Skip to main content
Use Sources → Documents & Websites for organization-owned knowledge that does not come from Zendesk Guide.

​
Upload documents

The current uploader accepts:
  • PDF
  • DOCX
  • XLSX
  • CSV
  • TXT
  • Markdown (.md)
The current per-file upload limit is 9 MB. You can also add an optional description to help identify the document.

​
Processing states

Documents can appear with states such as:
  • Processing
  • Ready
  • Error
A document should reach Ready before you rely on it for normal semantic retrieval. The document list also exposes information such as source type, tags, word/page counts where available, crawl timestamps, and embedding/chunk state.

​
Activate or deactivate a document

Use the active state to keep a document stored while excluding it from normal Assistant use. Deactivate stale knowledge instead of leaving conflicting versions active while someone decides which policy is real.

​
Reprocess a document

Reprocessing rebuilds the processed/indexed representation for a document. Use it after a processing failure or when the source needs to be rebuilt.

​
Add a website

The Website manager presents one Websites surface rather than making administrators choose a technical mechanism before entering a URL. RAUM supports two broad patterns:

​
Indexed/crawled website

Use a normal URL when you want RAUM to crawl and index a site or section for semantic retrieval. The crawler can maintain multi-page website sources and expose crawl/run state and synchronization controls.

​
Dynamic website lookup

A URL containing a supported {{variable}} placeholder is treated as a per-ticket dynamic web source where the feature is available. This is useful when the page should be fetched using ticket-specific input rather than pre-indexed as static knowledge.

​
Automatic synchronization

Website/document records can expose automatic synchronization settings where supported, including an interval for refreshing the source. Choose an interval appropriate to how often the underlying content changes. Hourly refresh for a policy that changes twice a year mostly produces electricity and optimism.

​
Search and filtering

The knowledge list supports filtering by source type and processing status, plus search across stored documents. Use tags and descriptions consistently so administrators can distinguish authoritative sources from temporary or imported material.

​
Embedding repair

The Assistant Sources workspace exposes embedding repair for procedures, Help Center articles, and guidelines whose vectors are missing. For uploaded/crawled documents, use the document processing/reprocessing controls and review the provider error when indexing fails.

AI models

Embedding configuration determines whether semantic retrieval can work normally.

Sources overview

Choose the right source type for each kind of knowledge.