Knowledge Base Q&A

Learn how to build a RAG (Retrieval-Augmented Generation) workflow using both Document Vector Store and GitHub Vector Store together, with query optimization.

This recipe shows you how to build an app that combines multiple Vector Stores for a powerful RAG (Retrieval-Augmented Generation) workflow. Learn to integrate Document Vector Store and GitHub Vector Store with query optimization for comprehensive knowledge base Q&A.

Overview

Purpose: Learn how RAG works by combining multiple Vector Stores.

Node	Role	Model
Document Vector Store	Vectorize PDF/text documents	text-embedding-3-small
GitHub Vector Store	Vectorize code repository	text-embedding-3-small
App Entry	Question input	-
Generator x2	Optimize search queries	gpt-5-nano
Query Node x2	Search for relevant information	-
Generator	Generate answer from search results	claude-haiku-4.5

What you'll learn: RAG (Retrieval-Augmented Generation), Document Vector Store, GitHub Vector Store, query optimization, integrating multiple Query nodes

Workflow Diagram

flowchart LR
    A[Question Input] --> B1[Doc Query Generator]
    A --> B2[Code Query Generator]
    B1 --> C1[Document Search]
    B2 --> C2[Code Search]
    D[Document Knowledge Base] --> C1
    E[GitHub Knowledge Base] --> C2
    C1 --> F[Answer Generator]
    C2 --> F
    A --> F

Node Configuration Details

1. Document Vector Store

Item	Value
Node Name	`Document Knowledge Base`
Embedding Model	OpenAI text-embedding-3-small

Pre-loaded Files:

File	Source URL	Content
`docs.giselles.ai-llms-full.txt`	`https://docs.giselles.ai/llms-full.txt`	Documentation (Text)
`giselles-terms.pdf`	`https://giselles.ai/legal/terms`	Terms of Service
`giselles-privacy.pdf`	`https://giselles.ai/legal/privacy`	Privacy Policy

2. GitHub Vector Store

Item	Value
Node Name	`GitHub Knowledge Base`
Repository	`giselles-ai/giselle`
Include Code	ON
Include Issues	ON
Include Pull Requests	ON
Embedding Model	OpenAI text-embedding-3-small

3. App Entry

Item	Value
Node Name	`Question Input`
Node Input Name	`question`
Prompt	- (User input only)

4. Generator Node (Doc Query)

Item	Value
Node Name	`Doc Query Generator`
Model	gpt-5-nano
Thinking	OFF
Web Search	-

Prompt:

Convert the following question into an optimized search query for documentation:

Question: @Question Input

Generate a query with relevant keywords and concepts for searching documentation.
Output only the query, nothing else.

5. Generator Node (Code Query)

Item	Value
Node Name	`Code Query Generator`
Model	gpt-5-nano
Thinking	OFF
Web Search	-

Prompt:

Convert the following question into an optimized search query for code:

Question: @Question Input

Generate a query with function names, file patterns, and technical terms for searching code.
Output only the query, nothing else.

6. Query Node (Document)

Item	Value
Node Name	`Document Search`
Source	`@Document Knowledge Base`
Query	`@Doc Query Generator`
Max Results	10
Similarity Threshold	0.3

7. Query Node (GitHub)

Item	Value
Node Name	`Code Search`
Source	`@GitHub Knowledge Base`
Query	`@Code Query Generator`
Max Results	10
Similarity Threshold	0.3

8. Generator Node (Answer)

Item	Value
Node Name	`Answer Generator`
Model	claude-haiku-4.5
Thinking	ON
Web Search	OFF

Prompt:

Answer the following question based on the provided content:

Question: @Question Input

Documentation:
@Document Search

Code & Repository:
@Code Search

Requirements:
- Answer based on both documentation and code repository content
- Reference specific files or passages when applicable
- If the answer is not found, clearly state that
- Be concise but thorough

On this page