Skip to content

About

A curated list of open-source Jev projects where every entry links to the line of code that calls Jev, and every performance number comes from an independent 2,390-question benchmark.

Resources

Readme

Contributing

Contributing
Activity

Stars

6 stars

Watchers

0 watching

Forks

0 forks
Report repository
BranchesTags

Repository files navigation

Awesome Jev — Verified

verified entries checked CC0

English · 简体中文

A curated list of open-source projects built on Jev, TypeSafe AI's System One model for typed decisions — where every entry links to the line of code that calls Jev, and every performance number comes from a measurement we ran ourselves.

Jev is not a chat model. You hand it state plus typed questions, and it hands back constrained answers with probabilities:

import { experimental_evaluate as evaluate } from 'ai';

const { answers } = await evaluate({
  model: 'typesafe-ai/jev',
  state: 'I was charged twice. Please refund the duplicate.',
  questions: {
    requestsRefund: { type: 'boolean', instructions: 'Is the customer asking for money back?' },
    dept: { type: 'choice', instructions: 'Which team?',
            criteria: { billing: 'Charges and refunds', technical: 'Bugs', account: 'Login' } },
    severity: { type: 'score', instructions: 'How severe?', criteria: ['Cosmetic', 'Degraded', 'Blocking'] },
  },
});
// answers.dept.choice === 'billing', with a probability distribution attached

Why another Jev list

There are already more than a dozen awesome-jev lists, and several are good — they are credited in Lists & resources. This one is built differently in two ways.

1. Entries carry receipts. Inclusion is not decided by reading a README. Every repository below was downloaded at a pinned commit and searched for an actual Jev call — the API endpoint, an official SDK import, a model identifier, or the AI SDK's evaluate interface. The kind badge on each entry links to that exact file and line. 765 of 788 entries cleared that bar; the 54 that matched a search but produced no code evidence are parked in data/leads.json instead of being quietly listed.

2. Claims are replaced by measurements. Most lists repeat TypeSafe's marketing numbers — "sub-100ms", "100× faster", "eliminates hallucination". We ran 2,390 labelled questions through Jev and four comparison models and checked. Some claims hold up well; several do not.

What we measured Jev GPT-4.1-mini GPT-5.6 Sol
True/false accuracy 92.5% 87.8% 92.1%
Multiple-choice accuracy 77.3% 74.3% 82.5%
Confidently wrong (≥90% sure, still wrong, of 800) 6 70 39
Prompt injections that worked (of 80) 1 73 1
Median latency, one question per call 456 ms 928 ms 1661 ms
Cost for all 2,390 questions $0.049 $0.246 $4.27

And claim by claim:

Claim Source What we measured Verdict
Probabilities are calibrated docs Boolean ECE 0.048 — level with Sol (0.043) and Terra (0.051), better than Luna (0.105). Multiple-choice ECE 0.124: overconfident, same as Sol. Partly
Extremely consistent on similar inputs docs 60 rephrased questions: zero answer flips (Sol 1, Terra 3, Luna 9). 120 polarity pairs: 1 self-contradiction (Sol 5, Terra 6, Luna 11). Best of the five. Holds
Deterministic — use it like a strongly typed object tweet Same question 5×: no answer ever flipped, max probability drift 0.09. But only 12 of 30 were bit-identical. Terra and Luna flipped outright (drift 0.81, 0.98). Close, not bit-exact
Under 100 ms — fits inside a 60 FPS game loop tweet Server-side median 248 ms, fastest single call 189 ms. Across 40 sequential calls, not one came in under 100 ms. End-to-end median 456 ms. Not reproduced
100× / 194× faster than language models tweet / blog One question per call: ~3–3.6× faster than GPT-5.6 (456 ms vs 1.3–1.7 s), ~2× vs GPT-4.1-mini. Batching 50 questions into one call gets to ~5 ms per question — that is where the order of magnitude lives. Only when batching
$42 per billion input tokens, output free docs 1,154,813 input tokens billed at $0.0485 — rate confirmed. Note Jev counts roughly 2× the tokens GPT does for the same text. Holds
445× cheaper than language models blog Same 2,390 questions: 88× cheaper than Sol, 45× than Terra, about 5× than Luna and GPT-4.1-mini. 445× did not reproduce against anything we ran. Depends who you compare to
Eliminates hallucination tweet 36 answers were wrong while ≥90% confident, out of 1,200 standard-dataset questions. That is the fewest of five models (Sol 84, Terra 88, Luna 140, GPT-4.1-mini 142) — but it is not zero. Overstated
Weaker in Chinese than English docs (TypeSafe's own caveat) Same 150 XNLI items: 79.3% in English, 68.7% in Chinese. All three GPT-5.6 variants dropped further (13–16 points) than Jev did (11). True, and not unique to Jev

