Skip to main content
Version: v0.1.0

Managing Knowledges

A Knowledge is the basic unit for transforming unstructured data into structured knowledge. It collects data from various sources such as web pages, document files, and manually written text, then processes and stores them in a searchable format. Each Knowledge serves as a knowledge base for a specific topic or domain.

The Knowledge management screen allows you to create Knowledges, collect documents, and ask questions about collected knowledge via RAG-based AI Chat. This document guides you through the entire management flow, from the Knowledge list screen layout, new Knowledge creation methods, the tab structure of the detail screen, to the deletion process.


Knowledge List Screen

You can view the full Knowledge list at the /knowledge path.

View Switching

The list screen supports two view modes:

View ModeDescription
Card ViewDisplays Knowledges in card format for visual overview. Great for quickly assessing status.
Table ViewDisplays the list in table format. Useful for comparing or sorting many Knowledges.

Click the view toggle icon in the upper right to switch to your preferred mode.

Search and Filtering

Use the search bar at the top of the list to quickly search by Knowledge name. Combined with tag filters, you can narrow down to Knowledges of specific categories.

info

As the number of Knowledges grows, systematically managing tags greatly helps search efficiency. It's recommended to establish tag naming conventions within your team in advance.

Knowledge Card Information

In card view, each Knowledge card displays the following information:

ItemDescription
NameThe unique identifying name of the Knowledge
Document CountTotal number of collected documents
Chunk CountTotal number of chunks the documents are split into
Created DateDate the Knowledge was created
TagsList of tags assigned for categorization

Creating a Knowledge

Click the + Knowledge button on the list screen to open the creation dialog.

Creation Fields

FieldRequiredDescription
NameThe unique name of the Knowledge. Lowercase English, numbers, and hyphens are recommended.
CollectionSelect a collection to link. Allows systematic resource management in conjunction with existing collections.
DescriptionDescription text about the Knowledge's purpose or contents.
Embedding ModelSelect the embedding model to use. Dimensions and provider information are displayed alongside each model.
Storage TargetSelect one or more from VECTOR, TEXT, GRAPH. Multiple selections are possible.
TagsAssign tags for categorization and search.

Storage Target Details

TargetEnginePurpose
VECTORVector DBSemantic similarity search based on embedding vectors
TEXTText search engineBM25-based keyword full-text search
GRAPHGraph DBEntity/relationship-based graph exploration
Storage Target Selection Guide

In most cases, the VECTOR + TEXT combination is most effective. Vector search ensures semantic similarity, while Text search supplements keyword accuracy to maximize Hybrid search quality.

After filling in all fields, click the Create button to create the Knowledge, which will automatically navigate to the detail screen.


Knowledge Detail Screen

You can view and manage individual Knowledge details at the /knowledge/:knowledgeId path.

Header Area

The top of the detail screen displays the following information:

  • Knowledge Name: Name of the current Knowledge
  • Document Count: Total number of collected documents
  • Chunk Count: Total number of split chunks

Tab Structure

The detail screen consists of 4 tabs:

Documents Tab (Default)

The default tab for managing the list of documents collected in the Knowledge. You can check each document's title, source type, chunk count, and status.

Documents can be added in three ways via the Add Document dropdown button:

MethodDescription
Web CrawlingEnter a URL to automatically collect web pages. Supports dynamic pages that require JavaScript rendering.
File UploadUpload document files in various formats including PDF, DOCX, PPTX, XLSX, HTML, TXT, and MD.
Manual InputWrite and register text chunks directly. Suitable for simple knowledge or FAQ-style data.
info

Detailed usage instructions for each document addition method can be found in the sub-documents.

Chat Tab

A RAG (Retrieval-Augmented Generation) based AI chat interface. You can ask questions and receive answers based on collected knowledge. Source document information is displayed alongside answers so you can verify the basis for responses.

tip

The Chat tab can also be used to conveniently validate the quality of collected documents. If expected answers are not being generated, try adjusting document chunking options or search modes.

Search Tab

A tab for testing the search quality of collected knowledge. You can enter search queries and check results using various search modes such as Vector, Text, and Hybrid. Verify knowledge quality by directly checking relevance scores and matched chunks in search results.

Settings Tab

A tab for managing Knowledge metadata and configuration options. You can manage the following items:

  • Basic Information: Modify name, description, and tags
  • Storage Settings: Check activated storage targets
  • Embedding Model: Check the currently applied embedding model and dimensions
  • Danger Zone: Irreversible operations such as Knowledge deletion

Deleting a Knowledge

You can delete Knowledges that are no longer needed. There are two deletion methods:

Deletion MethodDescription
Standard DeleteDeletes only the Knowledge metadata. Document and chunk data stored in storage is preserved.
Cascade DeleteCompletely deletes the Knowledge along with all sub-documents, chunks, and storage data.
Cascade Delete Caution

Executing a Cascade delete will permanently delete all documents and chunks belonging to that Knowledge. This action cannot be undone, so be sure to confirm before deleting. For Knowledges containing important data, backing up before deletion is recommended.


Next Steps

After creating a Knowledge, you can collect and utilize documents in various ways:

DocumentDescription
Web CrawlingURL-based automatic web page collection
File UploadDocument file upload and processing
Manual DocumentsWrite text chunks directly
AI ChatUtilize knowledge through RAG-based conversation