A headless Python voice-assistant daemon for Arch/Hyprland desktops and macOS: middle-click summon, DeepSeek-Flash reasoning, xAI Grok voice, about two dozen desktop tools, and optional Jev triage over memory.
A dated harness for running open System One typed-decision models — Laya, Decider, imajev, Winnow-12B, decider-12b — on a Snapdragon X Elite NPU, GPU, and CPU, benchmarked against TypeSafe's Jev API.
A logistics exception gate that reads carrier codes and free-text driver notes in one scan event and decides whether a shipment needs a human, which exception class it is, and how severe the delay looks.
A contextual brand-suitability scorer that judges each feed item against an advertiser's own versioned policy, returning a category, an exclusion flag, and a confidence for routing borderline items to review.
A localization pre-review gate that pairs source and target strings and judges meaning, terminology, tone, and placeholders as separate questions, so one failing dimension blocks a release without blending scores.
A pre-bid spend guard that scores a placement's expected relevance and flags whether it matches the campaign brief, letting deterministic code hold budget back before the auction instead of gating the creative itself.
A podcast chaptering tool that labels transcript windows and marks boundaries as separate local judgments, classifying content, sponsor, and housekeeping segments without generating prose for the titles.
A real-estate matching tool that applies hard-constraint gates before ranking a buyer's property preferences against listing features and remarks, returning a fit band and a per-feature score for the agent.
An insurance claim triage tool that classifies loss type, scores documentation completeness, and flags fraud indicators in a first-notice-of-loss narrative, routing claims without denying any or setting a payout.
An agricultural advisory demo that turns sensor readings, a forecast, and a crop stage into a next-action choice plus an urgency score, routing marginal readings to an agronomist instead of triggering an application.
A travel rebooking demo that classifies a trip disruption, checks itinerary feasibility, and ranks concrete rebooking options against a traveler's ordered preferences, returning ordered choices with probabilities.
A clinical intake router that picks a care pathway from a patient's reported reason for contact and flags whether the message needs a same-day response, without ever assigning a condition or a treatment.
A warehouse pick-verification step that compares a scanned item, the bin's expected contents, and the order line to decide between accept and re-scan, catching confusable picks a barcode alone would pass.
A Kubernetes admission webhook that asks Jev to allow, flag, or deny a manifest change for intent-shaped policies a Rego rule cannot express, keeping hard rules in Rego and failing closed by default.
TypeSafe's Series A announcement: $870M at a $7.5B valuation led by Andreessen Horowitz with Sequoia and DCVC, and what the post says the money changes for developers, businesses, and recruits.
A TypeSafe case study of a talent marketplace that swapped Gemini 3.1 Flash Lite for Jev in candidate matching: 88% lower cost, about twice as fast, and a projected $500K a year saved.
A TypeSafe media piece arguing that RLHF-trained language models please humans and assist rather than make reliable autonomous decisions, and asking what class of model comes next.
A TypeSafe media profile of co-founder and CEO Diogo Almeida, framed around why some AI applications over-promise while capable models can still under-deliver in practice.
A playable espresso bar where an XState machine models what the hardware can do next and Jev picks the move, from a plain-language order or after you break the machine.
A toy that hands Jev control of page layout: it picks the size, color, and position of on-screen components and rearranges them to its own preference, exposed as a live site.
A ComfyUI node pack that branches a generation workflow on a plain-English question such as whether an image is too graphic for kids, running locally on Laya by default or Jev on a dropdown switch.
A zero-shot text classification service over plain HTTP: send text and a list of labels, get the best-fitting label and a calibrated probability per label, with Jev as the default and fast tier.
Unsloth's guide to turning a generative LLM into a Jev-style decision model that returns a choice with a probability instead of text, plus a companion page for serving Laya behind the TypeSafe API.
A browser game experiment where Jev drives a non-player character in an open-world scene, choosing the NPC's next behaviour from the state it is given instead of a scripted tree.
