# Jev in production

Named products and companies running TypeSafe AI's Jev in production, or shipping a Jev-powered feature to real users. Drawn twice a day from TypeSafe's `#show-and-tell`, GitHub search and X posts found through web search. Every number is the source's claim, not a measurement of mine.

## In production

### New 2026-09-20
- **Clean Code Review** — https://clean-code-review.vercel.app — Judges each pull request file against Clean Code rules using Jev, answering every calibrated-probability question in about half a second per call. — [3/5] — #coding-agent #eval — via github — _added 2026-09-20_
- **any-auto** — https://github.com/jjyr/any-auto — Auto-approval pipeline for Antigravity and Pi coding agents lets Jev act as the AI reviewer backend that assesses tool-call risk before approving. — [3/5] — #coding-agent #guardrails — via github — _added 2026-09-20_
- **Juspay NeuroLink** — https://neurolink.ink — Decide primitive returns calibrated boolean/choice/score judgments via Jev; powers Tara, Yama and Clairvoyance products at Juspay in production. — [4/5] — #sdk #routing #classification — via github — _added 2026-09-20_
- **Light-Whisper** — https://github.com/sypsyp97/light-whisper — Optional Jev classification decides whether to skip the AI polishing step in this shipped Windows dictation app, shown with per-result processing times. — [3/5] — #classification #media — via github — _added 2026-09-20_
- **Sift** — https://github.com/bohutang/sift — Chrome extension labels every X post into six categories plus AI-written and off-topic flags using live Jev calls, vendor claims about $0.00003 per post. — [3/5] — #classification #browser — via github — _added 2026-09-20_
- **AnyFilter** — https://github.com/anyfilter/anyfilter — Chrome extension hides X posts matching user-described rules using live Jev classification calls per post, with a panel to undo misclassifications. — [3/5] — #classification #browser — via github — _added 2026-09-20_
- **av (Pixel ML)** — https://agentic.video — Shipped video-intelligence CLI uses Jev to filter source relevance and rank scenes for its refined `av ask` question-answering command. — [3/5] — #classification #media #data — via github — _added 2026-09-20_
- **JevProxy** — https://jevproxy.com/ — Drop-in reverse proxy routes OpenAI/Anthropic agent calls through Jev before an LLM call. Vendor claims 65% lower agent cost, 90% less loop latency, sub-25ms decisions. — [3/5] — #routing #coding-agent — via discord — _added 2026-09-20_
- **FakeCatch** — https://chromewebstore.google.com/detail/fakecatch-fake-review-det/jfkcpcmgneignanpnnggbfjljbpceelm — Chrome extension scores Amazon/Flipkart/Trustpilot reviews for authenticity using Jev, live on the Chrome Web Store. No numbers published. — [3/5] — #browser #classification — via discord — _added 2026-09-20_
- **Vettly** — https://vettly.dev/ — Content-moderation API for app developers; Jev runs parallel probabilistic signals across text, image and video with programmable policies. No numbers published. — [3/5] — #guardrails #classification — via discord — _added 2026-09-20_
- **eslint-plugin-jev (ShahriarBijoy)** — https://github.com/ShahriarBijoy/eslint-plugin-jev — ESLint plugin where each rule is a plain-English question Jev answers with a probability against a configurable threshold. No numbers published. — [3/5] — #coding-agent #classification — via github — _added 2026-09-20_
- **Jev-bot (venu)** — https://dm6rtlrn56ej8.cloudfront.net/ — Turns a plain-language idea into a Jev integration design: Jev evaluates a published research library, GLM 5.3 drafts the workflow and JSON. No numbers published. — [2/5] — #coding-agent #research — via discord — _added 2026-09-20_
