Skillget

awesome-jev-use-cases

mcp
Security Audit
Warn
Health Pass
  • License — License: CC0-1.0
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Community trust — 10 GitHub stars
Code Warn
  • Code scan incomplete — No supported source files were scanned during light audit
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

Awesome list of TypeSafe AI Jev use cases: 74 demos ranked by likes, 150+ GitHub repos, limits, cost and API examples. CC0, sponsored by AY Automate.

README.md

Awesome Jev use cases

Awesome PRs welcome License: CC0 1.0 Sponsored by AY Automate Open source

Awesome Jev use cases: TypeSafe AI Jev demos, repos, limits and examples

A list of things built with Jev, TypeSafe's model for typed decisions, with the numbers behind them: who posted each demo, how many followers they have, how many likes it got, and what the limits of the model are. This list is open source (CC0), free to copy and reuse, and sponsored by AY Automate. It is unofficial and is not affiliated with TypeSafe.

Every entry links to the original post or repository. Ideas that nobody has shipped are in their own section and marked as ideas.

Sponsored by AY Automate

Top 30 popular demos

The 30 most-liked demos, ranked. Click a card to open the original post. Each card shows the demo's rank, area, author, likes, reposts, and reach (likes divided by the author's followers). Preview frames are low-resolution stills from the builders' own videos and belong to them. If an author wants one removed, open an issue. Full metrics for every demo are in docs.

#1 Instant compaction for Claude #2 Flight search with Browser Use #3 Real-time slop detector as you scroll #4 724 competitor ads, broken down #5 Voice-controlled computer use on a Mac #6 jev-trader #7 Jev plays Doom #8 A canvas you control by pointing and speaking #9 Jev plays Subway Surfers #10 A real-time ad blocker #11 500 emails for 3.5 cents #12 Jev plays Smash Bros. against itself #13 Triage across 1,500 emails #14 700 leads scored in 40 seconds #15 Jev plays Super Mario Bros. #16 jev() for PostgreSQL #17 Game levels generated in real time #18 Keystroke oracle #19 jevlike #20 1kpapers #21 A model router on Jev #22 Computer use without screenshots #23 Every’s editorial vibe check #24 400 companies matched to one candidate #25 $10,000 in Jev’s hands #26 An always-on assistant with no wake word #27 Post scoring with SuperX #28 Doomscroll Filter #29 Predictive spreadsheets #30 The X algorithm, rebuilt with Jev

Contents

FAQ

Short answers to the questions people ask most, each with a source.

What is Jev?
Jev is a model from TypeSafe AI that answers typed questions instead of writing text. Each question is a Choice, a Score or a Noul (yes or no with a probability), and the answer comes back as a number or a pick with a confidence. It does not generate text. See What Jev is.

Is Jev the same as the "Jev" that searches show for Jevons paradox or Deltarune?
No. The word has other meanings. Search for "TypeSafe Jev" or "Jev AI model".

How do I call the Jev API?
Send a POST to https://api.typesafe.ai/v1/systemone with a bearer key. A full curl example is in docs/api-quickstart.md.

How much does Jev cost?
TypeSafe lists $42 per billion input tokens, and output tokens are free. Vercel says Jev is free on AI Gateway until Sept 25. See Reported cost and latency.

What can I build with it?
Routers, classifiers, judges, guardrails, triage and game agents. The Top 30 demos and Browse by area show real examples.

What are the limits of Jev?
It reads literally, is weak at math, counting and dates, and accuracy drops with irrelevant state. See Limits of Jev 1.13.

Is Jev better than an LLM?
It is a different tool. Use Jev for fast typed decisions and an LLM for writing. Many demos pair them.

Which open-source Jev projects exist?
More than 150 repositories. See Open source and Long tail.

Is this list official?
No. It is unofficial, open source under CC0, and sponsored by AY Automate.

Start here

Short on time? Read in this order.

  1. Most-liked demos shows what people actually built.
  2. Limits of Jev 1.13 tells you what it cannot do, so you do not build the wrong thing.
  3. Patterns shows how the working demos are put together.
  4. Reported cost and latency has the price and speed numbers.
  5. Contributing shows how to add yours.

Browse by area

Every tracked demo, grouped by what it does and ranked by likes inside each group.

Area Demos Top demo Likes
Content and growth 18 Real-time slop detector as you scroll 7,180
Apps and tools 17 Instant compaction for Claude 10,435
Agents and computer use 14 Flight search with Browser Use 8,723
Triage and routing 9 500 emails for 3.5 cents 3,853
Games and real time 7 Jev plays Doom 4,890
Research and data 7 jevlike 2,018
Trading and markets 2 jev-trader 4,913

Content and growth

Apps and tools

Agents and computer use

Triage and routing

Games and real time

Research and data

Trading and markets

Try Jev free right now

Vercel Developers announced on 2026-09-19 that Jev is free on Vercel AI Gateway until Sept 25. The model page is vercel.com/ai-gateway/models/jev. Outside that window, the price from TypeSafe is $42 per billion input tokens, and prices can change, so check before you build on it.

Vercel's earlier post says Jev was adopted faster than any other model in AI Gateway history. It reports about 13% of teams in the first day, 2x the GPT-5.6 family and 6x Fable 5.1. Those are Vercel's figures and I have not verified them. The free-window post had 2,043 likes and 552,601 views when I read it.

To use anything in this list, copy the pattern, get a key, and start with one question type. The Patterns section shows how.

Numbers at a glance

Number Value
Demo posts tracked 74
Total likes 127,162
Total reposts 6,998
Total replies 5,742
Median likes per demo 1,045
Demos over 1,000 likes 38
Median author followers 12,769
Authors under 1,000 followers 9
Open-source repos in the main list 37
Combined stars in the main list 21,456
Most common repo languages TypeScript (14), Python (12), JavaScript (6), Shell (1)
Most common licenses MIT (27), none listed (6), Apache-2.0 (3)
Busiest demo area Content and growth (18 demos)

Demos by area: Content and growth 18, Apps and tools 17, Agents and computer use 14, Triage and routing 9, Games and real time 7, Research and data 7, Trading and markets 2.

The first week in numbers

Snapshot of 2026-09-19.

  • 74 demo posts with video, published between 2026-09-15 and 2026-09-19, with 127,162 likes combined.
  • 102 builder accounts checked. The median has 6,615 followers. 47 have under 5,000 and 26 have under 1,000.
  • 37 open-source repositories below, with 21,456 GitHub stars combined. Each one mentions Jev or TypeSafe in its own README.
  • The four most-liked demos are a Claude Code plugin, a browser agent, an ad teardown and a Mac voice assistant. None of them generates text with Jev.

Most-liked demos

Bar chart of the most-liked Jev demos

The 15 demo posts with the most likes, with the follower count of whoever posted them. Numbers are a snapshot of 2026-09-19.

