Chat

Photos & voice notes

Customers send a product photo or a voice message.

Customers on Messenger and WhatsApp send pictures constantly — usually the photo from a boosted post, with "eta ache?" underneath. The agent can read those, and it can answer voice notes too.

Two steps: upload the file, then send the message that references it.

1. Upload

POST https://aiagent1.skotechlabs.com/api/v1/uploads — multipart/form-data

Field Required Description
file yes JPEG, PNG, WebP or GIF (photo), or an audio file (voice note).
conversation no The conversation UUID, if you already have one. 
curl https://aiagent1.skotechlabs.com/api/v1/uploads \
  -H "Authorization: Bearer sk_live_…" \
  -F "file=@saree.jpg"

{ "attachment": { "id": "0f2c…", "kind": "image", "url": "https://…/uploads/chat/…jpg", "mime": "image/jpeg", "size": 184320 } }

Photos are shrunk to 1568px on the long edge before the AI sees them — sending anything larger costs you upload time and buys nothing.

2. Send the message

curl https://aiagent1.skotechlabs.com/api/v1/chat \
  -H "Authorization: Bearer sk_live_…" \
  -H "Content-Type: application/json" \
  -d '{"conversation_id":"cust-42","message":"eta ache?","attachments":["0f2c…"]}'

message may be empty when there is an attachment — a photo on its own is a perfectly normal customer message.

Each attachment id can be used once, on one message, by the business that uploaded it.

What the agent does with a photo

  1. Reads any product code or text printed in the image.
  2. Describes what it sees and searches your catalogue with those words.
  3. Names the match and asks the customer to confirm it is the right item.
  4. If nothing matches, it says so and hands over to a person — it never guesses.

Prices and stock always come from the catalogue, never from the picture. A payment screenshot is never treated as proof of payment.

To make matching work well, fill in each product's "What the photo shows" — the console's Products & services → Describe photos button writes it for you from your catalogue images.

Voice notes

A voice note is turned into text first, and the text is then answered like any other message. When voice notes are not switched on for the account — or when the recording is unclear — the agent politely asks the customer to type instead. It never answers a guess.

The transcript comes back on the attachment, so your own UI can show what was heard:

{ "id": "7a19…", "kind": "audio", "url": "…", "transcript": "delivery charge koto?" }

Credits

An attachment costs a small number of extra credits on top of the reply — set per platform under Site settings → Photos & voice (1 each by default). A voice note that could not be transcribed is not charged.

← PreviousSend a message