An AI Knowledge Base That Speaks From Your Own Documents
Upload your files and we do the indexing. Your content feeds the reply suggestion that goes to the agent. You see on screen which document each suggestion came from. It works with documents in several languages.
Upload your PDF, Word, Excel or web page and we do the indexing. The content feeds the reply suggestion that goes to the agent, and the source document shows on screen. It works with documents in several languages. Your team is still the one who sends the reply.
Import your PDF, Word, Excel or web page. The system parses the text and indexes it by meaning. When a customer writes, a hybrid method matches both the words and the meaning. It finds the answer they are after. Behind it sits a multilingual template library prepared per sector.
Knowledge base, 24 documents
Indexedreturn-policy.pdf
3.2 MB · 14 pages · 47 chunks
product-catalogue-2026.docx
1.8 MB · 38 products · 142 chunks
treatment-knowledge-base.pdf
5.1 MB · 28 treatments · 218 chunks
kvkk-privacy-notice.pdf
0.4 MB · 4 pages · 12 chunks
24
Documents
1,847
Chunks
Multi
lingual
Customer query
"I do not like the product. Do I pay for the return shipping?"
Hybrid search results
3 chunks · 47 ms📄 return-policy.pdf · p.3
96%"Return shipping for unused items with tags still attached is covered by Chatinbox Store..."
📄 faq-general.pdf · p.7
82%"For damaged or used items the shipping cost is paid by the customer..."
AI draft (with sources)
Sorry to hear that 😔 If you want to return the item, we cover the shipping for unused items with the tags still on. If there is damage, the shipping is on you. Reply to start the return and we will send a shipping label.
Search that catches the meaning and the words
Keyword search (TF-IDF) and vector-based semantic search (pgvector) run at the same time. The customer writes "do I pay for the return". Your document says "return shipping is paid by the customer". The words differ, but the match is found by meaning. Next to the answer you see the match score, the source quote and the page it came from.
- Hybrid semantic search with the pgvector + ts_vector pair
- The document quote that feeds the suggestion shows on the agent's screen
- The match score sits next to the suggestion; the agent catches a weak match
Question and answer from multilingual documents
You can add documents in different languages to your knowledge base. When a customer writes in their own language, the answer is found in the content for that language. The source is shown. Useful for health tourism, hotels and e-commerce businesses that export.
- Meaning-based matching across multilingual documents
- Finds the relevant answer in documents written in other languages
- Frequent answers are cached and come back fast on repeat queries
Cross-language query flow
Customer question (Arabic)
ما هي مدة الإقامة بعد عملية زراعة الأسنان؟
→ "How long do I stay after dental implant surgery?"
Document match (Turkish)
"Implant cerrahisi sonrası 2-3 gün İstanbul'da kalmanızı öneririz. Kontrol için 7. günde tekrar muayene yapılır..."
treatment-knowledge-base.pdf · page 12 · 91% match
AI draft (Arabic)
ننصح بالإقامة في إسطنبول لمدة 2-3 أيام بعد جراحة الزرع. يتم إجراء فحص متابعة في اليوم السابع.
Pre-built FAQ sets for your sector
No blank page. A ready-made question set for your sector has been compiled for you. Import them, adjust them to your brand, publish right away.
Dental clinics
Ready FAQ set · multilingual
Aesthetics & clinics
Ready FAQ set · multilingual
E-commerce
Ready FAQ set · multilingual
Hotels & tourism
Ready FAQ set · multilingual
Real estate
Ready FAQ set · multilingual
Restaurants
Ready FAQ set · TR
Education & courses
Ready FAQ set · TR
Automotive
Ready FAQ set · TR
Ready sets are also available for finance, services, sports & fitness and veterinary clinics
Refresh the document, catch the missing answer
When you upload a new document, the old one moves to the archive on its own. The current content is indexed again. Content from the knowledge base feeds the reply suggestion that goes to the agent. The agent sees the suggestion and its source, and catches a weak match. On top, a report of the questions that found no answer shows you the gaps in your knowledge base.
- Version tracking: the old document is archived, the current one is used
- The agent sees the suggestion and its source, and catches a weak match
- A gap report that lists the questions left without an answer
Knowledge base gaps, this week
14 gaps"Do you treat children?"
7 timesBest score: 42% · Not enough
"What do I do about pain at night?"
5 timesBest score: 38% · Not enough
"Implant warranty period?"
4 timesNo answer found
Let Your Documents Feed the Reply Draft
Drop in your PDF or Word file and we do the indexing. The reply suggestion that goes to your team is fed by your document, and you see which document it came from. Your team is still the one who sends the reply.
How to do it
Step-by-step guides from the help center for this feature:
