Add content to a knowledge base
Upload files, paste text or add a web page, with the file types and limits that apply and what the statuses mean.
What you will achieve
Your documents will be indexed and searchable by an agent.
Prerequisites
- A knowledge base (see Create a knowledge base).
Step-by-step: upload a file
- Open the base and choose Upload (or Upload a file on an empty base).
- Pick one or more files.
Supported types: PDF, Word (.docx), and plain text formats: .txt, .md, .markdown, .rst, .csv, .json, .yaml, .yml.
Limits: 20 MB per file; PDFs up to 500 pages. Old .doc files are not supported: save them as .docx. Scanned PDFs with no text layer cannot be read.
Step-by-step: paste text
- Choose Paste text (on an empty base, or beside Upload once it has documents).
- Enter a Title (for example "Refund policy") and the Text.
- Choose to add it. This suits an FAQ answer, a policy or a script you would rather type than upload. Up to 500,000 characters.
Step-by-step: add a web page
- Choose Add URL.
- Enter the URL (for example
acme.com/help/refunds). - To follow links, tick Follow links on this site (same domain only, respecting robots.txt), then set Most pages (up to 100) and Depth (how many links deep, up to 3). Following links is opt-in, so a crawl of a whole site is a decision, not a default.
Pages are read as text. JavaScript-rendered pages are not supported. Private and internal addresses are refused.
Statuses
| Status | Meaning |
|---|---|
| Queued | Waiting to be processed. |
| Indexing | Being split into passages and embedded. |
| Indexed | Ready to be searched. |
| Failed | Could not be read or embedded; the row shows why. |
The Passages column counts the pieces a document was split into.
Keep web pages current
Set Automatic refresh in the base's Settings to the hours between re-fetching web pages (0 turns it off, up to 720). Unchanged pages are skipped. Refresh at the top re-fetches now.
Testing
When a document shows Indexed, open Test retrieval and ask a question it should answer.
Expected result
The document shows Indexed with a passage count, and its text appears in test answers.
Troubleshooting
See Knowledge troubleshooting.
Related guides
Something here does not match what you see in the dashboard? Let us know.