Xev — decisions, not paragraphs

Quickstart

One call, typed answers

Post the state — any text or JSON — and a set of named questions to /v1/systemone. Every answer comes back typed, with the probabilities behind it: nothing to parse, no prompt begging for JSON, no retries when it isn't.

$ curl -sS https://api.axforge.ai/v1/systemone \
    -H "Authorization: Bearer $AXFORGE_API_KEY" \
    -H "Content-Type: application/json" \
    -d '{
  "model": "xev-latest",
  "state": "I was charged twice for my subscription.",
  "questions": {
    "refund":     {"type": "noul",   "instructions": "Is the customer asking for money back?"},
    "department": {"type": "choice", "instructions": "Which team should handle this?",
                   "criteria": {"billing": "Charges and refunds", "technical": "Bugs and outages"}},
    "urgency":    {"type": "score",  "instructions": "How urgent is this?",
                   "criteria": ["Can wait a week", "Should be handled today", "Needs an answer within the hour"]}
  }
}'

{
  "model": "xev-latest",
  "answers": {
    "refund":     {"type": "noul", "noul": 0.8113},
    "department": {"type": "choice", "choice": "billing",
                   "probabilities": {"billing": 0.9464, "technical": 0.0536}, "confidence": 0.8928},
    "urgency":    {"type": "score", "score": 0.8526, "confidence": 0.3559, ...}
  },
  "usage": {"input_tokens": 224, "output_tokens": 0}
}

# pip install requests
import requests

r = requests.post(
    "https://api.axforge.ai/v1/systemone",
    headers={"Authorization": "Bearer YOUR_AXFORGE_KEY"},
    json={
        "state": ticket_text,
        "questions": {
            "spam": {"type": "noul", "instructions": "Is this spam?", "min_confidence": 0.9},
        },
    },
)
a = r.json()["answers"]["spam"]
if a["decided"] and a["noul"] > 0.5:
    archive(ticket)

The full request and answer format, every field and every error, is in the Xev docs.

Three kinds of question

Ask it the way your code needs it

Set min_confidence and Xev tells you when it isn't sure ("decided": false) — so the clear cases run automatically and the rest go to a person, or to a bigger model. Ask up to 32 questions about the same state in one call; the state is read once, however many questions you ask.

Coming from Jev?

Switch by changing two lines

Xev speaks TypeSafe's System One format: the same request, the same question types, the same answer shape. Point your client at https://api.axforge.ai/v1/systemone, use your AxForge key, and your parsing code keeps working — the model name jev-latest is accepted as it is.

The whole picture, side by side: Xev vs Jev.

Fit

What teams use it for

Why a decision model

Your code needs a value, not a paragraph

A chat model answers in prose. To turn that into a decision you write a prompt that begs for JSON, a parser, and retries for when it isn't JSON — and you still don't know how sure the model was. Xev skips all of it: a typed question in, a typed answer out, with the probability behind it. You decide the threshold for acting on its own.

Under the hood, a 27-billion-parameter model — Qwen3.8 27B — reads your state once and weighs every option you give it. It never writes text, which is why output is free.

Specifications

What is served

Data & privacy

Zero prompt retention

Your state and questions are processed in memory in Sweden — not written to disk, not logged, not retained, and never used to train anything. We keep only request metadata (token counts, timestamps, status) for billing and operations. The full policy is at axforge.ai/privacy.

Explore

More from AxForge

Discuss

Comments