What is @amatelic/jev?

A TypeScript package for one-line local AI decisions: it returns calibrated choice, yes/no (noul) and score answers from open models without API keys or cloud calls.

How do I make a decision with it?

Import createJev, await it, then call decide with a state string and typed questions; the README example classifies an event and reads answer.choice and answer.confidence.

Which models does it use?

Aliases cover kev-0.6b on Qwen3-0.6B at 383 MB q4 for fast prototyping, kev-4b on Qwen3-4B at 2.4 GB for better accuracy, and open-jev on DeBERTa-v3 at 800 MB.

Do I need to download models myself?

No. Models are auto-downloaded on first use from the HuggingFace CDN to a local cache with progress callbacks, and you can switch with createJev({ model: "kev-4b" }).

Does it need an API key or Docker?

No API keys and no cloud; the README says native ARM64 Mac or Linux works without Docker, which is only a fallback. Auto-downloading weights does require network access on first run.

Can I train a domain-specific model?

Yes, docs/TRAINING.md is linked for fine-tuning kev or open-jev, exporting to ONNX and loading the result through the same package.

What is the project's state?

MIT-licensed TypeScript created and last pushed 2026-10-13 with 0 stars and 0 forks; the README excerpt stops mid-configuration, so the complete option list is not shown here.

Jev directory

CommunityJev Repos & SDKs

Jev (@amatelic/jev)

A TypeScript package for one-line local AI decisions: calibrated choice, noul and score answers from open models such as kev-0.6b and kev-4b, with no API keys and no cloud, and models auto-downloaded on first use.

See highlights Open github.com

Stars: 0
Language: TypeScript
License: MIT
Reading time: 2 min
Published by: Community
Author: amatelic
Last verified: Oct 2026

amatelic/jev

User repository on GitHub

View on GitHub

One-line local AI decisions. Calibrated choice/noul/score from open models. No API keys, no cloud. Auto-downloads models. Native ARM64/Linux or Docker fallback.

Forks: 0
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

This is one of several Jev listings for the same project. The canonical entry is Jev — mayank953.

Tagscommunity github typescript local sdk

Works forJev for Classification Jev for Scoring Jev for Decision gating

Used inJev for Software & DevTools

Languages: TypeScript
Capabilities: Open sourceRuns locally

Highlights

createJev() downloads models on first use: kev-0.6b is 383 MB, kev-4b is 2.4 GB and open-jev is 800 MB.
One decide call takes a state plus typed questions and returns typed answers with confidence, e.g. sports at 0.998.
Aliases map to Qwen3-0.6B, Qwen3-4B and DeBERTa-v3 bases and are switchable with the model option.
Supports choice, yes/no (noul) and score answers; docs/TRAINING.md covers fine-tuning and ONNX export.
No API keys, no cloud and no Docker needed on ARM64 Mac or Linux; MIT-licensed, 0 stars, created and pushed 2026-10-13.

Quickstart

import { createJev } from "@amatelic/jev";

const jev = await createJev();

const [answer] = await jev.decide(
  "Event: HC Sparta Praha hockey at O2 arena",
  [{ kind: "choice", prompt: "What type of event?", options: ["concert", "sports", "exhibition", "general"] }]
);

Watch out

README and GitHub metadata reviewed 2026-10-17; the captured text ends mid-sentence in the configuration example, so the full options list, the Docker fallback, on-disk download sizes and any benchmark numbers are not visible.

Questions about this entry

Short answers drawn from the linked project or page.

From the Jev guides

Original write-ups on what this entry is for.

The Open Reimplementations: Local Models That Speak System One

Sep 22, 2026

Closed weights have not stopped the ecosystem from rebuilding the shape in the open. API-compatible servers, retrained heads on small models, and inference-time tricks that skip generation entirely.

Jev Repos & SDKs#open-models#research#local

Read article

From the Jev AI model community

Posts from builders shipping with Jev right now.

Follow @typesafeai