Demo By Followers Likes Reposts
Instant compaction for Claude @tamarajtran 12,739 10,435 631
Flight search with Browser Use @gregpr07 30,060 8,723 617
Real-time slop detector as you scroll @RBilgil 685 7,180 210
724 competitor ads, broken down @TheMattBerman 12,799 6,348 389
Voice-controlled computer use on a Mac @instantricecook 1,015 5,016 252
jev-trader @jarrodwatts 32,542 4,913 216
Jev plays Doom @CompleteSkeptic 122,369 4,890 240
A canvas you control by pointing and speaking @jackcheng 11,724 4,797 254
Jev plays Subway Surfers @_MaxBlade 22,962 3,956 253
A real-time ad blocker @iam_zachi 4,832 3,872 139
500 emails for 3.5 cents @rileybrown 244,870 3,853 96
Jev plays Smash Bros. against itself @maubaron 19,783 3,620 334
Triage across 1,500 emails @ryanvogel 18,403 3,538 106
700 leads scored in 40 seconds @romanbuildsaas 20,336 3,138 203
Jev plays Super Mario Bros. @faadilhshaik 192 2,860 248

Small accounts, big results

Likes per follower for Jev demos

Likes divided by followers, for demos with at least 1,000 likes. A high ratio means the post traveled far beyond the author's own audience.

Demo By Followers Likes Likes per follower
Jev plays Super Mario Bros. @faadilhshaik 192 2,860 14.9x
Real-time slop detector as you scroll @RBilgil 685 7,180 10.5x
Voice-controlled computer use on a Mac @instantricecook 1,015 5,016 4.9x
jevlike @vinnylarouge 1,392 2,018 1.4x
Game levels generated in real time @HugoDuprez 3,151 2,614 0.8x
Instant compaction for Claude @tamarajtran 12,739 10,435 0.8x

More demos by area

The next tier by likes, grouped by what the demo does.

Agents and computer use

Games and real time

Triage and routing

Trading and markets

Content and growth

Research and data

Apps and tools

Open source

Repositories with a working project and a README that mentions Jev or TypeSafe. Stars are a snapshot of 2026-09-19.

Browser and computer use

  • browser-use/jev-ultrafast. Browser agent that picks an operation and element per step, with a small LLM writing text only when typing is needed. 7,798 stars, Python, MIT.
  • awlevin/typesafe-computer-use. Drives a Mac toward a plain-English goal for about $0.0002 a step. It reads the screen with OCR and sends no screenshots. 456 stars, Python, MIT.
  • wy-coliney/jev-browser-use. Browser skill where Jev handles navigation and clicks while Codex keeps text input and final checks. Built at EZCollegeApp. 176 stars, JavaScript, MIT.
  • jkudish/jev-browser. Runs a real headless browser through an MCP server, a CLI or a library. It picks one action per step from the clickable elements on the page. 135 stars, TypeScript, MIT.
  • moritzkremb/jev-voice-browser. Node app controlling a headed Chromium window by voice. Jev picks intent and target on each partial transcript and Playwright acts. 110 stars, JavaScript, MIT.
  • droidrun/mobile-jev. Navigates a live Mobilerun phone with Jev. The demo opens Uber and enters a route in about 21 seconds for 9 actions. 209 stars, JavaScript, MIT.
  • realZachi/typesafe-adblock. Chrome extension that asks Jev whether a DOM element is an ad and removes it. A side project that needs your own key. 53 stars, JavaScript, MIT.
  • kitze/unclutter. WXT browser extension that removes page clutter using Jev, with reusable template rules. 129 stars, TypeScript, MIT.

Coding agents and developer tools

  • tamaratran/fast-jev-compaction. Claude Code plugin that scores each tool call and result at compaction, drops or truncates stale ones, and keeps the rest verbatim. 4,031 stars, TypeScript, MIT.
  • tamaratran/jev-pruner. Claude Code plugin that trims noisy Bash output with Jev after a command runs and before the main model sees the result. 80 stars, TypeScript, MIT.
  • devagrawal09/jev-review. Staged code-review workflow and local dashboard that reviews a Git diff or a whole codebase using focused Jev calls. 326 stars, TypeScript, MIT.
  • thruwire/foreman. Places Jev above slower coding agents to judge whether a Codex worker's implementation is complete for a ticket, spec or bug report. 359 stars, Python, MIT.
  • kitze/skillbox. Self-hosted, versioned skills library for AI agents with MCP, scoped clients and optional Jev recommendations. 188 stars, TypeScript, MIT.
  • DevMortimer/pi-warden. Guardrails for Pi that feed problems back to the agent. It judges irreversible or off-task tool calls, stuck loops and unverified done claims. 90 stars, TypeScript, MIT.
  • vinilana/jev-eval-agent. Compares how many steps an agent with 100 mocked tools needs to finish a task when the LLM picks tools versus Jev. 89 stars, HTML.
  • mrnugget/jev-shell-history. Zsh autosuggestions that have Jev rank your last 100 distinct history entries and show the best match in grey with its score. 63 stars, TypeScript.
  • vercel-labs/ai-cli. Terminal tool that generates text and media through AI Gateway and can also evaluate typed questions. 805 stars, TypeScript.
  • RafalWilinski/vibecheck. Chrome extension that rates a draft X post for virality and clarity before you publish. 41 stars, JavaScript.

Routing

  • gargpratyush/jev-router. Per-turn router for Claude Code and OpenAI Codex. Simple work goes to a fast tier and hard work to a strong tier. 191 stars, JavaScript, MIT.
  • BillionsBobby/JevRouter. Router that picks among models, subagents and tools from one candidate set. Jev answers a Choice question. Code enforces permissions. 81 stars, TypeScript, MIT.

Data and search

  • realZachi/pg-jev. PostgreSQL extension for asking your tables questions in plain language. 204 stars, Shell, NOASSERTION.
  • giuliosmall/pg_typesafe. Pre-alpha PostgreSQL extension that calls Jev from SQL for categorical classification. Tested on PostgreSQL 16 and 17. 77 stars, C, MIT.
  • superagents-lab/jev-search. Web search where Jev picks sources, time ranges and search terms, then ranks Search1API results with visible scores. No generated answers. 199 stars, TypeScript, MIT.
  • pithings/advocaat. Small type-safe client for asking questions about your data and getting typed answers in one request. 84 stars, TypeScript, MIT.

Games, trading and hardware

  • fhshaik/typesafe-mario. Experimental controller where Jev picks NES inputs for Super Mario Bros. from compact emulator telemetry instead of screenshots. 278 stars, Python.
  • vinnylarouge/jevlike. Trains a small model that takes text plus N options and returns one probability per option in a single pass. 961 stars, Python, MIT.
  • hr98w/jev-visual. Educational Apple Silicon project using Qwen3.5-0.8B with MLX to answer several questions about one image via browser UI, CLI or HTTP API. 138 stars, Python, MIT.
  • jarrodwatts/jev-trader. Trading bot that asks Jev for a buy or sell decision each Monad block on the Kuru MON-USDC order book and posts a post-only limit order. 1,184 stars, TypeScript, MIT.
  • RomanSlack/jev-drone. Simulated quadrotor flying a five-station MuJoCo obstacle course by camera alone, with Jev judging the situation at about 2.5 Hz. 71 stars, Python, MIT.

Open models and alternatives

  • TheoLeeCJ/SemIf. Semantic if-statements from open models on a home 3090 GPU, with a browser demo. Independent, not affiliated with TypeSafe. 1,839 stars, Python, MIT.
  • jaredpalmer/kev. LoRA adapter and small readout head on a Qwen base that answers many typed questions about a document in one prefill pass. 423 stars, Python, Apache-2.0.
  • razorback16/openjev. Open-source decision server that answers typed questions with probabilities and a confidence in tens of milliseconds. 113 stars, Python, Apache-2.0.
  • logan-markewich/jeff. Self-hosted stand-in for the Jev API on a 400M-parameter GLiFormer, usable with the official SDK. Less accurate on reasoning-heavy tasks. 87 stars, Python, MIT.
  • Mapika/decider. Language model fine-tuned from Qwen3.5-2B that returns calibrated probabilities for typed questions in one forward pass and generates no text. 86 stars, Python, Apache-2.0.

