Why not just hand-write Jev JSON?

Hand schemas take minutes and need exclusivity and sizing by hand; this parses intent, validates five checks, shows a confirmation card, and lets chip edits re-validate locally with no token cost.

What happens when generated options overlap?

Mutual-exclusivity check flags Billing versus Payment Error types, then the patcher collapses them into distinct buckets before execution.

Does editing chips burn another LLM call?

No. Chip add or remove triggers local fast re-validation with zero token cost; only cold misses or restricted mode invoke the generator engine.

Will it boot with no keys, and what does simulation mode fake?

It boots with no keys: docker compose up serves the Next.js frontend on port 3000 and FastAPI docs on port 8000/docs in high-fidelity deterministic simulation mode. The digest does not spell out which calls the simulator stands in for. Going live means JEV_API_KEY, JEV_API_URL, and an LLM_PROVIDER key (mock, gemini, groq, or openai) in backend/.env, plus CACHE_SIMILARITY_THRESHOLD, EXACT_CACHE_THRESHOLD, and DAILY_INQUIRY_LIMIT.

What is the intent cache actually keyed on, and when does it hit?

384-dimensional vectors from all-MiniLM-L6-v2, with an exact-hit threshold at 0.95 similarity. A hit in unrestricted mode skips the generator engine entirely and executes Jev straight away, which is what makes a repeat query free at zero credits. A cold miss or a restricted-mode request goes to the generator, which extracts a candidate schema and state before validation ever runs.

What happens when the patcher cannot repair a schema?

There is a stall-cycle detector, and after 2 retries it falls back to a graceful plain conversational response instead of looping. The repair itself is a targeted diff patcher, not a regeneration: a coverage failure inserts missing domain categories, an exclusivity failure collapses overlapping buckets, a type-fit failure converts between Choice, Score and Noul, and a sizing failure trims peripheral options to the typical 2 to 6 range.

Is there a security boundary in front of the prompt?

Yes, and it is the first stage of the pipeline. A prompt-injection barrier and sanitizer runs before the cache lookup, so user text is filtered before anything is embedded or sent. The verdict card then reports the certainty percentage, the execution latency and a collapsible probability distribution, which is what makes a generated schema auditable rather than a black box.

CommunityJev Models & Reimplementations

Simple-Jev

No-code layer that turns plain English into validated Jev schemas, confirming with chips before deterministic execution and verdict card.

See highlights

Open github.com

Stars: 0
Language: TypeScript
License: MIT
Published by: Community
Author: ihatecoding01
Last verified: Oct 2026

ihatecoding01/Simple-Jev

User repository on GitHub

View on GitHub

No-code conversational layer for Jev (TypeSafe AI's structured decision model). Turns plain-language questions into validated, cached Jev schemas no JSON, no developer required, no raw schema shown. Not affiliated with or endorsed by Jev or TypeSafe AI.

Forks: 0
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.Homepage

About this project

The No-Code Conversational Layer for TypeSafe AI's Jev Model Turn plain English intent into validated, deterministic System 1 decisions in milliseconds — without writing a single line of JSON schema. TypeSafe AI's Jev is a breakthrough System 1 decision model: sub-100ms, strictly typed, deterministic, and free of autoregressive hallucination across discrete choices (Choice), numeric scores (Score), and boolean assertions (Noul). The catch? Jev is an engine, not an interface. To use raw Jev today, an engineer must:

Quoted from the project's README. Read the full README.

Tagscommunity github

Works forJev for Classification

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

5-point validator checks coverage, exclusivity, type fit, scope, and state; patcher repairs or falls back after 2 retries.
Intent cache on 384-dim MiniLM vectors; exact hits over 0.95 run free with zero re-invocation on chip edits.
Boots with no keys in simulation mode; live Jev and LLM providers enabled via backend .env with thresholds and limits.
Stack is Next.js frontend on 3000 with FastAPI on 8000 and docs; Docker Compose or split hot-reload both work.

Watch out

Demo verdict example shows billing choice at high certainty; live behavior depends on keys, thresholds, and validation. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

jev-demo

Voice-driven UI prototyper for Next.js: local whisper.cpp transcribes mic audio on the Mac, Jev answers a batch of typed Choice/Noul questions per utterance, and json-render renders the design as shadcn/ui React.

Jev Models & Reimplementations#community#github

Community

julianoczkowski/jev-demoTypeScript

local-issue-classifier

Local-first web app that pulls a repo issues and README, asks Jev to score complexity, criticality, effort and relevance, then sorts and exports.

Jev Models & Reimplementations#community#github

Community

JeronimoRepetto/local-issue-classifierTypeScript

codriver

Per-turn opencode router that asks Jev which fleet model fits the turn, rewriting to that model or falling back on low confidence.

Jev Models & Reimplementations#community#github

Community

johnpozy/codriverTypeScript

LocalJev

823

A local Jev-compatible POST /v1/systemone server in TypeScript that maps typed questions onto oMLX classification prompts and normalizes logprob-style answers.

Jev Models & Reimplementations#community#typescript#local

Community

githubnext/localjevTypeScript

OpenJev

174

Express and Vite playground for System One-style decisions, taking a state plus typed questions and returning structured answers. Three swappable backends: per-option LLM micro-scorers, one structured JSON call, or a locally served decider model.

Jev Models & Reimplementations#community#jevbest

Community

SiliconLabAI/OpenJevTypeScript

system-one-foundation-models

Native Swift 6 package that evaluates @Generable structs and enums through System One decision models in 15-150 ms, with on-device Laya, self-hosted laya-serve, and TypeSafe Jev cloud backends.

Jev Models & Reimplementations#community#github

Community

peterfriese/system-one-foundation-modelsSwift

From the Jev guides

Original write-ups on what this entry is for.

Multimodal Decision Models: What Clef and d1 Change

Oct 12, 2026

Jev takes text only; Clef-omni adds image, audio, and video input and Liquid d1 adds images, each keeping the same typed-answer shape. Where that matters, where it does not, and the costs the vendors underplay.

Jev Models & Reimplementations#multimodal#open-models#decision-model

Read article

The Decision-Model Price War

Oct 11, 2026

One model in September, a dozen by October. Clef, Liquid d1, Perplexity, and OpenAI all ship the same typed-answer shape now. How to read the rival benchmarks and pick on your own task.

Jev Models & Reimplementations#decision-model#benchmarks#pricing

Read article

What Is Actually on Hugging Face

Oct 8, 2026

Hundreds of repositories carry the system-one tag, but the ecosystem is a dozen ideas in different sizes. A tour of the families, the licenses, and what a download does not include.

Jev Models & Reimplementations#open-models#hugging-face#local

Read article

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai