Manual Documents

Manual Documents is a method of adding text chunks directly to a Knowledge without file uploads or web crawling. Users can precisely control the content without any parsing or chunking process.

Creating a New Manual Document

Create a new manual document at the /knowledge/:knowledgeId/documents/manual/new path. Click the New Manual Document button from the Documents tab on the Knowledge detail page.

Input Fields

Field	Required	Description
Title	Required	The title of the document
Description	Optional	A description of the document
Initial Content	Optional	Initial text to be registered as the first chunk

If initial content is entered, the first chunk is automatically created along with the document. You can also leave it empty and add chunks later on the detail page.

Click the Create Document button to create the document and automatically navigate to the detail page.

tip

Initial content is optional when creating a manual document. It's also a good approach to create the document first and then add chunks one by one on the detail page.

Manual Document Detail

Manage the detail information and chunks of a manual document at the /knowledge/:knowledgeId/documents/manual/:docId path.

Document Information

Basic document information is displayed in the top area:

Item	Description
Type	MANUAL tag
Chunks	Total chunk count
Created	Creation date and time
Description	Document description (inline editable)

The Description can be edited directly on the page by clicking the Edit button without navigating away.

Chunk List

Chunks contained in the document are displayed in table format.

Column	Description
Checkbox	Multi-select (for bulk deletion)
#	Chunk sequence number
Content	Chunk content preview (2 lines)
Tokens	Token count (estimated)
Actions	Edit, Delete buttons

Pagination is handled server-side in units of 20. When chunks exceed 20, page navigation appears at the bottom.

Chunk Detail View

Click a chunk row in the table to open a Chunk Drawer side panel on the right:

• Navigation: Move to previous/next chunks (keyboard arrow keys supported)
• Content: View and copy the full chunk text
• Metadata: View metadata in Key-Value format
• Info: Chunk ID, storage (VECTOR/TEXT), creation/modification timestamps

Chunk Management

Adding Chunks

Click the Add Chunk button to open the chunk editing modal.

Modal input fields:

Field	Description	Limit
Content	Chunk text content	Max 10,000 characters
Metadata	Key-Value pair metadata	Dynamically add/remove

The estimated token count of the current text is displayed in real-time at the bottom of the modal.

Editing Metadata

Metadata consists of Key-Value pairs:

• Click the Add Metadata button to add a new entry
• Each entry has a Key input field and a Value input field
• Click the delete button on the right side of an entry to remove it
• Values support automatic type inference: true/false → Boolean, numbers → Number, JSON strings → Object/Array

Metadata Merging

When modifying existing chunks, metadata is updated using a merge approach. Assigning a new value to an existing key overwrites it, and new keys are added.

Editing Chunks

Click the Edit button in the chunk table or the Edit button in the Chunk Drawer to open the editing modal. Existing content and metadata are pre-loaded, and after saving, changes are automatically reflected in the respective storage (Vector/Text).

When content is changed, the embedding vector is automatically regenerated.

Deleting Chunks

Two deletion methods are provided:

• Individual Delete: Click the Delete button on a chunk row → Confirm (Popconfirm) and delete
• Bulk Delete: Select multiple chunks via checkboxes, then click the Delete (N) button

caution

Deleted chunks are removed from both Vector DB and Text DB. This action cannot be undone.

Deleting a Document

You can delete the entire manual document using the Delete button in the detail page header. Deleting a document also removes all contained chunks.

Use Cases

Manual documents are useful in the following situations:

• Registering FAQ data: Register question-answer pairs as individual chunks to ensure accurate search results
• Entering internal policies/regulations: Input policy content directly without a separate file
• Text data extracted from other systems: Register text copied from other systems chunk by chunk
• Saving AI chat responses: Save useful responses from AI Chat as manual document chunks for knowledge reuse
• Supplementing structured data: Add short texts or memos that are difficult to process via file upload

tip

Manual documents are suitable for small-scale text data. For large volumes of document data, use File Upload.

Next Steps

• File Upload - Document file upload and automatic processing
• Web Crawling - URL-based automatic web page collection
• AI Chat - Utilize registered knowledge through RAG-based conversation
• Search Test - Verify search quality of indexed chunks