Skip to content

Start in 3 steps

You need about 5 minutes, Python 3.10 or newer, and one OpenRouter API key. That single key works for both Jev and the LLM.

1. Get an OpenRouter key

Sign in at openrouter.ai/keys and create a key. It starts with sk-or-v1-. Add a few dollars of credit. The whole course costs less than $0.20 to run.

2. Clone and set up

git clone https://github.com/harshithsunku/learn-jev-end-to-end.git
cd learn-jev-end-to-end
./setup.sh

git clone https://github.com/harshithsunku/learn-jev-end-to-end.git
cd learn-jev-end-to-end
.\setup.ps1

git clone https://github.com/harshithsunku/learn-jev-end-to-end.git
cd learn-jev-end-to-end
uv sync --extra ui
cp .env.example .env

The setup script installs everything (with uv if you have it, otherwise with pip) and creates a .env file for your settings.

Open .env and paste your key:

.env
OPENAI_API_KEY=sk-or-v1-your-key-here

Then check that everything works:

uv run python scripts/doctor.py

OK   LLM chat completion                  1169 ms  'pong'
OK   LLM tool calling                      842 ms  get_time({"city":"Paris"})
OK   Jev System One (typesafe)             447 ms  model=typesafe/jev-1.13-20260917  is_scam=0.97  kind=phishing

All green. Next: uv run jupyter lab  ->  open 01_hello_jev.ipynb

Only the Jev check failed?

Your key may not have Jev access yet. Set JEV_BACKEND=adapter in .env and the notebooks will answer the same questions with your LLM instead. Everything except the benchmark notebook runs. More about the fallback.

3. Open the first notebook

uv run jupyter lab

Open 01_hello_jev.ipynb and run the cells from top to bottom (Shift+Enter). Every notebook is also readable on this site, with real outputs, if you just want to read first.

Your first Jev call, explained

This is the whole idea. Everything else in the course builds on it.

from typesafe_sdk import TypeSafeClient, Noul, Choice, Score

jev = TypeSafeClient(
    api_key="sk-or-v1-...",                  # your OpenRouter key
    base_url="https://openrouter.ai/api",    # Jev through OpenRouter
    model="~typesafe/jev-latest",
)

r = jev.system_one(
    "Help! Payouts have failed for 3 days and my team can't get paid.",     # the state
    {
        "urgent": Noul(instructions="Does the message convey urgency?"),
        "team":   Choice(instructions="Which team handles this?",
                         criteria={"billing": "payments, payouts", "technical": "bugs, outages"}),
        "mood":   Score(instructions="How frustrated is the customer?",
                        criteria=["calm", "annoyed", "frustrated", "furious"]),
    },
)

r.nouls["urgent"].noul      # 0.98       -> probability of "yes"
r.choices["team"].choice    # 'billing'  -> plus .probabilities and .confidence
r.scores["mood"].score      # 2.83       -> between "frustrated" (2) and "furious" (3)
Part What it is
state the thing being judged: text or JSON
Noul a yes/no question; you get the probability of yes
Choice pick one option from a list you define (up to 255); you get a probability for each
Score a position on an ordered scale (2-10 levels); the score can fall between levels

All three questions are answered in one call, in parallel.

Where to next?