Send context and a set of questions. blnkd returns a typed answer for each question, with probabilities you can inspect before taking action.
How a request works
The state contains the evidence. The questions define what to decide. For example, route a support ticket to one of three teams:
{
"model": "blink-1.0-experimental",
"state": "My card was charged twice.",
"questions": {
"team": {
"type": "choice",
"instructions": "Which team should handle this?",
"criteria": {
"billing": "Payment and invoice issues",
"shipping": "Delivery issues",
"other": "No matching team"
}
}
}
}The response includes the selected label and a probability for every option. Your code decides the next step — the API does not issue refunds or change a ticket.
Three answer types
Returns the probability assigned to true. Use it to check whether evidence meets a condition.
Returns one of your labels and the probability of every option. Use it to route or classify.
Returns a weighted mean across the ordered levels you define, plus their probabilities.
Text, image, and video inputs
The state can be text or structured JSON. You can attach one image or one short MP4 and ask any of the three answer types about it. Media is input — the answers are still Noul, Choice, or Score. See Image inputs and Video inputs.
A delivery photo can support a Choice question about visible damage, a Noul question about a policy condition, or a Score question using a damage rubric. It cannot establish the condition of hidden contents.
Use it in your application
- Quickstart — create a key and send an HTTP request.
- TypeSafe Python SDK — use typed questions and answers in Python.
- Playground — edit examples and compare results.
Test criteria and decision thresholds against your own labeled cases. High probability does not guarantee a correct answer. See Limits for the request and account limits.