# Jev radar

Curated twice a day from three sources: TypeSafe AI's `#show-and-tell` Discord channel (read in full from 2026-07-22), GitHub search, and X posts found through web search. Deliberately wide; use the search and filters.
Jev primitives: **Choice** (pick one), **Noul** (yes/no), **Score** (0 to 1). No temperature. No open-value extraction.

## Useful for the stack

### New 2026-09-20
- **jev-reranking** — https://github.com/carlaiau/jev-reranking — Benchmarks Jev zero-shot reranking against monoBERT and BM25 on TREC. — [4/5] — #eval #rag #search — via github — _added 2026-09-20_
- **clean-code-review** — https://github.com/frostney/clean-code-review — Judges PR files against Clean Code principles using Jev, summarized by Luna. — [4/5] — #coding-agent #eval — via github — _added 2026-09-20_
- **any-auto** — https://github.com/jjyr/any-auto — Auto-approves coding agent tool calls using Jev as a risk-reviewing backend. — [4/5] — #coding-agent #guardrails — via github — _added 2026-09-20_
- **super-jev** — https://github.com/Kevthetech143/super-jev — Experimental harness connecting evidence, Jev judgments, and verified actions. — [2/5] — #guardrails #sdk — via github — _added 2026-09-20_
- **SemDecide** — https://github.com/sharziki/semdecide — Unix CLI wrapping Jev for classification, scoring, filtering, and guarding. — [4/5] — #cli #classification #guardrails — via x — _added 2026-09-20_
- **TypeSafe on Neon** — https://github.com/andrelandgraf/typesafe-on-neon — Routes requests to Grok or GPT based on Jev's task classification. — [4/5] — #routing #classification — via x — _added 2026-09-20_
- **neo4jev** — https://github.com/jexp/neo4jev — Scores graph edges with Jev and beam-searches knowledge graph paths. — [3/5] — #rag #data — via x — _added 2026-09-20_
- **Prism** — https://github.com/irfndi/prism-liquidity-agent — Jev judges liquidity toxic flow and market stress in shadow trading mode. — [2/5] — #finance #classification — via x — _added 2026-09-20_
- **NeuroLink** — https://github.com/juspay/neurolink — One interface for 40 LLM providers, adds Jev-backed decide calls. — [4/5] — #sdk #routing #mcp — via github — _added 2026-09-20_
- **DocJev** — https://github.com/jerryjliu/docjev — Classifies and splits PDFs/DOCX into categorized documents using Jev. — [4/5] — #classification #rag #data — via github — _added 2026-09-20_
- **arc-cua** — https://github.com/shhivv/arc-cua — Executes bounded desktop UI subtasks click by click using Jev. — [3/5] — #browser #coding-agent — via github — _added 2026-09-20_
- **@receptron/laya** — https://github.com/receptron/laya — Runs the open Jev-compatible Laya decision model locally via ONNX. — [4/5] — #sdk #classification — via github — _added 2026-09-20_
- **JevGrep** — https://github.com/nassim-arifette/jevgrep — Finds code by behavior using Jev-scored search, returns exact excerpts. — [4/5] — #coding-agent #search #mcp — via github — _added 2026-09-20_
- **Jev Explained** — https://github.com/davila7/jev-explained — Interactive playground demonstrating Jev's three typed decision primitives live. — [2/5] — #education #demo — via github — _added 2026-09-20_
- **System One Harness** — https://github.com/HarnessRouter/SystemOneHarness — Turns any System One model into a confidence-gated agent loop. — [3/5] — #coding-agent #classification — via github — _added 2026-09-20_
- **ReadAloud (dasheng)** — https://github.com/wquguru/dasheng — Scores English read-aloud pronunciation errors locally using ASR plus Jev. — [3/5] — #education #media — via github — _added 2026-09-20_
- **arbiter** — https://github.com/0xBakeer/arbiter — Self-hosts typed-decision models like Laya behind a Jev-compatible API. — [3/5] — #sdk #classification — via github — _added 2026-09-20_
- **OpenJev (SiliconLabAI)** — https://github.com/SiliconLabAI/OpenJev — Approximates Jev by scoring each answer option in parallel calls. — [2/5] — #classification — via github — _added 2026-09-20_
- **dehydrator** — https://github.com/Arrmlet/dehydrator — Client-side tool search for LLM APIs, using BM25 or Jev. — [4/5] — #mcp #search #classification — via github — _added 2026-09-20_
- **jev-superpowers** — https://github.com/AkashPriyadarshii/jev-superpowers — Adds Jev typed decisions and package vetting to coding-agent workflows. — [3/5] — #coding-agent #guardrails — via github — _added 2026-09-20_
- **Jev Security Scan** — https://github.com/win4r/jev-security-scan — Reviews Skills and MCP code for injection risk using Jev. — [4/5] — #security #guardrails #mcp — via github — _added 2026-09-20_
- **Jevvy** — https://github.com/PanAchy/jevvy — Auto-approves harmless shell commands in coding agents via Jev. — [2/5] — #guardrails #coding-agent — via github — _added 2026-09-20_
- **JCR (Jev Capability Resolver)** — https://github.com/NiazMorshed2007/jcr — Finds deterministic commands in a capability tree using Jev search. — [3/5] — #coding-agent #mcp — via github — _added 2026-09-20_
- **Azdaja** — https://github.com/kubet/azdaja — Keeps source local and recurses into LLM calls, with optional Jev. — [2/5] — #coding-agent #sdk — via github — _added 2026-09-20_
- **Sift** — https://github.com/bohutang/sift — Labels every X post's substance and hides unwanted ones via Jev. — [2/5] — #classification #media — via github — _added 2026-09-20_
- **jevQL** — https://github.com/kylemclaren/jevql — Adds semantic jev() functions to plain SQL queries over Postgres. — [4/5] — #data #classification — via github — _added 2026-09-20_
- **OpenThai-SystemOne** — https://github.com/iapp-technology/openthai-systemone — Open Thai and English decision model mirroring Jev's API contract. — [2/5] — #classification #sdk — via github — _added 2026-09-20_
- **jev-sentinel** — https://github.com/harshwasan/jev-sentinel — Screens agent tool calls and replies for injection using Jev. — [4/5] — #guardrails #security #coding-agent — via github — _added 2026-09-20_
- **Jev vs ML** — https://github.com/QuicqDev/Jev-vs-ML — Benchmarks Jev against 11 classical ML pipelines on 8 datasets. — [4/5] — #eval #classification #research — via github — _added 2026-09-20_
- **jsort** — https://github.com/keltokhy/jsort — Sorts text by meaning via pairwise Jev comparisons, reports reliability. — [4/5] — #classification #eval #data — via github — _added 2026-09-20_
- **cascade-search** — https://github.com/zaidmukaddam/cascade-search — Parses queries locally, escalates only uncertain words to Jev. — [4/5] — #search #classification #routing — via github — _added 2026-09-20_
- **jev-cli (shaharia-lab)** — https://github.com/shaharia-lab/jev-cli — CLI turns Jev questions into exit codes for shell and CI. — [3/5] — #cli #classification #ops — via github — _added 2026-09-20_
- **snapjudge** — https://github.com/Micha0827/snapjudge — Reads typed-decision probabilities from local Qwen model logits, TypeSafe-compatible. — [3/5] — #classification #sdk — via github — _added 2026-09-20_
- **jev_jsonschema** — https://github.com/Kiln-AI/jev_jsonschema — Converts a JSON Schema into Jev questions and back into JSON. — [4/5] — #sdk #classification — via github — _added 2026-09-20_
- **Jev Cookbook** — https://github.com/nexibeo/jev-cookbook — Fifteen tested Jev recipes for triage, tagging, PII, and reranking. — [4/5] — #classification #rag #data — via github — _added 2026-09-20_
- **Jekhov** — https://github.com/jekhov/jekhov — Uses Jev only to pick Playwright targets, keeping flow deterministic. — [3/5] — #browser #coding-agent — via github — _added 2026-09-20_
- **SLO Router** — https://github.com/zeeshan8281/slo-router — Routes LLM requests to cheapest backend meeting a latency SLO. — [3/5] — #routing #eval — via github — _added 2026-09-20_
- **Jevals** — https://github.com/theyashwanthsai/jevals — Grades LLM/agent output with Jev, attaching a confidence to each. — [4/5] — #eval #classification — via github — _added 2026-09-20_
- **S18Share** — https://github.com/riteshverma/s18 — Multi-agent runtime with optional Jev-backed routing and skill matching. — [2/5] — #routing #coding-agent — via github — _added 2026-09-20_
- **av (Agentic Video Intelligence)** — https://github.com/PixelML/av — Video search CLI that uses Jev to filter and rank scenes. — [3/5] — #rag #media #classification — via github — _added 2026-09-20_
- **elons-job** — https://github.com/bugkiwi/elons-job — Chrome extension hides explicit X replies using Jev classification. — [2/5] — #browser #classification — via github — _added 2026-09-20_
- **typesafe-sdk-go** — https://github.com/Tangerg/typesafe-sdk-go — Official-style Go SDK wrapping the TypeSafe Jev API. — [2/5] — #sdk — via github — _added 2026-09-20_
- **awesome-jev-usecases** — https://github.com/aliaihub/awesome-jev-usecases — Sourced patterns and pitfalls collection for building with Jev. — [3/5] — #research — via github — _added 2026-09-20_
- **mcp-ui-poc** — https://github.com/iamadi11/mcp-ui-poc — Jev decides UI widget shape, LLM only writes when uncertain. — [3/5] — #mcp — via github — _added 2026-09-20_
- **pi-quiet-ask** — https://github.com/HyunjunJeon/pi-quiet-ask — Jev-backed rule packs gate destructive commands and secret leaks in pi. — [4/5] — #guardrails #security #coding-agent — via github — _added 2026-09-20_
- **pi-jev-router** — https://github.com/mejiasd3v/pi-jev-router — Jev picks model and reasoning effort once per session via AI Gateway. — [4/5] — #routing #coding-agent — via github — _added 2026-09-20_
- **jevlogs** — https://github.com/reachjalil/jevlogs — Scores OpenTelemetry logs with Jev before routing to expensive LLM analysis. — [4/5] — #ops #routing #data — via github — _added 2026-09-20_