MCP, skills and clients

  • itsmostafa/typesafe-mcp. MCP server that lets coding agents such as Claude Code and Codex call Jev and get probabilities to branch on. 97 stars, Go, MIT.
  • jkudish/jev-mcp. MCP server giving agents ten Jev judgment tools, such as verifying claims against evidence and screening content before it enters context. 92 stars, TypeScript, MIT.
  • dbreunig/building-with-jev-skill. Agent skill for writing and improving Jev programs. It teaches question design and how to diagnose wrong answers. 113 stars.

More projects

Found by searching GitHub for the model and company names on 2026-09-19. Stars are a snapshot. Each description comes from the repository's own summary and I did not open every README.

  • y0usaf/pi-jev. TypeSafe Jev as a decision layer for the Pi coding agent: a measured tool-call gate plus jev_ask for typed, calibrated answers. 76 stars, TypeScript, MIT License.
  • 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. 61 stars, JavaScript.
  • Dicklesworthstone/skillranker. Rust CLI powered by Jev from TypeSafe.ai that ranks agent skills for the next step using live session context. Includes Claude Code hooks, structured JSON, abstention, and local feedback. Requires a TypeSafe API key. 58 stars, Rust.
  • obie/ruby_decision_model. Ruby client for decision models such as Typesafe Jev. 44 stars, Ruby, MIT License.
  • devagrawal09/stanley-code. Bounded TypeSafe Jev workflows for coding agents. 28 stars, TypeScript, MIT License.
  • wfzyx/von. The open-source System One decision model. Sub-15ms, non-autoregressive, local drop-in alternative to TypeSafe Jev. 25 stars, Python, Apache License 2.0.
  • tacticocc/Jevbridge. ACP and MCP adapter that bridges TypeSafe Jev with any LLM , computer use and typed decisions alongside Codex, Claude, Grok, and OpenCode. 23 stars, TypeScript, MIT License.
  • AboveColin/HA-Jev. Ask your house a question, get a number back. Home Assistant integration for TypeSafe Jev: typed answers as sensors, four actions for automations, and a conversation agent for Assist. 21 stars, Python, MIT License.
  • peternguyen777/ai-icon-generator. WoofAI is an AI-powered micro-SaaS application allowing users to generate, share and download pet icon art. The project is built using the T3 stack (Typescript, TRPC, Tailwind), allowing end-to-end typesafe APIs and fast front-end prototyping. Under the hood, the project leverages OpenAI’s DALL-E2 and images hosted on Amazon S3. 20 stars, TypeScript.
  • TheoOliveira/pi-jev. Semantic tool routing and typed System One decisions for the Pi coding agent using TypeSafe Jev. 18 stars, TypeScript, MIT License.
  • dannote/jev. TypeSafe Jev for OTP: reply to Jev from a GenServer and pattern match on its answer. 16 stars, Elixir, MIT License.
  • phyous/tsai-sc. TypeSafe Jev controls original StarCraft shareware through keyboard and mouse with recorded action probabilities. 15 stars, Python, MIT License.
  • zhihz/openjev. Local bilingual probability decisions from context, questions, and candidate answers. Independent research preview inspired by TypeSafe Jev. 13 stars, 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 stars, JavaScript.
  • chopratejas/invalidate. The invalidation layer for AI memory. Every fact gets a lease; new evidence ends it. Built on TypeSafe Jev. 10 stars, Python, Apache License 2.0.
  • blakestone-x/jev-mcp. MCP server for TypeSafe Jev: typed classify, score, check, match and screen for any agent, with confidence on every answer. 10 stars, Python, MIT License.
  • AkashPriyadarshii/jev-seo. 100% free ₹0 agent-first SEO & GEO CLI suite and MCP server in Rust replacing Semrush and OpenSEO via DuckDuckGo and TypeSafe Jev System One. 10 stars, Rust.
  • yikangy873-gif/jev-desktop. TypeSafe Jev action selection inside Codex Computer Use. 9 stars, JavaScript, MIT License.
  • HyunjunJeon/pi-quiet-ask. TypeSafe Jev as the pi coding agent's quiet decision layer. 8 stars, TypeScript, MIT License.
  • 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 stars, Python, MIT License.
  • khordoo/jev-reflex-autonomy-lab. Multi-drone autonomy lab demonstrating TypeSafe Jev reflex decisions with optional System 2 strategy guidance. 8 stars, TypeScript.
  • GodsBoy/jev-agent-skill-router. Typed, confidence-aware agent skill routing with TypeSafe Jev. 7 stars, Python, MIT License.
  • matthewp/flue-jev-demo. Flue agent routing with TypeSafe Jev through Cloudflare AI Gateway. 7 stars, TypeScript.
  • abhixhek/jevcal. Stop guessing confidence thresholds: calibrate, threshold, and drift-check typed decision models (TypeSafe Jev) against an LLM teacher. 7 stars, Python, MIT License.
  • DECRUX9812/typesafe-skill-router. TypeSafe (Jev) skill routing for Hermes Agent: names the one skill worth loading, before the model call. Opt-in, stdlib only, ~$0.001 per routed turn. 7 stars, Python, MIT License.
  • keeltrace/hermes-jev. Typed System One decisions, ranking, verification, and an opt-in Hermes tool gate using TypeSafe Jev. 6 stars, Python, MIT License.
  • tumf/jev-cli. Small dependency-free CLI for TypeSafe Jev. 5 stars, Python, MIT License.
  • bohutang/sift. Chrome extension that labels every post on X (Substance · Humor · Chit-chat · Promo · Junk · AI-written) with TypeSafe Jev, and hides the ones you don't want. 5 stars, JavaScript, MIT License.
  • joelhooks/pi-fast-jev-compaction. Pi extension: verbatim context compaction with TypeSafe Jev decisions. 5 stars, TypeScript, MIT License.
  • Kevthetech143/super-jev. A small, extensible decision-to-action harness for TypeSafe Jev. 5 stars, Python, MIT License.
  • 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). 5 stars, Python.
  • altryne/jevify. An agent skill to discover TypeSafe Jev opportunities, design typed questions, and learn from recent community experiments. 5 stars, Python, MIT License.
  • raihankhan-rk/diffjury. DiffJury , TypeSafe Jev PR risk router + code review coach. 4 stars, TypeScript.
  • AkashPriyadarshii/jev-curate. High-throughput synthetic & pretraining dataset sifter powered by TypeSafe AI Jev (api.typesafe.ai). Stream, filter, and score Parquet & JSONL datasets at 1,500+ rows/sec using System One typed decisions (Choice, Score, Noul). 3 stars, Rust, MIT License.
  • xingwudao/OpenJev. OpenJev: an independent Jev-inspired System One decision API based on TypeSafe.ai concepts. Choice, score and noul primitives, local mock server, Python and TypeScript SDKs. Real inference planned; not affiliated with TypeSafe AI. 2 stars, Python.
  • FFatTiger/new-api-plugin-typesafe. TypeSafe AI System One (Jev) task plugin for QuantumNous/new-api , native /v1/systemone, synchronous evaluation, token billing. 2 stars, JavaScript, Apache License 2.0.
  • simota/tenbin. MCP server and agent skill for the TypeSafe AI System One API (Jev): decompose a judgment into Choice / Score / Noul questions, lint them, measure on labelled data, and put calibrated thresholds in code. 2 stars, TypeScript, MIT License.
  • chrishan17/claude-jev-mod. Typed decisions in Claude Code: adds $.jev over TypeSafe's Jev, through OpenRouter, Vercel AI Gateway, Cloudflare Workers AI, LiteLLM or the TypeSafe API. 1 stars, TypeScript, MIT License.
  • YIZY-API/yizy-web-app. Document and Build Typesafe JSON APIs with Instant Code Generation. Boost Development Productivity with AI. 1 stars, Svelte, GNU General Public License v3.0.
  • JacquesGariepy/ORIGIN-CIVILIZATION. AI life-and-civilization simulation: TypeSafe Jev makes every decision (typed, probabilistic, auditable); LLMs plan , OpenAI-compatible APIs, local models (Ollama, LM Studio), Claude Code, Codex. 1 stars, HTML.
  • everyinfra/jev-radar. 📡 全网最全 · The world's most full tracker of the Jev (TypeSafe AI System One) ecosystem , 220+ documented cases · 108 confidence-graded entries · verified & rescanned every 3 hours · API access guide included. 1 stars.
  • oceanByte/tsai-cli. Unofficial CLI for the TypeSafe AI System One API. 0 stars, TypeScript, MIT License.
  • jujumilk3/jev-calibration-audit. Independent API-only calibration audit of TypeSafe AI's Jev decision model. 0 stars, Python, MIT License.
  • twilwa/pi-typesafe. Pi coding-agent extension built on the TypeSafe AI System One API (Jev). 0 stars, TypeScript.
  • Olli0103/openclaw-typesafe-ai. Optional typed TypeSafe AI Jev decisions for OpenClaw, with SecretRef credentials and strict API validation. 0 stars, TypeScript, MIT License.

