Jev Yibie

A 144.3M-parameter bilingual typed decision model fine-tuned from jhu-clsp/mmBERT-small: it answers noul, choice, and score questions in one forward pass with calibrated confidence and zero output tokens, and it runs on CPU.

A native Swift implementation of OpenJev, the Jev-compatible typed-decision server: send a state and typed questions over a Jev-compatible HTTP API and read every answer from model probabilities, on Apple silicon Macs or inside iPhone and Mac apps.

Plugs Jev-style decision models into published rerankers as a fast primitive: one forward pass over k candidates, zero decoded tokens. On T2Ranking it reports up to 22x faster matches than a generative listwise reranker at 2.9x the nDCG@10.

Probabilistic control flow for TypeScript: Hunch asks a System One model such as Jev yes/no questions, thresholded booleans, picks and ratings, and returns typed answers you can branch on in ordinary code.

A Chrome extension that sorts Hacker News comments into buckets you define: Jev answers one Choice question per comment in about 300 ms, and each comment gets a coloured label with its bucket and probability.

A GitHub Action that reviews every pull request with TypeSafe Jev and fails the check when it finds a problem. It asks four yes/no questions about each part of the diff and compares files that look duplicated, catching secrets, vulnerabilities, needless complexity and prompt injection.

A testing app and CLI for chat endpoints: send messages to any chat API, watch replies stream in as a conversation, and check every reply with rule checks (JSONPath, latency, tokens, cost, tool calls) and judged checks including TypeSafe's System One. Export the suites as files and run them in CI.

A website companion's code: a page's actions are offered as WebMCP tools, a visitor's request becomes one Jev Choice, and a three.js voxel character hops to the button and does it.

A git pre-commit hook that asks TypeSafe Jev whether each new file belongs in your repo, catching scratch scripts, debug dumps, build output and local config before they land. Local rules run first, so .env and .DS_Store are matched locally and never sent anywhere.

A Claude Code mod that adds $.jev to the engine so other mods can ask TypeSafe Jev typed Choice, Score and Noul questions and get calibrated probabilities back. Answers are validated against their questions, keys are kept in the keychain, and calls time out at 4 seconds by default.

An e-commerce after-sales customer-service agent built on the learn-claude-code skeleton. Every turn asks Jev for the customer's intent with a confidence and a needs-human probability in one call, and a pure-code gate sends low-confidence cases to a human before the LLM runs.

An open-source no-limit Texas Hold'em game for two to six human and AI players, with TypeSafe Jev, Equity Rules, Upright Eevo and configurable LLM bots as opponents. The poker engine stays authoritative for cards, turns, betting, pots and winners.

A Codex and Claude Code Stop hook that detects overloaded multi-topic messages and forces the agent to present issues one at a time with context and missing info, using two Jev Choice questions behind confidence thresholds.

An OpenCode V2 plugin that asks TypeSafe Jev whether a prompt is deterministic before the LLM reasons, injecting a JEV PRE-ANALYSIS block and temperature 0 for deterministic turns so simple prompts skip chain-of-thought. Jev failures fail open and the session continues normally.

A pre-action gate for tool-using agents that decides allow, confirm or block with deterministic rules before a neural decider, plus a CDP browser loop whose decider swaps between hosted Jev and a local System One-compatible server through environment variables alone.

A benchmark comparing TypeSafe Jev with Claude Haiku 4.5 as input guardrails across jailbreak, evasion and benign prompts, measuring detection rates, false positives, latency and cost. At n=1,000 Jev traded some detection for a 328 ms median latency and $0.03 per 1K calls versus Haiku's 2,066 ms and $1.87.

Local spam and phishing triage for a Proton Mail Bridge inbox on macOS: Jev answers two questions per new message, answers are recorded in dry-run by default, and mail moves to a Jev Review folder only if you turn that on. Nothing is ever deleted.

The companion source code for the book Mastering TypeSafe Jev: 13 chapters plus two appendices build two capstones on one shared decision client — an autonomous incident and FinTech triage engine on hosted Jev, and a SmartPlant industrial-IoT engine on self-hosted open-source Kev.

ABTdomain's workflow design for brand-protection review: three layers of typed Jev questions turn each new domain that carries a brand word into Release, Watch or Flag, with 0.3 and 0.7 probability lines deciding what happens next.

An 11-page working paper on placing a small commercial text judge beside a live trading lab: six tests for a text signal, what the judge may and may never touch, and the plumbing failures that corrupted its readings. No profitability claims.

AnyEval's paired JevBench comparison of jev against gev-1.0 on 231 shared problems: the win table, an exact McNemar result, and end-to-end answer times for both models measured in one fixed stratum.

Benchmark Heaven's live JevBench board for typed decision models: florianstandhar's reproducible benchmark, presented with frozen versioned rankings, price bases, hosting and confidentiality filters, and per-model evidence requirements.

Blink enforces your team style guide, like AGENTS.md, by checking every file edit made by Claude Code and Codex and handing findings back to the agent inside the session in under half a second; it is mostly powered by Jev. Free public beta.

ChemIllusion's post on how its structure autocomplete works: RDKit builds and checks every candidate from your drawing, and Jev picks which checked cards to show and in what order, in one request of about 0.2 seconds.

A shared 1,000 by 1,000 emoji canvas where visitors paint and Jev paints alongside them: after each stroke it reads what is around it by name, weighs emoji that belong with it, and paints one or finishes a shape it is sure you meant.

Jevvie is the companion in the corner of every page on chriswijnia.com: ask it for something and Jev picks which of the page's actions you mean, one typed Choice at a time, then Jevvie hops over and does it. The same actions are WebMCP tools.

Microsoft's announcement of Microsoft-Decision-1, a model for fast decision scoring built for routing, classification, prioritization, verification and workflow control, available in Microsoft Foundry and through OpenRouter. Published 2026.10.09.

Perry Link's Zenodo preprint compares three judgment layers - a local Laya judge, TypeSafe Jev and DeepSeek-V4.1-Flash - on the same 120-checkpoint task, reporting that cost is not the binding constraint, that p50 latency spans about 25-32x, and that the judges' self-reported fields fail in clustered ways.

Stephen Solka's Hugging Face article tests whether Jev can filter political fundraising from an inbox: a frozen 100-email eval, one noul question per email, then follow-on experiments that shrink the classifier to CPU-only inference with SetFit and ONNX Runtime.

A Hugging Face Space that runs a tiny System One model entirely in the browser: type a prompt and the app produces output on the spot, with no server connection or installation required.

An independent lab for trying the Jev model: a customer-message example plus experiments for task routing, context selection and evidence checks. It shows choice, score and boolean answers with probabilities and can export the request as TypeScript.

A directory of 2,086 open-source projects built on Laya, the open System One decision model, with category and platform filters, star counts, a live laya-mlx demo, and a hosted Laya + Clef API with $1 free credit.

Nish Tahir's tutorial on how decision models like Jev work: mask an LLM's vocabulary to the option tokens, read the softmax as a probability, then measure calibration and add temperature scaling, using Qwen3-1.7B.

A revenue-focused AI product manager that watches growth metrics across your data, investigates why numbers moved, and answers questions in Slack, over MCP or in its own charts. Self-hosting in your own cloud is offered alongside a free trial.

RuntimeWire's report on Cloudflare extending its Clef decision models to audio and video: Clef-omni returns structured decisions from text, images, audio and video in one API call, while Clef-flash gets cheaper and its hosted context window shrinks.

Sayora's AI receptionist replaced production RAG with Jev for page picking and measured the change on 168 caller questions: wrong pages handed to the LLM fell from 33% to 3%, and the right page ranked first 88% of the time against 70%.

A social-scheduler founder's accuracy test of Jev: four yes/no questions over 522 captions, then the same questions for Fable on a random 100. Agreement reached 94/100 on promotion, and all 520 scored captions cost $0.011.

Tenuo's engineering write-up of safe-upgrade, a dependency-upgrade agent built on Jev, LangGraph and Tenuo: Jev turns repository evidence into typed probabilistic choices, the graph routes on confidence, and each action gets a short-lived task-scoped warrant.

Anton Zagrebelny's analysis: Jev's free output tokens and confidence-per-decision design push products toward two-model architectures and quietly break the way AI credits and entitlements are metered.

An independent playground for TypeSafe's Jev with clearly labeled modes: load Choice, Noul or Score examples for support routing, urgency and frustration, or run live calls against the real API when the page says live access is available.

A free browser-based Texas Hold'em game where Quick Play seats you against five AI poker bots, or you can build a custom table for two to six players and share its link. Supported opponents include Equity Rules, TypeSafe Jev and configurable LLM bots.

A Convex page for comparing models that answer verified multiple-choice questions about Convex, scoring them with source-weighted results and showing latency, cost and inspectable outcomes alongside the scores.

A walkthrough that pairs Unsiloed document extraction with Jev: extracted medication rows and their citations become Jev's state, one Noul question per row flags implausible or obscured doses, and the results decide which fields need review.

Viorant hosts TypeSafe's Jev as a remote MCP connector any MCP client can call: six tools cover typed decisions and compositions — jev_noul, jev_choice, jev_score, jev_verify, jev_screen and jev_find — with 2,500 Vio Credits in the Free Plan.

Yappy's macOS computer-use agent runs every step through Jev: the click, field and scroll are chosen from typed decisions over an accessibility text table, with no screenshots sent and a chat model called only when a step needs words.

elvex's Doyle Irvin walks through an internal prototype that wired Jev into the company's agent harness, from ambient classification while typing to attention filtering and risk-aware approvals; none of it is in production.

Trust middleware for Jev decision APIs: it logs every call, measures calibration in your domain as outcomes arrive, discounts stated confidence to what it is worth there, and signs the log for outside recomputation.

A DeepSeek Harness auto mode hosted on GitLab: full-access automation guided by TypeSafe System One at every end prompt, MIT-licensed and created in September 2026.

A third-party playground and API host for Jev: acceptance-style decision runs with readymade use-case templates, a model menu spanning Jev, Clef, Laya, and others, and 20 welcome credits to try it.