Launches on September 21, 2026

Benchmark harness evaluating TypeSafe's Jev model on six public safety benchmarks against Shieldstral-1.0-3B and top SOTA guard models

Title and abstract screening for systematic reviews with Jev: explicit criteria, include/exclude/maybe with reasons, PRISMA counts, RIS export, recall against human screeners.

Customer service bot that never generates text: Jev (TypeSafe AI) makes typed decisions with probabilities, plain Python does the rest. Chat UI, example data, Docker.

Test whether TypeSafe Jev answers your questions correctly before you let it decide anything.

A cached Jev prior for active learning: reusable rank fusion, matched ASReview controls, and a no-key evidence replay.

Jev plays Pokemon Showdown: 40 battles, 940 decisions, reproducible results, and an annotated replay.

Agent skill for TypeSafe AI's Jev decision model: fit assessment, integration recipes, calibration, multi-Jev, benchmarks

Empirical research, agent skills, and a production-grade swarm engine coordinating local SLMs under TypeSafe AI System One (Jev) supervision.

Fork of wuyoscar/jev-skill v0.2.0 adapted to call a local keyless Laya decision API by default (no cost); OpenRouter/TypeSafe routes unchanged

Jev-powered skill router & security auditor for any AI agent (Codex, Claude Code, OpenCode, Hermes): ONE cheap decision per request tells the model WHICH skill to load; scans skill libraries for prompt injection & dangerous commands.

Jev plays a slot machine until the money runs out. The local version of jevslots.live.

Smart paste: copy a whole resume, paste into a job application, and Jev (TypeSafe) routes only the relevant pieces into the right fields.

Snake auto-played by TypeSafe's Jev System One model: one typed decision call per tick, code owns legality and safety.

用 TypeSafe Jev（System One）驱动的自动贪吃蛇：每一步都是结构化判断，由代码组合。

Minimal Python starter for TypeSafe Jev typed decisions — UIbuckets

Jev parses and tracks states of AI models so only actions available to the AI model are chosen based on the models current state.

Use TypeSafe Jev to select subagent models and reasoning effort in Codex and Claude Code.

Technology-assisted review for legal teams with ranking, family propagation and elusion statistics.

Jev as a channel message router | Jev频道消息路由

Jev (TypeSafe System One) vs a heuristic scorer for book-candidate selection — harness, 500-book test, write-up

Simplified Tetris game using Jev AI model

A JEV POC

Replay changed DecisionPacks over historical data, reuse judgments and measure decision flips.

Jev (TypeSafe AI) decisions in Claude Code hooks plus a benchmark harness. Zero-dependency Node.

一个不停在做决定的小镇：每个居民每刻的判断都是一次 Jev（TypeSafe System One）调用，概率、犹豫和「哪些决定只是抛硬币」都看得见 · A town where every decision is a Jev call

Generate a reviewable requirements-to-code-and-tests matrix with coverage and change impact.

Chrome MV3 擴充：在 TradingView 圖表上按一次，把當前 300 根 K 棒送給 TypeSafe Jev 模型判斷多空，結果顯示於側邊面板（不自動下單、零依賴）

Confidence-gated creator inbox triage on TypeSafe's Jev (System One) model

Unofficial experiment: classify Chinese short texts with TypeSafe Jev and calibrate a confidence threshold from labeled samples. 用 TypeSafe Jev 给中文短文本分类并做置信度校准。Bring your own API key. Synthetic data only.

A TypeSafe Jev decision model flying U.N. Squadron (SNES, 1991) from raw WRAM via BizHawk + Lua + Python

Native Unreal Engine 5.8 StateTree tasks for finite, revision-guarded Jev decisions.

Auditable Jev-powered workflow gates with cache, ledger, policy, and stability measurement.

Semantic video scene search: embeddings retrieve, Jev decides. Measured on MSR-VTT — +9 pts R@1, and it can say 'nothing here matches'.

Jev + VLA: structured decisions, failure detection, and recovery for vision-language-action robot policies. Design stage.

Control your own Chrome by voice. Jev (TypeSafe System One) decides intent + target from typed probabilities in ~300 ms per spoken word; the extension acts before you finish the sentence. MV3 port of moritzkremb/jev-voice-browser.

Always-listening voice control for macOS: on-device speech → TypeSafe Jev decisions model (via OpenRouter) picks a command → the app clicks, types and drives apps. Config-driven commands, Accessibility + OCR clicking, live probability HUD.

Can an at-home model match Jev? TypeSafe Jev vs self-hosted Laya on Moroccan Darija sentiment, with reproducible zero-shot evaluation.

A benchmark comparing TypeSafe's Jev versus local open-weight and API LLMs, as well as making Jev chat :) (ish)

Resume↔JD apply-gate cost eval: TypeSafe Jev vs Claude/Grok/GPT tiers (10 real JDs)

Phishing email triage: TypeSafe Jev 1.13 vs GPT-5.6 Luna on 100 labelled emails, accuracy, speed and cost

Typed Jev review diagnostics anchored to exact selected VS Code lines.

Jev + WAM: structured decisions for World Action Models and world-model planning. Design stage.

A single web page for requesting Jev from TypeSafe AI, supporting TypeSafe official api or Vercel AI gateway.

Local-first WebUI for TypeSafe Jev: batch document evaluation, typed questions, formulas, and CSV export. Unofficial.

Rank or pick from more candidates than Jev can see in one call. The documented limit is 255 options; the real one is ~32,768 tokens. Naive chunk-merging throws away 83% of reranking's value -- measured on BEIR scifact, with controls.

An AI assisted storytelling party game. Everyone submits a line to add to the shared story, the AI bot ("Jev") picks the winner, and the story continues to grow. Everyone takes turns submitting a theme after each story. Jev scores every finished story out of 100.

Jevals (jevals.com): independent benchmark of TypeSafe's Jev vs LLMs

Command-line browser agent: Jev picks every action, a small LLM only writes text.

A chat interface built on Jev, TypeSafe's decision-only model: every word is a typed decision

Jev 中文社区