How I Skipped The Jev Waiting List (Free Access)

The Jev waiting list is dead as of 21 September 2026, and this is the step-by-step walkthrough for getting a working API key without waiting on anyone's approval email.

I sat in that queue myself.

I signed up on launch day, got nothing back for a while, and then discovered the same model had been callable through three gateways the whole time.

Six days after launch TypeSafe switched the queue off completely and handed every new account $5 of free credit.

So this post is the practical version — six numbered steps from zero to your first decision coming back, plus the code, plus the mistakes that cost me an hour.

If you want the background on what changed and why, I covered that separately in the Jev AI model explainer.

Step 1: Create the account at console.typesafe.ai

Go to console.typesafe.ai and sign up with your email.

This is the same URL that used to put you in the queue, which is why so many people assume they are still stuck.

Under the old system, creating an account and not being approved yet was the waiting list — there was no separate form to fill in.

That is no longer how it works.

Confirm your email and you should land inside the console rather than on a holding screen.

If you already made an account during early access, log back in with the same email instead of registering again.

Your $5 of credit should already be sitting there.

Step 2: Generate your API key

Open the API keys section of the console and create a new key.

Copy it immediately, because you will not be shown the full string a second time.

Name the key after the project you are building, not "test", because in a month you will not remember which is which.

If you are going to test from a laptop and deploy from a server, make two keys so you can revoke one without breaking the other.

Step 3: Store the key as an environment variable

Set it as TYPESAFE_API_KEY in your environment.

export TYPESAFE_API_KEY="your-key-here"

Do not paste the key into your source file, and do not commit it.

This sounds obvious and it is still the single most common way people leak API credentials.

Add the variable to your .env and make sure .env is in .gitignore before you write a line of code.

🔥 Want the exact setup I used to get these results? Inside the AI Profit Boardroom, I've got a full AI agent stack section with step-by-step video tutorials on API keys, routing and cost control. Plus weekly coaching calls + 2,800+ members building real automations. → Get access here

Step 4: Install the SDK

TypeSafe ships official Python and JavaScript SDKs, and those are only available if you go direct rather than through a gateway.

If you are already running LangChain, the langchain-typesafe package wires Jev in as a first-class component and picks up TYPESAFE_API_KEY automatically.

Install whichever matches your stack and nothing else.

You do not need a vector database, an orchestration framework or an agent runtime to make your first call.

Step 5: Send one tiny question to prove auth works

Do not start with the complicated thing.

Send a single Noul question — a plain yes or no — against a two-line piece of state.

A Noul question returns one probability that the answer is true, and nothing else.

If that comes back, your key works, your environment variable is loaded and your network path is fine.

If it does not, you have exactly one thing to debug rather than five.

Once that works, swap in a Choice question with three options and check the probability distribution looks sane.

Then, and only then, point it at real data.

Step 6: Watch the credit burn rate before you scale

Input costs $0.042 per million tokens.

Output costs nothing, because Jev's outputs are so small TypeSafe decided they were too cheap to meter.

Your $5 works out at roughly 120 million input tokens, which is a genuinely large amount of testing.

But the state you pass in is where the tokens go, and it is easy to shove an entire conversation history into every call out of habit.

Trim the state to what the question actually needs.

I went through the maths properly in the Jev pricing breakdown.

The three routes that never had a waiting list at all

This is the thing I wish somebody had told me on day one.

While the direct queue was running, the identical Jev 1.13 model was live on three gateways with no approval step.

Route	Model ID	Requirement	SDKs	Version pinning
TypeSafe direct	`jev-1.13`	Account at console.typesafe.ai	Official Python + JS	Yes
OpenRouter	`typesafe/jev-1.13`	Prepaid credits	Generic client	No
Vercel AI Gateway	`typesafe-ai/jev`	Card on the team	AI SDK 7 evaluate	No
Cloudflare Workers AI	`typesafe/jev`	Cloudflare account	Workers binding	No

All four give you the same model and the same answers.

Direct access buys you the official SDKs, version pinning, the status page and a bigger documented request budget.

Gateways buy you speed of setup, and if you are already on Vercel it is genuinely two lines of config.