- **JarvisCore (Prescott Data)** — https://github.com/Prescott-Data/jarviscore-framework — Jev drives agent decisions, subagent routing, model-tier selection, and RAG passage filtering. Maintainer claims 1.37s wall, $0.000074 for four concurrent live Jev calls. — [3/5] — #routing #rag #coding-agent — via github — _added 2026-09-20_
- **Awesome Jev (valentynkit)** — https://awesomejev.vercel.app/ — Directory of Jev projects where Jev itself classifies and reranks entries via 59 typed questions per repo. Author claims 323 projects, daily refresh. — [3/5] — #classification #data — via discord — _added 2026-09-20_
- **jev-use** — https://github.com/shitianfang/jev-use — Claude Code/Codex/pi plugin routes no-text-output agent steps to Jev instead of an LLM. Author claims p50 ~230ms, $0.02 per 1,000 judgments. — [3/5] — #coding-agent #routing — via github — _added 2026-09-20_
- **Revolve** — https://gitlab.com/porky11/revolve-agent — Open-source coding agent uses Jev to evaluate risky commands, set permission rules, flag broken instructions, and route external events. No numbers; author's daily driver. — [2/5] — #coding-agent #guardrails — via discord — _added 2026-09-20_
- **Alloy** — https://github.com/tlangridge/Alloy — Local multi-model CLI panel routes tasks between Claude, Codex, Gemini and Grok by complexity, model strength and quota via Jev. No numbers published. — [2/5] — #routing #cli — via github — _added 2026-09-20_
- **Community Signals (Joshua Poddoku)** — https://www.joshuapoddoku.com/community-signals/ — Jev adds precision to a live read of Bluesky/HN chatter about AI-generated code. Source release pending; no numbers published. — [2/5] — #classification #research — via discord — _added 2026-09-20_
- **Nola** — https://nola.sh/ — TypeScript superset (.tsi) with typed LLM inference; added Jev as a provider for Scale/Prob/Choice primitives alongside native TS types. No numbers published. — [3/5] — #sdk — via discord — _added 2026-09-20_
- **Enzyme** — https://github.com/byenzyme/enzyme — Local-first compiler over Markdown knowledge bases uses Jev for program synthesis, output in about 0.5s. Author claims 350x cost, 1000x speed vs frontier models. — [3/5] — #coding-agent #data — via github — _added 2026-09-20_
- **oko (bartlomein)** — https://github.com/bartlomein/oko — Jev reranks locally found matches so coding agents read less of the repo. Author's 108-session pilot claims up to 46% fewer agent tokens, 18% less time with a warm cache. — [4/5] — #coding-agent #search — via github — _added 2026-09-20_
- **musecases.app** — https://musecases.app — Live Jev-based auto-moderation of user submissions for harmful content and quality. No numbers published. — [3/5] — #guardrails #classification — via discord — _added 2026-09-20_
- **Venice (AskVenice)** — https://x.com/askvenice/status/2101095644467511578 — Jev added as an inference option on Venice's API: send app state and typed questions, get back a probability, choice, or score. Vendor beta launch, no numbers published. — [2/5] — #sdk — via x — _added 2026-09-20_
- **Fieldnote (ClawScience)** — https://fieldnote.clawscience.com/ — Live web tool pairs Jev with deterministic mathematics as a semantic physics/math solver. Author claims it solved 3 MIT Integration Bee 2026 qualifying questions. — [2/5] — #education #research — via discord — _added 2026-09-20_
- **OWLENA** — https://owlena.vercel.app/en/case-study — Vietnam warehouse-automation vendor uses Jev for structured decisions inside its picking and routing engine. Vendor claims 43.6% less picking travel time on one live fulfillment wave. — [3/5] — #robotics #ops — via discord — _added 2026-09-20_
- **hush (emreozyoruk)** — https://github.com/emreozyoruk/hush — GitHub Action triages issues with four calibrated Jev questions per issue, staying silent below maintainer-set confidence thresholds. Author claims ~300ms, ~$0.00002/issue, running on its own repo. — [3/5] — #ops #classification — via github — _added 2026-09-20_
