Content · Knowledge base
AI that answers
your way
For teams where the same questions come up again and again. Write your procedures and answers into the knowledge base, and the AI assistant draws on them when a colleague asks or when it drafts a ticket reply.
- Search by meaning, even across languages
- New findings from chat wait for human approval
- Outdated entries flag themselves for review
No credit card. Up and running in 5 minutes.
How do we handle a return when the customer has no receipt?
According to "Returns without a receipt": a bank statement or the e-shop order number is enough. You can find the order on the customer contact. With neither, we accept the return only with the store manager’s approval.
Source: Returns without a receipt · Curated article · verified 2 Oct
- Four entry typesQ&A pair, note, curated article and web snippet.
- Search by meaningFull-text and semantic search combined.
- ApprovalWhat AI saves from chat stays a draft until a person approves it.
- Freshness checksEntries move to "Needs review" after a while.
Why a knowledge base in BizKitHub
Know-how that lives only in heads leaves with people
A few experienced people know the answers to common questions. Everyone else asks them, and an AI without your data makes the answer up.
Today
- You answer the same question from a new colleague for the tenth time
- Procedures sit in a shared document nobody updates
- Generic AI answers confidently, but by someone else’s rules
- What was solved once in chat gets solved again next time
With BizKitHub
- The AI assistant answers from your entries and cites the source
- An entry is found even when the question is worded differently
- Save a good answer from chat as a new entry
- The system tells you which entries are due for review
Search
Finds the answer even when you ask differently
Every entry is split into chunks and embedded as vectors. At query time the system merges full-text and semantic results and ranks them by relevance and confidence. A Czech entry is found by a question asked in English.
- Full-text and meaning-based search in one ranking
- Works across languages
- Each entry’s confidence level affects its rank
- Entries due for review rank slightly lower
| Entry | Type | |
|---|---|---|
| Returns without a receipt | Article | Active |
| Refund to account or card | Q&A | Active |
| Return periods | Q&A | Needs review |
| Exceptions for businesses | Note | Active |
Review queue
AI suggests, a person decides
When the AI assistant reaches a useful answer in chat, it can save it to the knowledge base. That entry is only a draft and does not exist for search until an operator approves it in the queue. The same queue holds entries due for review and entries that mention an anonymised customer.
- Approve chat drafts in bulk or one by one
- Entries move to "Needs review" when their window runs out
- Flags for entries affected by contact anonymisation (GDPR)
- Archived entries are kept, but AI does not use them
Who uses it
One knowledge base for the assistant and the helpdesk
The AI assistant in the admin searches the knowledge base on its own when it needs to know your rules. In the helpdesk, an agent uses it to draft a reply to a new ticket, which an operator reviews, edits and sends.
- The assistant cites the entry it relied on
- A person always sends the ticket reply
- Write an entry in Czech and AI uses it in other languages too
- Tags to sort entries by topic
- 9:02Customer e-mail received"I lost the receipt, can I return the shoes?"
- 9:02Knowledge base searched2 relevant entries
- 9:03Reply draft readyWaiting for operator review
- 9:11Reply sentEdited and sent by Petra Malá
What it does
Know-how that does not get forgotten
Entry types
Q&A pair, note, curated article and web snippet.
Semantic search
Full-text and vector search merged into one ranking.
Across languages
A Czech entry answers a question asked in another language.
Save from chat
Keep a good assistant answer as a draft entry.
Review queue
Approve drafts, re-verify old entries, bulk actions.
Freshness checks
Notes after 180 days, curated articles after a year move to review.
Confidence
High, medium or low. It affects ranking in results.
GDPR
Flags on entries that mention an anonymised customer.
Tags and filters
Filter by type, status, origin, tag and text.
Getting started
Answers from your own knowledge base today
- 01
Write down common questions
Start with the ten questions you hear most often.
- 02
Ask the assistant
The AI assistant starts answering from your entries.
- 03
Approve what gets added
Review new findings from chat in the review queue.
Who it is for
Where the same questions keep coming
Customer support
Ticket reply drafts based on your rules, not generic advice.
Onboarding new people
Newcomers ask the assistant instead of the colleague next to them.
Multi-branch operations
One procedure for complaints, returns and exceptions in every branch.
Service and IT
Fixes for recurring problems written once and easy to find.
Connected modules
Runs on the same data as the rest of your company
AI assistant
AI that knows your data and can act on it.
Helpdesk
Support from your inbox. Mail gets sorted, you answer.
Posts & CMS
Blog, pages and glossary with translations and SEO.
Drive
Company files linked to customers and documents.
Contacts & CRM
Customers, companies and the full relationship on one card.
Frequently asked questions
Common knowledge base questions
Where does the AI get its answers?
From entries in the Active or Needs review status. Drafts and archived entries are not used. The assistant names the entry it relied on.
Can AI write to the knowledge base by itself?
It can suggest a new entry from a conversation, but it is saved as a draft. It only becomes searchable after a person approves it.
Do I have to write entries in every language?
No. Search works across languages, so an entry written in Czech answers a question asked in English.
How do I know an entry is outdated?
Each type has a review window: web snippets 30 days, notes and Q&A 180 days, curated articles a year. After that the entry moves to Needs review and appears in the review queue.
What happens to entries when a customer asks to be erased?
Entries mentioning the anonymised customer get a GDPR flag and appear in the review queue. An operator decides whether to delete them.
Teach the AI how things work at your company
Create a free account, write your first ten answers and ask the assistant.