Knowledge and documents
Under the answer you see which files it came from
Upload your price list, your policies and a page of your own site. The bot looks for the answer in them and shows the paragraphs it used, with the file name and the page.
Rates and policies
4 documents
PDFDelivery rates.pdf
86 fragments · 2.1 MB
DOCXReturns policy.docx
31 fragments · 240 KB
URLshop.example/help/delivery
site page · 12 fragments
MDSales scripts.md
processing · 40%
Drop a file here or paste a link
How much is delivery to Kazan and when does it arrive?
Found in the documents
Delivery rates.pdf · p. 4
«Kazan, Samara, Ufa: 390 ₽ for orders up to 5 kg. Above 5 kg at 78 ₽ per kilogram…»
Returns policy.docx · section 2
«Volga region: 2–3 working days from dispatch. Dispatch same day if paid before 15:00…»
Delivery to Kazan is 390 ₽ for orders up to 5 kg, and takes 2–3 working days. Pay before 15:00 and we dispatch today.
A question about delivery to Kazan. Three fragments found in 0.3 seconds, and under the answer you can see which file and which page each came from.
300 quarks a month once your address is confirmed, no card needed. Two knowledge bases, a hundred files and a hundred megabytes fit on the free plan: a price list and a dozen policies fit with room to spare.
Once the bot is answering customers
«Will it lie to my customer?»
«Not found» is an answer too
An empty search is not a failure, it is a branch. The bot can see that no paragraph fits, and what happens next is the branch you wrote yourself: apologise, call a manager, open a ticket.
A model can invent things when it answers from memory about everything in the world. Here it answers from the paragraphs you uploaded, and when there are none it has nothing to answer from. How the search works, and what to do when it finds the wrong thing, is covered in the documentation.
Do you deliver to Crimea?
Not found in the documents
0 fragments · the «not found» branch took over
Our delivery terms say nothing about Crimea, so I will check with a manager. Leave a phone number and we will call back today.
Next
Tables live somewhere else
Stock levels, order statuses, a list of branches: that is not text for a meaning-based search, it is rows where the answer has to be exact. The workspace database is for them: the scenario writes to it and reads from it on a condition, and the bot answers «order 10483 is being packed» with the value from a row rather than a retelling.
An XLSX or CSV you already have is loaded by its own node. The bot reads the text inside a file, so a scanned price list is put through recognition first and uploaded after.
Orders
| Number | Customer | Status | Amount |
|---|---|---|---|
| 10482 | A. Smirnova | dispatched | 4,390 |
| 10483 | Veter Ltd | packing | 18,120 |
| 10484 | I. Petrov | awaiting payment | 990 |
Try it on your own price list
Upload two files and ask the three questions you are asked most often. Twenty minutes later you will know whether this suits you.
Upload your documents