agentos.guide › blog
Free Jev API Key: Every Way In, And How To Wire It Into Your Agent OS
The fastest free Jev API key route right now is OpenCode Zen's jev-1.13-free model, and on my own test it answered with no key at all.
If you want your own key instead, you can create one in the TypeSafe console or go through Vercel AI Gateway or OpenRouter.
The question most people ask me is simple: can I get Jev for free, and is it safe to build on?
I have run every one of these doors on my own machine, so I will show you what each one costs, where it fits in an Agent OS, and the one mistake that makes a free key expensive.
What a free Jev API key actually gets you
Jev is a decision model from TypeSafe AI, and it launched on 15 September 2026.
It does not write text.
You hand it a "state", which is the email, ticket or log you care about, plus a set of typed questions.
It hands back an answer with probabilities and a confidence number attached.
There are three question types, and they cover most of the small calls an agent makes all day.
- A noul question is a yes-or-no check that returns a probability between 0 and 1.
- A choice question picks one option from a fixed list and shows the odds for every option.
- A score question places the input on a scale of two to ten levels.
That is why a free key is worth chasing.
Every small "is this a refund?" or "which agent should take this?" call stops costing you a full chat-model response.
🔥 Want the exact Jev decision layer setup I run inside my Agent OS?
Inside the AI Profit Boardroom you get the installable Agent OS, step-by-step video tutorials, four coaching calls a week and 3,400+ members building real automations.
Where the free Jev API key options stand in October 2026
This is the part that changes week to week, so here is the honest snapshot.
| Route | Do you need a key? | What it costs | Best for in an Agent OS |
|---|---|---|---|
OpenCode Zen jev-1.13-free | The Zen docs say you use your Zen key, but my own test on 21 September went through with no key at all. | Free for a limited time, and the reply showed a cost of zero. | Wiring a free decision step into a local agent pipeline today. |
| jevplayground.com | No key and no signup are needed. | Free, with usage limits. | Testing your questions before you write any code. |
| TypeSafe console | Yes, you create your own key at console.typesafe.ai. | $0.042 per million input tokens, and output is free. | Production work where you want the official source. |
| Vercel AI Gateway | Yes, a Vercel gateway key with a card on file. | Vercel's monthly free credit, then $0.042 per million input tokens. | Agent stacks already built on the Vercel AI SDK. |
| OpenRouter Decisions API | Yes, your normal OpenRouter key. | Pay as you go, which came to about $0.00002 a call in my runs. | Agent OS setups that already route every model through OpenRouter. |
Vercel's own model page now lists Jev at $0.042 per million input tokens, and the old promotional wording is gone.
That promo window closed on 25 September, exactly as Vercel said it would.
The free credit that Vercel gives every team is still described in its pricing docs.
Vercel's docs also say that once you buy credits, your account moves to the paid tier and the monthly free credit no longer applies.
So if you only want the free lane, add a card to unlock the credit and do not buy anything.
How to get the free Jev API key route working in five minutes
The quickest free lane is the OpenCode Zen model called jev-1.13-free.
The Zen docs describe it as available "for a limited time", so treat it as a free trial lane rather than a forever plan.
Here is the order I would do it in.
- First, open jevplayground.com and run one of the sample questions so you can see what the answers look like.
- Second, write your own question, such as "Was a refund issued to the customer?", and test it there until the answers look sensible.
- Third, send the same question to the Zen endpoint at
https://opencode.ai/zen/v1/systemonewith the model set tojev-1.13-free. - Fourth, read the probabilities and the confidence number that come back, not just the top answer.
- Fifth, decide on a confidence line below which the job goes to a human instead of an agent.
When I ran this on 21 September, my refund check came back at 0.98 and the reply showed a cost of zero.
The Zen docs say you should send your Zen API key, so create one anyway in case the keyless behaviour stops.
What a Jev request looks like in plain English
Every Jev call, free or paid, has the same three parts.
The first part is the model name, which is jev-1.13-free on the free Zen lane and jev-1.13 on the paid one.
The second part is the state, which is simply the text you want judged, such as a support note or a lead email.
The third part is your list of questions, and each question has a name, a type and plain-English instructions.
You can ask several questions about the same state in one call, which keeps the number of requests down.
What comes back is an answer for each question, with the probabilities and a confidence number alongside it.
On a choice question you see the odds for every option, not just the winner.
That full spread is what lets your agents behave sensibly when the model is torn between two answers.
TypeSafe quotes 70 to 500 milliseconds end to end for a call, and my free Zen calls took about a second from my desk, lag included.
The Agent OS angle: where a free Jev key sits in your stack
Most people treat a new model like a new chatbot, and that is the wrong mental model here.
Jev is a part, not a product.
In my Agent OS, the expensive brains like Claude and Hermes do the heavy thinking and the writing.
The tiny decisions in between are where the money leaks out.
Think about how many times an agent asks itself which folder an email belongs in, whether a task is finished, or which agent should pick up a card.
