Skip to main content

Knowledge Base

A Knowledge Base is the set of documents and web content your agents can draw on to answer callers accurately. Instead of cramming everything into a prompt, you add your material once and let the agent look things up during a call — pricing, policies, product details, FAQs, and more.

From your content to answers on the call — upload documents or scrape a website into the knowledge base, which the agent uses to answer questions during calls

Where knowledge lives​

Knowledge can be attached at two levels:

  • Workspace knowledge base — shared Knowledge items available to agents in the workspace.
  • Per-agent documents — files attached directly to a single agent.

Both places have the same two tabs: your content (Documents / Knowledge items) and URL scrape history.

Add knowledge​

1. Upload documents​

Upload files directly — PDF and HTML are supported. Drag and drop or pick files, and they're added to the knowledge base once processed.

tip

Well-structured documents work best. Clear headings and short sections help the agent find the right passage quickly.

2. Scrape a website​

Point Chatley at a URL and it will crawl the site and turn the pages into knowledge:

  • Scrape depth — how many links deep to crawl from the starting URL (default is 2). Higher depth pulls in more pages but takes longer.
  • Scrape history — every scrape job is recorded. Track progress in the URL scrape history tab.
  • Reuse — if a URL has already been scraped, Chatley can reuse the existing result instead of crawling again, saving time.

How agents use it​

Once knowledge is in place, an agent (or a workflow Knowledge Base tool) can query it mid-call to answer questions from your content rather than guessing. You add the knowledge once; every call the agent handles can use it.

Keeping knowledge fresh​

  • Re-scrape a site when its content changes (reuse keeps unchanged pages fast).
  • Remove documents that are out of date so the agent doesn't cite stale info.
  • Split very large documents into focused files for better retrieval.

Troubleshooting​

ProblemWhat to check
Upload rejectedConfirm the file is a supported type (PDF/HTML) and within the size limit.
Scrape returns little contentSome sites block crawlers or load content with JavaScript; try a more specific URL or a different depth.
Agent doesn't use the knowledgeMake sure the knowledge is attached to that agent (or its workspace) and the agent/workflow is set up to consult it.

Next steps​

  • Agents — attach documents to an agent.
  • Workflows — use the Knowledge Base tool inside a flow.