An open-source Los Angeles County ballot helper: answer a short quiz and enter an address to get a match for every contest on the November 3 ballot, with Claude doing the research and Jev the scoring.
Vercel's agent framework with Jev wired into its experimental evaluate path, so the scoring step of an agent returns typed decisions rather than prose the surrounding code has to parse.
A router playground that serves several System One decision models from one place: Jev, Perplexity's Decider, Bespoke Nimble, Mercury Decide, Clef, Kev 4B, and Decider 2B, with three of them free.
The Product Hunt listing for Jev: ranked second product of the day on 21 September 2026 with roughly 556 upvotes and a 5.0 rating, filed under API, Developer Tools, and Artificial Intelligence.
Cloudflare's first open decision models on Workers AI: Clef, a 27B model that reads text, JSON, images, or video, and Clef-flash, a 9B model for the latency-sensitive hop, each returning a probability per option.
A multimodal decision model on Workers AI that takes text, images, audio, or video in one request and returns schema-constrained probabilities, built on a 30B mixture-of-experts backbone with about 3B parameters active.
Cloudflare's decision models packaged for Ollama: pull clef for the 27B model or clef-flash for the 9B one and run classification, labelling, and routing locally with the same typed-question shape.
Liquid AI's first decision model, a hosted API that adds image input to the Choice, Score, and Noul shape. Liquid reports it matching or beating GPT-6.1 Sol on four of six tested applications at 19x to 200x lower cost.
A Claude Code plugin that replaces the built-in auto-mode permission classifier with Jev, deciding tool calls in about half the time while agreeing with the original on the calls it handles.
An open-source data analyst agent that keeps Qwen for generating SQL and uses Jev as a review layer at three checkpoints: question clarity, SQL fidelity to the request, and whether the final explanation is grounded in the returned rows.
A Python decorator that turns a Pydantic model into Jev questions: bool becomes a Noul, a Literal becomes a Choice, and a bounded int becomes a Score, with answers validated back into the model.
A field report on putting Jev inside a Playwright browser-testing loop: two Jev calls per probe, 63 probes in 24.3 seconds, and a 131 ms median round trip, with a candid account of near coin-flip decisions.
Katalon's guide to Jev in software testing: where bounded judgments help QA, why deterministic assertions stay first, the documented weak spots that hit test checks, and the caveat that a constrained answer can still be wrong.
A ranked comparison of thirteen decision-model APIs as of October 2026, including Jev, OpenAI Decisions, Clef, Liquid d1, Kev, Perplexity Decider, and Strands, scored on evidence, usage, and price.
A three-week timeline of the Jev clone war: the September 15 stealth launch, reported $10B valuation talks, an Amazon engineer's own build, and the $870M round, alongside OpenAI, Amazon, and Cloudflare shipping rivals.
GIGAZINE's report on Liquid AI's d1, which it describes as the first decision model to beat Jev on Hugging Face's Decision Index, and a short survey of the clone wave that followed Jev's September launch.
Jev Media & TalksArticle#community#news#benchmarks
A Japanese-published practical guide to Jev: request anatomy, runnable curl, Python, TypeScript, and Cloudflare Workers examples, confidence-gated branching, cost estimates, and twelve use cases.
The Hugging Face hub for Laya's open System 1 checkpoints: the 421M English model at the repo root, a 322M multilingual build in a subfolder, and the typed-decisions fine-tune, all Apache-2.0.
The fine-tuned Laya checkpoint for four typed-decision workflows: 0.766 accuracy on 2,000 held-out decisions, above Jev 1.13.0's published 0.727 and the teacher self-agreement ceiling it trained past.
Supersonic Labs' 144.3M-parameter decision model on mmBERT-small: choice, score, and noul answers from supplied options in one pass, 73.15% on 2,000 typed decisions, Apache-2.0.
A browser WebGPU export of Julia 1: the same 551 MB decision checkpoint, a Rust WASM tokenizer, and a JavaScript API that answers typed questions client-side with no server or key.