Full method, per-dataset accuracy, calibration curves and the failure cases are in benchmark/. The harness is open, so you can disagree with us in code.

Contents

Each section shows the top 10 by stars; the full category list is one click away. Stars were read on 2026-09-20 — they measure attention, not quality.

How an entry gets here

  1. Discover. GitHub search across repository names, descriptions and topics for Jev and TypeSafe System One terms.
  2. Verify. Download the repository at its current commit and search the source for api.typesafe.ai, /v1/systemone, an official SDK import, a jev-* model identifier, or experimental_evaluate. Config and CI files count for less than real source files.
  3. Classify. Jev itself sorts each repository into one of the categories above and rates how reusable and substantial it looks. Low-confidence calls are flagged in the data for human review. Using the model to build its own directory is partly a demonstration and partly an admission: the categories are a machine's opinion, not a verdict.
  4. Publish. scripts/build.mjs regenerates these pages from data/repos.json. Nothing in this README is hand-edited.

What a listing does not mean

A listing means one thing: on 2026-09-20, this repository contained code that calls Jev. It is not a review of quality, security, licensing or whether the project runs at all. Many of these repositories are days old and were published during a launch rush. Before depending on one, read the code, check the license, and run it yourself.

The list

Official

Repositories published by TypeSafe AI itself. · 4 entries

→ full list: categories/official.md

SDKs & clients

Libraries, bindings and CLIs for calling Jev from a language or shell. · 131 entries

  • Sac-Y/Jev-cu — 把 Computer Use 的「下一步点哪里」交给 Jev（TypeSafe System One）：Jev 从界面文字候选中选元素、动作、完成度与风险，Codex Computer Use 负责读取界面与执行，本地策略门槛拦截敏感操作。只传文字，不传截图。 ★ 355 · endpoint · JavaScript
  • sutro-sh/jev-align — Build calibrated AI classifiers from human feedback using Jev and GEPA. ★ 164 · sdk · Python
  • pithings/advocaat — A small, type-safe client for asking AI questions about your data, powered by TypeSafe Jev. ★ 85 · endpoint · TypeScript
  • obie/ruby_decision_model — Ruby client for decision models such as Typesafe Jev ★ 45 · endpoint · Ruby
  • bnsd55/jevmlx — Jev-style parallel constrained decisions for any MLX model on Apple Silicon. Typed, schema-valid JSON in one forward pass. ★ 40 · model_id · Python
  • mizchi/jev-lint — A linter for the things a linter could never check: whether a function does ★ 21 · endpoint · TypeScript
  • zhengxuyu/litjev — Turn any off-the-shelf LLM into a Jev -like decision layer ★ 21 · endpoint · Python
  • dannote/jev — TypeSafe Jev for OTP: reply to Jev from a GenServer and pattern match on its answer ★ 17 · endpoint · Elixir
  • shiftynick/jev-axi — Agent-ergonomic CLI for TypeSafe's Jev: fast calibrated judgments (pick, rate, check, rank, triage, guard) from the shell ★ 17 · endpoint · TypeScript
  • shantanugoel/mario-jev — A uv-managed Python prototype that plays NES Super Mario Bros. (level 1-1 by default). ★ 12 · sdk · Python

→ 121 more in categories/sdk.md

Agent tooling

Plugins, MCP servers and skills that put Jev inside a coding agent. · 154 entries

→ 144 more in categories/agent.md

Guardrails & review

Approve, reject, flag or escalate — safety gates, moderation, code review. · 47 entries