- **typesafe-skill-router** — https://github.com/DECRUX9812/typesafe-skill-router — Jev names the one relevant skill to load before the model call. — [4/5] — #routing #coding-agent — via github — _added 2026-09-20_
- **what-is-jev** — https://github.com/g0runmezadam/what-is-jev — Independent research site cataloging 947 rubric-scored Jev repositories. — [2/5] — #research — via github — _added 2026-09-20_
- **jev-review (MaxIvanyshen)** — https://github.com/MaxIvanyshen/jev-review — Code review filter tells coding agents where to look first. — [3/5] — #coding-agent — via github — _added 2026-09-20_
- **AIStock** — https://github.com/EthanAlgoX/AIStock — Multi-agent trading platform offers Jev as a decision backend. — [2/5] — #finance — via github — _added 2026-09-20_
- **jevmail** — https://github.com/fazlerocks/jevmail — Gmail triage sorts 1,000 emails a minute for 3 cents. — [4/5] — #classification — via github — _added 2026-09-20_
- **JevBench** — https://github.com/fstandhartinger/jevbench — Benchmark scoring Jev-class models on intelligence, calibration, speed, cost. — [3/5] — #eval — via github — _added 2026-09-20_
- **TypeSafe AI Playground** — https://github.com/TypeSafeAI/typesafe-playground — Community playground with 110 editable Jev classification and routing examples. — [3/5] — #classification #routing #education — via github — _added 2026-09-20_
- **OpenJev-Vision** — https://github.com/IamBusy/OpenJev-Vision — Open research toolkit answering multiple typed questions from one image encode. — [2/5] — #research #media — via github — _added 2026-09-20_
- **Jev Directory** — https://github.com/everyai-com/jev-directory — Directory of 1,300+ Jev builds with an MCP server and index. — [2/5] — #mcp — via github — _added 2026-09-20_
- **jev-skill-suggester** — https://github.com/win4r/jev-skill-suggester — Recommends which installed Claude or Codex skill fits a task. — [3/5] — #coding-agent #routing — via github — _added 2026-09-20_
- **invalidate** — https://github.com/chopratejas/invalidate — Flags stale agent memories when new evidence supersedes them. — [3/5] — #coding-agent — via github — _added 2026-09-20_
- **fast-browser-use** — https://github.com/APUS-AI-Lab/fast-browser-use — Local open-weight browser automation reproducing Jev's fast decision pattern. — [2/5] — #browser — via github — _added 2026-09-20_
- **jev-radar (everyinfra)** — https://github.com/everyinfra/jev-radar — Tracks the Jev ecosystem, rescanning 220+ documented builds every 3 hours. — [2/5] — #research #data — via github — _added 2026-09-20_
- **jevify** — https://github.com/fidecastro/jevify — Serves any logprob-capable LLM through Jev's typed-question API shape. — [3/5] — #sdk — via github — _added 2026-09-20_
- **jevwire** — https://github.com/Brainwires/jevwire — MCP server and Claude Code plugin gating tool calls with Jev. — [3/5] — #mcp #guardrails #coding-agent — via github — _added 2026-09-20_
- **jev-linkmap** — https://github.com/stas4000/jev-linkmap — Rebuilds a 566-page site's internal link map in 6 seconds. — [3/5] — #data — via github — _added 2026-09-20_
- **openvons** — https://github.com/genai-craft/openvons — Decision layer answering finite choices from text, image, or voice. — [3/5] — #classification #media — via github — _added 2026-09-20_
- **patdown** — https://github.com/tyler-dot-earth/patdown — Lints a codebase against fuzzy markdown rules using Jev as judge. — [3/5] — #coding-agent #eval — via github — _added 2026-09-20_
- **jev-rag-benchmark** — https://github.com/erendikmenn/jev-rag-benchmark — Measures whether Jev reranking actually improves RAG quality and cost. — [4/5] — #rag #eval — via github — _added 2026-09-20_
- **jev (BorisLeMeec)** — https://github.com/BorisLeMeec/jev — Claude Code plugin answers codebase questions without loading files into context. — [3/5] — #coding-agent — via github — _added 2026-09-20_
- **fable-jev** — https://github.com/imMamdouhaboammar/fable-jev — Sub-100ms Jev reflex layer for routing and context compaction. — [2/5] — #routing #coding-agent — via github — _added 2026-09-20_
- **jev-mail-classifier** — https://github.com/parth-kp/jev-mail-classifier — Config-driven inbox classifier tags, moves and flags mail via Jev. — [3/5] — #classification — via github — _added 2026-09-20_
- **findme** — https://github.com/marc2332/findme — Finds files by natural-language memory using beam search and Jev. — [2/5] — #search #coding-agent — via github — _added 2026-09-20_
- **hermes-jev-approvals** — https://github.com/anpicasso/hermes-jev-approvals — Jev reviews shell command approvals, 8.7x faster on 153 commands. — [3/5] — #guardrails #coding-agent #security — via github — _added 2026-09-20_
- **OpenSourceJev** — https://github.com/sabeel111/OpenSourceJev — Turns a local Qwen model into a sub-100ms decision engine. — [2/5] — #research — via github — _added 2026-09-20_
- **jev-fn (aaazzam)** — https://github.com/aaazzam/jev — Decorator turns a Python function signature into a typed Jev query. — [3/5] — #sdk — via github — _added 2026-09-20_
- **jevcal** — https://github.com/abhixhek/jevcal — Calibrates confidence thresholds and flags drift for typed decision models. — [4/5] — #eval — via github — _added 2026-09-20_
- **jev-recall** — https://github.com/samdotmak/jev-recall — filters AI-assistant memories by one calibrated yes/no per item, not embeddings — [3/5] — #coding-agent — via github — _added 2026-09-20_
- **Pi-Jev-Guide** — https://github.com/zszz3/Pi-Jev-Guide — Pi coding-agent guardrail plugin with configurable rules, redaction, stall detection — [3/5] — #guardrails #coding-agent — via github — _added 2026-09-20_
- **jev-arena (NanmiCoder)** — https://github.com/NanmiCoder/jev-arena — CSV-driven side-by-side Jev vs DeepSeek comment-labeling comparison with saved reports — [2/5] — #eval #classification — via github — _added 2026-09-20_
- **sarvam-jev** — https://github.com/SAGAR-TAMANG/sarvam-jev — generation-free typed decisions on Sarvam's Indic tokenizer instead of Qwen — [2/5] — #classification #research — via github — _added 2026-09-20_
- **jev-dsh-decision** — https://github.com/Devin-AXIS/jev-dsh-decision — Jev decision plugin for DeepSeek Harness, OpenCode and Codex Harness — [2/5] — #coding-agent — via github — _added 2026-09-20_
- **TypeLLM** — https://github.com/TypeLLM/TypeLLM — brings Jev-style typed decoding to open autoregressive models via constrained decoding — [3/5] — #classification #research — via github — _added 2026-09-20_
- **siftrank** — https://github.com/noperator/siftrank — LLM needle-in-haystack ranker adds a Jev provider option — [3/5] — #search — via github — _added 2026-09-20_
- **third-hand** — https://github.com/shhivv/third-hand — macOS menu bar assistant reads Accessibility tree, types and clicks via Jev — [2/5] — #robotics — 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 agents — [3/5] — #coding-agent #routing — via github — _added 2026-09-20_
- **openJev-verdict-2.0** — https://github.com/Heman10x-NGU/openJev-verdict-2.0 — open 150M model claims higher accuracy and better calibration than Jev — [3/5] — #eval #research — via github — _added 2026-09-20_
- **jev-skill (wuyoscar)** — https://github.com/wuyoscar/jev-skill — 90 catalogued Jev use cases and installable agent skills — [2/5] — #coding-agent — via github — _added 2026-09-20_
- **awesome-jev-gallery** — https://github.com/OmniJev/awesome-jev-gallery — curated papers, open reproductions and independent Jev evaluations — [2/5] — #research #eval — via github — _added 2026-09-20_
- **jev-use (savka777)** — https://github.com/savka777/jev-use — voice-controlled Mac computer use via Accessibility tree, no vision model — [3/5] — #robotics #media — via github — _added 2026-09-20_
- **LLM2Jev** — https://github.com/Yinsongxu/LLM2Jev — adapts local LLMs into Jev-style Choice, Score and Noul outputs — [3/5] — #sdk — via github — _added 2026-09-20_
- **pi-bifrost** — https://github.com/iamaamir/pi-bifrost — Pi model router with an optional Jev tier-selection backend — [3/5] — #routing #coding-agent — via github — _added 2026-09-20_
- **jegrep** — https://github.com/can1357/jegrep — semantic grep with calibrated per-path yes/no probabilities, no index — [3/5] — #search #cli — via github — _added 2026-09-20_
- **grok-bot-jev** — https://github.com/Bodila51/grok-bot-jev — Jev gates Grok Bot retries, caching and subagent spawn decisions — [2/5] — #guardrails #coding-agent — via github — _added 2026-09-20_
- **Astra** — https://github.com/matrixorigin/Astra — agent runtime with native Jev judgments and EXPLAIN ANALYZE for context — [3/5] — #coding-agent #eval — via github — _added 2026-09-20_
- **jev-desktop** — https://github.com/yikangy873-gif/jev-desktop — Jev picks the action inside Codex Computer Use sessions — [2/5] — #robotics #coding-agent — via github — _added 2026-09-20_
- **awesome-jev-zh** — https://github.com/yzfly/awesome-jev-zh — Chinese curated index with an independent-evaluation caveats section — [2/5] — #research #eval — via github — _added 2026-09-20_
- **jev-lint** — https://github.com/mizchi/jev-lint — flags function names, comments and tests that no longer match code — [3/5] — #coding-agent #eval — via github — _added 2026-09-20_
- **jev-as-a-judge** — https://github.com/danielgshea/jev-as-a-judge — benchmarks Jev as an agent-eval judge against three LLM judges — [3/5] — #eval — via github — _added 2026-09-20_
- **jev-mcp (burnigtm)** — https://github.com/burnigtm/jev-mcp — MCP server for coding-loop routing, review, verify and screen tools — [3/5] — #mcp #coding-agent — via github — _added 2026-09-20_