SDKs and clients

Unofficial libraries for calling the TypeSafe API from other languages.

  • burnigtm/jev-mcp. MCP server that puts TypeSafe Jev on the coding loop in Cursor, Codex, and any MCP client. 9 stars, TypeScript, MIT License.
  • Tangerg/typesafe-sdk-go. Go SDK for the TypeSafe AI API , typed questions in, probability distributions out. 8 stars, Go, MIT License.
  • saibimajdi/typesafeai-dotnet-sdk. Community .NET SDK for the TypeSafe AI System One API , typed noul, choice, and score questions with structured, confidence-scored answers. Not affiliated with TypeSafe AI. 5 stars, C#, MIT License.
  • gilljon/typesafe-ai-rs. Independent async and blocking Rust SDK for the TypeSafe AI System One API. 3 stars, Rust, MIT License.
  • codeitlikemiley/typesafe-sdk-rust. Rust SDK for the TypeSafe AI API. 2 stars, Rust, MIT License.
  • fgn/jevgo. Go client for TypeSafe AI's System One API (Jev), with optional Langfuse instrumentation. 2 stars, Go, MIT License.
  • chez-shanpu/typesafeai-go. Go SDK for TypeSafe AI API https://docs.typesafe.ai/api. 1 stars, Go, Apache License 2.0.
  • zhirschtritt/typesafe-go. Idiomatic Go SDK for the TypeSafe AI API. 1 stars, Go, MIT License.
  • abeldzan/jev-rs. Async-first Rust SDK for the TypeSafe AI API. 1 stars, Rust, MIT License.
  • yunusey/typesafe-sdk-cpp. Unofficial C++23 client for the TypeSafe AI API. 1 stars, C++, MIT License.
  • guillemus/jev-go. Unofficial Go SDK for TypeSafe AI's Jev API. 1 stars, Go.
  • lu-zero/systemone. Rust client for the TypeSafe AI systemone API. 0 stars, Rust, MIT License.
  • nirgal-soft/typesafe-rs. A rust client for the TypeSafe AI API. 0 stars.
  • mattneel/typesafe.zig. An idiomatic Zig client for the TypeSafe AI API. 0 stars, Zig, MIT License.
  • mattneel/typesafe. An idiomatic Elixir client for the TypeSafe AI API. 0 stars, Elixir, MIT License.
  • hnegishi/typesafe-ai-ruby. Ruby client for the TypeSafe AI(Jev) System One API. 0 stars, Ruby, MIT License.
  • community-ports/typesafeai-sdk-rust-community. Community-built Rust SDK for the TypeSafe AI API (System One / Jev). A port of typesafe-sdk-python. 0 stars, Rust, MIT License.
  • Shubham510/typesafe-go. Unofficial Go SDK for TypeSafe AI's System One API (Jev). 0 stars, Go, MIT License.
  • T-moz/typesafe-ai-dart. A pure Dart sdk wrapper around Typesafe AI API and JEV. 0 stars.
  • valksor/typesafe-sdk-go. Unofficial Go SDK for the TypeSafe AI System One API , 1:1 parity with the official JS and Python SDKs. Not affiliated with TypeSafe AI. 0 stars, Go, MIT License.
  • valksor/typesafe-sdk-php. Unofficial PHP SDK for the TypeSafe AI System One API , 1:1 parity with the official JS and Python SDKs. Not affiliated with TypeSafe AI. 0 stars, PHP, MIT License.

Long tail

A second GitHub sweep on 2026-09-19 found 111 more repositories that mention Jev or TypeSafe in their description. Those with at least one star are below, grouped by what they do. All 111 are in data/more-repos.csv. Descriptions come from each repository's own summary and I did not open every README.

Routers

  • yusukebe/hono-jev-router. Route HTTP requests by meaning. A semantic router for Hono powered by Jev. 34 stars.
  • mejiasd3v/pi-jev-router. Automatic model routing for Pi using TypeSafe's Jev through Vercel AI Gateway. 6 stars.
  • andrelandgraf/safer-with-jev. Neon Function proxy for the Neon AI Gateway with TypeSafe Jev routing. 4 stars.
  • nexibeo/jev-cookbook. Practical, tested recipes for TypeSafe's Jev decision model on OpenRouter: support triage, database indexing, file organizing, tagging, taxonomies, dedupe, PII detection, extraction, search re-ranking and a browser agent. 3 stars.
  • hamakyo/jev-starter. Typed, policy-driven decision workflows on top of TypeSafe AI Jev: confidence routing, fallbacks, evaluation, and RAG patterns for TypeScript apps. 2 stars.
  • rajdhakad9826/jev-router. Cost-aware LLM router that picks the cheapest model capable of handling a query, using TypeSafe's Jev for fast classification instead of an LLM call. 2 stars.
  • ajensenwaud/hermes-jev-plugin. TypeSafe Jev (System One) decision tools for Hermes Agent: jev_check / jev_route / jev_score / jev_evaluate. 2 stars.
  • iJ03l/jear. Jev-routed client for NEAR AI Cloud inference and IronClaw agents. 2 stars.
  • adarshmishra07/jcm-router. Local proxy that picks the Claude model and effort per message using TypeSafe Jev. Routes subagents, leaves your cached main chat alone. 2 stars.
  • prismhq/jev-router. Open-source LLM router that uses TypeSafe's Jev to pick a model, on top of LiteLLM. 2 stars.
  • Mandrilsquad1441/jev-model-router. Pick the best AI model and reasoning effort for any task in ~1s. Plugin for Claude Code, Claude Desktop and Codex, powered by TypeSafe's Jev decision model and live OpenRouter pricing. Balance intelligence, speed and cost, or choose your priority. 1 stars.
  • TokenTrim/jev-routing-experiment. Benchmarking TypeSafe's Jev decision model as a cost-efficient LLM router on RouterArena. 1 stars.
  • rsdkrasen/hermes-jev-router. TypeSafe/Jev router plugin for Hermes Agent , compact tool results, suppress duplicate tools, skip unnecessary main-model calls. 1 stars.
  • TheEleventhAvatar/triage-bot. Real-time support triage + response bot Jev routes the ticket to a specialist agent (general / account / billing / technical) and decides whether a human should take it instead , all as typed data, no text to parse. Cerebras then drafts the reply using whichever agent Jev picked. The script times both calls separately so you can see the split. 1 stars.
  • vinilana/jev-gateway-bench. Benchmark for jev-gateway: real coding agents on chess engine tasks, with Jev routing on and off. 1 stars.

