Macbrow
Voice-controlled macOS agent: Gradium hears you, Jev picks a tool in ~300 ms, then AppleScript or jev-ultrafast drives Chrome.
25 builds
Voice-controlled macOS agent: Gradium hears you, Jev picks a tool in ~300 ms, then AppleScript or jev-ultrafast drives Chrome.
Open-source AI email triage for Gmail: Jev sorts 1,000 inbox messages into Needs reply, Updates, Promos, Sales, and Spam in about a minute for 3 cents.
Elvis Sun's newsjack-detector on Jev: 384 morning headlines scanned for $0.19 vs Opus 5's 4 for $0.77 — about 390× cheaper per story.
Speech practice app: talk for 30s on a topic and Jev scores pauses, filler words, repetitions, confidence, and clarity with actionable feedback.
Chrome extension: paste copied text once and Jev maps blocks to form fields, verifies against the source, and fills only confident matches.
Enter your X handle and Jev matches you with makers to meet — 3,675 profiles scored in 40 seconds for $3.74 by bio, interests, product, and location.
Self-hosted bank statement classifier: drop a PDF in the web UI or Telegram bot, Jev categorizes transactions, asks on low-confidence rows, and returns year-to-date Excel.
Jev scores 16 ski clips from frame descriptions and picks intro/outro; Computer Use drives CapCut to cut a 16s Phonk montage in under 5 minutes.
沐阳 scored ~2700 AIHOT stories from the last 7 days with Jev — about 2 minutes and $0.21 — for topic selection and sentiment.
A macOS menu bar computer-use tool: press Control–Space and Jev picks the next click or typed text from accessibility labels — no LLM, sub-second.
Voice-driven Drape experiment: Jev reads the transcript and current outfit, picks from the closet, and changes clothes in ~620ms at $0.0011 per decision.

Two local-first projects for semantic if decisions: kev (Qwen2.5-0.5B LoRA on a MacBook) and SemIf (open-model scoring on a 3090 or in the browser).

Open data, model, and recipe for typed decisions without Jev distillation — LoRA on Qwen3.5-9B with contrastive curation, scoring answer tokens via logits.
Open replica of Jev System One: a 1.9B Qwen3.5 model that returns option probabilities in one forward pass—no decoding, no JSON parse, 4 ms on GH200.
Vercel Labs experiment wiring TypeSafe Jev into json-render so Generative UI specs compose from your catalog in milliseconds — your components, actions, and design system.

Flue agent demo where TypeSafe Jev routes support requests to billing, technical, or human teams via Cloudflare Workers AI — no app API keys required.
Codex skill that lets Jev pick the next UI element from text candidates while Computer Use reads the screen and executes — text only, no screenshots.

Local Qwen3.5-0.8B visual inference that maps answers to fixed labels, prefills once, then scores candidates — a 200ms three-color sorting demo on a 16GB M4 MacBook.
Talk to a real Chromium window; Jev scores intent and targets in ~300 ms per word and Playwright acts—often before you finish the sentence.

Paste a website URL into Shipper; jev-1.13.0 plans a native mobile app and Shipper submits it to the iOS and Android app stores.
A Chrome extension that classifies LinkedIn posts in real time with TypeSafe and hides cringe, clickbait, and low-value content.
A Claude Code plugin that replaces compaction summaries with Jev scores: keep, stub, or drop each tool call, and leave kept text verbatim.

Ego lite runs 20 Amazon product decisions with Jev and DeepSeek Flash in 3.71s — the same 10/10 result as GPT-5.6 Sol at 54.45s.
Virat added Jev to the AI Hedge Fund so trading decisions run in seconds — about 100x faster and cheaper than an LLM.