Integration
Jev on Vercel AI Gateway
The gateway with the most interesting adoption signal: Vercel does not just resell Jev, it uses it. The eve engine ships Jev as the default model behind its evaluation path, which says more about where a decision model belongs than any launch post.
Where it fits, and where it does not
A gateway is usually a drop-in: change a model string, keep your code. Jev is the exception, and it is worth being blunt about why. Jev does not take a messages array and does not return content. You send a state plus a map of named, typed questions; you get typed answers back under those same keys.
So swapping typesafe-ai/jev into an existing text-generation call is not a change of model, it is a change of interface. The tell is in Vercel's own tooling: both eve and the AI CLI reach for Jev in their evaluate paths — judging output — rather than in a generation path.
The pattern this gateway is good for
Vercel's adoption points at the highest-value shape for a decision model in a web stack: a cheap, fast judgement sitting in front of an expensive one.
Evaluate
Score or gate an LLM's output before it reaches the user. Sub-second and effectively free, so it can run on every response rather than on a sample.
Route
Pick which model handles a request. pi-jev-router does exactly this for a coding agent, per request, through this gateway.
Both keep Jev on the control plane and the LLM on the data plane, which is the split that justifies adding a second vendor at all — see agent patterns for the confidence-gating version.
Calling it
Because the request shape is Jev's rather than the chat shape, the honest starting point is an HTTP call you can read, not an SDK helper that hides a mismatch:
{
"state": "Help! My payouts have been failing for 3 days.",
"model": "jev-latest",
"questions": {
"department": {
"type": "choice",
"instructions": "Which team should handle this?",
"criteria": {
"billing": "Payments, invoicing, refunds",
"technical": "Bugs, outages, integrations",
"sales": "Pricing, upgrades, new accounts"
}
},
"is_urgent": {
"type": "noul",
"instructions": "Does this convey urgency?"
}
}
}That body is the documented first-party shape. When routing it through a gateway, the model field carries the gateway's id instead — and the gateway's billing, rate limits and privacy terms replace TypeSafe's. Check the current model id against the gateway's own model page before you ship; ids on gateways move more often than endpoints do.
Built on it
vercel/eve
Vercel's eve engine ships Jev as the default evaluation model (typesafe-ai/jev) in its experimental evaluate path.
vercel-labs/ai-cli
Vercel Labs CLI that can run Jev as the evaluation model for its evaluate command.
mejiasd3v/pi-jev-router
Adds automatic per-request model routing to the Pi coding agent through Jev decisions on Vercel AI Gateway.
ably-labs/jev-pong
Pong where the ball advances one step per model decision, putting Jev head to head with LLMs through Vercel AI Gateway.
jev-pong is the one to read if you want a feel for the latency difference rather than a number for it: the ball advances one step per model decision, so a slow model is visibly a worse Pong player.