How a knowledge base works on a call
When a caller asks a question, the agent searches the knowledge base for the passages that match it best, then writes a short spoken answer from them. This approach is often called retrieval-augmented generation, or RAG.
Here's a call to a college's admissions line:
- Caller: "B.Com ki fees kitni hai, aur installments mein de sakte hain?"
- Agent: "B.Com ki fees saal ki ₹85,000 hai, aur aap ise do installments mein de sakte hain. Kya main fee structure WhatsApp par bhej doon?"
Both answers came from the fee sheet the college uploaded, not from the model's general memory.
What makes a good knowledge base
- Short, clear passages, one topic each, written the way you'd say them out loud.
- Current facts. An old price list is worse than none.
- Clear limits. If a question isn't covered, the agent should say so and offer a person rather than invent an answer.
Why it matters
Without a knowledge base, an agent either can't answer or fills the gaps with plausible guesses. With one, answers stay accurate, and you update them in one place.
How Kallix uses it
Build a Kallix knowledge base from your website URL: Kallix crawls the page links and sitemap, you pick the pages, and it's ready in a couple of minutes. You can also upload documents and CSVs, such as FAQs, price lists and policies. A retrieval tester shows what the agent will answer before any customer hears it.