- **openpoke-meets-jev (0xShin0221)** — https://github.com/0xShin0221/openpoke-meets-jev — Fork of the open-source Poke reimplementation moves email screening, execution-agent guardrails, and search reranking from Sonnet to Jev. Author ran an A/B and an adversarial injection test; no numbers quoted. — [3/5] — #guardrails #search #coding-agent — via github — _added 2026-09-20_
- **Pi-Saver (amazingjoe)** — https://github.com/amazingjoe/pi-saver — Pi Coder extension where Jev dynamically prunes irrelevant conversation history per request while keeping full history locally. Author claims 75%+ less context. — [2/5] — #coding-agent #data — via github — _added 2026-09-20_
- **jev-usher (cvsgireesh)** — https://github.com/cvsgireesh/jev-usher — Jev-powered model routing and recoverable context filtering for Claude Code, with a local UI comparing answers, tokens and timing. No numbers published; early stage. — [2/5] — #routing #coding-agent — via github — _added 2026-09-20_
- **JevRouter.nvim (Mawfyy)** — https://github.com/Mawfyy/jev-router.nvim — Neovim plugin uses Jev to classify coding-assistant intent and route to file-read, edit, terminal, or Q&A handlers. Author claims under 200ms per decision. — [2/5] — #coding-agent #routing — via github — _added 2026-09-20_
- **jev-align (Sutro)** — https://github.com/sutro-sh/jev-align — CLI finds uncertain examples, collects human labels, and runs GEPA to improve a Jev classifier. No numbers published. — [2/5] — #classification #eval — via github — _added 2026-09-20_
- **SiftRank** — https://github.com/noperator/siftrank — Semantic ranking CLI added Jev as a provider alongside OpenAI-compatible APIs for ranking prose, code, and support tickets. No numbers published. — [2/5] — #search #cli — via github — _added 2026-09-20_
- **jegrep** — https://github.com/can1357/jegrep — Natural-language semantic grep answers a calibrated yes/no probability per file with Jev. Author claims $0.01-0.03 per search across a few thousand files. — [3/5] — #search #cli — via github — _added 2026-09-20_
- **jevcache** — https://github.com/hyperspaceai/jevcache — Local decision cache memoizes Jev calls keyed on model/schema/state so repeats are free and deterministic in CI. No numbers published. — [2/5] — #ops #cli — via github — _added 2026-09-20_
- **Grok Bot Jev Router** — https://github.com/Bodila51/grok-bot-jev — Jev classifies Grok Bot requests before research, browser, or retry work to reuse cache, stop retries, or ask a human. No numbers published. — [2/5] — #routing #browser — via github — _added 2026-09-20_
- **Pi-Bifrost** — https://github.com/iamaamir/pi-bifrost — Native model router for the Pi agent; an optional Jev backend selects a configured routing tier before generation starts. No numbers published. — [2/5] — #routing #coding-agent — via github — _added 2026-09-20_
- **Astra (MatrixOrigin)** — https://github.com/matrixorigin/Astra — Self-hosted agent runtime integrates Jev as an optional judgment layer over budgeted context pipelines, with a public research paper. No numbers quoted. — [3/5] — #coding-agent #research — via github — _added 2026-09-20_
- **Jev Desktop** — https://github.com/yikangy873-gif/jev-desktop — Jev selects an operation and target inside Codex Computer Use sessions for browser tabs and macOS apps. No numbers published. — [2/5] — #browser #coding-agent — via github — _added 2026-09-20_
- **jev-mcp** — https://github.com/burnigtm/jev-mcp — MCP server gives Cursor and Codex nine Jev tools for routing, patch review, claim verification, and content screening. No numbers published. — [2/5] — #mcp #coding-agent — via github — _added 2026-09-20_
