Hey community 
!
TypeSafe AI’s Jev model is now available in n8n, and it’s free on n8n Cloud (via Gateway credits) until October 10.
Unless you’ve been holidaying in a faraway galaxy, you’ve probably heard of TypeSafe AI’s new System One model, Jev. It’s spawned a million memes and I’ve had to personally coach @liam into saying it correctly - “It’s Jev. You’ve called it Jeff. Again!”
It’s a very different proposition from a regular LLM in that it doesn’t write text. But it also doesn’t hallucinate (though it can still pick the wrong option). You give it an input, a question, and your desired outcomes and it returns the probability for each one.
In n8n terms, it’s a smart If node. It handles decisions that are too fuzzy to be deterministic. It’s fast and cheap enough to run on every item and, at the time of writing this post, TypeSafe prices it at $42 per 1 billion input tokens (output is free), which they put at 238x lower cost than Claude Fable 5.1 (source). I suggest checking out the videos on TypeSafe AI’s website as the cost and latency numbers required me to do a double take.
Use cases
The obvious next question when you hear “smart If node” is, “so what are the use cases?”
I consider Jev to be useful anywhere a workflow has to choose between a set of options based on unstructured input.
As an example, I built an internal tool at n8n last week called Do I Need an Agent. It is designed for non-technical folk to be able to put their automation idea in as plain text, and Jev evaluates it as either yes or no. Where Jev adds value is with its low cost and low latency because you’d be right in thinking that the Do I Need an Agent example is something that we could do with a frontier LLM.
Liam gets hands on with Jev in n8n in this video and shows some really interesting use cases:
- Replacing an If node: Does this deal need an exec review? Pass in the deal, ask the question, and route to Yes or No.
- Replacing a Switch node: Is this support ticket critical, high, medium or low? Each priority gets its own output, going to the right team.
- Confidence gating: Only let a decision through automatically if Jev is at least 95% sure. Anything below that goes to a human.
Some other use cases that I think would be be very compelling:
- Is this inbound email a complaint, a question or spam?
- Is this lead a good fit for sales?
- Does this form submission need a follow-up?
It’s also worth saying when I wouldn’t reach for Jev. If you already have a clean rule (amount > 10000), stick with an If node. And if you need text to be output, you still want a regular LLM.
How to access it
n8n Cloud (Starter, Pro, or new trial): Jev runs on Gateway credits, so you don’t need a TypeSafe account or API key, and you don’t burn any credits. It’s free until October 10, 23:59 UTC. After that, standard Gateway credit rates will apply.
Self-hosted / Community Edition: Install the TypeSafe AI node and BYOK.
Can’t see the node? You may need to refresh your instance. It can take a little while to show up.
How it works
- Add the TypeSafe AI node and choose [Route item by System One question].
- State: the content you want Jev to evaluate. Map individual fields, or be lazy like @liam and pass the whole item with
{{ JSON.stringify($json) }}. - Instructions: your question, e.g.
Does this deal need an executive review? - Routes: one per outcome (
Yes/No, orCritical/High/Low). Each becomes a named output on the canvas, just like an If or Switch node.
Jev returns how confident it is so you can route the decision, accordingly
Be sure to checkout Liam’s video on YouTube for a more detailed overview.
How will you use Jev?
I would love to hear how you’re planning on using Jev in your workflows. Drop it in the thread below, along with any questions about the node or setup.