- **typesafe-mcp (PyModel)** — https://github.com/PyModel/typesafe-mcp — Go MCP server exposing Jev's Choice, Score and Noul as one tool — [2/5] — #mcp #sdk — via github — _added 2026-09-20_
- **Jevbridge** — https://github.com/tacticocc/Jevbridge — ACP and MCP adapter bridges Jev with Claude, Codex, Grok and OpenCode — [2/5] — #mcp — via github — _added 2026-09-20_
- **jev-bot** — https://github.com/venumadhav7484/jev-bot — turns an idea into a Jev integration design from a research library — [2/5] — #research — via github — _added 2026-09-20_
- **jevproxy** — https://jevproxy.com/ — reverse proxy claims 65% cost cut and 90% latency cut for agents — [2/5] — #ops #coding-agent — via discord — _added 2026-09-20_
- **FakeCatch** — https://x.com/CodeWithKolin/status/2101643875429580880 — Chrome extension scores review trust as real, generic, ad-like or short — [2/5] — #browser #classification — via x — _added 2026-09-20_
- **Vettly** — https://vettly.dev/ — content moderation API for text, image and video built on Jev — [2/5] — #security #media #classification — via discord — _added 2026-09-20_
- **eslint-plugin-jev** — https://github.com/ShahriarBijoy/eslint-plugin-jev — ESLint rules are plain-English questions Jev answers with a probability — [3/5] — #coding-agent — via github — _added 2026-09-20_
- **awesome-jev-tools** — https://github.com/v-modal/awesome-jev-tools — another curated index of public projects and patterns built on Jev — [2/5] — #research — via github — _added 2026-09-20_
- **Jeveloper** — https://jeveloper.dev/ — 100 example Jev use cases: routing, scoring, agents, support workflows — [2/5] — #education #coding-agent — via discord — _added 2026-09-20_
- **Jev on Venice API** — https://x.com/askvenice/status/2101095644467511578 — Jev added to Venice API in beta, no JSON parsing needed — [3/5] — #sdk — via x — _added 2026-09-20_
- **mefi-studio** — https://github.com/nateecho32-stack/mefi-studio — desktop AI workspace, Jev routes models by performance, speed and cost — [2/5] — #routing — via github — _added 2026-09-20_
- **jevfanity-api** — https://github.com/TickerDev/jevfanity-api — hosted profanity detector using Jev, no API key needed — [2/5] — #classification #security — via github — _added 2026-09-20_
- **jev-codex-pilot** — https://github.com/Charlyhno-eng/jev-codex-pilot — Codex web overlay picks model and reasoning depth, automated Kanban task sequencing — [2/5] — #coding-agent #routing — via github — _added 2026-09-20_
- **hunch** — https://github.com/carldaws/hunch — Ruby gem wraps Jev's three question types as chance, pick and rate — [3/5] — #sdk — via github — _added 2026-09-20_
- **pi-jev-wiki** — https://github.com/xAndreiLi/pi-jev-wiki — Pi package, agent-maintained project wiki holds only reasoning, not code-derivable facts — [3/5] — #coding-agent — via github — _added 2026-09-20_
- **oko** — https://github.com/bartlomein/oko — local search plus Jev reranking cuts what a coding agent has to read — [4/5] — #search #coding-agent #rag — via github — _added 2026-09-20_
- **jev-prune-kit** — https://github.com/CompleteTech-LLC-AI-Research/jev-prune-kit — capability-aware context-pruning installer for agent harnesses, experimental — [2/5] — #coding-agent — via github — _added 2026-09-20_
- **Jev Moderation Bot** — https://github.com/brainstormity/Jev-Moderation-Bot — Discord moderation with 4-stage escalation, audit logging, one-click pardon learning — [3/5] — #security #classification — via github — _added 2026-09-20_
- **jev-logtriage** — https://github.com/jyatesdotdev/jev-logtriage — Jev decides whether a batch of logs is worth acting on, confidence-gated — [2/5] — #ops #data — via github — _added 2026-09-20_
- **jevcache.sh** — https://jevcache.sh/ — memoizes Jev-class decisions so repeats are free and deterministic — [2/5] — #ops — via discord — _added 2026-09-20_
- **jev-java** — https://github.com/Olti1947/jev-java — idiomatic Java SDK for the Jev decision engine — [2/5] — #sdk — via github — _added 2026-09-20_
- **jev-builder** — https://collapseindex.github.io/jev-builder/ — browser form builds Jev requests from 34 templates, no JSON by hand — [2/5] — #browser — via discord — _added 2026-09-20_
- **Alloy** — https://github.com/tlangridge/Alloy — Jev routes tasks across Claude, Codex, Gemini and Antigravity by capability — [3/5] — #routing #coding-agent — via github — _added 2026-09-20_
- **opencode-plugin-variantizer** — https://github.com/Melivo/opencode-plugin-variantizer — Jev picks the cheapest capable OpenCode model and reasoning variant per task — [3/5] — #routing #coding-agent — via github — _added 2026-09-20_
- **enzyme** — https://github.com/byenzyme/enzyme — compiles Markdown-wiki reading rules into Jev queries, claims 350x cheaper than frontier — [3/5] — #classification #ops — via github — _added 2026-09-20_
- **jev-studio** — https://github.com/utk2103/jev-studio — MCP tools for Choice, Noul and Score plus prompt libraries and commands — [2/5] — #mcp — via github — _added 2026-09-20_
- **typesafe4s** — https://github.com/gruggiero/typesafe4s — type-safe Scala 3 SDK wrapping the Jev API — [2/5] — #sdk — via github — _added 2026-09-20_
- **Nola** — https://nola.sh/ — TypeScript superset adds Jev as a typed-inference provider alongside plain TS types — [2/5] — #sdk — via discord — _added 2026-09-20_
- **Jev 1.13 jaggedness (TypeSafe docs)** — https://docs.typesafe.ai/model-jaggedness/jev-1.13#math-and-numbers — TypeSafe's own list of known Jev weak spots, math first — [3/5] — #eval #research — via discord — _added 2026-09-20_
- **jarviscore-framework** — https://github.com/Prescott-Data/jarviscore-framework — multi-agent framework, Jev routes subagents and filters RAG passages by conflict/injection — [3/5] — #routing #rag #guardrails — via github — _added 2026-09-20_
- **pi-jev-sentinel** — https://github.com/harshwasan/pi-jev-sentinel — Pi agent extension screens tool calls and output for injection, scrubs secrets, pins task — [3/5] — #guardrails #security #coding-agent — via github — _added 2026-09-20_
- **jev-align** — https://github.com/sutro-sh/jev-align — CLI calibrates Jev to your own judgment criteria using GEPA — [3/5] — #cli #eval — via github — _added 2026-09-20_
- **jev-use** — https://github.com/shitianfang/jev-use — hands no-text-output agent steps to Jev, p50 ~230ms, ~$0.02 per 1,000 judgments — [3/5] — #coding-agent #classification — via github — _added 2026-09-20_
- **JevSwiftSDK** — https://github.com/NSStudent/JevSwiftSDK — unofficial Swift SDK, async/await, batching, zero dependencies, iOS/macOS/Linux — [3/5] — #sdk — via github — _added 2026-09-20_
- **revolve-agent** — https://gitlab.com/porky11/revolve-agent — coding agent uses Jev to gate risky commands, set permissions and flag instruction drift, still WIP — [2/5] — #coding-agent #guardrails — via discord — _added 2026-09-20_
- **JevFind** — https://github.com/Peu77/JevFind — semantic code search CLI, returns files, line ranges and confidence scores, author calls it a demo — [2/5] — #cli #search — via github — _added 2026-09-20_
- **jevrouter-ts** — https://github.com/rajdhakad9826/jev-router — TS router picks cheapest model tier per query using Jev classification — [2/5] — #routing #classification — via github — _added 2026-09-20_
- **JevNQL** — https://github.com/Adityakhalkar/JevNQL — natural-language database queries via Jev instead of SQL — [2/5] — #data #search — via github — _added 2026-09-20_
- **Jev CLI (vectorz)** — https://jevcli.vectorz.app/ — terminal CLI with skills and plugins for agent workflows — [2/5] — #cli #coding-agent — via discord — _added 2026-09-20_
- **Jev in production (iambraun)** — https://iambraun.com/jevreports/shipped/ — self-updating list of 18 products running Jev in production, scored 1-5 — [3/5] — #research — via discord — _added 2026-09-20_
- **jev-frontier-bench** — https://github.com/manjunathshiva/jev-frontier-bench — 200-decision benchmark of Jev vs five frontier models, calibration and cost data — [4/5] — #eval — via github — _added 2026-09-20_
- **Pi-Saver** — https://github.com/amazingjoe/pi-saver — Jev prunes Pi Coder conversation history per turn, claims 75%+ token cut — [3/5] — #coding-agent #ops — via github — _added 2026-09-20_
- **JevRouter.nvim** — https://github.com/Mawfyy/jev-router.nvim — Neovim plugin routes prompts to file, edit or terminal handlers via Jev, under 200ms — [3/5] — #routing #coding-agent — via github — _added 2026-09-20_
- **hush** — https://github.com/emreozyoruk/hush — GitHub Action triages issues with Jev, stays silent below a confidence threshold — [3/5] — #ops #classification — via github — _added 2026-09-20_
- **jev-usher** — https://github.com/cvsgireesh/jev-usher — model routing plus recoverable context filtering for Claude Code, local compare UI — [3/5] — #routing #coding-agent — via github — _added 2026-09-20_
- **openpoke-meets-jev** — https://github.com/0xShin0221/openpoke-meets-jev — moves OpenPoke's email screening, tool guardrail and reranking to Jev — [3/5] — #guardrails #classification — via github — _added 2026-09-20_
- **JevArena** — https://github.com/chenmingtang830/jevarena — open BYOK arena blind-votes Jev against other models acting as judge — [3/5] — #eval — via github — _added 2026-09-20_
- **Jev For Dummies** — https://github.com/brainstormity/Jev-For-Dummies — wraps Jev primitives as plain HTTP GET endpoints for newcomers — [2/5] — #sdk #education — via github — _added 2026-09-20_
- **jev4k** — https://github.com/pambrose/jev4k — Kotlin DSL and client for the Jev API — [2/5] — #sdk — via github — _added 2026-09-20_
- **code-compact-jev** — https://github.com/joa/code-compact-jev — Jev trims excess comments out of code — [2/5] — #coding-agent — via github — _added 2026-09-20_