  • devagrawal09/jev-review — A staged code-review workflow and local dashboard built with TypeSafe Jev. ★ 367 · sdk · TypeScript
  • lakeday-org/perch — Semantic code linting with Jev ★ 161 · endpoint · JavaScript
  • DevMortimer/pi-warden — Guardrails for Pi built on pi-typesafe that steer the agent instead of interrupting you: Jev judges irreversible and off-task tool calls, detects stuck loops, checks unverified done claims, flags slop ★ 96 · endpoint · TypeScript
  • brainstormity/Jev-Moderation-Bot — A Discord moderation bot built with Python and TypeSafe AI (Jev System One). It filters spam and scam links in real time, escalates offenses automatically, and lets moderators prof ★ 36 · endpoint · Python
  • Eriskii/ErisLint — Rust linter powered by configurable Jev rules, with a VS Code extension. ★ 15 · endpoint · Rust
  • wobsoriano/oxlint-plugin-jev — Lint rules written in plain English. Oxlint finds the code, TypeSafe Jev answers the question. ★ 14 · endpoint · TypeScript
  • huntedman/JevLint — Configurable semantic linting powered by Jev, with file-level NOUL judgments and a magic-strings plugin. ★ 10 · endpoint · TypeScript
  • tyler-dot-earth/patdown — typesafe's jev as a "fuzzy linter". give your code an ocular patdown. ★ 10 · endpoint · TypeScript
  • anpicasso/hermes-jev-approvals — PoC: TypeSafe Jev as the reviewer for Hermes Agent smart command approvals. 8.7x faster, 4.4x fewer prompts, measured on 153 real commands. Approvals only. ★ 8 · endpoint · Python
  • frostney/clean-code-review — Every code file in a pull request, judged against Uncle Bob's Clean Code by TypeSafe's Jev, then reviewed by Luna. Built on eve and Next.js. ★ 6 · ai_sdk · TypeScript

→ 37 more in categories/guardrail.md

Context engineering

Compaction, pruning, memory and deciding what an agent should read or keep. · 25 entries

→ 15 more in categories/context.md

Browser & computer use

Browser automation, GUI and device control driven by typed decisions. · 42 entries

  • browser-use/jev-ultrafast — i. am. speed. ★ 9.4k · endpoint · Python
  • wy-coliney/jev-browser-use — 5–10x faster browser operations: Jev clicks, Codex thinks and verifies. Built at EZCollegeApp. ★ 207 · endpoint · JavaScript
  • jkudish/jev-browser — Browser use using Typesafe's Jev model ★ 156 · sdk · TypeScript
  • moritzkremb/jev-voice-browser — Control a real browser by voice. Jev (TypeSafe System One) decides intent + target in ~300 ms per spoken word; Playwright acts — often before you finish the sentence. ★ 126 · sdk · JavaScript
  • savka777/jev-use — Say it, and your Mac does it. A computer-use harness on Jev that reads the screen through Accessibility. Fast, no vision model ★ 36 · endpoint · Swift
  • Ying-Kai-Liao/jev-browser — Browser automation where an LLM plans and Jev (Typesafe System One) decides. Library, CLI and MCP server. ★ 22 · endpoint · JavaScript
  • razaanstha/ulka — Experimental browser agent powered by FX, Jev, and Vercel AI Gateway. Bring your own API key to read pages and automate browser tasks. ★ 18 · ai_sdk · TypeScript
  • jcpsimmons/jev-macos-loop — Open-source macOS AI computer use and native GUI automation on Apple silicon. Jev + OmniParser CoreML + Apple Vision OCR. Bring your own OpenRouter, Vercel AI Gateway, or TypesafeAI token. ★ 13 · endpoint · JavaScript
  • jekhov/jekhov — Policy-bounded Jev target selection for resilient Playwright workflows ★ 7 · endpoint · TypeScript
  • romaluev/jev-ego — Fast browser agent for ego lite. One TypeSafe request per step; an agent or Jev picks the move. ★ 7 · endpoint · TypeScript

→ 32 more in categories/browser.md

Infrastructure

Gateways, routers, caches, database extensions and self-hosted endpoints. · 31 entries