MCP servers and agent skills

  • arunav25/jev-mcp. Connect JEV to MCP clients and compare its judgments against general-purpose LLMs using shared datasets and measurable accuracy. 5 stars.
  • harshil1712/slidepilot. Voice-driven semantic auto-advance for Slidev, powered by Cloudflare Agents and TypeSafe AI Jev. 4 stars.
  • rashedInt32/jev-mcp. MCP server exposing TypeSafe Jev as typed, calibrated judgment tools: classify, score, check, batched ask. Ships as a Claude Code plugin. 3 stars.
  • samtay32/jev-system-architect. System-architecture skill for TypeSafe AI Jev/System One , find fuzzy semantic judgment and turn it into small Choice/Score/Noul primitives. 2 stars.
  • abhishekashokvkumar/jev-mcp-dispatcher. Natural-language MCP tool dispatcher powered entirely by TypeSafe's Jev , no general-purpose LLM. Discovers a simple MCP server's tool signatures at runtime and uses Jev's typed primitives (Choice/Noul) to pick the right tool and extract its arguments straight out of the sentence. 2 stars.
  • forvela/jev-agent-browser. Fast, bounded browser agents powered by Jev and agent-browser , typed actions, research, classification, and safe orchestration. 2 stars.
  • raihankhan-rk/jevarena. JevArena , two Jev agents duel in click-only browser games (Browser Use + TypeSafe Jev). 2 stars.
  • siddicky/omp-typesafe. TypeSafe AI (Jev) adversarial reviewer and typesafe_ask tool for the omp coding agent. 1 stars.
  • BYK/jev-mcp. An eval-first MCP server for TypeSafe's Jev, a System One model that returns typed judgments (noul, choice, score) with probabilities instead of generated text. 1 stars.
  • TokenTrim/jev-agent-failure-benchmark. Benchmarking Jev (Typesafe.ai) against a strong LLM on the Who&When Pro agent-failure-attribution benchmark (text subset). 1 stars.

SDKs and clients

  • Olti1947/jev-java. Idiomatic Java SDK for TypeSafe AI Jev System One decision engine. 3 stars.
  • Stumble/jev-go. Community Go SDK for TypeSafe AI Jev / System One. 2 stars.

Games and experiments

  • Heman10x-NGU/Verdict-open-jev. Non-autoregressive decision engine on ModernBERT (151M) with calibrated uncertainty (RLCD), TypeSafe AI Jev benchmark audit, and in-browser WebGPU playground. 22 stars.
  • vinilana/live-jev. 2D autonomous car simulation in the browser, driven by TypeSafe's Jev decision model. 10 stars.
  • Dimesio/typesafe-chess. FUn little experiment with Typesafe AI Jev Model playing chess against stockfish :). 2 stars.
  • sysadarsh/zerosweep. Autonomous System-One Triage Engine & Benchmark powered by TypeSafe AI (Jev). 75ms inference, $0 output tokens, and RLCD epistemic safety gates. 2 stars.
  • tedliou/decision-model-playground. A local playground for comparing Laya and Jev decision models with article recommendations. 1 stars.

Other

  • 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 stars.
  • gtaras7/typesafe-jev. Screen a folder of CVs with the TypeSafe Jev decision model: typed judgments, an editable policy, free re-scoring. 2 stars.
  • noetion/dsh-jev. DSH bundle that registers jev_ask for TypeSafe Jev noul, choice, and score answers. 2 stars.
  • thezem/jev-one. A vocabulary-driven TypeScript runtime for safe, stateful applications powered by TypeSafe AI Jev. 1 stars.
  • ItisShikhar/gg-friggin-ez. Fast, drop-in profanity and toxicity screener for Node.js, powered by TypeSafe AI Jev. Catches leetspeak, character spacing, and romanized profanity across languages including Kannada, Telugu, Tamil, Hindi, and Bengali. ~50-500ms latency. 1 stars.
  • AkashPriyadarshii/jev-scout. Zero-hallucination open-source repo and crate scout powered by TypeSafe AI Jev System One scoring. 1 stars.
  • AkashPriyadarshii/jev-git. Sub-second Git pre-commit & pre-push semantic reflex gate powered by TypeSafe AI Jev. 1 stars.
  • kitze/pagegrade. Grade page sections for clarity, writing and on-page SEO. WXT + TypeSafe AI Jev. 1 stars.

Other lists

  • sontakey/awesome-jev. Unofficial list of insanely useful TypeSafe AI Jev / System One projects. 1 stars.

Search demand

Deeper keyword research, live results-page checks, a forecast from X and YouTube data and a content plan are in docs/keyword-research.md. Views for 65 Jev YouTube videos are in data/youtube.csv. The full list of 112 keywords is in data/keywords.csv.

Monthly search volume for Jev related keywords

Google search volume in the US, from a keyword-data provider called through treg on 2026-09-19. These are monthly averages for the last 12 months.

Keyword Monthly searches Competition Note
jev 4,400 Low Ambiguous. The word has other meanings, so do not read this as demand for the model.
typesafe ai 320 Low Up about 1,080% year over year. It sat near 50 a month in late 2025 and reached 590 to 720 a month from March 2026.
typesafe 170 Low Also the name of a Scala type-safety idea, so mixed intent.
ai router 880 Low Fits the router projects in this list.
open router ai 2,900 Low Adjacent demand for model routing.
ai classifier 140 Low Fits classification demos.
llm classifier 40 Low Small.
content moderation ai 30 Low High cost per click of about $19, so advertisers value it.
jev model, jev api, awesome jev, system one model, slop detector No data None Too new or too small to report.

What this says: people who look for the company by name are few but growing fast, and nobody yet searches for the model as "Jev" in a way that isolates it. Terms that describe the job, such as router and classifier, carry the volume. A page that ranks for those and shows a Jev example has more room than a page fighting for "jev". Volume for the newest terms is blank because the provider has no data yet, not because it is zero.

Suggested repository topics: jev, typesafe, system-one, llm, classification, routing, awesome-list.

Full tables

Every entry with its source. Sorted by likes and by stars. Snapshot 2026-09-19.

