Knowledge Management

This chapter covers building and operating collections of knowledge assets that agents can reference.

What Is a Collection

A collection is a knowledge storage unit grouping related documents. The retrieval node in an agentflow searches at the collection level.

Term	Description
Collection	A group of documents
Document	An individual asset within a collection (PDF, Word, text, etc.)
Chunk	A small text segment a document is split into for embedding
Embedding	A search-ready vector derived from a chunk

See the Glossary for full terminology.

Collection List

Select Knowledge Management → Knowledge Collections in the left sidebar.

Tab	Shows
Storage	Collections you created
Shared	Collections others shared with you
All	All collections you can access

Creating a Collection

1. Click + New Collection at the top right
2. Enter:
   • Name: identifiable name (Korean or English)
   • Description (optional): one-line summary
   • Encryption (optional): set a password if enabling password protection
   • Expiration date (optional): auto-delete date. Leave empty for permanent retention
3. Create

Button label

The actual solution button label is "새 컬렉션 생성" / "Create New Collection".

Document Upload

1. Collection detail → Upload button
2. Select files (PDF, DOCX, TXT, MD, etc.) — multi-select supported
3. Configure embedding options (defaults recommended)

Option	Korean	Meaning	Default
Chunk Size	청크 크기	Max characters per chunk	1000
Chunk Overlap	청크 오버랩	Overlap between adjacent chunks	200
Ontology	온톨로지	Auto-extract concepts and relationships	Enabled
PII Scan	PII 스캔	Auto-detect and mask PII	Enabled

1. Start Upload → progress bar appears

Processing Time After Upload

Large documents take time to embed. Track progress in the Upload History tab. You can continue other work during processing.

Upload History

Check status and results of uploaded documents.

Status	Meaning
Queued	In the processing queue
Processing	Chunking and embedding underway
Completed	Searchable
Failed	Error occurred (check logs)

Ontology

A collection uploaded with the Ontology option enabled carries an 온톨로지 (Ontology) badge on its card. Click the card to enter the collection detail, then click the 온톨로지 (Ontology) button at the top right to visualize the concepts and relationships automatically extracted from the uploaded documents as an interactive graph.

Visualization Layout

Region	Content
Top stats	Triples / Classes / Properties / SCS counts — a summary of the extracted knowledge assets
Top-right actions	Rebuild · View Profile · Fullscreen
Body	Interactive graph of Class (blue) / Instance (green) / Property (orange) nodes connected by edges
Bottom-left legend	Node types (Class · Instance · Property) and edge types (instanceOf, subClassOf, datatypeProperty, ObjectProperty)
Top-left search	Node search — filter graph nodes by keyword

GraphRAG Query

Type a question in the GraphRAG Query area at the bottom of the screen to generate an answer grounded in the ontology above. When the Context ON toggle at the top right is enabled, the graph's nodes and relationships are included in the LLM context to improve answer quality.

Operational Guidance

• When to rebuild — When uploaded documents are partially refreshed the graph may reflect a stale snapshot. After bulk additions or replacements, run Rebuild to keep the graph current.
• Working with dense graphs — When the graph is too crowded to read at a glance, use View Profile and the Node search at the top left to narrow the scope.
• Cost of Context ON — GraphRAG context can improve answer accuracy but also raises LLM call cost. Monitor both response quality and cost together when defining policy for the production environment.

Sharing a Collection

Grant other users access:

1. Collection detail → Share button
2. Search and select users
3. Choose permission (Read / Read·Write)
4. Save

File Storage

Besides uploaded files, File Storage (system file resources) can serve as a source for collections. From Knowledge Management → File Storage in the left sidebar, click + New Storage at the top right to open the creation modal and enter the storage name, description, and encryption toggle.

Field	Description
Storage name	A one-liner identifiable to others
Description	A paragraph about what this storage holds
Encryption	Whether to protect the storage with a password

DB Integration

DB Integration (tables / views from external databases) is also available as a source. From Knowledge Management → DB Integration in the left sidebar, click + New Connection at the top right to open the database connection registration modal.

Field	Description
Connection name	A one-liner identifiable to others
Description	A paragraph about which database this is and its purpose
Custom password	An optional access password specific to this connection
Database type	PostgreSQL, MySQL, etc. — pick from the dropdown

Table- and column-level documentation (descriptions, sample values, policies) is managed on a separate DB Documentation screen and is only reachable once at least one DB connection has been registered.

Operational Recommendations

• Separate collections by purpose — Different audience or classification criteria warrant separate collections. Cramming too much into one degrades retrieval quality.
• Periodic cleanup — Remove or set expirations on outdated and duplicate documents.
• Verify PII policy — For documents containing personal information, confirm that PII Scan is enabled.

Contact

For knowledge management questions, please contact the Xgen Solution Administrator.