An open distillation of Jev 1.13 onto Qwen3.8-27B: a 108.9M-parameter decision block answers typed questions in one prefill pass and adds a POST /v1/decide route beside a full System 2.
The vision sibling: the same text-trained System 1 head on Qwen3.8-27B's image encoder, taking images zero-shot, ranked first of 20 vision decision models on the community Jev Decision Index 0.3.
The 9B member of AutoTrust's Jev students: a 40.2M-parameter decision block on Qwen3.5-9B with 0.019 mean KL to Jev 1.13 and 70.7% HumanEval left untouched, small enough for one modest GPU.
The Jev Day release post: Jev 1.14.0 under jev-preview plus three danger-zone features — composable images on both models, the box question type on 1.14 only, and opt-in 262k context.
The box question type: one bounding box per question, returned as four edges that are fractions of the image, plus confidence, with no criteria field and no not-found value.
LiteLLM's Day 0 post for TypeSafe Jev 1.14: the proxy adds box questions, image input and long-context mode on both POST /v1/systemone and POST /v1/decisions, forwarding TypeSafe-Danger-Zone headers only to TypeSafe deployments.
Every entry is reviewed against the same bar, whether it comes from TypeSafe or the community.
Public and working — no paywalls, sign-up walls, or dead links.
Concrete — it has to be about Jev and the System One API, not generic LLM content.
Useful today — something a developer can learn from or use right now.
Distinct — if a similar entry exists, the description has to say what makes this one different: a different runtime, a cheaper path, a measured result, or a narrower focus.
Documented — repos state their license, and docs are in English or carry an English summary.
Spam, affiliate links, and thin AI-generated content are rejected. New entries are expected to ship with a deep dive: two to five highlights, caveats, a quickstart snippet where one helps, and a verification date. Star counts are refreshed when an entry is touched.
Common questions
Short answers about Jev and how this directory works.
What is the Jev AI model?
Jev is TypeSafe AI's first System One model — an AI model that takes a state plus typed questions and returns one structured answer per question: a choice, an ordered score, or the probability of yes, with calibrated confidence on Choice and Score answers. Code can branch on the result without parsing generated text.
What is the difference between Choice, Score, and Noul?
They are the three question primitives. Choice picks one of up to 255 options, Score rates on 2 to 10 ordered levels, and Noul answers yes or no. Choice and Score answers carry a confidence between 0 and 1; a Noul answer is the probability of yes.
Which Jev SDKs and client libraries are available?
TypeSafe maintains official JavaScript/TypeScript and Python SDKs plus a Python adapter. The community has built clients for Go, .NET, PHP, Laravel, Rust, Elixir, Ruby, and Scala, along with framework integrations such as DSPy, LlamaIndex, and Home Assistant. All of them are listed in the Repos & SDKs category.
Is the directory free, and is it affiliated with TypeSafe AI?
The directory is free to browse: every entry is reviewed, and spam, affiliate links, and sign-up walls are rejected. JevDirectory.org is community-run and not affiliated with TypeSafe AI — Jev and System One are TypeSafe AI trademarks.
How often is the directory updated?
Entries are added as new resources appear and re-checked when they are touched. The directory currently holds 3919 entries from 293 source sites, last updated Nov 6, 2026.
What is a verified deep dive?
Most entries include a deep dive: two to five highlight bullets someone would only know from using the resource, a caveats note about limits and requirements, an optional quickstart snippet, and the date the link was last checked. 3919 of the 3919 entries carry one today.
nodaysidle/kureksistantPython
statelyai/jevspressoTypeScript
dante01yoon/ComfyUI-SystemOnePython
aaazzam/jevPythonDo Multimodal Models Decide Consistently on Equivalent Optical Spectra?Jev Benchmarks & Evaluations
@harlan-zw/jevJev Repos & SDKs
Jev-AD: Safe Language-Guided Autonomous Driving via a Language World ModelJev Demos & Experiments
Jev GUI Training PreviewJev Models & Reimplementations