All demos

# Demo Author Followers Likes Reposts Posted Source
1 Instant compaction for Claude @tamarajtran 12,739 10,435 631 2026-09-17 post
2 Flight search with Browser Use @gregpr07 30,060 8,723 617 2026-09-17 post
3 Real-time slop detector as you scroll @RBilgil 685 7,180 210 2026-09-19 post
4 724 competitor ads, broken down @TheMattBerman 12,799 6,348 389 2026-09-17 post
5 Voice-controlled computer use on a Mac @instantricecook 1,015 5,016 252 2026-09-18 post
6 jev-trader @jarrodwatts 32,542 4,913 216 2026-09-16 post
7 Jev plays Doom @CompleteSkeptic 122,369 4,890 240 2026-09-15 post
8 A canvas you control by pointing and speaking @jackcheng 11,724 4,797 254 2026-09-17 post
9 Jev plays Subway Surfers @_MaxBlade 22,962 3,956 253 2026-09-17 post
10 A real-time ad blocker @iam_zachi 4,832 3,872 139 2026-09-17 post
11 500 emails for 3.5 cents @rileybrown 244,870 3,853 96 2026-09-17 post
12 Jev plays Smash Bros. against itself @maubaron 19,783 3,620 334 2026-09-18 post
13 Triage across 1,500 emails @ryanvogel 18,403 3,538 106 2026-09-16 post
14 700 leads scored in 40 seconds @romanbuildsaas 20,336 3,138 203 2026-09-18 post
15 Jev plays Super Mario Bros. @faadilhshaik 192 2,860 248 2026-09-16 post
16 jev() for PostgreSQL @iam_zachi 4,832 2,738 182 2026-09-17 post
17 Game levels generated in real time @HugoDuprez 3,151 2,614 208 2026-09-18 post
18 Keystroke oracle @dabit3 194,714 2,368 132 2026-09-18 post
19 jevlike @vinnylarouge 1,392 2,018 168 2026-09-16 post
20 1kpapers @nutlope 100,148 1,962 139 2026-09-17 post
21 A model router on Jev @ephraimduncan 6,705 1,858 66 2026-09-17 post
22 Computer use without screenshots @milindlabs 4,025 1,823 97 2026-09-17 post
23 Every’s editorial vibe check @danshipper 123,938 1,818 90 2026-09-15 post
24 400 companies matched to one candidate @sarvagya_kul 5,832 1,688 74 2026-09-18 post
25 $10,000 in Jev’s hands @abolbuild 2,188 1,606 47 2026-09-17 post
26 An always-on assistant with no wake word @_MaxBlade 22,962 1,538 73 2026-09-18 post
27 Post scoring with SuperX @robj3d3 61,401 1,341 63 2026-09-17 post
28 Doomscroll Filter @robj3d3 61,401 1,213 47 2026-09-18 post
29 Predictive spreadsheets @dabit3 194,714 1,169 55 2026-09-18 post
30 The X algorithm, rebuilt with Jev @leojrr 22,025 1,167 26 2026-09-17 post
31 SEO and GEO fixes, 90% cheaper @irabukht 19,093 1,146 63 2026-09-18 post
32 Jev plays Slay the Spire 2 @coolish 64,731 1,137 93 2026-09-17 post
33 A chat bot with no LLM @CodingGarden 13,749 1,115 63 2026-09-17 post
34 Simple Jev @picocreator 5,783 1,105 120 2026-09-18 post
35 A Downloads folder that sorts itself @marcelpociot 70,603 1,092 45 2026-09-18 post
36 Hide posts on X in plain language @marcelpociot 70,603 1,090 41 2026-09-17 post
37 YouTube sponsor skipper @tdinh_me 201,821 1,078 40 2026-09-18 post
38 End-to-end tests run by agents @o_kwasniewski 8,788 1,012 74 2026-09-18 post
39 openjev-sglang @ekzhang1 22,899 972 60 2026-09-17 post
40 Live viral post analyzer @rileybrown 244,870 915 30 2026-09-17 post
41 Intent-based search in Gmail @dabit3 194,714 886 34 2026-09-18 post
42 A second-hand shopping agent @AlanDaitch 49,142 873 35 2026-09-18 post
43 Fraud detection with Jev and Kimi K3 @nutlope 100,148 846 47 2026-09-17 post
44 Ad creatives from filtered assets @higgsfield_ai 232,876 824 108 2026-09-19 post
45 openjev on Qwen 4B @justALEXWORTEGA 635 774 47 2026-09-16 post
46 Stagehand on a remote browser @kylejeong 8,105 743 43 2026-09-17 post
47 3,282 posts, eight questions each @iannuttall 81,192 740 34 2026-09-17 post
48 A visual reference finder @albicodes 10,546 654 27 2026-09-17 post
49 TypeSafe Typewriter @stevekrouse 11,267 587 28 2026-09-16 post
50 Lurk @mxfp4 602 559 32 2026-09-18 post
51 askjev.ai @waynesutton 68,076 511 27 2026-09-17 post
52 jev-review @niazmorshed_ 1,422 492 28 2026-09-17 post
53 900 images in 40 seconds @fayazara 8,177 433 8 2026-09-18 post
54 A local Jev @wmoto_ai 1,441 387 49 2026-09-17 post
55 Website to App @chddaniel 27,842 382 25 2026-09-18 post
56 Jev Calc @thekitze 102,836 318 15 2026-09-18 post
57 Jev Detector @jozef_gherman 144 296 22 2026-09-17 post
58 An agent with a Jev model router @rileybrown 244,870 281 10 2026-09-17 post
59 A filter for reply-guy comments @iannuttall 81,192 250 4 2026-09-18 post
60 Headless Chromium agent @mormonnegro 7,026 213 8 2026-09-17 post
61 Agentic browsing in Chrome @razaanstha 525 194 12 2026-09-17 post
62 A prompt box that fills itself in @sawyerhood 18,482 193 8 2026-09-18 post
63 A chief of staff for bots @milindlabs 4,025 182 10 2026-09-17 post
64 Real-time Clippy @sotak 6,030 175 12 2026-09-17 post
65 A Slack agent, twice as fast @johnyeo_ 3,943 148 3 2026-09-18 post
66 Which outreach signals book demos @pierreeliottlal 9,774 113 13 2026-09-18 post
67 DiffJury @raihankhan_rk 3,470 81 4 2026-09-17 post
68 X timeline labeler @the_cyw 1,199 79 3 2026-09-18 post
69 jev-job-hunter @hqmank 12,060 75 4 2026-09-18 post
70 700 live ads in 40 seconds @Yarilo7brigada 1,993 59 5 2026-09-18 post
71 Jev plays Tetris @AlanDaitch 49,142 48 4 2026-09-17 post
72 One-click invoice finder @FarouqAldori 459 43 5 2026-09-17 post
73 Jev as an agent safety monitor @isNickMa 590 1 0 2026-09-17 post
74 AI slop detector @kraayenJon 151 0 0 2026-09-19 post

