Collect User Form

When you need to collect several pieces of structured information (name + contact + order number…), the model can natively render a form in the current channel and send it to the user, instead of having the user type freely in the chat (which is prone to vagueness, or several items crammed into one sentence).

Core behavior: the turn ends when the form is sent, without waiting for answers

After the model sends the form, it ends the turn immediately and does not hang waiting for the user to fill it in. It usually follows up with a line like "Please fill in the form above." What the user submits re-enters the conversation as a new message.

So: once the form is sent, the turn is over; only after the user fills it in and submits does a new turn trigger — this is by design, not a lack of response.

Prerequisites and field capabilities

• Form collection only appears to the model when the current channel supports native forms; on channels that don't, the model falls back to collecting one item at a time in plain text.
• Each form has at most 10 fields, each choice field has at most 10 options, and anything beyond is truncated.
• Attachment fields: you can let users upload images, invoices, IDs, and so on (you can set the allowed file types and a count of 1–9).
• The form's title and description are generated on the fly by the model in the conversation's language, not preset in the backend, so the wording differs across languages and cannot be "edited in the backend."
• If you want to control when the form is triggered, you can use an @ directive in the persona to summon the form-collection tool and add prompt text about when to send the form and which fields to send.

Channel differences

• Currently all channels are fill-the-whole-form-at-once; there is no "field-by-field" experience.
• When a channel does not support forms, the model automatically falls back to text and will never send a dead link that can't be opened.

Attachment troubleshooting

• The customer uploaded an image / file, but the agent can't read its content: first confirm the agent's multimodal input settings have the corresponding category enabled (images / documents, etc.), and that the attachment field's allowed types cover it. Otherwise the attachment only enters as a link, and the model won't actually recognize the image / read the file.
• Telegram / LiveChat hosted page: submit on the popped-up form page as soon as possible (normally done within seconds); submitting too long afterward may cause the attachment to be treated as plain text.