Vivollo
Demo

Visual AI

Your agent sees what customers send.

When someone sends a photo — a product, a damaged item, an error screen — Vivollo passes it to a vision model so the agent can recognize it and reply in context. The image joins the same agentic loop, so a snapshot can match a catalog item, start a return or hand off.

Request a demo
visionlive
customer photo
Do you have this one?
detectedhandbag · quiltedhardware · gold chaincolour · black
That's our Mia Quilted Bag in Black — it's in stock. Want me to add it to your cart?Mia Quilted BagBAG-MIA-BLK · €189in stock

The agent sees what your customer sees — and acts on it.

A photo comes in. Here is what the agent read.

Customers rarely know the model name, but they can take a picture. The image goes to a vision model; what it reads joins the conversation like text.

What the customer sent
customer photo · 1.2 MBDo you have this one?
What the agent read

vision.read · 0.92

  • handbag · quilted
  • hardware · gold chain
  • colour · black
  • logo · none visible
Mia Quilted Bag · BlackBAG-MIA-BLK · 4,890 TLin stock

01Loop

What it sees runs the same tools as what it reads.

A product photo becomes a catalogue search; a damaged item becomes a return; an error screen becomes a fix. The image is another input to the same agent — not a separate feature.

FIG 1Panel · the thread with cardsWhat the customer saw
The inbox with product cards the assistant sent in a web-chat thread
vision
customer photoIs this one still available?vision.read→quilted handbag · blacksearch_catalog→1 match · in stockThat's our Mia Quilted Bag in black — in stock. Want me to add it to your cart?Matched from a photo · 6 s
Photo → search
Objects, text and detail read from the image feed the catalogue search.
Damage → return
A broken item in a photo can open the return with the picture attached.
Screen → fix
An error screenshot is read like text; the answer comes from your docs.

02Limits

When it isn't sure, it asks — it doesn't guess.

Vision is tier-gated and looks at the recent images in a thread. Below the confidence threshold the agent asks for a clearer shot or the label; if a provider fails, the conversation carries on in text.

  • Sees

    Product, label, serial, damage, error screens — on every channel that carries attachments.

  • Acts

    The read feeds the same tools: search, return, handoff with the image attached.

  • Degrades gracefully

    Low confidence asks a question; a provider error never breaks the reply.

vision
customer photo · blurryvision.read→confidence 0.41 · under thresholdI can't quite make out the label — could you send a photo of the tag, or tell me the colour and size?Asked instead of guessing

Proof of control

One confident match, one honest question.

Two photos, one afternoon. The first is read at high confidence and matched in the catalogue. The second is blurry and lands under the threshold — the agent asks for the tag instead of picking a product.

  1. 1vision.read1.4 sInputphoto #1quilted handbag · black · 0.92
  2. 2search_catalog230 msInput'quilted bag black'Mia Quilted Bag · in stock
  3. 3vision.read1.3 sInputphoto #20.41 · under threshold
  4. 4ask_customer20 msInputtag photo or colour + sizequestion sent · no guess
  5. 5vision.read1.1 sInputphoto #3 · tagserial read · 0.97
Example conversation; latencies are illustrative.

Connected channels and stores

All integrations
  • WhatsApp
  • Instagram
  • Messenger
  • Web widget
  • Shopify
  • TiTicimax

Ready to meet your AI agent?

Book a demo and we'll build a working agent on your real data — across WhatsApp, Instagram and your website. Live in days.

Request a demo