### New 2026-09-19
- **Jev on Vercel AI Gateway** — https://vercel.com/changelog/typesafe-ai-jev-now-available-on-ai-gateway — Jev added to Vercel AI Gateway, one line to call it — [3/5] — #sdk — via discord — _added 2026-09-19_
- **Jev on Cloudflare AI Gateway** — https://developers.cloudflare.com/ai/models/typesafe/jev/ — Jev live on Cloudflare AI Gateway as a typed decision endpoint — [3/5] — #sdk — via discord — _added 2026-09-19_
- **flaviocopes Jev deep dive** — https://flaviocopes.com/jev/ — practitioner walkthrough of the API, primitives and basic usage patterns — [2/5] — #education — via discord — _added 2026-09-19_
- **jev-capability-atlas** — https://github.com/Zaious/jev-capability-atlas — evidence-based map of where Jev's calibration holds up or fails — [4/5] — #research #eval — via github — _added 2026-09-19_
- **jev-reviewer** — https://github.com/choxos/jev-reviewer — systematic-review data extraction, every answer a verbatim cited quote — [4/5] — #research #data — via github — _added 2026-09-19_
- **jev-column-race** — https://github.com/goodrahstar/jev-column-race — Jev vs Gemini labelling 1,000 reviews, 4.1x faster, 7x cheaper — [4/5] — #eval #classification — via github — _added 2026-09-19_
- **pi-jev-auto-mode** — https://github.com/jomatsu/pi-jev-auto-mode — Jev auto-approves Pi bash/write/edit calls, fails closed, 171 tests — [4/5] — #guardrails #coding-agent — via github — _added 2026-09-19_
- **winnow** — https://github.com/GhalebDweikat/winnow — calibrated context sieve for Claude Code, stubs blocks Jev judges unneeded — [3/5] — #coding-agent #ops — via github — _added 2026-09-19_
- **yoshi** — https://github.com/compozy/yoshi — Jev-judged context-pruning proxy for Claude Code and Codex, POC — [3/5] — #coding-agent #ops — via github — _added 2026-09-19_
- **pi-dcp** — https://github.com/Davidcreador/pi-dcp — dedups and prunes Pi coding-agent context, experimental Jev selection — [3/5] — #coding-agent #ops — via github — _added 2026-09-19_
- **pi-jev** — https://github.com/TheoOliveira/pi-jev — semantic tool router and skill finder for the Pi coding agent — [3/5] — #routing #coding-agent — via github — _added 2026-09-19_
- **pi-typesafe** — https://github.com/DevMortimer/pi-typesafe — batched Jev eval tool, terminal playground, shared client for Pi extensions — [3/5] — #eval #cli #sdk — via github — _added 2026-09-19_
- **stanley-code** — https://github.com/devagrawal09/stanley-code — Jev routes requests to deterministic coding workflows, falls back to an agent — [3/5] — #routing #coding-agent — via github — _added 2026-09-19_
- **blink** — https://github.com/ellipsis-dev/blink — Jev-driven file-system walkers search a codebase by natural language — [3/5] — #search #coding-agent — via github — _added 2026-09-19_
- **ErisLint** — https://github.com/Eriskii/ErisLint — Rust linter with Jev-judged rules for complexity, naming, comment value — [3/5] — #coding-agent — via github — _added 2026-09-19_
- **ts-browser-agent** — https://github.com/ndrezn/ts-browser-agent — LangChain browser agent using Jev as the tool-picking model — [3/5] — #browser #routing — via github — _added 2026-09-19_
- **ruby_llm-typesafe** — https://github.com/kieranklaassen/ruby_llm-typesafe — TypeSafe provider for RubyLLM, Noul/Choice/Score via structured output — [3/5] — #sdk — via github — _added 2026-09-19_
- **jev (Elixir/OTP)** — https://github.com/dannote/jev — GenServer replies to Jev, pattern-match the typed answer — [3/5] — #sdk — via github — _added 2026-09-19_
- **effect-questions** — https://github.com/saiashirwad/effect-questions — semantic judgment as an Effect-TS primitive: is, choose, rank, branch — [3/5] — #sdk #classification — via github — _added 2026-09-19_
- **Verdict (OpenJev)** — https://github.com/Heman10x-NGU/Verdict-open-jev — open 151M non-autoregressive decision model, calibrated uncertainty, WebGPU demo — [3/5] — #research #demo — via github — _added 2026-09-19_
- **litjev** — https://github.com/zhengxuyu/litjev — turns any Qwen checkpoint into a Jev-style typed decision layer — [3/5] — #research — via github — _added 2026-09-19_
- **jev-on-a-laptop** — https://github.com/rorshopping/jev-on-a-laptop — reproduces Jev's parallel constrained decoding on a 1.5B model locally — [3/5] — #research — via github — _added 2026-09-19_
- **open-jev (DiffusionGemma)** — https://github.com/JoshuaSP/open-jev — typed JSON via diffusion-model denoising, benchmarked against saved Jev decisions — [3/5] — #research — via github — _added 2026-09-19_
- **jev_local** — https://github.com/Argos1111/jev_local — local Jev-style API on LFM or ModernBERT, JGLUE-benchmarked — [3/5] — #research #eval — via github — _added 2026-09-19_
- **jevscan-evm** — https://github.com/devtooligan/jevscan-evm — heat map of likely bugs across a repo, vibe-coded proof of concept — [2/5] — #security #coding-agent — via github — _added 2026-09-19_
- **ulka** — https://github.com/razaanstha/ulka — experimental browser extension, Jev selects actions from the accessibility tree — [2/5] — #browser #demo — via github — _added 2026-09-19_
- **ainode** — https://github.com/getainode/ainode — self-hosted GPU appliance with a typed decision endpoint alongside inference — [2/5] — #sdk #ops — via github — _added 2026-09-19_
- **awesome-jev** — https://github.com/kraayenjon/awesome-jev — another curated Jev index, links to madewithjev.com use cases — [2/5] — #research — via github — _added 2026-09-19_
- **phuryn/experiments (jev-decisions-api)** — https://github.com/phuryn/experiments — hardened invoice test: Jev ties Haiku 4.5, beats Opus 115x on cost — [4/5] — #eval #finance — via github — _added 2026-09-19_
- **Sando** — https://github.com/yuzushi-dev/Sando — bounds Claude Code and Codex tool output, TypeSafe shadow judge optional — [4/5] — #guardrails #coding-agent — via github — _added 2026-09-19_
- **jev-codex-router** — https://github.com/0xNatoshi/jev-codex-router — per-turn Codex model routing, measured 60% savings on 237 turns — [4/5] — #routing #coding-agent — via github — _added 2026-09-19_
- **jev-rules** — https://github.com/EliaAlberti/jev-rules — Jev picks which CLAUDE.md rules apply to each prompt — [3/5] — #coding-agent #classification — via github — _added 2026-09-19_
- **jev-pruner** — https://github.com/tamaratran/jev-pruner — trims noisy Bash output before Claude sees it, keeps text verbatim — [3/5] — #coding-agent #ops — via github — _added 2026-09-19_
- **skillranker** — https://github.com/Dicklesworthstone/skillranker — Rust CLI ranks Claude Code skills for the next step — [3/5] — #cli #coding-agent — via github — _added 2026-09-19_
- **save-token-jev-clean** — https://github.com/IAmUnbounded/save-token-jev-clean — Jev picks which tool calls survive compaction across five agent hosts — [3/5] — #coding-agent #ops — via github — _added 2026-09-19_
- **jev-semgrep** — https://github.com/uehaj/jev-semgrep — grep by meaning across languages, no translation step needed — [3/5] — #search #coding-agent — via github — _added 2026-09-19_
- **supercov** — https://github.com/supercorp-ai/supercov — scores code quality with Jev, pairs with test coverage gaps — [3/5] — #coding-agent #eval — via github — _added 2026-09-19_
- **jev-sift** — https://github.com/kbhuw/jev-sift — classifies files and URLs by relevance before the agent reads them — [3/5] — #classification #coding-agent — via github — _added 2026-09-19_
- **advocaat** — https://github.com/pithings/advocaat — typed client for asking Jev questions about your data — [3/5] — #sdk #data — via github — _added 2026-09-19_
- **decider** — https://github.com/Mapika/decider — open one-pass typed decision model, fine-tuned Qwen3.5-2B, HF benchmarks included — [3/5] — #research #eval — via github — _added 2026-09-19_
- **jeff (GliFormer)** — https://github.com/logan-markewich/jeff — self-hosted drop-in Jev replacement, cheaper but less accurate — [3/5] — #sdk — via github — _added 2026-09-19_
- **fabricioctelles/skills** — https://github.com/fabricioctelles/skills — 26 agent skills, four use Jev for quality judgment — [3/5] — #coding-agent #eval — via github — _added 2026-09-19_
- **commit-miner** — https://github.com/devanshbatham/commit-miner — classifies commit diffs by bug fix, CWE and change type — [3/5] — #classification #security — via github — _added 2026-09-19_
- **vibecheck** — https://github.com/RafalWilinski/vibecheck — scores draft X posts on virality, cringe and regret risk before posting — [3/5] — #writing #classification — via github — _added 2026-09-19_
- **JevScout** — https://github.com/hqman/JevScout — Jev scores job links and postings while Chrome drives the browsing — [2/5] — #browser #classification — via github — _added 2026-09-19_
- **typesafe-ai-benchmark** — https://github.com/iammrduncan/typesafe-ai-benchmark — compares Jev against LLM-native structured output on latency, cost and quality — [3/5] — #eval — via github — _added 2026-09-19_
- **mini-Jev** — https://github.com/r-ms/mini-jev — preregistered test reads option-letter logits instead of generating JSON — [3/5] — #research #eval — via github — _added 2026-09-19_
- **building-with-jev-skill** — https://github.com/dbreunig/building-with-jev-skill — Claude Code skill for writing and debugging Jev-calling programs — [3/5] — #coding-agent #education — via github — _added 2026-09-19_
- **fast-jev-compaction** — https://github.com/tamaratran/fast-jev-compaction — Jev scores each tool call, drops stale ones, keeps rest verbatim — [3/5] — #coding-agent #ops — via github — _added 2026-09-19_
- **compact-adviser** — https://github.com/kunchenguid/compact-adviser — Jev judges when a Claude Code session is safe to compact — [3/5] — #coding-agent #ops — via github — _added 2026-09-19_
- **claude-code-trace** — https://github.com/delexw/claude-code-trace — Jev scores Claude Code sessions on progress, focus and token efficiency — [3/5] — #coding-agent #eval — via github — _added 2026-09-19_
- **jev-review (dashboard)** — https://github.com/devagrawal09/jev-review — staged Jev code review workflow with a local dashboard — [3/5] — #coding-agent — via github — _added 2026-09-19_
- **perch** — https://github.com/lakeday-org/perch — semantic linter flags defects with confidence scores, CLI and agent skill — [3/5] — #coding-agent #cli — via github — _added 2026-09-19_
- **foreman** — https://github.com/thruwire/foreman — Jev watches Codex workers, judges completion and drift, labeled an experiment — [2/5] — #coding-agent #eval #ops — via github — _added 2026-09-19_
- **evot** — https://github.com/evotai/evot — terminal coding agent, Jev prunes stale context instead of summarizing it — [3/5] — #coding-agent #cli — via github — _added 2026-09-19_
- **openagents (Coder)** — https://github.com/OpenAgentsInc/openagents — terminal coding agent, Jev routes turns and judges shell output — [2/5] — #coding-agent #cli #routing — via github — _added 2026-09-19_
- **vexjoy-agent** — https://github.com/notque/vexjoy-agent — Jev routes plain-English requests across 43 specialist Claude Code agents — [3/5] — #routing #coding-agent — via github — _added 2026-09-19_
- **jev-router (per-turn)** — https://github.com/gargpratyush/jev-router — routes each Claude Code or Codex turn to the cheapest capable model — [2/5] — #routing #coding-agent — via github — _added 2026-09-19_
- **jev-mcp (10 tools)** — https://github.com/jkudish/jev-mcp — verify, screen, rerank, classify, review and gate as MCP tools — [4/5] — #mcp #guardrails #classification — via github — _added 2026-09-19_
- **tax-doc-classifier** — https://github.com/kyotofin/tax-doc-classifier — classifies pages across 261 IRS forms, 100% accuracy, $0.001 per page — [4/5] — #classification #finance — via github — _added 2026-09-19_
- **classifier.dev** — https://github.com/mrmps/classifier-dev — zero-shot text classification over HTTP, no key, Jev backend — [3/5] — #classification #sdk — via github — _added 2026-09-19_
- **llm-rankers (Jev arm)** — https://github.com/ielab/llm-rankers — academic reranker benchmark adds Jev as a pointwise and listwise reranker — [3/5] — #eval #search #research — via github — _added 2026-09-19_
- **jev-eval-agent** — https://github.com/vinilana/jev-eval-agent — compares agent step counts when Jev picks the tool versus the LLM — [3/5] — #eval #coding-agent — via github — _added 2026-09-19_
- **jev-ultrafast** — https://github.com/browser-use/jev-ultrafast — Jev picks click target and action, small LLM only types text — [3/5] — #browser — via github — _added 2026-09-19_
- **jev-browser** — https://github.com/jkudish/jev-browser — Jev picks one browser action per step; Wikipedia nav in 4s — [3/5] — #browser — via github — _added 2026-09-19_
- **jev-browser-use** — https://github.com/wy-coliney/jev-browser-use — Jev clicks and navigates, Codex verifies; 5-10x faster in production — [3/5] — #browser #coding-agent — via github — _added 2026-09-19_
- **SemIf** — https://github.com/TheoLeeCJ/SemIf — open 4B model reproducing Jev's typed-decision interface, runs via WebGPU — [2/5] — #research #sdk — via github — _added 2026-09-19_
- **kev** — https://github.com/jaredpalmer/kev — tiny Jev-like model on Qwen2.5, trains and runs on a MacBook — [2/5] — #research #sdk — via github — _added 2026-09-19_
- **laya** — https://github.com/NandhaKishorM/laya — non-autoregressive multilingual decision model, 33ms per call on a T4 — [3/5] — #research #classification — via github — _added 2026-09-19_
- **LangWatch Instant Evals** — https://langwatch.ai — Jev judges your whole production history on demand; 97% agreement with human labels on 300 support chats, $0.32 per 10k conversations (vendor's claim) — #eval #ops — via discord — _added 2026-09-19_
- **JevSubRouter** — https://github.com/leftspace89/jevsubrouter — hook routes each Claude Code subagent dispatch to the cheapest model that can finish it; ~300ms warm — #routing #coding-agent — via github — _added 2026-09-19_
- **toolgate** — https://github.com/RiskAverseTech/toolgate — Claude Code hook and MCP proxy blocks, confirms or allows tool calls by risk — #guardrails #mcp #security — via github — _added 2026-09-19_
- **jev-code** — https://github.com/rhighs/jev-code — TypeScript coding CLI using Jev typed decisions and constrained AST generation — #coding-agent #cli — via github — _added 2026-09-19_
- **jevusecases** — https://www.jevusecases.com/ — crowdsourced directory of what people replaced with Jev — #demo — via discord — _added 2026-09-19_
- **saypage** — https://saypage.vercel.app/ — describe a page in plain language, builds a standalone site in under 0.4s — #demo #media — via discord — _added 2026-09-19_
- **jev-cvss** — https://github.com/Red5d/jev-cvss — CVSS 3.1/4.0 vectors from a plain description, matches NVD on sample set — #security #classification — via github — _added 2026-09-19_
- **jev-browser (MCP)** — https://github.com/Ying-Kai-Liao/jev-browser — LLM plans, Jev decides click target and action, Playwright executes — [3/5] — #browser #mcp — via github — _added 2026-09-19_
- **von** — https://github.com/wfzyx/von — open-source non-autoregressive decision model, sub-15ms local alternative to Jev — [3/5] — #research #sdk — via github — _added 2026-09-19_
- **jevbetter** — https://github.com/olanotolu/jevbetter — stronger one-pass option scorer, benchmarked head-to-head against jevlike — [3/5] — #eval #research — via github — _added 2026-09-19_
- **OpenDecision** — https://github.com/deepanwadhwa/OpenDecision — open-source Choice/Noul/Score engine over zero-shot models — [3/5] — #sdk #classification — via github — _added 2026-09-19_
- **duckdb-jev** — https://github.com/colliber/duckdb-jev — DuckDB extension, Jev answers typed as real SQL columns per row — [3/5] — #data #sdk — via github — _added 2026-09-19_
- **openjev** — https://github.com/zhihz/openjev — local bilingual probability decisions on a frozen Qwen3-4B backend — [3/5] — #research #sdk — via github — _added 2026-09-19_
- **tidepool** — https://github.com/tidepool-heavy-industries/tidepool — Haskell notebook agent harness, Jev supplies System 1 judgments — [3/5] — #coding-agent #research — via github — _added 2026-09-19_
- **oxlint-plugin-jev** — https://github.com/wobsoriano/oxlint-plugin-jev — plain-English yes/no lint rules, Oxlint finds the code, Jev judges it — [3/5] — #coding-agent #guardrails — via github — _added 2026-09-19_
- **jev-benchmarks** — https://github.com/AbdelStark/jev-benchmarks — calibration, selective risk and latency eval, Jev vs GLiNER2.5 — [4/5] — #eval #classification — via github — _added 2026-09-19_
- **jevil** — https://github.com/grabbou/jevil — Jev + agent-device QA agent, reads a mobile app and picks the next action — [3/5] — #browser #eval — via github — _added 2026-09-19_
- **jev-macos-loop** — https://github.com/jcpsimmons/jev-macos-loop — native macOS GUI automation, OmniParser plus Jev for action choice — [3/5] — #browser #ops — via github — _added 2026-09-19_
- **jgrep** — https://github.com/keltokhy/jgrep — grep by natural-language description, pipeable, ~200ms per line — [4/5] — #cli #search — via github — _added 2026-09-19_
- **macbrow** — https://github.com/timpratim/macbrow — voice-controlled Mac, Jev picks the AppleScript or browser tool — [3/5] — #browser #cli — via github — _added 2026-09-19_
- **choosekit** — https://github.com/NotXf1le/choosekit — typed choices and probabilities from local llama.cpp models — [3/5] — #sdk — via github — _added 2026-09-19_
- **beam-cli (AgentBeam)** — https://github.com/whyashthakker/beam-cli — hooks and local policy layer across Claude Code, Codex, Jev — [3/5] — #coding-agent #guardrails #cli — via github — _added 2026-09-19_
- **Turn** — https://github.com/ekizito96/Turn — compiled language, Choice/Score/Noul decisions as durable VM effects — [2/5] — #sdk #research — via github — _added 2026-09-19_
- **system-one** — https://github.com/sgoedecke/system-one — batched single-token choice inference for open models, Jev-compatible — [3/5] — #sdk #research — via github — _added 2026-09-19_
- **open-alternative-jev** — https://github.com/ikermoel/open-alternative-jev — open System One layer for open-weights LLMs, benchmarked on RACE-H — [4/5] — #research #eval — via github — _added 2026-09-19_
- **jot** — https://github.com/runta-dev/jot — general-purpose agent loop where Jev picks the tool and arguments — [3/5] — #coding-agent #routing — via github — _added 2026-09-19_

