Knowledge Base – Files
On this page
Purpose
Knowledge Base - Files reads uploaded files and returns processed file content for the rest of a flow. It can provide a DataFrame for file-by-file processing or raw content for a message-based step.
When to use it
- Extract text from uploaded documents.
- Prepare files for splitting, parsing, or type conversion.
- Send one uploaded document's raw content to a prompt or model.
Required setup
Upload one or more supported files. Choose the output that matches the next component: Files for DataFrame-aware steps or Raw Content for a message-based step. Use Structured File when CSV, Excel, Parquet, JSON, or YAML must preserve real columns or nested object structure.
How to use it
- Add Knowledge Base - Files and upload the files.
- Connect Files to a DataFrame-aware component, or use Raw Content for a message-based component.
- Change the file-processing controls only when the defaults do not fit the workflow.
Configuration
| Setting | What it controls | Recommended starting point |
|---|---|---|
| Files | Selects the files to process. Server File Path takes precedence when connected. | Start with one representative file. |
| Server File Path | Reads server-hosted files instead of the picker selection. | Leave empty unless an upstream step saved the file. |
| Processing Concurrency | Controls how many files process at once. The default is 1. | Keep 1 until file processing time is a problem. |
| Silent Errors | Skips standard parsing failures instead of raising them. It is disabled by default. | Leave it off until a skipped file is acceptable. |
| Delete Server File After Processing | Removes uploaded server files after processing. It is enabled by default. | Leave it enabled unless later flow steps need the server file. |
| Ignore Unsupported Extensions | Skips unsupported files instead of stopping the run. It is enabled by default. | Leave it enabled for mixed uploads. |
| Separator | Defines the boundary when content is combined. | Use a visible separator when a downstream prompt needs to distinguish files. |
| Advanced Parser | Exposes document-processing controls for one compatible file. | Enable only when basic extraction is not enough. |
Expected result
The component returns processed file data or raw content. Raw Content joins the parsed files and fails above 10 MiB rather than truncating. Unsupported files are ignored or reported according to the selected controls. Advanced document processing can take more resources than basic extraction.
Reference details
Use Advanced Parser for one compatible file when you need Docling processing or OCR options. The current VLM selection falls back to the default converter, and advanced parser diagnostics can include file paths or parser details. Test a representative non-sensitive file before a larger upload.
Limitations
Knowledge Base - Files does not index, embed, search, or retrieve by itself. Files selected in Builder are included with the published version. Publishing stops if a selected file is unavailable; reselect the file and try again. Use Server File Path only when an upstream flow step owns the file lifecycle.
Works well with
Use Split Text to chunk extracted text, Parser to structure a result, and Type Convert when the next component needs a different value type.