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 706,048-parameter form-filling scorer from Cua AI: given a UI element and candidate options it returns one probability per option, including check, click, and skip actions.
A dependency-free JavaScript library that maps uploaded CSV columns onto your destination schema: a strict normalized-name pass first, then one Jev boolean per remaining source-destination pair plus a guard question per column, leaving uncertain fields visible instead of guessed.
A browser extension plus Cloudflare Worker that scores search results before you click: Jev answers six questions per page (relevance, depth, SEO filler, sales page, page type, key passage) and code combines them into a 0-100 badge that re-ranks Google, Bing, DuckDuckGo and 104.com.tw.
Alpha playground that answers plain-English questions over SQLite by composing small Jev decisions -- table, column, comparison and operand selection -- into inspectable typed plans, deterministic SQL and previewed guarded writes. Bundles Cyberchase, DESE and Moneyball databases.
Zero-dependency Python CLI that routes small semantic judgments to Jev for agents and CLI pipelines: choose (Choice), check (Noul), score (ordered ratings), batch (1-32 questions in one request) and purify (verbatim passage extraction). GPL-3.0 licensed.
A Jev harness for PostgreSQL: 41 semantic operators plus a native Rust extension evaluate row-level predicates with Noul questions and keep the returned probabilities for explicit decision thresholds, leaving joins and aggregation to SQL.
grep, sort, cut and uniq that match by meaning: each line whose Jev Noul clears the threshold survives grev, sibling filters route by Choice and rank by Score, and the output is always your own input, never generated text.
Jag, a Jev-inspired visual grounding model: a Qwen3.5-0.8B backbone plus a lightweight MLP predicts normalized cxcywh bounding boxes for an image and referring expression in a single forward pass, removing autoregressive coordinate generation from the base model.
Grounded information extraction library: deterministic code proposes candidate spans with exact offsets and Jev answers one Choice per span plus sentence-level Nouls, so every extraction is a substring of the source. Ships a bilingual benchmark and paper.
Zero-shot named entity recognition library on Jev: text is split at punctuation, candidate windows are offered as one Choice question per entity type, then verified and boundary-polished by further Choice calls. Types are defined in plain words; no training data or GPU is needed.
Project memory for Claude Code, Cursor, and Codex. Jev decides whether a turn is a decision, rule, bug, or injection, and code writes or supersedes a line in JEVMEM.md.
The index of TypeSafe's worked examples, from parallel questions and reranking to guardrails, date extraction, and self-consistency, each with datasets and measured results.
A hand-written Go client for the System One API, built against the vendored OpenAPI 3.1 specification with no third-party dependencies and typed Noul, Choice, and Score questions.
A Home Assistant integration that turns Jev questions about entity state into sensors and automation actions, with an Assist conversation agent and a daily token budget.
A small demo app that fetches a URL, turns the page into clean Markdown, and asks Jev ten bounded Choice questions about what a first-time visitor is likely to understand.
A CLI that reranks, filters, or extractively compresses JSON retrieval results by asking Jev separate Noul questions, then lets Rust apply the order and thresholds, reading stdin and writing stdout.
SQLite C extension that exposes Jev as SQL functions (jev_noul, jev_choice, jev_score) so queries can classify states such as ticket bodies directly, with only libcurl as a dependency.
WXT Chrome extension that grades readable page sections with Jev through Vercel AI Gateway, giving each section ten rubric scores and each page an inspectable A-E grade.
Text-only classifier for 261 federal tax forms that sends each PDF page's text to Jev, which returns a probability over forms and page kinds behind a confidence gate; no model is trained.
An observable stealth game where Jev supplies guards' split-second judgments while deterministic code owns the 30 Hz simulation, validates every proposal, and exposes each decision in a Decision Lens.
A browsable atlas of 1,018 AI papers from August 2025 to August 2026, sorted into 24 topic collections with labs, citations, and code links, plus a reproducible cost benchmark for summarizing 1,000 papers.
News analysis arguing Jev's refusal to generate text is the feature, covering the $40M seed, the OpenAI-veteran founding team, and what structured, thresholdable decisions mean for agent pipelines.
An independent measurement spanning 5,477 test rows and 34.1M input tokens, comparing one direct Jev question per row with 12-14 scored dimensions fitted to local labels.
A maintained list of jev-1.13's known failure modes, literal reading, unreliable counting, dates as text, indirection, context rot, and contradictory criteria, each with a guardrail.
A two-stage extraction cascade: a mini model extracts, a Noul battery verifies each field in one request, and a 0.7 gate escalates to a reasoning model, sitting on the cost/quality frontier.
Decide which of 450 candidate product pairs from two beer catalogues describe the same entity: one three-level Score per pair splits 40 merges, 50 curator reviews, and 360 non-matches.
Extract dates by asking seven Choice questions about a date's shape and parts, then resolving them in code: five of six examples auto-accepted and missing dates flagged at 0.46 confidence.
Turn wine tasting notes into 67 numeric columns for a CatBoost regressor: five rounds of reading its own errors cut held-out RMSE from 3.09 to 1.77 across 800 unseen reviews.
Regex finds candidate emails, phone numbers, and amounts; Jev picks the requested span so code copies a verbatim value that cannot be invented, with no free-text extraction required.
An agent-first CLI for System One: ask, rank, extract, verify, screen, find, and decide from the shell, with an installable agent skill. Published on PyPI.
A maintained list of jev-1.13's known failure modes, literal reading, unreliable counting, dates as text, indirection, context rot, and contradictory criteria, each with a guardrail.
Most Jev accuracy problems are state problems. What belongs in a state, what is a pre-made judgment to leave out, and why naming the slice you ask about matters more than the wording.
A Noul returns one number and no separate confidence. How to read the value, pin its boundary with criteria, split compound questions, and set a per-action threshold that matches the cost of being wrong.
A Score is a weighted position, not a ruler. Why the number can land between levels, why levels are judged separately, and how to combine several atomic scores in code instead of one crowded rubric.
Jev Practices & Patterns#score#primitives#scoring
Read article
Watch: Jev in action
Video walkthroughs and demos from the community.
YouTube: CJ from Syntax demos browser use, classification, code review, a model router, and a chat bot with no LLM.
YouTube: Ryan Vogel runs Jev on 100 and then 1,000 emails: category, priority, spam, and reply predictions.
YouTube: AICodeKing tests support routing, refund detection, prompt-injection resistance, and browser automation.
Discussions about Jev
Threads from Reddit and Hacker News about building with Jev.
Reddit: Reranking rebooking options with a decision model, ranked with reasons.
Hacker News: Pick Verify: a single decision at the warehouse pick station catches confusable picks.