Each of those used to be a full model call that wrote a sentence when the answer was one word.
A decision layer flips that around.
The decision model answers the small question in well under a second, and only the jobs that need writing go to the big model.
That is the slot a free Jev API key fills.
Three places I would plug it in first
The first place is inbox triage, where every incoming email gets sorted into a folder before an agent touches it.
On a 48-email sample inbox, the free model put 45 of 48 in the right folder in 8.4 seconds and cost nothing.
The second place is lead grading, where every form fill gets marked hot or cold the second it lands.
On 24 contact-form messages, the free model put all 24 on the right side.
The third place is the "is this task done?" check that sits at the end of an agent run.
In my test, a thin one-line log scored 0.22, a failed import scored 0.02 and a clean import scored 0.79.
That spread is exactly what you want, because it tells the agent when to retry and when to stop.
How to use the confidence number so the free key does not burn you
Here is the trap that catches almost everyone in week one.
Jev always answers inside the box you give it, but inside the box can still be wrong.
When I gave it a list with no "other" option, it filed "Can I bring my dog to the workshop?" under sales at 98%.
The answer was the right shape and completely wrong.
The fix is simple, and it is the habit I want you to build before you put this anywhere near a customer.
- Always add an "other" or "none of these" option to every choice question.
- Read the confidence number, which is a separate thing from the top probability.
- Set three lanes: act automatically, act with a check, or send to a human.
- Move the line based on the cost of being wrong, so refunds get a stricter line than folder sorting.
TypeSafe's own docs describe the same three paths of acting automatically, proceeding with caution or routing to a human.
Free Jev API key limits you should know before you build on it
Free routes come with strings, and I would rather you know them now.
The Zen free model is labelled limited time, so build your code so the model name lives in one config value.
That way, switching from jev-1.13-free to the paid jev-1.13 is a one-word change.
Read the Zen data policy before you send anything sensitive through a free lane.
Jev is text only, so it will not read images, audio or video.
It cannot write prose, and it is not built for maths, counting, dates or pulling exact values out of text.
The state plus your longest question needs to fit in roughly 32,000 tokens, and a choice question can hold up to 255 options.
The benchmark numbers TypeSafe publishes come from workflows TypeSafe wrote, so run it on your own data before you trust it.
Is the paid key worth it once the free Jev API key runs out?
For most small businesses, the paid price is so low that the free tier is really just a convenience.
TypeSafe lists Jev at $42 per billion input tokens, and output tokens are free.
My OpenRouter runs worked out at around $0.00002 per call.
A thousand decisions a day at that rate costs you a couple of cents.
The real cost is never the key.
The real cost is a confident wrong answer that an agent acts on without a check.
So spend your energy on the confidence lanes, not on hunting for another free key.
What to do next
Open the playground and run one question you actually care about in your business.
Then wire that same question into one agent workflow, with a human lane for anything under your confidence line.
If you want the whole decision layer already wired into an Agent OS, with the tutorials and the coaching calls, it lives inside the AI Profit Boardroom.
My full walkthrough of the three free doors, with screen recordings, is in the Jev AI free guide.
Also On Our Network
FAQ
Is there a free Jev API key?
Yes, there are free routes. OpenCode Zen offers a limited-time free model called jev-1.13-free, jevplayground.com needs no key at all, and Vercel AI Gateway gives every team a monthly free credit once a card is on file.
Does the Jev free tier on OpenCode Zen need a key?
The Zen docs say to use your Zen API key. In my own test on 21 September 2026 the request went through without one, but create a key anyway in case that changes.
How much does Jev cost after the free tier?
TypeSafe lists Jev at $0.042 per million input tokens, and output tokens are free. My OpenRouter runs came to roughly $0.00002 per call.
Is Vercel's free Jev promo still running?
No. Vercel's promotional pricing ended on 25 September 2026, and its model page now shows the normal input price. The standard monthly free gateway credit still applies until you buy credits.
Where should Jev sit in an Agent OS?
Put it in front of your big agents as a decision layer. It answers the small yes-or-no, pick-one and score questions, and only the jobs that need writing go to Claude, Hermes or another large model.
Final word on the free Jev API key
Start with the free lane, test it on your own data, and set a confidence line before any agent acts on an answer.
That is how you turn a free Jev API key into a decision layer you can actually trust.
About Julian
I'm Julian Goldie, an AI entrepreneur, SEO expert and the founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
My YouTube channel has 400,000+ subscribers, and I run Goldie Agency, a seven-figure SEO agency.
Related guides on agentos.guide
→ 3 Free Ways To Use Jev AI Right Now
→ Jev AI: The Fastest-Adopted AI Model on Vercel
→ 10 Things I'm Building With Jev AI
Get the whole Agent OS
The installable system behind every guide on this site — dashboard, agents, memory and pipelines — updated as new tools land, with 3,400+ members and four coaching calls a week.
Join the AI Profit Boardroom →📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