- **typesafe-mcp** — https://github.com/PyModel/typesafe-mcp — Stdio MCP server wraps the TypeSafe API with automatic 429/529 retries so Claude Code and Codex can call Jev directly. No numbers published. — [2/5] — #mcp #sdk — via github — _added 2026-09-20_
- **SmartMoney-Cub** — https://github.com/myc0576/SmartMoney-Cub — Read-only trading journal harness uses Jev for typed noul/choice/score judgments inside a reproducible finance benchmark; no order execution. No numbers published. — [2/5] — #finance #eval — via github — _added 2026-09-20_
- **jev-use (savka777)** — https://github.com/savka777/jev-use — Voice-driven macOS computer use: Jev picks the next action from the Accessibility tree, no screenshots. Author claims about 0.3-1.5s per step. — [2/5] — #browser #demo — via github — _added 2026-09-20_
- **Hermes Jev Skills** — https://github.com/kerpopule/hermes-jev-skills — Jev handles model routing, memory compaction, and skill selection for Hermes/Claude Code/Codex agents. Author claims roughly 0.4s per decision. — [3/5] — #routing #coding-agent — via github — _added 2026-09-20_
- **Jev X Sentiment Analysis** — https://github.com/brainstormity/Jev-X-Sentiment-Analysis — Crypto decision terminal: Jev issues a buy/sell/hold call from live price data and pre-processed tweet sentiment across up to 1,000 tweets. No numbers published. — [2/5] — #finance #classification — via github — _added 2026-09-20_
- **Jev Recall (samdotmak)** — https://github.com/samdotmak/jev-recall — Live demo reranks AI-assistant memories with one calibrated yes/no per memory via Jev instead of embedding similarity. No numbers published. — [2/5] — #search #rag — via github — _added 2026-09-20_
- **Pi Jev Guard (zszz3)** — https://github.com/zszz3/Pi-Jev-Guide — Pi coding-agent plugin where Jev judges destructive commands, unauthorized exfiltration, and task drift before allowing tool calls. No numbers published. — [2/5] — #guardrails #coding-agent — via github — _added 2026-09-20_
- **Jev DSH Decision (Devin-AXIS)** — https://github.com/Devin-AXIS/jev-dsh-decision — Plugin for DeepSeek Harness and iPolloWork where Jev picks tools, skills, and task owners and scores output quality. No numbers published. — [2/5] — #routing #coding-agent — via github — _added 2026-09-20_
- **is-malicious (luantak)** — https://github.com/luantak/is-malicious — CLI scans a codebase's source, config, build, and CI files with Jev for hidden or data-stealing behavior before you run it. No numbers published. — [2/5] — #security #coding-agent — via github — _added 2026-09-20_
- **jev-code (rhighs)** — https://github.com/rhighs/jev-code — TypeScript coding-agent CLI and SDK where Jev picks one option from a typed router or decision tree; code runs the effects. No numbers published. — [2/5] — #coding-agent #sdk — via github — _added 2026-09-20_
- **jev-seo** — https://github.com/AkashPriyadarshii/jev-seo — Free SEO/GEO CLI and MCP server where Jev does semantic intent classification and AI-visibility scoring in place of Semrush/OpenSEO. No numbers published. — [2/5] — #search #mcp #classification — via github — _added 2026-09-20_
- **Transcript Lens** — https://github.com/sensahin/transcript-lens — Turkish-language YouTube transcript explorer where Jev classifies chapters, claims, opinions, and topic matches without checking factual accuracy. No numbers published. — [2/5] — #classification #media — via github — _added 2026-09-20_

### New 2026-09-19
- **Jev Codex Router** — https://github.com/0xNatoshi/jev-codex-router — Per-turn model/reasoning routing for Codex, classified by Jev. Author claims ≈60% cost savings vs a full-frontier baseline on a 7-day replay of 237 real turns. — [4/5] — #routing #coding-agent — via github — _added 2026-09-19_