### Do now
- **jev-watch** — https://github.com/akanthed/jev-watch — baseline today's answers, flag drift when TypeSafe updates the model — #eval #ops — via github — _added 2026-09-19_
- **Jev on OpenRouter** — https://openrouter.ai/~typesafe/jev-latest — $0.042/M input, $0 output; plugs into the broker with no new adapter — #routing #sdk — via discord — _added 2026-09-19_
- **jev-sec-bench** — https://github.com/Gaurav-Gosain/jev-sec-bench — 96.5% on deepset prompt-injection set, 325ms p50; second opinion beside the intake regex — #security #eval — via github — _added 2026-09-19_
- **jev-belay** — https://github.com/valentynkit/jev-belay — Stop hook that catches "tests pass" when nothing ran — #guardrails #coding-agent — via github — _added 2026-09-19_
- **jev-commit** — https://github.com/valentynkit/jev-commit — commit-msg hook: message vs diff, leaked creds, debug leftovers — #guardrails #security #coding-agent — via github — _added 2026-09-19_

### Coding-agent guardrails (Claude Code hooks)
- **pi-warden** — https://github.com/DevMortimer/pi-warden — judges every tool call, output and reply in ~250ms — #guardrails #security — via github — _added 2026-09-19_
- **second-thought** — https://github.com/rodriveiga01/second-thought — shell commands judged pre-execution, blocks past 85% confidence — #guardrails #security #cli — via github — _added 2026-09-19_
- **jev-guard** — https://github.com/ClemensSchartmueller/jev-guard — auto-approval layer for Claude Code and Codex tool calls, ships a bypass CTF — #guardrails #security #coding-agent — via github — _added 2026-09-19_
- **omp-auto-mode** — https://github.com/alexsatch/omp-auto-mode — safe/unsafe/ask tool-call classifier emulating auto mode — #guardrails #classification — via github — _added 2026-09-19_
- **pi-heed** — https://github.com/Nyarlathoteppppp/pi-heed — enforces "never touch prod" constraints on tool calls, 46 tests — #guardrails #ops — via github — _added 2026-09-19_
- **tenet** — https://github.com/zoidsh/tenet — judges every agent commit against plain-English rules in AGENTS.md — #guardrails #coding-agent — via github — _added 2026-09-19_
- **abide** — https://github.com/coldteadotai/abide — makes the coding agent obey project rules — #guardrails #coding-agent — via github — _added 2026-09-19_
- **Reaper** — https://github.com/Eliran-Turgeman/repear — semantic linter for silent failures, weakened tests, scope creep — #coding-agent #guardrails — via github — _added 2026-09-19_
- **jevlint** — https://www.jevlint.com/ — plain-English convention rules checked in the agent loop — #coding-agent #guardrails — via discord — _added 2026-09-19_
- **commentlint** — https://github.com/ari-becker/commentlint — checks that agent-written comments still match the code — #coding-agent #guardrails — via github — _added 2026-09-19_
- **wince** — https://github.com/TinyFrontier/wince — scores diffs by blast radius and auth/data-write for review routing — #security #coding-agent #routing — via github — _added 2026-09-19_
- **Migration Guardian** — https://github.com/opaielsheikh/typesafe-migration-guard — halts destructive SQL migration plans — #guardrails #data #ops — via github — _added 2026-09-19_
- **is-malicious** — https://github.com/luantak/is-malicious — screens a repo or PR for obviously malicious code before running it — #security #guardrails — via github — _added 2026-09-19_
- **Gorgona** — https://github.com/adrianpeticila/gorgona — zero-dependency Python guardrail engine, halts agents on leaked keys or loops — #guardrails #security — via github — _added 2026-09-19_

