01
Route before you call
Assess task difficulty first, send simple requests to a faster model, and escalate only when needed.
Jev · a decision tool for software teams · Jev AI Model
Jev handles classification, routing, scoring, and safety checks—give it state and get a result your code can use.
Three deploys have failed and production is returning 500s. Should this be escalated to a person?
0.94
Intent
0.91
Confidence
typed
Output
Online playground
Enter a real scenario, define typed questions, and inspect a typed decision with probabilities.
The playground is free with unlimited runs.
Where it connects
Jev handles the decision itself. Your product still decides which model to call, when to queue, and when to ask a person.
Assess task difficulty first, send simple requests to a faster model, and escalate only when needed.
Check intent before a tool executes. Ask for confirmation around deletion, payment, or other sensitive actions.
Classify intent, urgency, and human-review needs together so tickets take the right path.
Decide which tool results still matter and preserve source details when context is compressed.
From input to action
Jev does not ask you to rewrite your system. Separate state, questions, and the next action; the model only handles the decision in the middle.
Use a ticket, message, form fields, or the structured context your agent already sees.
state = ticket.textDescribe the decision with choice, score, or noul. Ask multiple questions in one request.
questions = { intent, urgent }Use the result and its probability to queue, route, block, or request review. Your system keeps control.
if result.urgent: escalate()Where it fits
The hardest product work is often not generating a paragraph. It is making small decisions many times a second without losing control.
Define possible answers up front and pass the result into functions, workflows, and databases.
Classify, score, and answer yes-or-no questions together instead of chaining separate calls.
Use probability or confidence to decide whether the system should continue automatically or ask a person.
Use it inside if, route, guardrail, and queue logic without another chat surface.
A consistent request shape, explicit schemas, and manageable API keys take you from test to production.
Start with question types and state inputs, then follow an SDK or REST example into your system.
Designed for developers
Understand Jev in the playground, create an API key, and connect a validated decision workflow to your product.
Explore API docsWhat a request looks like
Drop the decision into an existing service
from typesafe_sdk import Choice, Noul
response = client.system_one(
state=ticket,
questions={
"intent": Choice(criteria={...}),
"needs_human": Noul(
instructions="Does this need a person?"
),
},
)
# response.answers["intent"].choiceAgent integration
Install the Jev Agent Skill, configure one API key, and let your coding agent use typed decisions for routing, guardrails, verification, and completion checks.
Read the agent guideAdd the public Jev Agent Skill to Codex, Claude Code, Cursor, or another compatible agent.
Set a Jev AI Model API key and choose English or Simplified Chinese for the onboarding and examples.
Give the agent a bounded question; Jev returns a choice, score, or yes probability that code can use.
Start with the Jev Agent Skill
English by default · Chinese supported
npx skills add jev-ai/jev-agent-skill
export JEV_API_KEY="sk_your_key_here"
export JEV_LANGUAGE="en-US"The agent still keeps control
Jev judges the decision. Your agent, application permissions, deterministic rules, and human approvals still control the final action.
Documentation
The docs follow the path developers actually take: understand the model, define questions, send state, and wire probabilities into code.
Open full docsCreate an account, prepare your first state and questions, and inspect the typed decision in the response.
Use choice, score, and noul for classification, continuous scoring, and yes-or-no decisions. Ask multiple questions in one request.
Create an API key, follow the SDK or REST example, and use probabilities and confidence to drive your application logic.
Jev currently accepts text, JSON objects, and arrays of text. Images, audio, and video are not supported yet; validate accuracy separately for non-English inputs.
Jev AI Model turns unstructured state into typed decisions with probabilities for classification, routing, scoring, and safety checks. Choose monthly, yearly, or one-time access for your usage.
One-time plans do not auto-renew or require a long-term subscription; the $10, $100, and $1,000 tiers include credits that remain available after purchase.
Starter
Validate one real workflow, from the playground to your first API call
Pro
Connect classification, routing, and safety decisions to a production product
Enterprise
For multiple workspaces, team collaboration, and custom production integrations, with 10% extra credits
A few quick answers before you start with Jev AI Model.
Validate one decision first, then connect it to your workflow.