- **Supercov (Supercorp)** — https://supercov.com — Scores source-code quality with Jev inside a coding-agent test/coverage loop. Vendor claims about $0.01 per megabyte of source scored. — [3/5] — #coding-agent #eval — via discord — _added 2026-09-19_
- **Prokop** — https://prokopai.dev — Experimental opt-in setting where Jev selects relevant memories and skills for each response in a persistent coding workspace. No numbers published; off by default. — [2/5] — #coding-agent #rag — via discord — _added 2026-09-19_
- **Kyotofin tax-doc-classifier** — https://github.com/kyotofin/tax-doc-classifier — Classifies pages across 261 IRS forms in a production tax pipeline. Author claims 100% accuracy, $0.001/page, 34x cheaper, 6x faster than prior LLM. — [4/5] — #classification #finance — via github — _added 2026-09-19_
- **OpenAgents Coder** — https://openagents.com — Terminal coding agent whose Jev "System One" judges respond/clarify/end and pass/retry/stop each turn before slower generation runs. — [3/5] — #coding-agent #guardrails — via discord — _added 2026-09-19_
- **typesafe-computer-use** — https://github.com/awlevin/typesafe-computer-use — Mac computer-use agent: Jev picks the next click from OCR'd text. Author claims about $0.0002 per step. — [3/5] — #browser — via github — _added 2026-09-19_
- **classifier.dev** — https://classifier.dev — Public zero-shot text classification API; Jev is one tier in its live classification and LLM-fallback chain. No numbers published. — [3/5] — #classification — via discord — _added 2026-09-19_
- **EZCollegeApp (via jev-browser-use)** — https://github.com/wy-coliney/jev-browser-use — Jev drives clicks/navigation while Codex verifies, in production browser workflows. Author claims 5-10x faster browser operations at EZCollegeApp. — [3/5] — #browser #coding-agent — via github — _added 2026-09-19_
- **LangWatch Instant Evals** — https://langwatch.ai — On-demand evals over a product's whole production trace history. Vendor claims 97% agreement with human labels on 300 support chats and $0.32 per 10k conversations. — [4/5] — #eval — via discord — _added 2026-09-19_
- **Near Here** — https://nearhere.events — Event-listing validation. Published methodology against Mistral Small 4 and Gemini 3.5 Flash-Lite: claims up to 5.7x faster, 98% cheaper, 12 points more accurate. — [4/5] — #eval #data — via discord — _added 2026-09-19_
- **Swamp (Watson Labs)** — https://blog.watson-labs.co.uk/typesafe-ai-alert-fatigue/ — Jev screens alerts before routing in a live alert-triage system. Author claims 91% cheaper and 14.8% faster triage. — [4/5] — #ops #routing — via discord — _added 2026-09-19_
- **Vercel eve agent triage** — https://x.com/danmana/status/2100545412780220877 — Failure-reason classification over real agent session logs. Author claims 320 sessions, 2M tokens, classified in 14 seconds for $0.08. — [4/5] — #classification #eval #coding-agent — via x — _added 2026-09-19_
- **Pinecone Cultivar** — https://github.com/pinecone-io/cultivar — Grades agent skills and docs in Pinecone's eval harness. Maintainer claims 18x faster and about 38x cheaper than Claude for the grading step. — [4/5] — #eval #coding-agent — via github — _added 2026-09-19_
- **Fuel** — https://x.com/ashleyhindle/status/2100553892639473951 — Classifies the first message of a chat to pick the model tier. Author claims 3x faster and 65% cheaper on their benchmark. — [4/5] — #routing #classification — via x — _added 2026-09-19_
- **Maxfusion** — https://x.com/vladdubchak_x/status/2100870244004683886 — Scores every shot of scraped competitor video ads and rates each ad out of 100. Author claims 450+ ads in under 3 minutes, about 1.5s per shot. — [4/5] — #media #classification — via x — _added 2026-09-19_
- **AATF AI News Daily** — https://news.aatf.ai/ — Article-relevance gate inside a 50+ agent pipeline. Author claims 69 articles sorted in 0.7s at near-zero cost. — [4/5] — #classification #coding-agent — via discord — _added 2026-09-19_
