jevwrapperv0.1.0

OPEN SOURCE. HUMAN FIRST.

Human in.
Precision out.

You speak English. JEV speaks decisions.
One small wrapper makes the connection.

Turn everyday requests into explicit criteria, typed questions, and structured answers you can actually use.

Explore the sandbox Get the source
MIT licensedBring your own keysBuilt by Ken Cox
ENGLISH → STRUCTURE01 / 03
YOUR WORDS

“Route this customer request
and tell me if it’s urgent.”

decision.jsonTYPED OUTPUT
{
  "department": "billing",
  "urgent_probability": 0.64,
  "status": "needs_review"
}
Explicit answers. Visible uncertainty.
A conversation with a contract.ILLUSTRATIVE EXAMPLE
01 Describe the intent02 Review the criteria03 Get a typed decisionNo hidden leap from words to action.

THE DECISION SANDBOX

From “I need…” to exactly that.

Checking live service

See the questions before you trust the answers. Start with an illustrative example, or connect to this instance’s live service.

01

Make yourself understood.

Live mode uses this server’s configured LLM and JEV accounts. Your text is sent to those providers when you run it.
START WITH
Owner access Optional

Kept only in this tab’s memory and sent to this server. Provider API keys belong on your server, never here.

You review the plan. JEV evaluates the questions.
This wrapper does not carry out external actions.

02

See the decision take shape.

AWAITING INPUT

A little structure goes a long way.

Your request becomes a plan you can inspect, edit, and approve before JEV evaluates it.

CHOICESCORETRUTH ESTIMATE

A SMALL BRIDGE. A CLEAR CONTRACT.

Language is flexible.
Your decision rules shouldn’t be.

The conversational model helps express the task. JEV evaluates the structured questions. You can inspect the handoff between them.

01 / TRANSLATE

Intent becomes a plan.

A small language model turns your English into typed questions and allowed answers. Missing criteria become follow-up questions.

English → structured plan
02 / EVALUATE

JEV gets the specifics.

Review the plan, adjust the criteria, then send the questions and context to JEV for evaluation. The questions define what it judges.

Explicit inputs → typed results
03 / INSPECT

Keep the human in it.

Read the results and available confidence. Low-confidence or ambiguous answers get a review flag. Export the full response for your workflow.

Clear answers + visible uncertainty

YOURS TO RUN. YOURS TO BUILD ON.

A useful idea.
An open starting point.

Run your own instance with your LLM and JEV API keys. Keep your credentials server-side, define the decision rules, and build the interface your team needs.

Explore the repository JEV documentation ↗

MIT license · Node.js · Vanilla frontend · No build step

YOUR TERMINAL
git clone https://github.com/bouncerguy/jevwrapper.git
cd jevwrapper
cp .env.example .env
# Add your API keys and access token to .env
npm start
Setup, security notes, and API examples are in the README.

Make the conversation natural.
Make the decision explicit.

↗