Skip to main content
The Knowledge Base stores reference content for retrieval during conversations. Adding content does not train a model or automatically assign it to every agent.

​
Add a source

Open Configure > Knowledge Base and choose: The current upload path supports text files; convert PDF or DOCX content to a supported text format before uploading. Uploads and total storage are checked against server limits. The reviewed upload limit is 20 MB per file, and URL content has a separate size limit. Use the application’s usage display and error messages for your account’s available capacity. A URL imports the fetched page content. Do not assume it recursively crawls a whole website or continuously refreshes when that page changes.

​
Check indexing

Knowledge items can be Pending, Processing, Completed, or Failed. Wait for successful indexing before checking retrieval. Search by name and filter by source type. Use Preview to inspect indexed content, and delete obsolete items when appropriate. If indexing fails, verify that the file contains readable text or the URL is accessible. Recreate or revise the source as needed, then verify the resulting item before assigning it.

​
Assign and verify

Select relevant items in the voice agent’s knowledge settings or a Chat Agent’s Knowledge tab, then save the agent. Retrieval-augmented generation (RAG) searches assigned sources when relevant; it does not guarantee that every response includes a knowledge result. Use a question whose answer exists in the source and a question whose answer is absent. Check that the agent answers the first accurately and handles the second according to its instructions. Knowledge access and synchronization can differ across providers and browser test modes; review any readiness error instead of assuming a failed test proves the source was ignored.