What happens when a customer sends a photo in WhatsApp or Telegram:

1. When your account keeps customers' files (see below), the photo is uploaded to private storage (7-day signed URL).
2. It is forwarded to your operator Telegram group as an actual photo. When files are not kept, a WhatsApp photo reaches the group as a short notice that the customer sent one.
3. If Vision AI is enabled (see below), the assistant reads the photo — what it is, what is readable on it, a status such as "result not ready" — and answers it the way it answers a typed message: in the same conversation, remembering what was said before, from your own facts (prices, addresses, hours). Addresses and prices printed on a customer's document are never repeated as yours, and personal data on the photo (names, ID numbers) is not copied into the chat text.
4. The reply is mirrored into the operator Telegram group like any other AI reply, so the team sees what was said.
5. A PDF sent in WhatsApp (an invoice or a contract, for example) is stored and shown to your team; with Vision AI on it is also read and answered. A video, a location or a contact card reaches your team in the operator group and in Chats.
6. The photo + caption + AI reply all appear in Chats as an inline thumbnail in the customer's thread.

Photos sent in Instagram Direct are not read or forwarded — open the conversation in the Instagram app to see them.

How to enable Vision AI:

Settings → "Image recognition (Vision AI)" checkbox. Off by default — each account switches it on. Photos always reach the operator group regardless of this setting, so a wrong AI guess is never the only signal.

Keeping customers' files. Settings → Customer Attachments decides whether photos and files customers send are stored, and for how many days; leave the days empty to keep them. With storing off, nothing is saved, but your operators are still told that the customer sent something.

Tuning what the bot says:

A photo is answered by the same assistant, with the same instructions, as a typed message. Add a "Photo handling" block to the prompt to tell the AI what to do with different image types. Example:

## Photo handling

When the customer sends a photo, identify what it is and reply briefly:

### Document / receipt / contract
If you see a document, a receipt or a contract:
"Got your document. Please share your name and a good number — a manager will check it and reply. If it's urgent: "

### A photo of a product or a problem (damage, a defect, an error on a screen)
"Thanks, we can see it. Tell us in a few words what happened and we'll tell you what to do next."

### Off-topic
"Got the photo. What can we help you with today?"

### Rules
- Reply MUST be short (2-4 sentences) — this is a chat, not an email.
- Never draw a conclusion a specialist has to make (a diagnosis, a damage estimate).
- Do NOT echo back personal data (name, ID, date of birth) — privacy.

One photo, one copy:

A photo the messenger delivers more than once is shown and answered once. If you see the same picture twice, the customer sent it twice.