Question 1

Does the chatbot search the internet if my knowledge does not cover a question?

Accepted Answer

No. It answers only from the sources on this chatbot's Knowledge tab. If the fact is not there, it says it does not know rather than inventing an answer.

Question 2

Why can't the chatbot answer from a file I uploaded a minute ago?

Accepted Answer

Extraction and training run in the background after the upload finishes. Watch the badge on the row — the chatbot can only use a source once it reads trained.

Question 3

Which source type should I start with?

Accepted Answer

Pasted text. It has no extraction step to go wrong, costs the least of your knowledge cap, and proves the chatbot is answering before you add anything larger.

Question 4

Can two chatbots share the same knowledge?

Accepted Answer

No. Each source belongs to one chatbot. Add the material to each chatbot that needs it, and remember it counts against both knowledge caps.

Question 5

Does adding knowledge use message credits?

Accepted Answer

No. Credits are spent when the chatbot answers a visitor. Adding and training knowledge costs capacity, not credits.

Question 6

Is pasted text as good as uploading the same content as a PDF?

Accepted Answer

It is usually better. There is no extraction step to fail, and it uses far less of your knowledge cap because designed documents carry repeated headers and footers.

Question 7

How should I word a text source so the chatbot finds it?

Accepted Answer

Write the question the way a customer asks it, then answer immediately underneath. Keep each answer self-contained so it still makes sense on its own.

Question 8

Can I edit a text source after it is trained?

Accepted Answer

Yes. Open the row, edit, and save. The chatbot keeps answering from the previous version until the updated one finishes training, so nothing goes dark.

Question 9

Why does the chatbot ignore my new FAQ?

Accepted Answer

Either the row is still training, or you asked using words that never appear in the text. Test with a phrase from the source, then add the customer's phrasing to it.

Question 10

What are the exact upload ceilings?

Accepted Answer

10 MB per file, 20 files per batch, and 90 MB for the batch in total. These are transfer limits and are separate from your plan's knowledge cap.

Question 11

Why is PowerPoint not accepted?

Accepted Answer

Slide decks are not in the supported list. Export the deck as PDF, or paste the speaker notes as text — usually the better result, since slides are bullet fragments.

Question 12

My file uploaded but the row shows almost no size. Why?

Accepted Answer

Almost nothing was extracted. That is the signature of a scanned or image-only document, which goes through text recognition instead of ordinary extraction.

Question 13

How do I replace a document with a newer version?

Accepted Answer

Use Replace file on the existing row. It swaps the document while keeping the title and collection, so you do not end up with two versions competing.

Question 14

Can I upload the same file to two chatbots?

Accepted Answer

Yes, but you upload it twice and it counts against both knowledge caps. Sources cannot be shared or moved between chatbots.

Question 15

Which files get transcribed instead of read directly?

Accepted Answer

Every image (.png, .jpg, .jpeg, .webp), and any PDF with no usable text layer — a pure scan, a near-empty text layer, or an Arabic-dominant document.

Question 16

Why are Arabic PDFs transcribed even when they contain text?

Accepted Answer

Arabic text layers are frequently scrambled — letters reordered, words reversed — so copying them produces gibberish. Reading the page as an image gives clean Arabic.

Question 17

How do I know whether the transcription worked?

Accepted Answer

The row reaches trained rather than failed, and Test Chatbot can answer a question using a phrase printed on the page and cite your document by title.

Question 18

My scan failed with 'No trainable content was found'. What now?

Accepted Answer

Nothing readable was found. Re-take the photo flat and well lit, rescan at a higher quality, or type the content in as text knowledge instead.

Question 19

Is a phone photo of a page good enough?

Accepted Answer

It is the worst of the options. A blurred or angled photo produces missing and misread words, and the chatbot answers confidently from that mangled text.

Question 20

Does an imported file stay in sync with Drive?

Accepted Answer

No. It is a one-time copy. Editing the document in Drive changes nothing here — import it again and remove the older row so both versions do not compete.

