Jev on Hugging Face
"jev huggingface" is a search with a surprising answer: there is no official Jev on Hugging Face. Jev (TypeSafe) is a hosted, closed-weight model you call over an API — there are no weights to download. What Hugging Face does host are the open, Jev-like decision models people reach for when they want to self-host. Here's what's real, what isn't, and how to use each.
Every example needs a jv_live_ key. Grab one on the pricing page, export it as JEV_API_KEY, and you're ready. Get a key →
The reason "jev huggingface" and "github jev" are rising together is that people assume a model this talked-about must have downloadable weights. It doesn't. Jev is TypeSafe AI's hosted System One model: closed-weight, served from their infrastructure, and calibrated with RLCD so its probabilities can be trusted. There is no official Jev checkpoint on Hugging Face and no official Jev download. What's genuinely on Hugging Face is the open ecosystem that grew up around the same idea.
What is NOT on Hugging Face
- Official Jev weights — none published. Jev is hosted and closed-weight.
- A Jev checkpoint or GGUF/ONNX of the real model — doesn't exist; don't trust a repo claiming to be "official Jev weights".
- A way to run the actual TypeSafe model offline — the only way to use real Jev is the hosted API.
What IS on Hugging Face — the Jev-like open models
If you want open weights you can download and run, you want the Jev-like projects, not Jev itself. These are separate models that do the same job — typed, calibrated decisions instead of free text — under open licenses:
- Laya (Convai Innovations) — an open-weight System One decision model, with MLX builds for Apple Silicon and ONNX builds, hosted on Hugging Face.
- OpenJev — a community open-weight take on the Jev-like decision model.
These are real, downloadable, and self-hostable — but they are not Jev, and their quality and calibration are their own. Validate any open build on your own labelled data before you wire a threshold to it.
Which should you use?
If you need the calibrated, pinned, zero-ops decision contract, use hosted Jev — nothing to download, and the probabilities are trained to be honest. If you need weights on your own hardware (privacy, offline, no per-call bill) and will own calibration yourself, pull a Jev-like model from Hugging Face. Because both speak the same primitives — choice, score, noul — the interface your code depends on can stay the same while you swap what answers it.
Use hosted Jev without any download
There's nothing to install for real Jev — it's one HTTP call. Grab an instant hosted jv_live_ key, and you're making calibrated typed decisions in minutes, billed per input token with output free:
curl https://jevtypesafeai.com/api/v1/decide \
-H "Authorization: Bearer $JEV_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"state": "Ticket: refund not received after 10 days",
"questions": { "urgent": { "type": "noul", "instructions": "Escalate now?" } }
}'Want the self-host path instead? The Jev vs Laya and Running Jev locally pages walk through the open, downloadable route end to end.
FAQ
Is Jev on Hugging Face?
No. There are no official Jev weights on Hugging Face — Jev is a hosted, closed-weight model you call over an API. What's on Hugging Face are the open, Jev-like decision models (Laya, OpenJev), which are separate projects, not Jev.
Can I download Jev weights?
No official Jev download exists. If a repo claims to be 'official Jev weights', treat it with suspicion. For downloadable weights, use the Jev-like open models such as Laya (MLX/ONNX) instead.
What's the difference between Jev and the open models on Hugging Face?
Jev is hosted, closed-weight, and calibrated with RLCD so its probabilities are trustworthy out of the box. The open models (Laya, OpenJev) are downloadable and self-hostable, but you own their calibration and quality validation.
Is there a Jev GitHub repo?
The model itself isn't open-sourced, but the tooling around it is — the open jev-mcp server and SDKs live on GitHub. The open-weight decision models (Laya, OpenJev) publish on both GitHub and Hugging Face.
How do I use real Jev if it's not downloadable?
Call the hosted API. Grab an instant jv_live_ key and POST to the decision endpoint — it's one HTTP call, no install, billed per input token with output free.
See also: Jev vs Laya (open/local) · Running Jev locally · Is Jev open source?
Try Jev before you wire it in
Run a real, typed decision in the browser — free, no signup — then drop your key into the example above.