Upload document files
Upload document files such as PDF and DOCX, or scanned images, from a knowledge detail screen. After processing finishes, the extracted text is available to search and AI Chat.
Supported file formats
| Format | Extension |
|---|---|
.pdf | |
| Word | .docx |
| PowerPoint | .pptx |
| Excel | .xlsx |
| CSV | .csv |
| HTML | .html |
| Text | .txt |
| Markdown | .md |
| Image | .png, .jpg, .jpeg, .tiff, .webp, .bmp |
A file with an unsupported extension, an empty file, or a file larger than 100MB is not added to the upload list.
Upload screen
On the Documents tab of a knowledge detail page, select Add document → File upload.
Upload uses a three-step workflow.
| Step | Name | Description |
|---|---|---|
| Step 1 | File upload | Select files and configure parsing options. |
| Step 2 | Chunking settings | Configure a chunking strategy and values. |
| Step 3 | Review and submit | Review indexing options and submit the files. |
Step 1: Enter files and metadata
Drag files into the selection area or select the area to choose files.
- Upload multiple files: Select or drag multiple files at once.
- Remove a file: Remove an added file individually from the list.
- Format filtering: Unsupported file formats are excluded from the list.
- Metadata by file: Under Document metadata, enter a required name and optional alias, description, and tags for each file. Select an AI suggestion button for the alias, description, or tags to display a draft. The entered metadata appears in the document list.
This step configures common parsing options together with per-file metadata.
Preserve options between steps
Moving between steps 1–3 preserves parsing options, chunking options, and per-file metadata. If you change chunking settings in an earlier step, the preview shows the updated result.
Processing options
Parsing options (Step 1)
Parsing options control how text is extracted from a document.
| Option | Description | Default | Notes |
|---|---|---|---|
| Parse table structure | Preserves table structure while extracting content. | Enabled | On or off |
| Table mode | Selects the table extraction method. | Accurate | Fast or Accurate |
| Enable OCR | Recognizes text in images. | Disabled | On or off |
| OCR backend | Selects a traditional OCR engine or a vision model. | Traditional engine | A vision model is suggested first for image files |
| OCR engine / Vision model | Selects the engine or registered model used by the chosen backend. | Auto | See the following sections |
| Timeout (seconds) | Sets the maximum wait time for parsing. | 90 | 10–600 |
Capture: Step 1 after selecting a scanned image, with OCR backend suggesting Vision model and a collection-scoped model selector visible.
OCR engine options
Select an engine for optical character recognition (OCR). Auto chooses an engine based on the file format.
| Engine | Description |
|---|---|
| Auto | Selects a processing engine by file format. |
| Tesseract | Uses the Tesseract engine. |
| EasyOCR | Uses the EasyOCR engine. |
| RapidOCR | Uses the RapidOCR engine. |
Read scanned documents with a vision model
To extract text and structure from a scanned PDF or image without a text layer, turn on Enable OCR and select OCR backend > Vision model.
- When an image is added, Vision model is selected automatically unless you already changed the backend. Removing the last image restores the traditional-engine suggestion.
- Under Vision model, select an image-capable model registered in the same collection as the knowledge. If none is available, review collection scope and capabilities under LLM models.
- The vision model converts each page image to Markdown. Parse table structure and OCR engine do not apply to this path.
This path does not preserve source images as searchable results or support querying the image itself. It sends converted text and chart descriptions through the existing chunking and search flow.
To reduce synchronous processing cost, the pre-upload chunk preview uses a traditional engine. After upload completes, review the vision model's final Markdown in the document chunks.
Chunking options (Step 2)
Chunking options determine how extracted text is divided into searchable chunks. See Chunking and search options for strategy and value selection guidance.
The Chunk preview panel in Step 2 shows the result produced by the current settings. Review chunk size and split positions before indexing.
Indexing options (Step 3)
Under indexing options, select the search modes applied to the document.
The available methods depend on the knowledge settings. See Chunking and search options for their differences.
Monitor status after upload
Submitting files starts parsing, chunking, and indexing. Review the current state of each document in the document list.
Document states
| State | Color | Description |
|---|---|---|
| Pending | Blue | Waiting for processing. |
| Preparing | Blue | Preparing to process. |
| Processing | Blue, animated | Parsing, chunking, and indexing. |
| Complete | Green | Indexing has finished. |
| Paused | Yellow | Processing was paused by a user. |
| Failed | Red | Processing stopped because of an error. |
| Canceled | Gray | Processing was canceled by a user. |
| Reindexing | Blue, animated | Reindexing is in progress. |
Documents in an active state—Pending, Preparing, Processing, or Reindexing—refresh automatically every five seconds.
Actions by state
- Processing: A Stop button appears.
- Paused: Resume and Cancel buttons appear.
- Complete: A Reindex button appears.
- Failed: The error message and a Try again button appear.
Document detail page
Select a completed document to open its detail page. The page provides the following information and actions.
- Indexing status card: Current processing state and progress information.
- Chunk list: Generated chunks, divided into pages. Change the number of chunks per page below the list.
- Download file: Downloads the original file.
- Reindex: Processes the document again with its saved options.
For a document with tables or images, compare the Fast and Accurate table modes and OCR results in Chunk preview before choosing settings.
If processing fails after exceeding the limit, increase Timeout (seconds) and upload the file again.
Next steps
- Create a manual document — Add a document by entering text directly.
- Collect web pages — Collect documents from a URL.
- Review search results — Confirm that indexed documents are retrieved as expected.