Knowledge base documents
Add source material that Nviti can process into knowledge for your assistants. Open Knowledge Base → KB Documents, or select Documents from the knowledge overview.
Choose a source
Start at source selection and continue through upload and processing checks.
| Source | What to provide |
|---|---|
| URL Link | A descriptive title and an accessible web URL for each source. |
| Upload Document | A title and a TXT, HTML, Markdown, PDF, or CSV file. The current upload limit is 20 MB per file. |
| Upload Sitemap | An XML or JSON sitemap for importing URL sources. |
Supported filename extensions are .txt, .html, .htm, .xhtml, .md, .markdown, .pdf, and .csv. The current document-upload form does not accept Word or Excel formats. Export those files to a supported format first. For PDFs, use readable text and check the processed result; scanned images may not yield useful text.
Processing states
| State | Meaning |
|---|---|
| Pending | Waiting for a worker. |
| Processing | Source extraction and knowledge processing are underway. |
| Processed | Processing completed. Test the relevant assistant before relying on the answer. |
| Failed | Processing did not complete; open the record for the reported error. |
Processing time depends on source size, accessibility, and queue workload. Open a document to review its source, filename, size, uploader, timing, chunk progress, and any failure information.
Keep sources current
A URL is fetched for processing; it is not a promise that every chat reads the latest website content. After changing a source, use Reprocess when available. Failed and processed documents can be queued again; bulk reprocessing skips documents already pending or processing.
For an uploaded file, add the revised source and review the old version so assistants do not receive conflicting guidance. Check processing status, assistant associations, and a realistic test question after an update.
Manage the list
Search by title or filename, filter by status, and use the available view, feature, reprocess, and delete actions. Feature/unfeature bulk actions help organize important documents.
Provide only information appropriate for the assistants that will use it. See Articles and AI Assistants for writing and attaching focused knowledge.
Company-private source files
New knowledge sources and supported new article attachments use company-private, metered storage. Extraction uses an owned temporary local spool, not an assumption that a cloud object has a local path. Published reader access verifies publication and channel/article membership; legacy public files remain subject to their existing compatibility path.
The staff mobile KB shows permitted published content; its compact navigation label is KB. Access to a source file and access to a published article are different checks.