### RAG, grading, classification (GWEDU)
- **Jev vs rerankers** — https://anessbelbati.com/blog/i-gave-jev-a-rerankers-job — ties Cohere rerank-4-pro on NDCG@10 across 8 datasets — #search #eval — via discord — _added 2026-09-19_
- **GBrain Jev reranker** — (PR by Daniel Andrade, see channel) — Score-based reranking, 1.9s vs 14.4s LLM at similar cost — _added 2026-09-19_
- **ReadyBase rerank eval** — https://www.promptforce.ai/readybase — 28-scenario gold-labeled harness for code-context selection — #eval #search #coding-agent — via discord — _added 2026-09-19_
- **extraction-marker-recovery** — https://github.com/chrismoseley/jev-extraction-marker-recovery — restores OCR-flattened footnote and citation markers in passages — #data #research — via github — _added 2026-09-19_
- **jev-document-classification** — https://github.com/Charlyhno-eng/jev-document-classification — document classes plus injection detection — #classification #security — via github — _added 2026-09-19_
- **book-aurora** — https://github.com/dani1005/book-aurora — 601 passages, 9 emotion scores, 25s, $0.034 — #classification #media — via github — _added 2026-09-19_
- **AI Elo ranker** — https://github.com/opaielsheikh/ai-elo-ranker — tournament scoring for text, reusable for rubric grading — #eval — via github — _added 2026-09-19_
- **sqlite-jev** — https://github.com/mgaitan/sqlite-jev — batched judgments as SQL functions in SQLite — #data #sdk — via github — _added 2026-09-19_
- **pg-redact** — https://github.com/rishi-raj-jain/pg-redact — tags PII spans in Postgres text, redacts only the span — #security #data — via github — _added 2026-09-19_
- **FairCheck** — https://fair-housing-guardrail.roberto-94c.workers.dev/ — RAG-grounded guardrail citing regulation paragraphs; pattern for cited refusals — #guardrails #rag — via discord — _added 2026-09-19_
- **sentence-relationship classifier** — https://jev-demo.corca.ai/ — non-LLM segmentation plus 3 Jev calls, ~0.5s — #classification — via discord — _added 2026-09-19_

### Eval and adoption method
- **Sniff Test** — https://github.com/DanRWilloughby/snifftest — prose linter as CLI, pre-commit, Action and Claude Code skill; 182ms, $0.013 per 100 paragraphs — #writing #cli — via github — _added 2026-09-19_
- **shadow-mode adoption write-up** — https://x.com/yonyoniz/status/2101241236107542549 — run Jev beside the LLM for 3 days, promote only where it wins — #ops #eval — via x — _added 2026-09-19_
- **Working with Jev** — https://rodrigopsasaki.com/blog/working-with-jev/ — practitioner notes from a real PR reviewer — #coding-agent #writing — via discord — _added 2026-09-19_
- **Cultivar** — https://github.com/pinecone-io/cultivar — Pinecone's agent-skill eval harness with Jev grading — #eval #coding-agent — via github — _added 2026-09-19_
- **dinostomp** — https://github.com/collapseindex/dinostomp — verifies the scorer, not just the score, with a blind control per run — #eval — via github — _added 2026-09-19_
- **Vercel eve-agent failure classifier** — https://x.com/danmana/status/2100545412780220877 — 320 sessions, 2M tokens, classified in 14s for $0.08 — #classification #eval #coding-agent — via x — _added 2026-09-19_
- **Empryo harness write-up** — https://empryo.com/blog/jev-and-the-harness — five decision points in a coding agent benchmarked against frontier models — #eval #coding-agent — via discord — _added 2026-09-19_
- **Swamp alert triage** — https://blog.watson-labs.co.uk/typesafe-ai-alert-fatigue/ — 91% cheaper triage before routing to Claude — #routing #ops #classification — via discord — _added 2026-09-19_
- **ZeroSweep** — https://github.com/sysadarsh/zerosweep — Jev vs LLM inbox triage, live comparison — #classification #eval — via github — _added 2026-09-19_
- **optimaizr** — https://optimaizr.com/ — ranks LLM spend waste by dollar savings on real traffic — #finance #ops — via discord — _added 2026-09-19_

### SDKs, adapters, glue
- **JevRouter** — https://github.com/BillionsBobby/JevRouter — Jev-powered router for models, tools and subagents — #routing #coding-agent — via github — _added 2026-09-19_
- **daf-jev** — https://github.com/docxology/daf-jev — Python toolkit: confidence gates, tiered routing, calibration module, MCP server — #routing #mcp #guardrails — via github — _added 2026-09-19_
- **dspy-typesafeify** — https://github.com/typesafeainate/dspy-typesafeify — one decorator for Jev-backed DSPy signatures — #sdk #coding-agent — via github — _added 2026-09-19_
- **jev-mcp** — https://github.com/blakestone-x/jev-mcp — MCP server exposing classify, score, check, match, screen — #mcp #classification — via github — _added 2026-09-19_
- **typesafe-mcp** — https://github.com/itsmostafa/typesafe-mcp — drop-in MCP connector for Claude Code — #mcp #coding-agent — via github — _added 2026-09-19_
- **jev-cli** — https://github.com/Nasrallah-AL/jev-cli — verify, screen, classify, extract, route, rerank from the shell — #cli #classification — via github — _added 2026-09-19_
- **jev-axi** — https://github.com/shiftynick/jev-axi — pick, rate, check, rank, triage, guard shell commands — #cli #guardrails — via github — _added 2026-09-19_
- **simple-jev** — https://github.com/featherless-ai/simple-jev — turn any open model into a classifier endpoint; fallback if TypeSafe goes away — #classification #sdk — via github — _added 2026-09-19_
- **System One Search** — https://s1s.iar.dev/ — Jev-only codebase retrieval and reference graph — #search #coding-agent — via discord — _added 2026-09-19_
- **every** — https://github.com/sufianetaouil/every — ask a yes/no question of every function in a codebase — #coding-agent #research — via github — _added 2026-09-19_
- **jev-browser** — https://github.com/vlad-terin/jev-browser — cheap Playwright browser agent driven by Jev — #browser — via github — _added 2026-09-19_
- **typesafe-computer-use** — https://github.com/awlevin/typesafe-computer-use — OCR, classify, click loop at ~$0.0002 per step — #browser #classification — via github — _added 2026-09-19_
- **vgi-typesafe** — https://github.com/Query-farm/vgi-typesafe — DuckDB extension, Jev as LATERAL joins — #data — via github — _added 2026-09-19_
- **pg-jev** — https://pgjev.zachi.dev/ — Postgres extension, plain-English per-row WHERE judgments — #data — via discord — _added 2026-09-19_
- **HA-Jev** — https://github.com/AboveColin/HA-Jev — Home Assistant sensors and automations — #ops — via github — _added 2026-09-19_
- **awesome-jev-typesafe** — https://github.com/valentynkit/awesome-jev-typesafe — 250-entry curated index — #research — via github — _added 2026-09-19_
- **awesome-jev-projects** — https://github.com/logicrw/awesome-jev-projects — 287-entry index with English view — #research — via github — _added 2026-09-19_
- **TypeSafe primitives quiz** — (dball9, `quiz_handover.md` in channel) — self-test on Choice, Noul, Score and the cascade patterns — _added 2026-09-19_