Question 21

What are the size limits on a cloud import?

Accepted Answer

The same as a direct upload: 10 MB per file, 20 files per import, and 90 MB in total. They are checked on the server as each file downloads.

Question 22

What happens to Google Docs, Sheets, and Slides?

Accepted Answer

They are exported first — Docs to Word, Sheets to Excel, Slides and Drawings to PDF — then read normally. Slide speaker notes are not included.

Question 23

Does Agentency keep access to my whole drive?

Accepted Answer

No. It receives short-lived access to the files you explicitly picked, and that access expires. This is also why nothing can sync afterwards.

Question 24

The Drive card is missing. Can I unlock it by upgrading?

Accepted Answer

No. Cloud import is an installation setting, not a plan feature. Use the Files tab instead — the resulting knowledge is identical.

Question 25

What does the Reading Document badge mean?

Accepted Answer

A scanned or image-based document is being transcribed into text before it can be trained. The progress bar has no percentage because there is no honest halfway figure.

Question 26

Does editing a trained source take the chatbot offline?

Accepted Answer

No. The previous version keeps answering while the updated one is processed, so there is never a window where the chatbot has forgotten the topic.

Question 27

A row has been pending for ages. Should I delete and re-add it?

Accepted Answer

No — that puts you at the back of the queue again. Stranded rows are picked up automatically by a recovery pass, so waiting is the better move.

Question 28

Why do all my rows say Not trained?

Accepted Answer

They were added with Start training switched off. Use the train action on each row, or the bulk train action in the banner above the table.

Question 29

Can I undo a removal?

Accepted Answer

No. There is no recycle bin. Keep a copy of the original file or text outside Agentency if there is any chance you will want the material back.

Question 30

Should I remove or replace an outdated document?

Accepted Answer

Replace when it is a newer version of the same document — it keeps the title and collection. Remove when the material should simply be gone.

Question 31

Why does the row stay visible after I confirm?

Accepted Answer

Cleanup of the search index runs in the background. The removing badge is honest about that: until it clears, the chatbot may still find the passage.

Question 32

Does deleting a crawl from history remove the pages it collected?

Accepted Answer

No. The history entry is the record of the run. The pages are ordinary sources and stay on the Knowledge table until you remove them there.

Question 33

Why is my small file using so much of the cap?

Accepted Answer

The cap measures extracted text, not bytes on disk. A text-dense document costs far more than a large PDF that is mostly photographs.

Question 34

Is the cap per chatbot or per workspace?

Accepted Answer

Per chatbot. Two chatbots on Standard have 20 MB each. Extra knowledge add-ons top up a shared pool the workspace owner can draw on.

Question 35

My crawl stopped early. Is that a bug?

Accepted Answer

No. A crawl halts when the knowledge budget runs out, so you get a partial trained crawl rather than a failure. Free space or narrow the crawl, then run it again.

Question 36

Is running out of knowledge space the same as running out of credits?

Accepted Answer

No. Credits stop the chatbot answering visitors; the knowledge cap stops you adding material. Neither one causes the other.

Question 37

What happens if I downgrade below my current usage?

Accepted Answer

Nothing is deleted, but you cannot add more knowledge until you are back under the new ceiling. Prune sources or upgrade again.

Question 38

Do collections give me extra capacity?

Accepted Answer

No. They are purely organisational. Your knowledge cap is per chatbot no matter how many collections you create.

Question 39

Why can't I move a source to another chatbot?

Accepted Answer

Every source belongs to one chatbot by design, so it is always clear which material produced an answer. Remove it there and add it again on the right chatbot.

Question 40

A teammate cannot see a file I added. Why?

Accepted Answer

Usually they are in a different workspace, on a different chatbot, or their role does not include knowledge access. It is almost never the collection.

Question 41

Does filtering the table by collection limit what the chatbot searches?

Accepted Answer

No. The filter changes your view only. The chatbot always searches every trained source on that chatbot.

Knowledge base FAQ