Open-weights 4B decision model you self-host to get validated JSON answers with calibrated probabilities.

SelfJev homepage hero
Overview
What is SelfJev?
Your decisions, typed and certain.
SelfJev is an open-weights 4B decisions model you host yourself. Instead of generating text, it answers typed questions about any text — yes/no, labels, categories, numbers, lists — and returns validated JSON with calibrated probabilities and per-decision thresholds.
It's built for pipelines and agents where a wrong guess costs more than no answer: the model abstains on uncertain calls, batches hundreds of decisions in parallel, and exposes a Jev-compatible interface so existing applications can point at your own server unchanged. Weights are on Hugging Face and the model runs on infrastructure you control, down to a single GPU.
Free and open source, for developers and AI engineers building agents that need structured, typed decisions.
Platforms and languages
SelfJev Availability
Platforms
Languages
Capabilities
SelfJev Key Features
Typed decisions, not text
Ask yes/no, label, category, number, or list questions and get validated JSON back — no parsing raw model text.
Calibrated probabilities
Every answer ships with a calibrated probability and thresholds, so you know how much to trust it.
Abstention over guessing
The model refuses uncertain calls instead of hallucinating an answer.
Built-in batching
Run hundreds of decisions in parallel without retrying, streaming, or re-parsing.
Jev-compatible API
Point your application at your own SelfJev server with no code changes.
Best for
Who uses SelfJev?
Developers and AI engineers building agents or pipelines
Get structured, typed decisions from text with calibrated confidence
Plans and access
SelfJev Pricing
Free
Free and open source. Weights on Hugging Face; runs on your own infrastructure.
Common questions
SelfJev FAQs
What kinds of questions can it answer?
Typed questions: yes/no, labels, categories, numbers, and lists — each returning calibrated probabilities in validated JSON.
How much does it cost?
Nothing for the model itself. It's open weights on Hugging Face; you pay only for the infrastructure you run it on.
What does Jev-compatible mean?
Applications written against the Jev interface can point at your own SelfJev server with no code changes.
Does it generate text?
No. It reads text once and returns structured answers your code can use — no generated prose to parse.
Reviews
See what the community thinks and share your experience.
—
Based on 0 ratings
Rating distribution
Community reviews
No reviews
No written reviews yet.