Sources are how you teach your agent about your business. Everything in Sources becomes searchable knowledge the agent draws on when answering customers.
Four kinds of sources
- Files — PDFs, Word docs, spreadsheets, Markdown. Best for static reference: policies, manuals, product specs.
- URLs — webpages crawled directly. Best for frequently changing content like pricing, blog posts, help articles.
- Text — paste anything directly. Best for short policies, brand guidelines, tone of voice notes.
- Q&A pairs — explicit question/answer pairs. Best for high-stakes answers that must be exact.
Tip
Use Q&A pairs for the questions where being right matters most: pricing, refund policy, hours of operation, contact info. They give you deterministic answers.
How training works
When you add a source, Chatzuri does three things:
- Extracts the text content (parses PDFs, scrapes webpages, reads files)
- Chunks it into searchable passages
- Embeds each chunk into the agent's knowledge index
At chat time, the agent searches this index for the most relevant passages and grounds its answer in them.
Where to find Sources
On any agent, click the Sources tab. You'll see four sub-tabs (Files, URLs, Text, Q&A) and a counter for each.
How much is enough?
For most agents, 5–20 well-chosen sources are far better than 200 random ones. Quality beats quantity:
- Your top 5 most-asked questions as Q&A pairs
- Your pricing page (URL)
- Your product overview (URL or PDF)
- Your refund / cancellation policy (text)
- Your tone of voice notes (text)
Re-training
Sources index automatically when added. If a webpage changes, click Re-crawl to refresh it. See the Re-training and metrics guide for details.