→ 21 more in categories/infra.md

Search & retrieval

Semantic grep, ranking, filtering and classification over documents and data. · 25 entries

  • kyotofin/tax-doc-classifier — Tax document page classifier built on Jev decisions. 100% strict accuracy across 261 IRS forms, ~$0.001 per page. ★ 272 · endpoint · TypeScript
  • superagents-lab/jev-search — Search the web with TypeSafe's Jev: source selection, query understanding and relevance ranking. Built with Search1API. ★ 248 · endpoint · TypeScript
  • uehaj/jev-semgrep — grep by meaning, across languages. TypeSafe Jev scores every line against a meaning; combine meanings with AND/OR/NOT. 意味で探す grep。日本語で英語を、英語で日本語を検索できる ★ 96 · endpoint · JavaScript
  • can1357/jegrep — Semantic grep: find code by describing what you're looking for, powered by Jev. ★ 29 · endpoint · Rust
  • devanshbatham/commit-miner — Classify Git commit diffs and messages with Jev. Bug fixes, security fixes/CWEs, and change types. ★ 26 · endpoint · Rust
  • ellipsis-dev/blink — Codebase search powered by Jev from @typesafe-ai ★ 24 · sdk · TypeScript
  • keltokhy/jgrep — grep, but the pattern is a description. Filters lines by meaning with TypeSafe's Jev decision model: ~200 ms and a thousandth of a cent per line. ★ 12 · endpoint · Python
  • hev/reranker — Use Jev (TypeSafe's System One model) as a calibrated reranker: one call, up to 30 documents, a probability per document. Apache-2.0. ★ 6 · sdk · Python
  • kyu1204/jgrep — grep for what code does, not what it's called. Semantic code search powered by TypeSafe Jev. ★ 6 · endpoint · TypeScript
  • 1jehuang/jev-pr-labeler — Semantic GitHub PR labels using Jev's typed decisions, with conceptual scope instead of line counts ★ 4 · model_id · Python

→ 15 more in categories/retrieval.md

Applications

End-user products: trading, social, recruiting, home automation, productivity. · 106 entries

  • jarrodwatts/jev-trader — One AI trade decision every Monad block. Jev on Kuru MON-USDC. ★ 1.4k · ai_sdk · TypeScript
  • droidrun/mobile-jev — ▶ Watch the demo — Jev opens Uber, enters a route from San Francisco Airport to the Golden Gate Bridge, and reaches payment selection. The recorded task timer shows about 21 second ★ 241 · endpoint · JavaScript
  • kitze/unclutter — WXT browser extension: Jev-powered page clutter removal with reusable template rules. ★ 136 · endpoint · TypeScript
  • trungdq88/youtube-sponsor-detection — Detect youtube sponsor segment with live audio and transcript powered by Jev ★ 74 · endpoint · JavaScript
  • ChetasLua/jevmeter — Put a live Jev (TypeSafe) meter on any video: every sentence scored, rendered as a 16:9 edit ★ 70 · endpoint · Python
  • mrnugget/jev-shell-history — Fish-style zsh history autosuggestions ranked by Jev (TypeSafe) ★ 67 · sdk · TypeScript
  • realZachi/typesafe-adblock — 🧹 Fun project: a Chrome extension that asks a tiny AI decision model (TypeSafe Jev) "is this DOM element an ad?" and pops it off the page. BYOK, no backend, not a real ad blocker. ★ 58 · endpoint · JavaScript
  • w3cj/jev-chat — A tool calling chat bot built with Jev and no LLM. ★ 46 · sdk · TypeScript
  • RafalWilinski/vibecheck — Chrome extension: vibe-check your X posts with TypeSafe's Jev before you hit Post ★ 42 · endpoint · JavaScript
  • brainstormity/Jev-X-Sentiment-Analysis — An on-demand crypto market intelligence and decision-support terminal powered by TypeSafe AI's System One model (Jev). ★ 41 · sdk · Python