All repositories

Repo Stars Language License Last push Source
browser-use/jev-ultrafast 7,798 Python MIT 2026-09-18 GitHub
tamaratran/fast-jev-compaction 4,031 TypeScript MIT 2026-09-18 GitHub
TheoLeeCJ/SemIf 1,839 Python MIT 2026-09-19 GitHub
jarrodwatts/jev-trader 1,184 TypeScript MIT 2026-09-17 GitHub
vinnylarouge/jevlike 961 Python MIT 2026-09-16 GitHub
vercel-labs/ai-cli 805 TypeScript - 2026-09-19 GitHub
awlevin/typesafe-computer-use 456 Python MIT 2026-09-18 GitHub
jaredpalmer/kev 423 Python Apache-2.0 2026-09-19 GitHub
thruwire/foreman 359 Python MIT 2026-09-19 GitHub
devagrawal09/jev-review 326 TypeScript MIT 2026-09-17 GitHub
fhshaik/typesafe-mario 278 Python - 2026-09-16 GitHub
droidrun/mobile-jev 209 JavaScript MIT 2026-09-17 GitHub
realZachi/pg-jev 204 Shell NOASSERTION 2026-09-18 GitHub
superagents-lab/jev-search 199 TypeScript MIT 2026-09-19 GitHub
gargpratyush/jev-router 191 JavaScript MIT 2026-09-19 GitHub
kitze/skillbox 188 TypeScript MIT 2026-09-19 GitHub
wy-coliney/jev-browser-use 176 JavaScript MIT 2026-09-19 GitHub
hr98w/jev-visual 138 Python MIT 2026-09-18 GitHub
jkudish/jev-browser 135 TypeScript MIT 2026-09-19 GitHub
kitze/unclutter 129 TypeScript MIT 2026-09-18 GitHub
razorback16/openjev 113 Python Apache-2.0 2026-09-18 GitHub
dbreunig/building-with-jev-skill 113 - - 2026-09-17 GitHub
moritzkremb/jev-voice-browser 110 JavaScript MIT 2026-09-17 GitHub
itsmostafa/typesafe-mcp 97 Go MIT 2026-09-18 GitHub
jkudish/jev-mcp 92 TypeScript MIT 2026-09-19 GitHub
DevMortimer/pi-warden 90 TypeScript MIT 2026-09-19 GitHub
vinilana/jev-eval-agent 89 HTML - 2026-09-17 GitHub
logan-markewich/jeff 87 Python MIT 2026-09-19 GitHub
Mapika/decider 86 Python Apache-2.0 2026-09-19 GitHub
pithings/advocaat 84 TypeScript MIT 2026-09-18 GitHub
BillionsBobby/JevRouter 81 TypeScript MIT 2026-09-19 GitHub
tamaratran/jev-pruner 80 TypeScript MIT 2026-09-19 GitHub
giuliosmall/pg_typesafe 77 C MIT 2026-09-18 GitHub
RomanSlack/jev-drone 71 Python MIT 2026-09-17 GitHub
mrnugget/jev-shell-history 63 TypeScript - 2026-09-18 GitHub
realZachi/typesafe-adblock 53 JavaScript MIT 2026-09-17 GitHub
RafalWilinski/vibecheck 41 JavaScript - 2026-09-18 GitHub

By the maintainer

  • will-it-hit. A live LinkedIn draft scorer. One call asks eight Score questions (hook, specificity, emotion, clarity, repostability, authority, algorithm fit, expected engagement) and one Choice question for post type. The rubrics carry real engagement numbers from LinkedIn posts. A separate LLM writes rewrites, and Jev scores each rewrite again so you see both numbers. The source is in a private repository for now.
  • linkedin-slop-blocker. A browser extension that scores every post in your LinkedIn feed as you scroll. Each post gets a "Slop" or "Not slop" pill with a percentage, and slop posts get a fading text treatment and a rotated SLOP stamp you can click away. That scroll style follows the real-time slop detector demo by @RBilgil. A free pattern scan runs first, then one Jev call per scroll returns a spam probability and a quality score for each post. It keeps your own API key in the browser and learns from posts you hide or dismiss. The source is in a private repository for now.
  • Jev tagger (inside linkedin-slop-blocker v0.5). A floating corner panel where you write your own categories, each with a name and a plain-English description, then pick colors. As you scroll LinkedIn, one Jev Choice question per post tags it with the best-fitting category, or leaves it untagged when confidence is low. Each category can be shown, dimmed or hidden, and clicking a counter chip focuses the feed on that category. It follows a demo by @nateherk, who tags X posts as breaking, golden nugget or slop. Two presets ship with it, "Signal or slop" and "By intent". The source is in a private repository for now.

What Jev is

Jev answers typed questions about a piece of context. You send one state (a string, an object, or an array) and a map of named questions to a single endpoint. There are three question types.

  • Choice picks one option from a set you define and returns the probability of every option.
  • Score rates the state on an ordered rubric and can land between two levels.
  • Noul answers yes or no as a probability.

Jev does not generate text. Output tokens are free, and questions asked over the same state run in parallel. For anything that needs prose, pair it with a normal LLM.

The smallest request:

curl https://api.typesafe.ai/v1/systemone \
  -H "Authorization: Bearer $TYPESAFE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "state": "Help! My payouts have been failing for 3 days.",
    "model": "jev-latest",
    "questions": {
      "is_urgent": {"type": "noul", "instructions": "Does this convey urgency?"}
    }
  }'

The API reference shows this response for that request:

{
  "model": "jev-latest",
  "answers": { "is_urgent": { "type": "noul", "noul": 0.92 } },
  "usage": { "input_tokens": 312, "output_tokens": 48 }
}

The models page lists Jev 1.13 at $42 per billion input tokens, a limit of 250,000 tokens per second and 1,200 requests per minute, and a cap of 32k tokens on the state plus the longest question. The model value jev-latest currently points to jev-1.13.0.

Cookbooks from TypeSafe

TypeSafe's own worked examples. Numbers below are the ones TypeSafe reports in each cookbook.

  • Parallel questions. Runs a 13-question regulatory briefing over one document. Batching every question into one call is reported as 12.2x cheaper and 10.0x faster with no change in answers.
  • Re-ranking. Asks one question per query and candidate pair on 30-passage shortlists for 40 legal queries. Top-1 accuracy goes from 5% to 18% and top-10 from 38% to 62%.
  • Line-by-line search. Semantic search over GitHub's Terms of Service. One request scores 218 line ids against a plain-language query.
  • Structure recovery. Rebuilds Markdown from plain text that lost its formatting, in two requests.
  • Function calling. Turns natural-language trading requests into calls to ordinary typed functions.
  • Skill suggestion. Picks at most one skill for an agent turn out of the 182 in Nous Research's Hermes catalog, and can reject every candidate.
  • Knowledge graph entity alignment. Decides which of 450 candidate pairs from two beer catalogues are the same product. One Score question carries the decision, and its levels are merge, leave unlinked, and hand to a curator.
  • Classifying RAG passages. Scores each retrieved passage in one request, then code decides which ones reach the answering model. A passage carrying a hidden instruction can be dropped.
  • Double-checking citations. One Choice question decides whether the quote's context supports the claim, and low confidence flags the citation for review.
  • Guardrails for LLMs. Screens every message going into and out of an LLM app with one request that scores hazards such as jailbreak attempts. Your code applies the thresholds.
  • SDE cascade. A two-stage structured-data-extraction cascade that gets most of a large reasoning model's quality at a fraction of the cost.
  • Date extraction. Asks Jev for the parts of a date named in a document, then resolves and validates them in code.
  • Pre-parsed value extraction. Regexes find candidate emails, phone numbers, and amounts, then Jev selects the requested span so code can normalize it.
  • Hierarchical classification. Classifies documents through deep patent, retail product, biomedical, and source-code hierarchies with parallel beam search over Choice probabilities.
  • Autoresearch feature discovery. A loop that proposes questions, turns free text into numeric features, and uses model errors to improve a supervised CatBoost regressor.
  • Classification using confidence. Classifies SEC annual reports into 75 industry groups with one Choice each, then reads the answer's confidence to decide whether to report the group or the broader division above it.
  • Self-consistency with nouls and with choices. Route uncertain probabilities to human review, and add an uncertain outcome to moderation decisions.

