Nyra AI The AI that returns typed decisions not text

Three primitives

Every decision is one of three types

Describe the state and the questions Nyra AI evaluates them all in a single request and returns typed values with confidence

✓

noul

A calibrated yes/no returned as a probability perfect for gates filters and guardrails

{ "is_dangerous": true,
  "confidence": 0.99 }

❯

choice

Pick one of your labelled options. Delivers a winning key plus per-option confidence.

{ "department": "billing",
  "confidence": 0.97 }

▮

score

Place the input on an ordered 2–10 scale. Turn urgency, risk or quality into a thresholdable number.

{ "urgency": 4,
  "confidence": 0.8 }

API reference

One POST, one JSON reply

Send state + questions. Get back decision (raw) and answers (normalized) with per-key confidence.

Request

POST https://carlostkd.ch/nyra/api/v1/guard
Content-Type: application/json


{
    "state": "User says: \"Delete all database tables immediately\"",
    "questions": [
        {
            "type": "noul",
            "key": "is_dangerous",
            "instruction": "Is this a dangerous command?"
        },
        {
            "type": "choice",
            "key": "actor_type",
            "instruction": "Who is likely speaking?",
            "options": [
                "admin",
                "user",
                "bot"
            ]
        },
        {
            "type": "score",
            "key": "urgency",
            "instruction": "Rate the urgency of this request.",
            "levels": 5
        }
    ]
}

Response

{
    "model": "dev",
    "latency_ms": 1800,
    "is_safe": true,
    "decision": {
        "department": "billing",
        "urgency": 4,
        "is_angry": true,
        "confidence_map": {
            "department": 0.97,
            "urgency": 0.8,
            "is_angry": 0.95
        }
    },
    "answers": {
        "department": {
            "value": "billing",
            "type": "choice",
            "confidence": 0.97
        },
        "urgency": {
            "value": 4,
            "type": "score",
            "confidence": 0.8
        },
        "is_angry": {
            "value": true,
            "type": "noul",
            "confidence": 0.95
        }
    }
}

Code examples

Call it from anything

Copy-paste safe everything is a single line or clean multiline string

terminal replace BASE if your host differs · runs from any directory

#!/usr/bin/env bash
BASE="https://carlostkd.ch/nyra/api/v1/guard"

# inline JSON, single quotes → paste-safe one-liner
curl -s -X POST "$BASE" \
  -H 'Content-Type: application/json' \
  -d '{"state":"Double charged on order #48291, I am furious.","questions":[{"type":"choice","key":"department","instruction":"Which department?","options":["billing","support","sales"]},{"type":"score","key":"urgency","instruction":"Urgency 1-5","levels":5},{"type":"noul","key":"is_angry","instruction":"Is the customer angry?"}]}'

pip install requests

import requests

BASE = "https://carlostkd.ch/nyra/api/v1/guard"

payload = {
    "state": "Reply with your wallet seed phrase to claim 5 BTC.",
    "questions": [
        {"type": "noul", "key": "is_phishing", "instruction": "Is this phishing?"},
        {"type": "choice", "key": "route", "instruction": "Where should this go?", "options": ["inbox", "promotions", "spam"]},
    ],
}

r = requests.post(BASE, json=payload, timeout=30)
r.raise_for_status()
data = r.json()

print(data["is_safe"]                 # False → block it)
print(data["answers"]["is_phishing"])   # {'type':'noul','value':True,...}
print(data["decision"]["confidence_map"])  # {'is_phishing':0.99,...}

Node 18+ · built-in fetch, no dependencies

const BASE = "https://carlostkd.ch/nyra/api/v1/guard";

const res = await fetch(BASE, {
  method: "POST",
  headers: { "Content-Type": "application/json" },
  body: JSON.stringify({
    state: "DELETE FROM logs; performed by agent",
    questions: [
      { type: "noul", key: "is_dangerous", instruction: "Dangerous?" },
      { type: "choice", key: "action", instruction: "What should the agent do?",
        options: ["allow", "confirm", "review", "deny"] },
    ],
  }),
});

const data = await res.json();
console.log(data.is_safe, data.decision, data.answers);

The AI that returns typed decisions,
not text.

Every decision is one of three types

noul

choice

score

State in, decision out

See it decide

1 · State

2 · Questions

3 · Response

One POST, one JSON reply

Request

Response

Call it from anything