> For the complete documentation index, see [llms.txt](https://docs.clickai.vn/clickai-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.clickai.vn/clickai-docs/clickai-docs-en/database/database-and-knowledge.md). # Database & Knowledge
## Table of Contents · \[Introduction to RAG]\(#introduction-to-rag) · \[Practical Use Cases]\(#practical-use-cases) · \[Creating a Knowledge Base]\(#creating-a-knowledge-base) · \[Management & Optimization]\(#management--optimization) · \[Integrating into Applications]\(#integrating-into-applications) · \[Retrieval Strategies]\(#retrieval-strategies) ## Introduction to RAG RAG (Retrieval-Augmented Generation) is the core architecture of the Knowledge system in ClickAI: 1\. Retrieval — When a user asks a question, the system first retrieves the most relevant information from the incorporated knowledge. 2\. Augmented — This retrieved information is then combined with the user's original query and sent to the LLM as augmented context. 3\. Generation — The LLM uses this context to generate a more precise answer. 💡 TIP: RAG enables AI to answer based on your up-to-date data rather than relying solely on pre-training knowledge, significantly reducing hallucination. ## Practical Use Cases
Use CaseDescription
Customer Support BotAccurate answers from product docs, FAQs, troubleshooting guides
Internal Knowledge PortalAI search & Q&A for employees accessing company policies and procedures
Content GenerationGenerate reports, articles, emails based on background materials
Research & AnalysisRetrieve and summarize from research repositories, market reports
## Creating a Knowledge Base ClickAI provides 3 ways to create a Knowledge Base: ### Method 1: Quick Create (Recommended for beginners) The fastest way — import data, define processing rules, and ClickAI handles the rest. Steps: 4\. Go to Knowledge from sidebar → Click "+ Create Knowledge" 5\. Name and describe your Knowledge Base 6\. Upload documents — Supported formats:
FormatType
.pdfPDF documents
.docx .docMicrosoft Word
.txtPlain text
.mdMarkdown
.csvData tables
.htmlWeb pages
.xlsxExcel spreadsheets
7\. Configure chunking — How to split documents:
MethodDescriptionWhen to Use
AutomaticClickAI optimally splitsMost cases
CustomCustom separator, chunk sizeSpecially structured documents
Parent-ChildLarge chunks containing smaller chunksNeed broader context
8\. Select Embedding Model — Model to convert text into vectors 9\. Choose Index Method: · High Quality: Uses Embedding Model (more accurate, uses tokens) · Economical: Uses keyword indexing (saves costs, less accurate) 10\. Click "Save & Process" → Wait for indexing to complete ### Method 2: Knowledge Pipeline (Advanced) Orchestrate more complex data processing workflows with custom steps and plugins. · Suitable for complex data pre-processing needs · Supports ETL (Extract-Transform-Load) plugins · Reusable workflows ### Method 3: External Knowledge Base Connect directly to external knowledge bases via APIs. · Leverage existing data without migration · Supports vector databases: Pinecone, Weaviate, Qdrant, ... · Real-time synchronization ⚠️ IMPORTANT: Knowledge Base quality directly determines AI response quality. Invest time in preparing clean, well-structured documents. ## Management & Optimization ### Content Management
ActionDescription
View documentsSee list of uploaded documents
Add documentsUpload new documents
Edit chunksEdit chunk content
Enable/DisableToggle document or chunk
DeleteRemove unneeded documents
Re-indexRe-index when settings change
### Test & Validate Retrieval Test retrieval quality before integrating into applications: 11\. Open Knowledge Base → Click "Test Retrieval" 12\. Enter a test query (simulating user questions) 13\. Click "Run Test" 14\. Review results: · Matched chunks — Retrieved document segments · Relevance score — Relevance rating (0-1) · Source document — Source document reference 💡 TIP: Test with diverse questions to ensure the Knowledge Base covers all topics. If relevance scores are low (below 0.5), consider improving document content. ### Metadata Enhancement Add metadata to documents for improved retrieval accuracy: \# Example Metadata\ document: "Warranty Policy 2024.pdf"\ metadata:\ category: "policy"\ department: "customer-service"\ effective\_date: "2024-01-01"\ language: "en" Metadata enables filter-based search, e.g., "Only search within category=policy documents." ### Adjusting Settings You can change at any time: · Index method — Switch between High Quality and Economical · Embedding model — Change to a different embedding model · Retrieval strategy — Semantic Search, Full-text Search, or Hybrid
## Integrating into Applications ### In Chatbot / Agent 15\. Open the app in Studio 16\. In the Context section, click "+ Add" 17\. Select the Knowledge Base to connect 18\. Configure Retrieval Settings:
SettingDescriptionRecommended Value
Top KNumber of chunks returned3-5
Score ThresholdMinimum score threshold0.5-0.7
Retrieval ModeSearch methodHybrid Search
### In Workflow / Chatflow Use the "Knowledge Retrieval" node: 19\. Drag the Knowledge Retrieval node onto the canvas 20\. Select a Knowledge Base 21\. Configure the query variable (typically {{sys.query}}) 22\. Connect the output to an LLM node as context
## Retrieval Strategies ### Semantic Search Searches based on contextual meaning using embedding vectors. · ✅ Understands synonymous questions · ❌ May miss exact keyword matches ### Full-text Search Searches based on exact keyword matching. · ✅ Precise with specific terminology · ❌ Doesn't understand context ### Hybrid Search (Recommended) Combines both Semantic + Full-text, using Rerank to sort results. · ✅ Leverages advantages of both methods · ✅ Highest accuracy 📝 NOTE: \*\*Rerank\*\* is an additional processing step that re-sorts retrieval results by relevance. Enable Rerank when using Hybrid Search for the best results. *📖 Previous: \[Monitor]\(./03-monitor-en.md) · Next: \[Workspace]*