Skip to content
learnjev

Start here

Two hours, end to end

Jev is small. 13 tutorials cover essentially all of it, and they are meant to be read in order — each one assumes the one before. If you only have five minutes, the call below is the whole idea in one request.

The five-minute version

needs a key from console.typesafe.ai
one decision
curl -X POST https://api.typesafe.ai/v1/systemone \
  -H "Authorization: Bearer $TYPESAFE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "state": "My card was charged twice for order A-104.",
    "model": "jev-latest",
    "questions": {
      "refund_requested": {
        "type": "noul",
        "instructions": "Does the customer request a refund?"
      }
    }
  }'

You get back {"refund_requested": {"type": "noul", "noul": 0.94}}. No prose to parse, no schema to validate, no branch for “the model apologised instead”. That is the entire pitch, and everything else on this site is about what to build on top of it.

Stage 1

Foundations

~26 min

Jev's whole surface area: one endpoint, three question types, one state field. Get this right and the rest is architecture.

  1. 01
    Your first Jev call

    Get a key, send one request, and read a typed answer back — in cURL, Python and TypeScript.

    Beginner · 6 min

  2. 02
    Noul, Choice and Score

    Jev has exactly three question types. Picking the wrong one is the most common early mistake — here is how to tell them apart.

    Beginner · 11 min

  3. 03
    Designing state Jev can actually use

    State is not a prompt. Shape it like a record you'd hand an expert, filter it in code first, and name the parts your questions point at.

    Beginner · 9 min

Stage 2

Patterns

~69 min

The shapes that recur in every real system — fan-out, confidence gates, composite scores, cascades, agent harnesses and real-time loops.

  1. 04
    Ask every question at once

    Questions run in parallel and output tokens are free, so asking something you might not need is close to free. Branch in code afterwards.

    Intermediate · 8 min

  2. 05
    Confidence-gated routing

    The answer tells you what. Confidence tells you whether you're allowed to act on it without asking anyone.

    Intermediate · 12 min

  3. 06
    Composite scoring: move the weights into code

    Stop asking one model one big question. Ask several small ones and own the weighting yourself.

    Intermediate · 9 min

  4. 07
    Ask for features, not answers

    The one lesson every practitioner reached independently in launch week: Jev is a bad judge and an excellent sensor.

    Intermediate · 10 min

  5. 08
    The cascade: Jev in front of an LLM

    Let a fast, cheap decision decide who handles the work — plain code, a specialist model, or a person.

    Intermediate · 10 min

  6. 09
    Building an agent harness

    Tool selection, dangerous-action gates and loop detection — the control layer that every coding agent already has, and nobody could build cheaply until now.

    Advanced · 11 min

  7. 10
    Real-time loops: the Doom lesson

    Ten decisions a second for about $7 an hour. What that unlocks, what it does not, and the discipline a control loop demands.

    Advanced · 9 min

Stage 3

Reliability

~34 min

Where it breaks, what the benchmarks actually say, and the operational details that decide whether this survives contact with production.

  1. 11
    Read this before you ship

    The nine documented ways jev-1.13 gets things wrong, and what to do instead of each one.

    Intermediate · 13 min

  2. 12
    What the numbers actually say

    193.6× faster, 445× cheaper, 100×, 200×, 40×, 20× — six multipliers from one vendor, and what independent testers measured instead.

    Advanced · 12 min

  3. 13
    Shipping it

    Errors, retries, rate limits, version pinning and the request ID you will wish you had logged.

    Advanced · 9 min

If you only read three pages