→ 96 more in categories/app.md

Games, robotics & simulation

Games, simulations, drones and other embodied or playful demos. · 49 entries

  • fhshaik/typesafe-mario — A TypeSafe/Jev agent that plays Super Mario Bros. from structured emulator state. ★ 287 · sdk · Python
  • rokbenko/quackd — One CLI for all your robots. Connect them, command them, and let them work together, each with an LLM for a brain, Jev for cheaper steps. Microduck, Open Duck Mini, LeRobot, XLeRobot, AlohaMini, ToddlerBot or any ROS base. Claude, OpenAI, Gemini, Grok, or local via Ollama or vLLM. Simulator, .duck safety contracts, MCP, memory between runs, flocks. ★ 214 · sdk · Python
  • standardagents/jevpilot — A playable Three.js driving simulator with Jev-powered autopilot ★ 99 · endpoint · JavaScript
  • RomanSlack/jev-drone — Camera-only autonomous drone in MuJoCo with a small judgment model (TypeSafe Jev) in the loop at 2.5Hz ★ 79 · sdk · Python
  • sorrycc/typesafe-snake — Snake auto-played by TypeSafe's Jev model: one System One choice per tick, legal moves and facts generated in code ★ 18 · sdk · TypeScript
  • phyous/tsai-sc — TypeSafe Jev controls original StarCraft shareware through keyboard and mouse with recorded action probabilities. ★ 17 · endpoint · Python
  • Dimweaker/jev-libero — Fine-grained robot control with Jev, physics previews, and configurable LIBERO tasks. ★ 10 · endpoint · Python
  • kavehmz/typesafe-playground — Interactive experiments with TypeSafe Jev, from support routing to 3D driving simulations with real AI decisions and visible sensor inputs. ★ 10 · endpoint · JavaScript
  • vinilana/live-jev — 2D autonomous car simulation in the browser, driven by TypeSafe's Jev decision model ★ 10 · sdk · JavaScript
  • AbdelStark/heist-one — Observable browser stealth game: Jev makes typed guard judgments while deterministic code owns the world. ★ 6 · sdk · TypeScript

→ 39 more in categories/game.md

Open reproductions

Open-source clones, alternative System One models and local runtimes. · 35 entries

  • TheoLeeCJ/SemIf — Semantic ifs from open models, on a 3090 at home. Independent; not affiliated with Jev or TypeSafe. ★ 2.0k · primitives · Python
  • TianyuCodings/NanoJev — A nano replica of Jev: parallel decisions, dynamic candidates, and an end-to-end training pipeline. ★ 1.1k · ai_sdk · Python
  • jaredpalmer/kev — tiny Jev-like model built on top of Qwen2.5-0.5B you can train and run on your MacBook ★ 617 · endpoint · Python
  • razorback16/openjev — Open, Jev-compatible System One decision server on DiffusionGemma ★ 146 · endpoint · Python
  • logan-markewich/jeff — A self-hosted drop-in replacement for TypeSafe's jev, powered by GliFormer. ★ 125 · endpoint · Python
  • Heman10x-NGU/openJev-verdict-2.0 — Calibrated 151M Non-Autoregressive Decision Engine beating TypeSafe Jev & Laya on LocalLLaMA/typed-decisions (77.10% acc, 0.0636 Brier, 0.0144 ECE) ★ 80 · model_id · Python
  • kshetrajna12/reflex — A small open decision model: state + typed questions -> calibrated probabilities. A Jev / System One re-creation on Qwen3.5. ★ 80 · endpoint · Python
  • daseinlabs/open-jev — Open Jev implementation with custom finetuning ★ 58 · endpoint · Python
  • wfzyx/von — The open-source System One decision model. Sub-15ms, non-autoregressive, local drop-in alternative to TypeSafe Jev. ★ 48 · endpoint · Python
  • deepanwadhwa/OpenDecision — OpenDecision is an open-source semantic decision engine like typesafe's jev. ★ 29 · endpoint · Python

→ 25 more in categories/repro.md

Benchmarks & research

Evaluations, calibration studies and test harnesses that measure behaviour. · 83 entries

