Your bot can read a photo a customer uploads in web chat. It looks for the device or product name, the model, printed labels, serial numbers and indicator lights, tells the customer what it sees, and uses that to help before asking them to type the model.

Turn it on

Image Analysis is an add-on. Ask Velaro support to enable it for your site, or check Settings, Packages. When it is off, the bot asks the customer to type the details instead.

Limits

• Monthly limit per site (default 100 photos a month, shown in your plan; extra blocks of 250 can be added). Photos are not charged twice: the same image is read once and remembered.
• 10 photos per conversation and 30 per visitor per month.
• JPEG, PNG, GIF or WebP, up to about 3.7 MB. Photos sent through WhatsApp, Facebook, Instagram, text messages and Apple Messages cannot be read yet. Ask those customers to type the model or use the website chat.
• Photos are read by OpenAI vision. The photo itself is not kept; only the short text result is stored for a short time.

Teach the bot what to match

The bot matches only what it knows. You extend that knowledge in Knowledge, Upload files:

1. Troubleshooting guides and manuals: upload PDFs or Word files. Name each file after the product or model it covers, for example "Inogen Rove 4 troubleshooting.pdf", so a photo of that model finds it.
2. Spreadsheets (xlsx or csv): upload a matrix with one row per product. Useful columns are your SKU, model name, competitor equivalent, key specs and the guide that applies. The bot treats each row as a fact it can look up exactly.
3. Similar SKUs that share one troubleshooting guide: put every SKU or model name on its own row and give them the same guide name in the spreadsheet, or list all of the models in the first lines of the guide itself. Then a photo of any of those models finds the same guide.
4. Pick the right index on your bot (Bot, Knowledge) so the uploaded files are searched.

Tips

• Ask customers to photograph the label or screen straight on with good light.
• If a model cannot be read with confidence the bot reads it back and asks the customer to confirm.
• The bot never says two products are the same just because they look alike. It needs a row in your spreadsheet or a statement in your guide.