- **glutenornot** — https://github.com/amr05008/glutenornot.com — Gluten flagging from labels and menus. Author benchmarked Jev against Opus on 1,000 labels: equal accuracy, $0.03 vs $8.45, 0.17s vs 2.89s median. Production still on Opus at posting. — [3/5] — #classification #health #eval — via github — _added 2026-09-19_
- **Atisbo** — https://www.atisbo.dev/ — Classifies customer feedback as problem, idea, question or noise in a live feedback-to-roadmap pipeline, dogfooded daily. Head-to-head against their LLM classifier posted without final numbers. — [3/5] — #classification #data — via discord — _added 2026-09-19_
- **RuleRaven** — https://github.com/ddalcero/ruleraven — Kubernetes incident-triage controller, running on the author's company cluster. Jev classifies ambiguous signals beside deterministic rules. No numbers. — [3/5] — #ops #classification — via github — _added 2026-09-19_
- **Slaide** — https://getslaide.com/ — Intent gate, scope enforcement and fabrication guard in a live Markdown-presentation product. No numbers. — [3/5] — #guardrails #classification — via discord — _added 2026-09-19_
- **Everyframe Try-on** — https://tryon.everyframe.studio/ — Picks matching pieces to complete an outfit in a live AI try-on app. Author claims about 420ms per styling decision. — [3/5] — #classification #demo — via discord — _added 2026-09-19_
- **BlackVeil Security** — https://blackveilsecurity.com/ — Integrated into a DNS and email security scanner with NIST-aligned checks. No numbers. — [3/5] — #security — via discord — _added 2026-09-19_
- **Empryo** — https://empryo.com — Five decision points in a coding harness, benchmarked against frontier models in a public write-up. Founder calls it a positive experiment. — [3/5] — #coding-agent #eval — via discord — _added 2026-09-19_
- **Lectio** — https://uselectio.com — Glossary term-sense selection and idiom flagging in a translation pipeline. Speed comparison posted, numbers in the linked test only. — [3/5] — #writing #classification — via discord — _added 2026-09-19_
- **Monti tariff triage** — https://www.linkedin.com/pulse/testing-ai-tariff-triage-cost-speed-accuracy-matthew-hartman-gqe3c/ — Tariff-classification triage on real trade-compliance data. Author reports Jev beat Sonnet on his classification set. — [3/5] — #classification #finance — via discord — _added 2026-09-19_
- **Peakflo** — https://getpeakflo.com — General-ledger invoice coding on the fly inside an AP/AR automation product. No numbers. — [3/5] — #finance #classification — via discord — _added 2026-09-19_
- **Jev Reviewer** — https://jevreviewer.xera.ac — Browser tool for systematic-review data extraction: Jev locates verbatim-quoted answers with page/paragraph against RoB 2, ROBINS-I, QUADAS-2 or TIDieR templates. No numbers published. — [3/5] — #research #health #browser — via discord — _added 2026-09-19_
- **duckdb-jev** — https://github.com/colliber/duckdb-jev — DuckDB extension: jev_choice/jev_score SQL functions call Jev per row, returning a typed SQL column instead of generated text. No numbers published. — [3/5] — #data #sdk — via github — _added 2026-09-19_
- **oxlint-plugin-jev** — https://github.com/wobsoriano/oxlint-plugin-jev — Oxlint plugin sends each matched function, call or JSX element to Jev as a yes/no question and flags matches over a configurable cutoff. No numbers. — [3/5] — #coding-agent #classification — via github — _added 2026-09-19_
- **JevLint** — https://jevlint.com — npm CLI that sends source files to Jev for semantic rule checks, reporting a per-file violation probability against an 80% default threshold. No numbers. — [3/5] — #cli #coding-agent — via discord — _added 2026-09-19_