## Cool

### New 2026-09-20
- **1v1 Jev** — https://github.com/emrickgarrett/OneVOneJev — FPS bot where Jev decides movement, aim, and firing at 9Hz. — [1/5] — #game #demo — via x — _added 2026-09-20_
- **Jev's Fly** — https://github.com/webdevcody/jevs-fly — Jev steers a Three.js flying-game character in real time. — [1/5] — #game #demo — via github — _added 2026-09-20_
- **RoboJEV** — https://github.com/lykycy123/RoboJEV — Two-stage Jev control of a simulated Franka Panda arm, evaluated. — [3/5] — #robotics #eval — via github — _added 2026-09-20_
- **ST-jeved** — https://github.com/mossyfield/ST-jeved — Jev reads roleplay replies and triggers narrator rules or rerolls. — [1/5] — #game #media — via github — _added 2026-09-20_
- **J++** — https://github.com/Towow-ai/jpp — Experimental language where questions and methods compose as values. — [1/5] — #research — via github — _added 2026-09-20_
- **JEVfire** — https://github.com/kikoncuo/jevfire — Batches typed-variable decisions in parallel on CUDA LLMs, benchmarked. — [3/5] — #classification #game #research — via github — _added 2026-09-20_
- **RuneBench with Jev (jevscape)** — https://github.com/Skyvern-AI/jevscape — Lets Jev play RuneScape through a bounded 50-action harness. — [1/5] — #game #demo — via github — _added 2026-09-20_
- **xtags** — https://github.com/manifoldor/xtags — Chrome extension tags X posts with Jev-derived intent labels. — [2/5] — #browser #classification — via github — _added 2026-09-20_
- **syft-listening** — https://github.com/tpaulshippy/syft-listening — Real-time speech analysis using Jev, tested live in browser. — [2/5] — #media #research — via github — _added 2026-09-20_
- **slop-filter** — https://github.com/adamnroman/slop-filter — Chrome extension hides AI-generated posts on X and LinkedIn. — [2/5] — #browser #classification — via github — _added 2026-09-20_
- **Jev plays F-Zero** — https://x.com/GothLupita/status/2101487063413543405 — Jev plays F-Zero from raw screen pixels, no game code. — [1/5] — #game — via x — _added 2026-09-20_
- **crush-monitor** — https://github.com/FerryCorleone/crush-monitor — Analyzes WeChat chat sentiment and rates reply quality with Jev. — [1/5] — #classification #media — via github — _added 2026-09-20_
- **Nemotron_Jev** — https://github.com/pst2154/Nemotron_Jev — Serves a diffusion model behind a Jev-shaped decision API. — [1/5] — #media #sdk — via github — _added 2026-09-20_
- **jev-reflex-autonomy-lab** — https://github.com/khordoo/jev-reflex-autonomy-lab — Multi-drone sim where fast Jev reflexes escalate to a slower planner. — [1/5] — #robotics — via github — _added 2026-09-20_
- **feelings** — https://github.com/BoundaryML/feelings — Adds a typed .feels() method to any value via Jev. — [2/5] — #sdk — via github — _added 2026-09-20_
- **Jevthoven** — https://github.com/cocktailpeanut/jevthoven — Generates editable multitrack MIDI music one Jev decision at a time. — [2/5] — #media — via github — _added 2026-09-20_
- **PlayJev** — https://github.com/OmniJev/PlayJev — 0.8B model plays ten browser games directly from raw pixels. — [2/5] — #game — via github — _added 2026-09-20_
- **Jevinik** — https://github.com/unicodeveloper/jevocks — Stock terminal predicts 30-day price direction using Jev over evidence. — [2/5] — #finance — via github — _added 2026-09-20_
- **jev-robot-control (openroboto-ai)** — https://github.com/openroboto-ai/jev-robot-control — Jev vs GPT-6 Astra vs GPT-4.1 mini placing an apple — [3/5] — #robotics — via github — _added 2026-09-20_
- **transcript-lens** — https://github.com/sensahin/transcript-lens — Jev finds chapters, claims and key passages in YouTube transcripts, Turkish UI — [2/5] — #media #research — via github — _added 2026-09-20_
- **jev-seo** — https://github.com/AkashPriyadarshii/jev-seo — free SEO/GEO CLI, Jev classifies intent instead of paid Semrush-style scoring — [2/5] — #cli #search — via github — _added 2026-09-20_
- **jev-game-tools (Brotato)** — https://github.com/Eniip/jev-game-tools — Jev picks every-frame movement, Claude handles long-term shop strategy — [2/5] — #game — via github — _added 2026-09-20_
- **Jev Othello** — https://x.com/4ba_ba_baba/status/2100103803282551151 — Jev plays Othello against random moves, heatmap vs minimax comparison — [1/5] — #game — via x — _added 2026-09-20_
- **Jev + Stagehand browser use** — https://x.com/kylejeong/status/2100622054945095934 — accessibility tree state, Jev decides next click, task cost $0.001 — [2/5] — #browser — via x — _added 2026-09-20_
- **minecraft-agent** — https://github.com/rmalde/minecraft-agent — Astra plans, Jev picks moves in a verified 8m43s dragon kill — [2/5] — #game #coding-agent — via github — _added 2026-09-20_
- **embodied-jev** — https://github.com/FBddcz/embodied-jev — MuJoCo robot arm workbench compares Jev against local and cloud models — [2/5] — #robotics — via github — _added 2026-09-20_
- **SmartMoney-Cub** — https://github.com/myc0576/SmartMoney-Cub — read-only trading journal uses Jev for typed judgments, no live orders — [2/5] — #finance — via github — _added 2026-09-20_
- **MKUltraScale logic circuits** — https://x.com/mkultrascale/status/2101615647625257385 — built logic circuits out of Jev gates — [1/5] — #research — via x — _added 2026-09-20_
- **Shader from text (0xFotex)** — https://x.com/0xFotex/status/2101636257264619900 — turns any text into a visual shader with Jev — [1/5] — #media — via x — _added 2026-09-20_
- **Jev color palette generator (fran)** — https://x.com/fran_rimoldi/status/2101084847548850410 — Jev scores hue, warmth and energy, then picks a font — [1/5] — #media #demo — via x — _added 2026-09-20_
- **jevmoji** — https://github.com/cheeaun/jevmoji — type anything, get related emojis scored 0-3 by Jev — [1/5] — #media #demo — via github — _added 2026-09-20_
- **Adventures of Jev** — https://bschoolland.dev/jev-adventure — Jev plays an RPG adventurer across 43 places and 31 people — [1/5] — #game — via discord — _added 2026-09-20_
- **Jev's Sprint Planning** — https://jevs-sprint-planning.vercel.app/ — Jev negotiates sprint scope, coordinates four developers in a 3D office — [1/5] — #ops #demo — via discord — _added 2026-09-20_
- **Fieldnote** — https://fieldnote.clawscience.com/ — semantic physics and math solver, Jev solved 3 MIT Integration Bee questions — [2/5] — #education #research — via discord — _added 2026-09-20_
- **jevchat** — https://github.com/kyle-pena-nlp/jevchat/ — turns Jev into a chatbot — [1/5] — #demo — via github — _added 2026-09-20_
- **jev-libero** — https://github.com/Dimweaker/jev-libero — Jev controls a simulated robot arm to close a microwave and drawer — [2/5] — #robotics — via github — _added 2026-09-20_
- **Quarrag and the Sun-Heart of Mordanne** — https://looselyconnected.wordpress.com/2026/09/19/quarrag-and-the-sun-heart-of-mordanne/ — Sonnet writes the story, Jev makes every choice — [1/5] — #writing #game — via discord — _added 2026-09-20_
- **Jev-X-Sentiment-Analysis** — https://github.com/brainstormity/Jev-X-Sentiment-Analysis — scores crypto tweets bullish or bearish, combines with funding rate and RSI — [2/5] — #finance #classification — via github — _added 2026-09-20_
- **jev-ncr-demo** — https://github.com/Alexandre-Borghi/jev-ncr-demo — suggests defect codes from a plain-English non-conformance report, Rust/Leptos/Axum — [2/5] — #classification #demo — via github — _added 2026-09-20_
- **jev-311-heatmap** — https://github.com/CompleteTech-LLC-AI-Research/jev-311-heatmap — NYC 311 complaint heatmaps classified and mapped by Jev — [1/5] — #classification #data — via github — _added 2026-09-20_
- **HN reranker (danprice.ai)** — https://danprice.ai/hn — reranks Hacker News with a plain-English query like "jealous author" — [2/5] — #search #routing — via discord — _added 2026-09-20_
- **ORIGIN-CIVILIZATION** — https://github.com/JacquesGariepy/ORIGIN-CIVILIZATION — inspectable life-and-civilization sim, every voluntary NPC action needs a Jev decision — [1/5] — #game #research — via github — _added 2026-09-20_
- **JevLM Studio** — https://www.jevlm.studio/ — playground compares three experimental models against ten prompts — [1/5] — #eval — via discord — _added 2026-09-20_
- **canyoubeatjev.fyi** — https://canyoubeatjev.fyi/ — unfinished dashboard for comparing Jev against other models head to head — [1/5] — #eval #demo — via discord — _added 2026-09-20_
- **HazAlerts near-me** — https://demo.noxidsoft.com/hazalert-near-me/ — live AU fire and emergency map triaged by Jev, concept demo — [1/5] — #classification #demo — via discord — _added 2026-09-20_
- **Odyssey (Jev flight director)** — https://fly.rahmanyoonus.com/ — moon and Mars launch sim narrated by Jev's decisions with confidence scores — [1/5] — #demo #media — via discord — _added 2026-09-20_
- **jamsesh** — https://github.com/kmooney/jamsesh — MIDI jambox agent follows your lead while you play music — [1/5] — #media — via github — _added 2026-09-20_
- **jev-mermaid** — https://github.com/JordanDalton/jev-mermaid — Jev-driven Mermaid diagram generator — [1/5] — #media #coding-agent — via github — _added 2026-09-20_
- **flappyai** — https://flappyai.app/ — reversed Flappy Bird, Jev controls the missiles, you move the guardrail — [1/5] — #game — via discord — _added 2026-09-20_
- **memefy** — https://memefy.lol/ — finds a matching meme for your text using Jev — [1/5] — #media #demo — via discord — _added 2026-09-20_
- **Vesper living-town demo** — https://www.youtube.com/watch?v=LZilhFs34Wo — Ultima Online fan demo, Jev picks each NPC's next action — [1/5] — #game #demo — via discord — _added 2026-09-20_
- **Jev vs Jev Ultima Online duels** — https://www.youtube.com/watch?v=tjX3UoD7HPQ — two Jev-controlled characters duel each other — [1/5] — #game — via discord — _added 2026-09-20_
- **jevfish** — https://jevfish.patebryant.com/ — chess engine, every move a Jev judgment call, no search, 4-0 vs humans so far — [2/5] — #game — via discord — _added 2026-09-20_
- **jev-chess (SyedZawwarAhmed)** — https://github.com/SyedZawwarAhmed/jev-chess — Jev picks a move from a code-generated legal-move list — [2/5] — #game — via github — _added 2026-09-20_
- **City Dispatch** — https://diurnalproductions.com/games/city-dispatch — driving game, bot cars use Jev and sensor input to drive — [2/5] — #game — via discord — _added 2026-09-20_
- **Jevilish** — https://jev.ldlework.com/ — word game, guess the phrase Jev mangled into synonyms — [1/5] — #game — via discord — _added 2026-09-20_
- **From a movie** — https://from-a-movie.vercel.app/ — type a line or scene, Jev finds the movie — [1/5] — #search #media — via discord — _added 2026-09-20_
- **Jev MIDI phrase picker** — https://www.youtube.com/watch?v=2ALKgcocx6s — Jev selects the next musical phrase from generated candidates — [1/5] — #media — via discord — _added 2026-09-20_
- **The Trolley Problem** — https://gpu.studio/trolley — put anything on the tracks and find out whether Jev pulls the lever — [1/5] — #demo #guardrails — via discord — _added 2026-09-20_

### Older
_Folded from 2026-09-19 to stay under the line budget; names link to the original project or post._
- [Jev live demo (val.run)](https://typesafe-demo.val.run/), [killmyidea](https://github.com/monteduro/killmyidea), [typesafe-snake](https://github.com/sorrycc/typesafe-snake), [jev-voice](https://github.com/kevinbadi/jev-voice), [tsai-sc](https://github.com/phyous/tsai-sc), [refgarden](https://github.com/AlbionaHoti/refgarden), [jev-playground](https://github.com/mizchi/jev-playground), [jevpilot](https://github.com/standardagents/jevpilot), [jev-recruiter](https://github.com/skeptrunedev/jev-recruiter), [jev-trade (Hyperliquid)](https://github.com/aowang-ai/jev-trade)
- [ghosthands](https://github.com/affirmitv/ghosthands), [jev-chat](https://github.com/w3cj/jev-chat), [jev-paint](https://github.com/achimala/jev-paint), [hono-jev-router](https://github.com/yusukebe/hono-jev-router), [neo4jev](https://github.com/jexp/neo4jev), [jev-shell-history](https://github.com/mrnugget/jev-shell-history), [jevmeter](https://github.com/ChetasLua/jevmeter), [typesafe-adblock](https://github.com/realZachi/typesafe-adblock), [youtube-sponsor-detection](https://github.com/trungdq88/youtube-sponsor-detection), [jev-trader](https://github.com/jarrodwatts/jev-trader)
- [mobile-jev](https://github.com/droidrun/mobile-jev), [jev-voice-browser](https://github.com/moritzkremb/jev-voice-browser), [unclutter](https://github.com/kitze/unclutter), [jev-visual](https://github.com/hr98w/jev-visual), [typesafe-mario](https://github.com/fhshaik/typesafe-mario), [Jev Search](https://github.com/superagents-lab/jev-search), [Jev-Trades](https://github.com/zadescoxp/Jev-Trades), [live-jev (driving sim)](https://github.com/vinilana/live-jev), [typesafe-playground](https://github.com/kavehmz/typesafe-playground), [mario-jev](https://github.com/shantanugoel/mario-jev)
- [live-jev (Ableton)](https://github.com/okinaaudio/live-jev), [x-scanner](https://github.com/oso95/x-scanner), [Piano in real time](https://x.com/chandamamz/status/2101066294816920063), [Jev vs Opus competitive Pokémon](https://x.com/sid19arya0/status/2100458351440048258), [JevPlaysPokemon](https://github.com/anxkhn/JevPlaysPokemon), [Mario from NES memory](https://x.com/rherton/status/2101005942913729001), [TerraBlind](https://github.com/Reisenbug/TerraBlind), [Wordle x32](https://x.com/RichardSuwandi), [Jev vampire survivors](https://github.com/oldmoldycake/jev_vampire_survivors), [Jev arena](https://jev-arena.vercel.app/)
- [Coup](https://coup.clairbridge.tech/), [jev-chess](https://jev-chess.tgr.rs), [Doom](https://x.com/CompleteSkeptic/status/2099925687465570372), [Minecraft zombie escape](https://youtu.be/3G14OU00crI), [Ultima Online combat AI](https://www.youtube.com/watch?v=cP-GWKteHKo), [Ember Hollow living village](https://lifelike.build/), [NPC judgment village](https://x.com/haiderTheDev/status/2100419750459519156), [Coffee-shop sim](https://x.com/soya_da_yoot/status/2100451279222755616), [DJ/VJ director](https://x.com/shannon_fano/status/2100639390468833379), [Real-time video effects](https://x.com/ponyo877/status/2101139914419290345)
- [JevPixelArt](https://github.com/rivianpratama/JevPixelArt), [jev-drone](https://github.com/RomanSlack/jev-drone), [SO-101 arm via quackd](https://github.com/rokbenko/quackd/blob/main/docs/jev.md), [Robot manipulation gating](https://x.com/JonyShaik/status/2100827590781214981), [jev-mobile](https://github.com/Friedjof/jev-mobile), [WikiRace](https://x.com/princecaarlo/status/2100313645116952701), [Football, Jev vs Jev](https://x.com/attractmodeio/status/2100723340021276739), [date-with-jev](https://date-with-jev.vercel.app/), [Piano, failed](https://x.com/BaselAshraf81/status/2100400028711805242)

