Skip to content

Manage Documents and Chunks

After creating a dataset, you can check document processing status, manage document enablement, and open the document detail page to review and adjust chunks.

Document List

After opening a dataset, the related files list appears below:

FieldDescription
File nameDocument name, extension, and file size.
WordsParsed word count.
RetrievalsNumber of times the document was retrieved.
Uploaded atTime when the document was added to the dataset.
StatusPending, processing, completed, or failed.
EnabledControls whether the document participates in retrieval.
ActionsView, download, rename, archive, or delete.

Dataset related files list

Manage Documents

You can perform these actions from the document list:

  • View: Open the document detail page to review metadata and chunks.
  • Download: Download the original file.
  • Rename file name: Update only the display file name. The original uploaded file is not changed.
  • Archive: Archive the document.
  • Delete: Delete the document.
  • Enable / Disable: Control whether the whole document can be retrieved.

Document Details

After clicking a document row, the document detail page shows:

  • File size
  • Uploaded time
  • Last updated time
  • Source
  • Paragraph length
  • Average paragraph length
  • Retrieval count
  • Embedding time
  • Embedding token count

Manage Chunks

The lower part of the document detail page shows the chunk list. Each chunk displays:

  • Chunk number
  • Word count
  • Retrieval count
  • Enabled status
  • Content preview

You can:

  1. Click the chunk content to open the edit dialog.
  2. Update the chunk text.
  3. Click Update chunk. EZChat updates the chunk and regenerates its embedding vector.
  4. Use the switch to disable chunks that should not be retrieved.
  5. Click the delete button to remove unwanted chunks.
  6. Click Add chunk to manually add a new chunk.

Document detail chunk list

Note

After you edit or add a chunk, EZChat regenerates its embedding vector. Before production use, review retrieval results to confirm the updated content behaves as expected.