Node Details

1. Text Input

• Purpose: Entry point for user questions.

• How it works: User types a message, which is sent to both the Knowledge Base and the LLM.

2. Knowledge Base

• Purpose: Searches your knowledge base for relevant information.

• How it works: Receives the user’s question and returns the most relevant context chunks.

3. OpenAI (LLM)

• Purpose: Generates a response using both the user’s question and the knowledge base context.

• Prompt Example:

• Model: gpt-4o-mini

• Provider: OpenAI

4. Output

• Purpose: Displays the AI’s answer to the user.

Node Connections

• Text Input → Knowledge Base

• Text Input → OpenAI (LLM)

• Knowledge Base → OpenAI (LLM)

• OpenAI (LLM) → Output

How to Use

1. User enters a question in the Text Input node.

2. Knowledge Base node retrieves relevant info.

3. OpenAI LLM uses both the question and KB context to generate a response.

4. Output node shows the answer.