My rule is simple — prototype on whatever gateway you already pay for, then move to direct before anything touches customers.

The full gateway comparison is in the Jev AI API post.

What you can actually ask Jev

Worth being blunt here, because a lot of people queued up for the wrong reasons.

Jev does not write text.

It takes state plus questions and returns typed decisions with calibrated probabilities.

There are exactly three question types.

Choice picks one option from a list you supply, and returns a probability for every option plus an overall confidence number.

Score rates the state against ordered levels you define, and returns a continuous score plus a distribution across those levels.

Noul answers yes or no, and returns a single probability that the answer is true.

Choice tops out at 255 options per query.

If you need to pick from thousands of items, score candidates in batches first and push the finalists through one Choice question.

That constraint shapes more designs than any other limit in the API.

Why the speed matters more than the intelligence

End-to-end response time runs between 70ms and 500ms.

TypeSafe puts that at 40x to 200x faster than frontier models on the same System One tasks.

That is the actual product.

A decision you can afford to make ten thousand times a minute changes what you are willing to build.

Routing every request through a model becomes reasonable when the model answers in 90ms and costs four cents a million tokens.

LangChain shipped a router built exactly on this — you describe each model's strengths in plain English and Jev picks per request, which is just a Choice question under the hood.

🔥 Building an agent stack that needs cheap routing? The AI Profit Boardroom covers the full build — model routing, fallbacks, cost control and the workflows I run in my own agency. Weekly coaching calls and 2,800+ members. → Join the Boardroom

The gotchas nobody mentions

Rate limits still move around.

TypeSafe has been adjusting them dynamically based on available GPU capacity, and it has not published fixed numbers.

Build a retry with backoff into your client from the start rather than bolting one on after your first 429.

There is no published credit-expiry policy, so do not assume your $5 sits there forever.

TypeSafe has also not said whether removing the waiting list means full general availability or an open beta, which matters if you are putting it on a critical path.

And on accuracy, TypeSafe's own disclosed numbers put Jev at 67.8% aggregate against 74.1% for a comparable LLM on the same benchmarks.

The self-published JevBench score is 75.3.

That is fine for routing and triage where a wrong call is cheap to correct, and not fine for anything irreversible.

Test it on your own data, not the demo prompts.

The first three things I'd build with the free credit

Start with a model router, because it pays for itself immediately.

Describe your models in plain English, let Jev pick per request, and watch your spend on the expensive model drop.

Second, build a triage layer for whatever inbox or ticket queue you already have, using Score rather than Choice because severity is an ordered scale.

Third, put a Noul question in front of anything your agent does that is expensive or irreversible, as a cheap sanity gate.

Each of those is under fifty lines of code and each one saves real money.

More patterns are in the Jev use cases post and the open-source side is covered in OpenJev.

Jev waiting list FAQ

Is the Jev waiting list still active?

No. TypeSafe removed the Jev waiting list on 21 September 2026, six days after the 15 September launch. Signup at console.typesafe.ai is now open to anyone, with $5 of free credit per account.

How do I get off the Jev waiting list if I'm still on it?

You are not on it any more. Log back into console.typesafe.ai with the email you originally registered, and you should be able to create an API key straight away.

How many people were on the Jev waiting list?

Roughly 140,000 signups were processed within about 36 hours around launch, before TypeSafe removed the queue entirely on 21 September 2026.

Can I use Jev without a TypeSafe account at all?

Yes. OpenRouter, Vercel AI Gateway and Cloudflare Workers AI all serve the same Jev 1.13 model. You lose the official SDKs, version pinning and the larger request budget, but the model output is identical.

What does the $5 free credit get me?

About 120 million input tokens at $0.042 per million. Output tokens are not charged at all, so the credit goes much further than it would on a text-generating model.

Does Jev replace my existing LLM?

No. Jev only returns typed decisions — choices, scores and yes/no answers with calibrated probabilities. Keep a generator model for prose and use Jev for routing, scoring and verification around it.

About Julian

I'm Julian Goldie — AI entrepreneur, SEO expert, and founder of the AI Profit Boardroom (2,800+ members). I help business owners scale with AI agents, automation, and SEO.