Jev Tutorial
English
English Español हिन्दी Français 日本語 한국어 Deutsch Português Italiano አማርኛ Русский Polski Svenska Српски 简体中文 繁體中文
Submit a project
Learn · System One

Jev Tutorial: How to Use TypeSafe Jev

Follow one complete path: decide whether Jev fits the task, choose an output type, write the decision contract, make the call, and handle uncertainty with confidence-aware policy.

WRITTEN & REVIEWED AgentBuff LAST UPDATED 2026-09-26 PRIMARY SOURCES TypeSafe Docs ↗ Editorial and review method →
5-MINUTE QUICKSTART

Make your first typed Jev call

Jev accepts state plus closed questions and returns Choice, Score, or Noul instead of free-form text. This is the shortest Python path; the six chapters below explain type selection, question design, and production policy.

  1. 01
    Install the clientpython -m pip install typesafe-sdk
  2. 02
    Set the server-side keyexport TYPESAFE_API_KEY="your-api-key"
  3. 03
    Send one Choice questionThe result includes the winner, full probability distribution, and confidence.
from typesafe_sdk import Choice, TypeSafeClient

with TypeSafeClient() as client:
    result = client.system_one(
        state={"message": "Payment failed for three days."},
        questions={
            "route": Choice(
                instructions="Which team should handle this?",
                criteria={
                    "billing": "Payments or subscriptions",
                    "technical": "Bugs or integrations",
                },
            )
        },
    )

answer = result.answers["route"]
print(answer.choice, answer.confidence)
Open the complete Jev Python SDK tutorial →
START BY GOAL

What do you want to do with Jev?

How do I set up Jev?

Configure TYPESAFE_API_KEY on the server, then install the official Python SDK or call /v1/systemone. Start with one bounded Choice question and validate the response in application code.

Open the setup and API guide

How do I use Jev for coding?

Use Jev to classify a change, estimate a risk band, or decide whether human review is required. Keep Git operations, tests, permissions, and merge policy deterministic.

Read the Jev Git workflow tutorial

Is Jev open source?

TypeSafe Jev is a hosted model, not an open-weight model. For local or open-source deployment, compare System One alternatives such as AnyJev, JevK5, Winnow, and Laya.

Open the System One model directory
01
Frame the task

Decide whether Jev fits

Jev is not a chat model. It turns existing state into a closed Choice, Score, or Noul decision and returns a probability distribution.

A good fit

Use Jev for bounded decisions

  • Route one request to one named queue.
  • Place a record on an ordered risk or quality scale.
  • Estimate whether one clearly stated condition is true.
Not a good fit

Use a generative model instead

  • Write, summarize, translate, or rewrite open-ended text.
  • Hold a natural multi-turn conversation.
  • Create an answer when the valid outputs cannot be named in advance.

The official mental model

STATE

One state to evaluate

Pass a string, object, or array containing the text and related facts the judgment needs. Use named object fields for most real requests.

QUESTIONS

One snap judgment per question

Each question should ask one focused thing a knowledgeable person could judge quickly from the supplied state.

PARALLEL

Independent questions in parallel

Questions in one request see the same state, run independently, and do not leak one answer into another.

CODE

Compose answers in code

Weight, threshold, branch, and combine typed answers in ordinary code instead of hiding workflow logic in one prompt.

Decomposition test: If the judgment weighs several independent factors or requires extended reasoning, split it into atomic questions and combine the answers in code.
After this chapter:You can separate content-generation tasks from structured-decision tasks.

Official references: Introduction · State · Primitives

01 / 06