Knowledge base

It answers from what you gave it.

The knowledge base is what an agent answers from. You point it at your website and drop in documents; retrieval is heading-aware and works natively in Chinese and Japanese rather than degrading to substring matching.

Start delegating The whole list

Knowledge basesources

harbourdental.com — 14 pages crawl recrawled 2d ago
Price list 2026.pdf pdf 38 passages
Aftercare instructions.docx doc indexing

Three callers this week asked about weekend appointments. No source covers it.

You pick which pages become knowledge rather than ingesting the careers page along with the prices. Answers cite the source they came from.

Or just ask for it

Feed it by forwarding, not by uploading.

The price list you were going to attach anyway.

Gmailfrom:octo knowledge

OctoIngested: Autumn price list.pdf — 14 pages, live on two agents9:52am

OctoQuality review: 3 contradictions — Two documents disagree on the cancellation feeWed

OctoRe-crawled northbay-dental.com — 9 pages refreshed, hours changedMon

InboxIngested: Autumn price list.pdf

Dme9:50 AM

New prices from Monday — attached. Use these from now on.

OOcto9:52 AM

Ingested all 14 pages and retired the old sheet. Both the receptionist and the email agent answer from it now. Heads up: the cancellation fee contradicts your terms page — which is right?

Octoonline

does it know we do saturday mornings now?4:15 PM ✓✓

No — the site still says Mon–Fri and that’s where it’s answering from.4:15 PM

we do 9 to 1 on saturdays4:16 PM ✓✓

Written in as a manual entry, so it’s answerable straight away without waiting for a re-crawl.4:16 PM

someone asked about parking on a call yesterday and it didn’t know4:18 PM ✓✓

I found that call. Tell me the answer and I’ll add it permanently, so it never asks again.4:18 PM

Message➤

#omniocto4 members

DWDana2:20 PM

@Octo what do we actually tell people about the deposit?

OOctoAPP2:20 PM

On it 🔧(runningkb_search…)

“$50, refundable up to 48 hours before.” Source: Autumn price list.pdf, page 3. (edited)

✅ 2

2 replies · last reply 15m ago

MRMarcus2:44 PM

that’s out of date, it’s $75 now — can you fix just that doc?

Message #omniocto

Email octo@agent.omniocto.com · WhatsApp 1-MAN-ASK-OCTO · Slack #omniocto

Website crawl with a page picker: Crawl your site and choose which pages become knowledge, rather than ingesting the careers page and the cookie policy along with the prices.
PDF, Word, Excel and menu ingestion: Price lists, menus, service sheets and spreadsheets, up to very large documents, become answerable material.
Heading-aware retrieval: Passages are chunked at heading boundaries and a heading match is weighted more heavily, so a question about a section finds that section.
CJK-native search: Chinese and Japanese are handled with script-aware boundaries rather than whitespace splitting, which is the difference between retrieval working and not working in those languages.
Quality review with per-document fixes: The knowledge base is reviewed for gaps and contradictions, with fixes applied per document rather than by re-uploading everything.
Manual entries: Write an answer directly when there is no document to point at.
One-command re-crawl: Refresh from the source site in one step when your prices or hours change.
Learns from live calls: A question a human answered on a live conversation can be promoted into the knowledge base, so the agent knows it permanently rather than asking again.
Instant-answer tier: Smaller knowledge bases are answered from directly with no lookup pause, which is what keeps a voice call from having a silence in the middle of it.
Per-agent source status: Which sources an agent has, and whether each one is current, is visible rather than assumed.

What this powers

The jobs these capabilities do.

Start

Put one of these to work.

Start delegating See pricing