  • hr98w/jev-visual — An educational Jev-like visual inference experiment on Apple Silicon: shared context, direct candidate scoring, and local visual demos. ★ 152 · primitives · Python
  • myc0576/SmartMoney-Cub — Read-only trading journal and review harness: Jev typed judgments, agent integration, and a reproducible finance benchmark. No orders, no advice. ★ 24 · endpoint · Python
  • Zaious/jev-capability-atlas — Independent, evidence-based map of when TypeSafe's Jev actually holds up vs. breaks down — real API-call receipts, not a leaderboard. 中文為主的雙語 repo。 ★ 19 · sdk · Python
  • JoshuaSP/open-jev — Typed JSON inference with DiffusionGemma, with Every and Jev benchmark results ★ 17 · model_id · Python
  • mizchi/jev-playground — TypeSafe AI の System One モデル Jev を MoonBit から触るためのプレイグラウンド。 ★ 17 · endpoint · TypeScript
  • goodrahstar/jev-column-race — Jev vs Gemini 3.8 Flash: labelling 1,000 app reviews, 4.1× faster and 7× cheaper ★ 16 · endpoint · JavaScript
  • danielgshea/jev-as-a-judge — Using Jev as an evaluator. ★ 12 · model_id · Python
  • AbdelStark/jev-benchmarks — Probability-aware evaluation for typed decision models: calibration, selective risk, latency, and reproducible benchmarks. ★ 11 · sdk · Python
  • NanmiCoder/jev-arena — Jev 模型介绍与实测：通过 Choice / Score / Noul 将自然语言转为带类型的判断与概率，用于分类、评分和路由；支持与 DeepSeek 等模型对比评论打标、速度与结果，含 CSV/Excel 导入、原速回放与离线报告。 ★ 9 · model_id · JavaScript
  • abhixhek/jevcal — Stop guessing confidence thresholds: calibrate, threshold, and drift-check typed decision models (TypeSafe Jev) against an LLM teacher. ★ 8 · endpoint · Python

→ 73 more in categories/research.md

Lists & resources

Other curated lists, directories, guides and cookbooks. · 56 entries

  • Anil-matcha/awesome-jev-by-typesafe — Evidence-backed use cases, patterns, prompts, and starter code for TypeSafe Jev — a System One model for fast, typed, confidence-aware decisions in software. ★ 643 · sdk · Python
  • yibie/awesome-jev — A curated list of public projects, integrations, and discussions built on Jev — TypeSafe AI's System One model for typed decisions. ★ 420 · resource · Python
  • AbdelStark/awesome-typesafe — A curated list of official resources and community projects for TypeSafe, System One models, and Jev. ★ 352 · resource · CSS
  • v-modal/awesome-jev-tools — A curated list of tools built for Jev — TypeSafe AI's System One model for typed decisions. ★ 292 · resource
  • cobanov/awesome-jev — A curated, source-backed list of projects built with Jev, TypeSafe AI's System One model for typed decisions. ★ 221 · resource
  • fatwang2/awesome-jev — A source-backed Jev project directory with a reusable Jev-only GitHub review workflow. ★ 166 · endpoint · JavaScript
  • logicrw/awesome-jev-projects — Awesome Jev: source-backed open-source ecosystem radar, plain-language project discovery, and automatic GitHub sync ★ 135 · endpoint · JavaScript
  • AnotiaWang/awesome-jev — A curated list of awesome Jev / TypeSafe System One applications, libraries, and resources. ★ 87 · resource
  • hellogumbo/awesome-jev — A community directory of projects built on Jev, TypeSafe AI's System One model. ★ 77 · endpoint · JavaScript
  • OmniJev/awesome-jev — 🔥🔥 Papers, open reproductions and independent evaluations behind System One models and Jev. ★ 71 · resource · JavaScript

→ 46 more in categories/list.md

Contributing

Additions, corrections and removals are all welcome — see CONTRIBUTING.md. Because entries are generated, a pull request should change data/ or the pipeline, not the README. If we listed a project that does not belong here, open an issue and it will be removed.

License

CC0 1.0 — public domain. The pipeline under scripts/ is MIT. This is an independent community project, not affiliated with or endorsed by TypeSafe AI.

About

A curated list of open-source Jev projects where every entry links to the line of code that calls Jev, and every performance number comes from an independent 2,390-question benchmark.

Resources

Readme

Contributing

Contributing
Activity

Stars

6 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages