Knowledge Base
Your docs, talking back to your callers
Stellar's knowledge base lets you upload your existing docs, policies, FAQs, and pricing pages, and your AI agent answers callers from those exact sources. No training step, no reformatting, no rewriting your help center for an LLM. Whatever your best support agent would say, your AI agent says.
Upload anything
Five formats, all processed the same way. Drag and drop or paste a URL.
Product guides, whitepapers, manuals
.docx
SOPs, scripts, training documents
.txt
Plain text notes and transcripts
.csv
FAQ lists, pricing tables, contact data
.url
Web pages, help docs, knowledge bases
How does Stellar turn a PDF into something an AI agent can answer from?
When you upload a document, three things happen behind the scenes. None of them require any work from you, but all three matter for the answer the caller hears.
Split into searchable passages
Your document is broken into smaller passages, each one big enough to carry the context of a paragraph but small enough to retrieve precisely. The agent never has to read the whole 60-page PDF on a call. It just pulls the parts that match.
Indexed by meaning, not keywords
Every passage is processed by an embedding model that captures what the text means, not just which words appear. So a caller asking "do you do estimates over the phone?" matches a doc that says "phone consultations are free" even when the words don't overlap.
Retrieved in under 200ms per call
When your agent needs an answer mid-call, it pulls the most relevant passages in under 200 milliseconds. The caller hears no pause. Your agent answers from your content instead of guessing.
Test it before you go live
The built-in Q&A panel lets you type the questions your callers actually ask and see exactly which sections of your docs the agent pulled to answer them. If a passage is missing or the answer is wrong, you fix the source doc and re-test. No surprises on your first real call.
Your agent should sound like your team
Upload your docs and test a call. It takes about two minutes.