Skip to the demo

Hesperan

Model checking…

Hesperan

Calibrated judgments for software.

Hesperan 1 reads text or application state and answers typed questions — choice, yes/no, score — with a probability for every outcome. One HTTP call, a number your code can put a threshold on.

Get an API key →See it work ↓

Free plan · no card · email sign-in

(A)

See it decide

A situation, a typed question, an answer with a probability for every outcome. These are recorded answers — run your own in the playground.

The situation

Hi, I was charged twice for order #48213 yesterday. The second charge still shows as pending on my card. Can you refund it? The parcel itself arrived fine.

The question · choice

Which team should handle this ticket?

  • billing
  • shipping
  • technical
  • account

Hesperan 1 answers

billing

p = 0.78

Probability per option
billing77.7%
shipping15.9%
technical3.2%
account3.2%

Recorded 2026-09-22 · 80 ms decision · from the mock API while the model is offline

Try your own →See the API call

(B)

Three types

Every answer is typed and schema-stable. Ask several questions about the same state in one request; they are answered together.

01

Choice

Pick one of up to 26 options you describe in plain language. You get the winner and a probability for every option.

{ "type": "choice",
  "choice": "billing",
  "probabilities": {
    "billing": 0.91,
    "shipping": 0.04, … } }

02

Noul

State a claim; get the probability that it holds. Yes/no decisions with a number you can put a threshold on.

{ "type": "noul",
  "noul": 0.97 }

03

Score

Rate on your own ordinal scale. You get the expected level and the full distribution across levels.

{ "type": "score",
  "score": 2.64,
  "probabilities": {
    "0": 0.01, "1": 0.06,
    "2": 0.21, "3": 0.72 } }
Judgment types in the docs →

(C)

Benchmarks

Measured on public benchmarks, next to Jev 1.13. Strong on decisions; weaker on knowledge-heavy tasks such as MMLU-Pro (61.2 vs 82.9).

8 / 12
decision benchmarks at or above Jev 1.13
88.7%
JevBench public · Jev 86.6%
77.5%
JevBench hard · Jev 73.0%
All 16 benchmarks →

(D)

Use cases

Anywhere software has to make a judgment call and a person should only see the uncertain cases.

  1. 01

    Ticket routing

    Send each message to the right queue, and to a human when the probability is low.

  2. 02

    Phishing & fraud screening

    Score emails, sign-ups and payments before they reach a person.

  3. 03

    Incident triage

    Decide from an alert and its context whether someone must be paged now.

  4. 04

    Content moderation

    Apply your written policy to posts and reviews, with a confidence per rule.

  5. 05

    Agent guardrails

    Let an agent ask “should I proceed?” and act only above a threshold you set.

  6. 06

    Lead qualification

    Read a form or an email and rate fit on your own scale.

Discuss your use case →Pricing

(E)

Start building

From sign-in to the first answer takes about two minutes. Only answered questions count.

  1. 01

    Sign in

    With your email — no password. The free plan starts right away.

  2. 02

    Create a key

    In the console, one per app or environment.

  3. 03

    Ask a question

    POST your state and typed questions to /v1/systemone.

Get an API key →Quick startUse with AI agents

Enterprise & on-prem

A dedicated model in the EU, or Hesperan inside your own infrastructure. Tell us what you want to decide.