Jev
Jev is not a Gonka model. It is a real decision model. Send the facts in state and the judgments you need in questions. Jev returns structured answers your code can act on. It does not write chat text, so do not send jev to /v1/chat/completions.
It costs $0 and is not counted against the proxy rate limit. An account can call it only after 100,000,000 tokens in the last 30 days (UTC), including today. Below that, there is no access. The dashboard shows those tokens. Eligibility is checked on every call. When the window moves on and the account is under the line, the call is refused and nothing is forwarded.
Deep research uses Jev inside its loop to keep or drop pages. That internal use is not a Jev API call: it is not billed, and the 100,000,000-token line does not apply. The report is written by the model you named.
Make a call
Use the same API key as chat completions. Keep it on your server.
POST https://api.proxy.gonka.gg/v1/systemone
Authorization: Bearer sk-...
Content-Type: application/jsonPOST /v1/decisions is the same endpoint. GET /v1/systemone describes the model, the price, and the eligibility rules.
curl https://api.proxy.gonka.gg/v1/systemone \
-H "Content-Type: application/json" \
-H "Authorization: Bearer sk-your-api-key-here" \
-d '{
"model": "jev",
"state": "My card was charged twice. Please help ASAP.",
"questions": {
"team": {
"type": "choice",
"instructions": "Which team should handle this?",
"criteria": {
"billing": "Payments and refunds",
"technical": "Bugs and integrations",
"sales": "Pricing and new accounts"
}
}
}
}'const res = await fetch("https://api.proxy.gonka.gg/v1/systemone", {
method: "POST",
headers: {
Authorization: `Bearer ${process.env.GONKA_API_KEY}`,
"Content-Type": "application/json",
},
body: JSON.stringify({
model: "jev",
state: "My card was charged twice. Please help ASAP.",
questions: {
team: {
type: "choice",
instructions: "Which team should handle this?",
criteria: {
billing: "Payments and refunds",
technical: "Bugs and integrations",
sales: "Pricing and new accounts",
},
},
},
}),
})
const data = await res.json()import os, requests
res = requests.post(
"https://api.proxy.gonka.gg/v1/systemone",
headers={"Authorization": f"Bearer {os.environ['GONKA_API_KEY']}"},
json={
"model": "jev",
"state": "My card was charged twice. Please help ASAP.",
"questions": {
"team": {
"type": "choice",
"instructions": "Which team should handle this?",
"criteria": {
"billing": "Payments and refunds",
"technical": "Bugs and integrations",
"sales": "Pricing and new accounts",
},
}
},
},
timeout=60,
)
data = res.json()The values below are a sample of the response shape, not a live result.
{
"model": "jev",
"answers": {
"team": {
"type": "choice",
"choice": "billing",
"probabilities": {
"billing": 0.94,
"technical": 0.04,
"sales": 0.02
},
"confidence": 0.85
}
}
}Read answers.team.choice for the selected team.
Request fields
| Field | Type | What to send |
|---|---|---|
state | String, object, or array. Required | The text and facts needed to answer. Put the relevant records or policy here. |
questions | Object. Required | A non-empty map of question IDs to question definitions. Your IDs map each result back to your code. |
model | String | jev |
Put the question in instructions. An ID such as team labels the answer. It does not replace an explicit question.
Every question sees the same state. Send independent questions together in one call. Keep each question focused on one judgment. You can mix all three types below.
Choice
Pick one option from a set you define. Criteria can name up to 255 options. The answer includes choice, probabilities, and confidence.
{
"model": "jev",
"state": "My card was charged twice. Please help ASAP.",
"questions": {
"team": {
"type": "choice",
"instructions": "Which team should handle this?",
"criteria": {
"billing": "Payments and refunds",
"technical": "Bugs and integrations",
"sales": "Pricing and new accounts"
}
}
}
}{
"model": "jev",
"answers": {
"team": {
"type": "choice",
"choice": "billing",
"probabilities": { "billing": 0.94, "technical": 0.04, "sales": 0.02 },
"confidence": 0.85
}
}
}Score
Evaluate something on an ordered scale of up to 10 levels. The answer includes score, legend, probabilities, and confidence.
{
"model": "jev",
"state": "My card was charged twice. Please help ASAP.",
"questions": {
"severity": {
"type": "score",
"instructions": "How severe is this request?",
"criteria": {
"0": "No impact",
"1": "Minor inconvenience",
"2": "The customer is blocked",
"3": "Money or data is at risk"
}
}
}
}{
"model": "jev",
"answers": {
"severity": {
"type": "score",
"score": 1.6,
"legend": [
"No impact",
"Minor inconvenience",
"The customer is blocked",
"Money or data is at risk"
],
"probabilities": { "0": 0.05, "1": 0.40, "2": 0.45, "3": 0.10 },
"confidence": 0.72
}
}
}Noul
Measure a yes-or-no judgment. criteria can describe what true and false mean, and can be omitted. The answer is noul, a number from 0 to 1. Near 1 is yes, near 0 is no, and near 0.5 is uncertain. Noul has no separate confidence field.
{
"model": "jev",
"state": "My card was charged twice. Please help ASAP.",
"questions": {
"urgent": {
"type": "noul",
"instructions": "Does this need a response today?",
"criteria": {
"true": "The customer is blocked or money is at risk",
"false": "It can wait"
}
}
}
}{
"model": "jev",
"answers": {
"urgent": {
"type": "noul",
"noul": 0.92
}
}
}Use the answer
Read a result by its question ID, then apply your own rules. Choice and score include probabilities and confidence from 0 to 1. Confidence is a signal, not a guarantee. Validate thresholds on your own examples.
const { team } = data.answers
if (team.confidence < 0.8) {
queueForReview(ticket)
} else {
routeTicket(ticket, team.choice)
}Who can call it
100,000,000 tokens in the last 30 UTC days, including today. The window moves forward each day. Below that line, the account cannot call Jev.
The overview page in the dashboard shows those tokens and whether Jev is included right now. There is no per-account request cap on this route.
If it fails
Check the HTTP status before using answers.
| Status | What to do |
|---|---|
| 401 | The API key is missing or invalid. See Authentication. |
| 400 or 422 | The JSON is invalid. Include state, a non-empty questions map, and model jev. |
| 403 | This account is not eligible. Check the dashboard, or see the rules above. |
| 429 | Jev is busy. Honor Retry-After when it is present, then retry with backoff. This is not the proxy's 512 req/s cap. |
| 502 or 503 | Jev is temporarily unavailable. Retry with backoff. |