curl --request POST \
--url https://api.tokz.dev/decide \
--header 'Authorization: Bearer <token>' \
--header 'Content-Type: application/json' \
--data '
{
"inputs": "I was charged twice for my May invoice and the app keeps logging me out.",
"decisions": {
"queue": {
"type": "choice",
"question": "Which team should handle this ticket?",
"options": [
"billing",
"technical",
"account",
"sales"
]
},
"urgent": {
"type": "bool"
}
}
}
'{
"id": "<string>",
"model": "<string>",
"results": [
{}
],
"usage": {
"billed_tokens": 123,
"cost_usd": 123
}
}{
"error": "<string>",
"message": "<string>",
"requestId": "<string>"
}{
"error": "<string>",
"message": "<string>",
"requestId": "<string>"
}{
"error": "insufficient_credits",
"balance": 123
}{
"error": "<string>",
"message": "<string>",
"requestId": "<string>"
}{
"error": "<string>",
"message": "<string>",
"requestId": "<string>"
}{
"error": "<string>",
"message": "<string>",
"requestId": "<string>"
}{
"error": "<string>",
"message": "<string>",
"requestId": "<string>"
}Answer typed decisions about texts
Asks one or more typed questions (choice, multilabel, bool, score) about up to 64 texts and returns, per text and question, the best option, the selected options, a probability distribution over the options, and for scores the expected value. This runs an OpenDecision model: the text is sent to a Tokz-run model service. It is not stored or logged; only counters are kept. Billed per token at the price of the chosen model, from the prepaid credit balance, and only when the request succeeds. Give the name of a fine tune your organization owns as model and leave out decisions to run that model on the API server’s own CPU, up to 16 texts per request. The answer is then under the key decision.
curl --request POST \
--url https://api.tokz.dev/decide \
--header 'Authorization: Bearer <token>' \
--header 'Content-Type: application/json' \
--data '
{
"inputs": "I was charged twice for my May invoice and the app keeps logging me out.",
"decisions": {
"queue": {
"type": "choice",
"question": "Which team should handle this ticket?",
"options": [
"billing",
"technical",
"account",
"sales"
]
},
"urgent": {
"type": "bool"
}
}
}
'{
"id": "<string>",
"model": "<string>",
"results": [
{}
],
"usage": {
"billed_tokens": 123,
"cost_usd": 123
}
}{
"error": "<string>",
"message": "<string>",
"requestId": "<string>"
}{
"error": "<string>",
"message": "<string>",
"requestId": "<string>"
}{
"error": "insufficient_credits",
"balance": 123
}{
"error": "<string>",
"message": "<string>",
"requestId": "<string>"
}{
"error": "<string>",
"message": "<string>",
"requestId": "<string>"
}{
"error": "<string>",
"message": "<string>",
"requestId": "<string>"
}{
"error": "<string>",
"message": "<string>",
"requestId": "<string>"
}Authorizations
Your API key: Authorization: Bearer tokz_live_<id>_<secret>. Only a SHA-256 hash of the secret is stored.
Body
One text or up to 64, each at most 20,000 characters. A fine tune takes up to 16.
1 - 20000A catalog model (opendecision-small, opendecision-large-packed, opendecision-large), or the name of a fine tune your organization owns.
Questions keyed by name, 1 to 16 per request. Required for a catalog model, and refused for a fine tune, which already knows its question and labels.
Show child attributes
Show child attributes
Response
One object per input, keyed by decision name. A fine tune uses the fixed key decision.