Patterns

From TypeSafe:

  • Speculative fan-out. Send many questions in one call. Speculative ones are fine, and your code decides what is relevant.
  • Confidence-gated routing. The answer tells you what, and confidence tells you whether to act.
  • Composite scoring. Break a complex judgment into atomic scores and combine them with weights you control in code.
  • Intent routing. Classify incoming requests and send each to a deterministic handler, a specialist LLM, or a human.

Found while building the projects above:

  • Generate with an LLM and judge with Jev. In 1kpapers the summaries cost $3.99 and the classification cost $0.08, so judging 1,018 papers cost about 50 times less than summarizing them. In will-it-hit, a separate LLM writes the rewrite because Jev cannot write text, and Jev then scores the result.
  • Put many items in one state with one question per item. Send an array as the state, key the questions like post_0 and post_1, and refer to items by path such as posts[0] in the instructions. One request covered a full feed scroll. The cost is accuracy, because TypeSafe's docs warn that unrelated material in the state lowers it. Keep batches small and check results by hand.
  • Write rubrics from real outcomes. Score levels should describe concrete situations, and the instructions can carry observed numbers, such as how many reactions a flop and an outlier received, so scores do not drift upward.
  • Do the arithmetic in code. will-it-hit combines its eight Score answers with fixed weights in code and uses thresholds for the verdict labels, which matches the advice on TypeSafe's limits page.
  • Send requests from the extension's service worker. On linkedin.com a fetch from the page context to localhost failed in testing. A background service worker is not bound by the page's Content Security Policy.
  • Select by test ids on LinkedIn, not class names. The feed uses hashed class names that change on every deploy. In September 2026 these worked: div[role="listitem"][componentkey^="update-card-focus"] for a post card, [data-testid="expandable-text-box"] for the body text, and the aria-label of button[aria-label^="Open control menu for post by "] for the author name. Expect them to break when LinkedIn changes its markup.

Limits of Jev 1.13

From TypeSafe's jaggedness page, last reviewed 2026-09-17, and the models page. Read the source pages before you build.

  • It reads instructions literally. Write the exact condition and put boundary cases in the criteria.
  • It is not a calculator. Keep math in code, which includes counting and date comparison, and use Jev to extract the parts.
  • Use Score outputs for thresholds and ranking. Do not interpolate an exact number between two levels.
  • Accuracy falls when the state carries content unrelated to the question. Filter first.
  • Text in the state can steer the answer. An injected instruction or a misleading framing can move it, so test edge cases before deploying to many users.
  • Separate questions are not guaranteed to agree. A Noul and a yes or no Choice about the same thing can return different numbers, so do not carry a threshold from one type to the other.
  • It does not generate text. Use a generative model for that.
  • English works best. Test other languages on your own content.
  • The state plus the longest question is capped at 32k tokens and a whole request at 64k. Input is text only.

Reported cost and latency

Except where a row names TypeSafe, these are numbers from the builders themselves and have not been independently verified.

Source Report
TypeSafe models page $42 per billion input tokens ($0.042 per million). Output tokens are free.
TypeSafe launch post End-to-end response time of 70 ms to 500 ms, and 40x to 200x faster than frontier LLMs on System One tasks. It says the 193.6x faster and 444.6x cheaper figures on its home page are on the higher end of real-world gains.
TypeSafe parallel-questions cookbook 12.2x cheaper and 10.0x faster than one call per question, with the same answers, on one document.
@nutlope 8 cents to classify 1,018 papers, 256 ms median end-to-end per paper. The summaries from another model cost $3.99.
@cjzafir $3.40 spent over 24 hours of testing. Repeats TypeSafe's 70 to 500 ms and 193.6x and 444.6x figures.
@walidboulanouar About $0.001 spent across roughly 100k tokens in one day of building.

Ideas nobody has shipped yet

These are hypotheses from a brainstorm on 2026-09-19, not products. Where a documented limit applies, the note says so.

Consumer and content:

  • A tone meter for any text box. Score clarity and tone as you type in Gmail, Slack, or X. The hard part is attaching to other sites' inputs without breaking them.
  • Live chat moderation. Classify each message in a fast Twitch or Discord chat with Noul questions for toxicity and spam. Messages written to fool the classifier are the main risk.
  • A feed reranker that asks only when unsure. Score every item in an infinite feed for interest fit and ask the user a question only when confidence is low.

Developer tools:

  • A model router. Score prompt complexity and choose a model tier before the call, so the router costs less than the call it routes. Watch for flapping between tiers on near-identical prompts.
  • CI test selection. Choose which test tier to run from a commit diff. Diffs are large, so filter to changed paths and hunks before sending them.
  • Abuse scoring at an API gateway. Ask a Noul question about request text. Rates and timing must be computed in code. Request text is controlled by the attacker, so this is the riskiest idea here.

Browser extensions:

  • A job board scam filter. Score each listing card as real, ghost, or scam. Selectors differ by site and change often.
  • A marketplace price-trap detector. Flag likely bait listings, and compare prices in code instead of asking the model.
  • A fake review flagger. Score each review as the list loads. Text alone is a weak signal for reviews written to look organic.

Operations:

  • Support ticket triage. A Choice for department, a Score for urgency, and a Noul for refund requests. Calibrate the criteria on the client's own ticket history.
  • Inbound lead scoring on form submit. Needs closed-won and closed-lost examples to calibrate the rubric.
  • An outbound message compliance check. One Noul per policy rule in one request before send. Tune the confidence thresholds so false positives do not push people to turn it off.
  • A CRM hygiene sweep. Score record quality and possible duplicates. A duplicate check needs both records in the state.

Tools

How this list was made

Snapshot taken 2026-09-19. Post metrics (likes, reposts, replies) come from the posts themselves, read with yt-dlp. Follower counts come from treg. Repository stars, languages and licenses come from the GitHub API, and every repository was opened to confirm its README mentions Jev or TypeSafe. Descriptions are written from the original posts and READMEs and checked so that no number appears that the source did not contain. Metrics change by the hour, so treat them as a dated snapshot. Followers are counted today, after most of these posts went out, so the follower ratios understate how small the accounts were at the time.

The two CSV files in data hold the same numbers, and docs has one page per demo.

Contributing

This list grows by pull request. Three ways to help, from smallest to largest.

  • Report a broken link, a wrong number or a dead repo. Open an issue with the "Fix an entry" form.
  • Suggest something new. Open an issue with the "Add an entry" form, or send a pull request.
  • Add a limit you hit. A failure with a reproduction is worth more than another success story.

Where things live:

Path What it holds
README.md The list itself
docs/demos/ One page per demo, with every metric and the source link
data/demos.csv, data/repos.csv The raw numbers behind the tables
assets/ Banner, charts and preview frames

Full rules for an entry are in CONTRIBUTING.md. Every entry needs a source link that opens, numbers named to whoever reported them, and one to three sentences.

Sponsor

Sponsored by AY Automate

Sponsored by AY Automate, an AI-native engineering company. This list stays free and open source.

License

CC0 1.0

Reviews (0)

No results found