Skip to main content

Knowledge Base Guide

Knowledge Base enables you to store, organize, and retrieve documents for RAG (Retrieval-Augmented Generation), allowing AI agents to answer questions based on your specific documents.

🎯 Overview​

What is Knowledge Base?​

Knowledge Base is a document storage and retrieval system that enables RAG. It allows AI agents to search through your documents and provide answers grounded in your specific content.

Core Benefits​

BenefitDescription
Grounded ResponsesAI answers based on your documents
Source CitationsEvery answer includes references to source documents
Smart ChunkingAutomatic document splitting optimized for retrieval
Flexible RetrievalVector, keyword, or hybrid search modes

πŸ†• Creating Knowledge Bases​

Step 1: Navigate to Knowledge Base​

  1. Log in to Wegent
  2. Navigate to Knowledge Base section in the sidebar
  3. Click Create Knowledge Base

Step 2: Choose Knowledge Base Type​

TypeDocument LimitChat SupportBest For
Notebook50 documentsβœ… YesInteractive research, Q&A sessions
ClassicUnlimited❌ NoLarge document collections, archives

See Knowledge Base Types for detailed comparison.

Step 3: Configure Basic Settings​

FieldDescriptionExample
NameDisplay name (1-100 chars)"Product Documentation"
DescriptionOptional description (max 500 chars)"Internal product docs and guides"

Step 4: Configure Retrieval Settings​

SettingOptionsDescription
Retrieval ModeVector / Keyword / HybridSearch method
top_k1-10 (default: 5)Number of results
score_threshold0.0-1.0 (default: 0.5)Minimum relevance score

Step 5: Upload Documents​

  1. Click Add Document
  2. Choose source type (File / Text / URL / External Table)
  3. Configure chunking settings (optional)
  4. Click Upload

πŸ“„ Document Management​

Supported Sources​

SourceDescription
File UploadUpload from local computer
Text PastePaste text content directly
External TableImport from DingTalk/Feishu
Web ScrapingScrape content from URL

Supported Formats​

  • .txt - Plain text files
  • .md - Markdown files
  • .pdf - PDF documents
  • .doc, .docx - Word documents

See Document Management for detailed guide.

πŸ”ͺ Chunking Strategies​

Available Strategies​

StrategyBest ForDescription
Smart ChunkingGeneral documentsAuto-detect structure
Sentence-basedFAQ, Q&A contentSplit by sentence boundaries
SemanticLong articlesSplit by semantic similarity

Key Parameters​

ParameterRangeDefaultDescription
chunk_size128-8192500Characters per chunk
chunk_overlap0-204850Overlapping characters

See Chunking Strategies for detailed guide.

πŸ” Retrieval Test​

For smaller knowledge bases, the system may also use an all-chunks direct injection path to load the full chunk set into the model. This exists mainly to compensate for cases where vector retrieval recall is not stable enough.

See Retrieval And Direct Injection for the design rationale and permission model.

Before saving retrieval configuration, you can test the retrieval effectiveness.

How to Use​

  1. Go to Knowledge Base Retrieval Settings
  2. Configure retrieval parameters
  3. Enter a test query in the Retrieval Test area
  4. Click Test button
  5. Review returned chunks and relevance scores
  6. Adjust parameters based on results
  7. Click Save when satisfied

πŸ“ Summary Features​

Document Summaries​

When enabled, the system automatically generates summaries for each document:

  1. Document is uploaded and indexed
  2. Summary model processes content
  3. Summary stored with document metadata
  4. Available for quick reference

Knowledge Base Summary​

Generate an aggregate summary of the entire Knowledge Base:

  1. Navigate to Knowledge Base settings
  2. Click Generate Summary
  3. View combined summary of all documents

Manually Editing Knowledge Base Summary​

Knowledge base summaries can be edited manually to correct AI-generated content, add missing context, or rewrite the summary in a form that better fits team usage.

  • Manual summary takes priority in UI display
  • Manual summary takes priority when injected into knowledge-base chat context
  • AI summary generation continues running in the background
  • Use Restore AI Summary to switch back to the latest AI-generated summary

Notes:

  • The current version only supports manual editing of the knowledge base long summary
  • Document summaries are still generated automatically
  • If Auto Generate Summary is disabled, new AI summaries stop updating, but saved manual summary content can still be displayed

Summary Retry​

If summary generation fails:

  1. Find the document with failed summary
  2. Click Retry Summary button
  3. System will attempt to regenerate

πŸ€– Integration with Agents​

Selecting Knowledge Base in Chat​

  1. Click the context selector near the chat input
  2. Select Knowledge Base from options
  3. Choose your desired Knowledge Base
  4. Send your question

Notebook Mode Chat​

In Notebook mode, you can chat directly within the Knowledge Base:

  1. Open a Notebook-type Knowledge Base
  2. Use the chat panel on the right
  3. Select specific documents for context (optional)
  4. Ask questions about your documents

πŸ“‘ Citations and References​

Citation Format​

AI responses include numbered citations linking to source documents:

Based on your product documentation [1], the feature works by...
Additionally, the troubleshooting guide [2] suggests...

Sources:
[1] product-overview.pdf (Section 3)
[2] troubleshooting-guide.md (Chunk 15)

Viewing Source Content​

  1. Click on a citation number in the response
  2. View the original chunk content
  3. See surrounding context
  4. Navigate to full document if needed

πŸ” Permission Management​

Knowledge bases support fine-grained access control through the Share Service architecture. You can manage members, assign roles, and authorize external entities (e.g., groups) to access your knowledge bases.

Member Roles​

RolePermissions
OwnerFull control, can transfer ownership and manage all members
MaintainerCan manage documents, settings, and invite members
DeveloperCan read and contribute documents
ReporterRead-only access

Managing Members​

  1. Open a knowledge base and click Permission Management
  2. Under the Personal tab, invite users by username or email
  3. Assign a role from the dropdown (Owner, Maintainer, Developer, Reporter)
  4. Click Add to send the invitation

Entity-Level Authorization​

In addition to individual users, you can authorize entire groups or namespaces:

  1. Switch to the Group tab in the permission dialog
  2. Search for a group or namespace
  3. Select the desired group and assign a role
  4. All members of that group inherit the assigned permissions

Note: Entity-authorized knowledge bases appear in the members' Shared with Me section rather than under the group's native knowledge bases.

Permission Source Visualization​

When viewing members, the system displays how each member gained access:

  • Direct β€” Added directly as a member
  • Entity β€” Access granted through a group or namespace
  • Link β€” Access obtained via a share link

Role Conflict Resolution​

When a user has multiple access paths to the same knowledge base (e.g., direct membership and group membership), the system automatically resolves conflicts by selecting the highest-privilege role.

Ownership Transfer​

Owners can transfer ownership to another member:

  1. In Permission Management, locate the target member
  2. Click Transfer Ownership next to their name
  3. Confirm the transfer in the dialog

The previous owner is downgraded to Maintainer, and the new owner gains full control.

❓ Troubleshooting​

Upload Issues​

ProblemSolution
File upload failsCheck file size (max 50MB) and format
Document stuck in "Processing"Wait for large documents or re-upload

Retrieval Issues​

ProblemSolution
No results returnedLower score_threshold or try different query
Irrelevant resultsReduce chunk_size or try hybrid mode

πŸ’‘ Best Practices​

Document Organization​

PracticeDescription
Meaningful namesUse descriptive document names
Consistent formatStandardize document formatting
Regular updatesRe-index when documents change
Clean contentRemove irrelevant headers/footers

Retrieval Tuning​

ScenarioModetop_kthreshold
Precise answersVector3-50.8
Broad searchHybrid8-100.6
Exact matchingKeyword50.7

πŸ”— Related Resources​