## Findings and gotchas

### New 2026-09-20
- **intikhab49**: An open 150M reproduction scored 0.697 vs Jev's 0.727 on the same benchmark, while being 2.5x better calibrated and 4x faster.
- **QuicqDev**: Benchmarked Jev 1.13 against 11 classical ML pipelines on 8 datasets: Jev hit 96.3% balanced accuracy on IMDb sentiment vs 88.4% for the best classical pipeline, but classical pipelines won on all 4 tabular datasets.
- **zeeshan8281**: Routing through Jev preserved the same routes and accuracy as local deterministic features but raised p95 end-to-end latency from 77.93ms to 490.38ms.
- **DECRUX9812**: Jev-based skill routing for Hermes Agent costs about $0.001 per routed turn.

- **openroboto-ai**: on a single seed-0 xArm7 pick-and-place trial (one apple, one plate), Jev 1.13 placed it in 226 API calls for $0.019 and 182s wall time; GPT-6 Astra also placed it but took 707s at $5.93, 312x the cost; GPT-4.1 mini hit the 160-cycle limit without finishing.
- **Kyle Jeong**: driving Stagehand from an accessibility-tree state instead of screenshots, letting Jev pick the next click completed a browser task for $0.001 at near-instant speed.
- **Bartosz Mikulski**: converting 400 hand-drawn doodles into SVG coordinate text, Jev named the drawing correctly about 35% of the time versus 10% chance, but lost to Claude Sonnet 5 and guessed "airplane" for more than half.
- **yuwakisa benchmark (via hypnoticfuzzwave)**: on a two-step benchmark of stating a principle from examples then applying it cold in a new context window, Jev scored ahead of every other tested model.
- **bartlomein**: oko's 108-session pilot on ranking local grep matches before an agent reads them saw up to 46% fewer agent tokens and 18% less wall time with a warm cache.
- **Fatalfencer**: swapping a Monte Carlo bot's option-picking step for Jev beat a "hard" difficulty tactics-game bot but not the hardest hand-tuned one, at about $0.10 per game.
- **loop (dev rel @ openrouter)**: OpenRouter's Ori Eval judging benchmark found Jev over 5x faster than the next fastest model, and Jev's slowest requests still beat every other model's median latency.
- **_FailSafe**: a WIP Jev-scored retrieval triage sidecar over Noema's hybrid search kept 2 of 8 candidates on a clear query (relevant scores ~0.95/0.91 vs ~0.03-0.06, ~3700 in/666 out tokens) and 4 of 8 on a messier preference query (preference-flagged items scored 0.86-0.98), with a soft decision band identified around 0.55.
- **Zawwar**: playing chess by having Jev pick from a code-generated list of legal moves, it found the correct move on 16% of 600 rated Lichess puzzles versus 5% for random guessing, and accuracy did not fall as puzzles got harder (16% at 800-rated vs 23% at 2200-rated, the opposite of a human), because Jev isn't calculating; it found mate-in-one only about 1 time in 10.
- **patebry**: jevfish's move-by-move confidence score cannot separate a mate-in-2 from a waiting move; up a queen against a human it took 65 moves to deliver mate.
- **askmuyukani**: four concurrent Jev calls filtering real FAISS retrieval results in JarvisCore completed in 1.37s wall time for $0.000074.
- **shitianfang**: routing agent steps that need no text output straight to Jev instead of an LLM measured p50 ~230ms and ~$0.02 per 1,000 judgments.
- **Zaious**: on the same historical multiple-choice question, Jev gave a wrong answer at 0.90 confidence with no supporting passage, then a correct answer at 0.97 confidence once the background text was included in `state`; it is only as good as the facts you give it.
- **goodrahstar**: labelling 1,000 Android app reviews across 4 typed questions each, Jev finished in 4.6s for $0.023 versus Gemini 3.8 Flash's 18.8s and $0.158 (4.1x faster, 7x cheaper), with near-identical sentiment agreement against star ratings (rho 0.80 vs 0.82).
- **JoshuaSP**: an open DiffusionGemma reproduction of Jev's typed-decision interface matched 48/48 relevance labels and 6/6 top-1 retrieval on a code-retrieval eval, and hit 138/144 (95.8%) agreement with saved Jev decisions on a customer-voice eval, using a single denoising step.
- **Argos1111**: a local zero-shot LFM 1.2B backend scored JNLI 17% / JComQA 69% on JGLUE, versus a fine-tuned ModernBERT cross-encoder backend at JNLI 93% / JComQA 92%; zero-shot small models lag furthest on entailment judgments, not commonsense QA.
- **phuryn**: hardened Jev's own invoice-classification showcase to 50 documents; Jev scored 50/50 at $0.025 per 1,000 decisions, tied by Claude Haiku 4.5 and nearly matched by open 48/50 self-hosted models, while Opus 5 landed one answer behind at 115x the cost; strip the written definitions out and every Jev mistake falls below 0.80 confidence.
- **0xNatoshi**: per-turn Jev routing for Codex cut cost roughly 60% against an all-frontier baseline across a 7-day replay of 237 real turns, at about $0.00003 and 0.6s per routing decision.
- **yuzushi-dev**: deterministic tool-output bounding (Sando) cut Codex shell tokens 90.9% across 40,000+ recorded results and Claude PostToolUse tool results 14.5-18.9% across two machines, fitting roughly 2.4x more files in a 200,000-token window.
- **coldteadotai**: replaying 93 real Claude Code sessions found agents break an unwritten project rule on 1 turn in 13; Jev flagged 39 edits and 15 turns, an independent reviewer confirmed 10 and 11.
- **kyotofin**: a Jev tax-form classifier hit 100% strict accuracy across 261 IRS forms at $0.001 per page, 34x cheaper and 6x faster than their prior LLM pipeline.
- **jaz**: toolgate's held-out adversarial set (injection, forged approval, encoded payloads) scored 23/24 with zero permissive errors, both frozen sets 20/20; about a second and a fraction of a cent per call.
- **Suraj Phanindra**: a Choice with no matching option still answers at confidence 1.00. Always include an "other/none" option.
- **ubhillutd, Eddy Vinck**: phrasing shifts probabilities on identical state ("save money" vs "save money long term" flips the answer). Per-call is stochastic, the average is stable.
- **Zurban**: calibration under domain shift is unproven. A miscalibrated gate is worse than no gate.
- **parallax**: 1,565 emails, 10 classes. All 737 predictions at 99%+ confidence were correct; mistakes cluster at low confidence. Trailed Gemini slightly on accuracy, 10 to 22x cheaper.
- **Cleista**: ~90% accuracy at ~12k classifications per dollar; the last 3% of hard cases still need an LLM at 10 to 100x the cost.
- **identityTorn**: zero-shot Jev within ~5 points recall of a fine-tuned Qwen 9B; the fine-tune still wins on latency and cost.
- **Atanas**: 94% tool-pick accuracy on typos vs 28 to 34% for keyword rules. Cannot fill open values; hand those to a small LLM (0.2s added).
- **venkat (LessWrong)**: one yes/no call reaches ~0.97 AUROC catching backdoors, but accuracy drops once an attacker can generate and score variants.
- **Sub**: EU to Jev infra adds up to 320ms network latency. Killed a real-time NPC use case.
- **stbenjam**: no temperature setting, so repeated generative-style tasks come out repetitive.
- **Tawfek Sraj**: knowledge cutoff reverse-engineered to between 2024-07 and 2025-01.
- **sh0rtythegreat**: token counts ran unexpectedly high on a line-by-line task; check the bill, not just the per-token price.
- **Sid Arya**: Jev is greedy in multi-turn setups. Frame whole actions, not letter-by-letter steps.
- **Yonatan Gross**: hand-rolled client sent Score criteria as a dict and got silent 422s. Use the SDK.
- **azakhary**: 60% of a 10x-speed game playtest was Jev thinking; effective 3x. Server capacity is the bottleneck at scale.
- **timpratim**: an early version of a voice-controlled Mac agent, told to "clean up my desktop," moved every file on the Desktop into one folder; the project now ships an explicit safety policy after that run.
- **cramforce**: swapped Jev into a classifier eval that previously ran on Gemini 2.5 Flash Lite; Jev matched or beat it on quality, saturating the eval, and ran roughly 6x faster.
- **Manjunath Janardhan**: a 200-decision benchmark against Claude Fable 5.1, GPT-6 Astra, Kimi K3, MiniMax M3 and DeepSeek V4.1 Flash on BANKING77, BoolQ, Yelp and ChaosNLI found only the two flagships clearly ahead of Jev (11.5 and 6.5 points), while the mid-tier models were within noise but cost 6-150x more and took 2-15x longer; Jev tied every model on BoolQ at 94% (AUROC 0.970); on Yelp its errors are nearly all one star off but arrive at 0.98-0.99 confidence, with 18 of 55 errors above 0.9 confidence; on ChaosNLI, where 100 people split their labels, Jev's probabilities sat further from the human distribution (JSD 0.149) than a blind guess (0.127), while Claude Fable 5.1 hit 0.043.
- **ASHu2**: an independent 8-dataset benchmark against classical ML found Jev strong on sentiment classification but mixed across other tasks, with fine-tuned ML staying cheaper wherever fine-tuning is an option; few-shot prompting did not reliably improve Jev's results.
- **yzfly (awesome-jev-zh)**: an independent evaluation cited in the list found that asking Jev a single direct question about a phishing email scored 62.6% accuracy, while two lines of regex reached 91.8%.