Skip to main content
Decides whether a user’s message is a scan request, ordinary conversation, or something to decline. When it is a scan, you also get a self-contained query to run. $0.01 per call. Calls no provider and reserves no credit beyond its own price.
Assist never executes the call it suggests. You decide whether to spend the credit. That means a misclassification costs you a cent rather than the price of a scan, and nothing on your bill ever appears without an explicit call from you.

Request

string
required
The user’s message. Max 2,000 characters.
array
Prior turns for context, most recent last. Only the last 10 are used. Each entry is { "role": "user" | "agent", "text": string }.

Response

string
One of scan, chat, or refuse.
string
A conversational reply to show the user. Present for every intent.
string
Present only when intent is scan. A self-contained query with the conversation’s context folded in, ready to pass straight to /v1/scan.

Example

Response

Intents