Knowledge

Add knowledge from the bot's Knowledge tab with Add knowledge. Indexing uses your own embedding key; content that has not changed is not embedded again, so re-syncs cost almost nothing.

• robots.txt is always respected, and pages marked noindex are skipped. The crawler identifies itself as KothadeskBot, so you can allow or block it by that name.
• Sitemaps on your site are found and used automatically.
• Pages behind a login cannot be crawled. Upload them as files or paste them as text instead.
• A page that disappears from your site stays in use until two full crawls in a row miss it, or it returns 404.

Each FAQ entry is kept whole and ranks above crawled text, so FAQs are the most reliable way to get an exact answer, such as a price, a timing or a policy. Edit entries in the table or import a CSV.

question,answer,url,category
What are your opening hours?,"Monday to Saturday, 9 am to 8 pm. Closed on Sundays.",https://example.com/contact,Visiting
Do you accept UPI?,"Yes, we accept UPI, cards and cash.",,Payments

Columns question and answer are required; url (linked from the answer's source) and category are optional. Tick Replace all existing entries to overwrite the table instead of adding to it.

To attach a picture to an answer, add it in Markdown: . The alt text replaces the image in the answer text, and the picture itself can be shown with the answer.

When a picture helps, the assistant shows it inside the answer, right after the sentence it illustrates, with its caption and a link to where it came from. Visitors tap it to see it full size. It only uses images from your own knowledge, never from elsewhere, and shows at most the number you allow per answer (2 by default, up to 3).

• Images are copied into Kothadesk's storage when a source syncs, so answers keep working if your site changes. They count towards your plan's knowledge storage; the same picture used on many pages is stored once.
• Skipped automatically: icons, logos, SVGs, avatars, tracking pixels, images smaller than 120 px, and images repeated on so many pages that they are decoration (shown as Repeated on many pages).
• Up to 20 images per page or file and 2000 per source are kept, in reading order.
• Turn extraction and display on or off per bot under Bot settings, Retrieval, Images.

Tip: Images are found by their words: alt text, caption, title and description. A photo named IMG_2041.jpg with no alt text is nearly invisible to search. Write alt text on your site, or turn on Describe images with AI in Bot settings.

1. 01

   Add knowledge, Images

   Give the source a name, then drop up to 50 photos (JPEG, PNG, WebP, GIF or HEIC, up to 10 MB each).

2. 02

   Title and describe each photo

   A title is suggested from the file name. Describe what the picture shows in the words a visitor would use, for example “Red 2024 hatchback, front view, alloy wheels” or “2 BHK floor plan, 950 sq ft, east facing”.

3. 03

   Upload

   Each photo becomes one document with one searchable entry. Uploaded originals are re-encoded and stripped of location and camera data.

Open any source and choose the Images tab to see every image taken from it, with its alt text, caption, your description, the AI description and the document it came from. Filter by description status.

• Hide an image to stop it appearing in answers without deleting it, for example an out-of-date photo. Show brings it back.
• Edit description to add or correct your own words; it is used ahead of the AI description.
• Describe with AI asks the bot's own key for a description of one image.
• Re-extract images reprocesses the whole source, for example after turning on Extract images from sources. Text that has not changed is not embedded again.

• The list shows each source's type, documents, status and last sync, and refreshes by itself while indexing runs.
• Re-sync a website at any time, or Cancel sync if you started the wrong one.
• Switch a source off to stop using it for answers without deleting it.
• Open a source to see every document with its status, chunks and when it was indexed, and the history of sync jobs with any errors.
• Answers keep working while a source re-indexes: each document is swapped in only when it is ready.
• Deleting a source stops it being used at once, then removes its documents and files in the background. This cannot be undone.

Answers carry numbered markers such as [1] and a Sources list. Website sources link to the page, FAQ entries to their url if you set one; text sources have no link. An image in an answer shows the citation number of its source when the answer cites that source. Choose how citations appear in Bot settings. When nothing in your knowledge answers a question, the assistant says so instead of guessing.

• Use clear headings: one topic per section. Headings help the assistant find the right passage.
• Write numbers exactly: "Returns within 7 days of delivery", not "about a week". Exact numbers let the promises check allow the answer.
• Use the words customers use as well as your own terms: "cancel my plan" as well as "subscription termination".
• Put common questions in an FAQ table with short, complete answers.
• Remove outdated pages with exclude patterns so old prices never appear.
• Add the question to an evaluation set whenever you fix a wrong answer.