# Awesome Jev — full export

> Everything built on Jev — TypeSafe AI's System One model, the first model that answers with typed judgments and calibrated probabilities instead of prose. Generated from one data file in three languages, maintained by deterministic scripts and a free-AI reviewer.

352 entries, 93 460 stars, 19 rules, refreshed 2026-10-01.
Source: https://github.com/hdjekuue/awesome-jev

Format: one H3 per entry, followed by a field list.

---

## 001 · Start Here

The spec sheet, the three primitives and the docs — read these before building.

### Introduction

- url: https://docs.typesafe.ai/introduction
- description: The mental model in two pages: state plus typed questions in, typed answers with probabilities out.
- official: true
- rule: 001

### Quick start

- url: https://docs.typesafe.ai/introduction/quickstart
- description: First request in Python, TypeScript, or curl.
- official: true
- rule: 001

### Primitives

- url: https://docs.typesafe.ai/primitives
- description: Choice, Score, and Noul, and when each one fits.
- official: true
- rule: 001

### State

- url: https://docs.typesafe.ai/concepts/state
- description: How to package what Jev judges, and why less is more.
- official: true
- rule: 001

### API reference

- url: https://docs.typesafe.ai/api
- description: The request and response contract.
- official: true
- rule: 001

### Models

- url: https://docs.typesafe.ai/models
- description: Aliases, current version, price, and rate limits.
- official: true
- rule: 001

### Model jaggedness: jev-1.13

- url: https://docs.typesafe.ai/model-jaggedness/jev-1.13
- description: Known failure modes, straight from the vendor.
- official: true
- rule: 001

### System One

- url: https://docs.typesafe.ai/concepts/system-one
- description: What the category means and how it differs from a chat model.
- official: true
- rule: 001

### How to build with System One

- url: https://docs.typesafe.ai/concepts/how-to-build-with-system-one
- description: Decompose a judgment into atomic questions and keep the control flow in code.
- official: true
- rule: 001

### Confidence

- url: https://docs.typesafe.ai/confidence
- description: What the confidence field means and how to turn it into act, review, or fall back.
- official: true
- rule: 001

### Patterns

- url: https://docs.typesafe.ai/patterns
- description: Speculative fan-out, confidence routing, composite scoring, intent routing.
- official: true
- rule: 001

### Use-case map

- url: https://docs.typesafe.ai/concepts/use-case-map
- description: The vendor's own catalogue of where Jev fits and where it does not.
- official: true
- rule: 001

### Cookbooks

- url: https://docs.typesafe.ai/cookbooks/parallel_questions
- description: Worked recipes, starting with batching many questions into one request; the sidebar has the rest.
- official: true
- rule: 001

### Workflow evals

- url: https://evals.typesafe.ai/
- description: The vendor's benchmark on four workflows, with the caveats printed on the page.
- official: true
- rule: 001

### llms.txt

- url: https://docs.typesafe.ai/llms.txt
- description: Every documentation page as plain Markdown, for feeding to an agent.
- official: true
- rule: 001

### Console

- url: https://console.typesafe.ai/
- description: Waitlist, API keys, and usage.
- official: true
- rule: 001

## 011 · Official SDKs & Framework Support

Maintained by TypeSafe AI, plus the frameworks that ship Jev natively.

### vercel/eve

- url: https://github.com/vercel/eve
- description: Vercel's agent framework; Jev is the typed judge in its evaluate step.
- stars: 5 418
- language: TypeScript
- license: Apache-2.0
- pushed_at: 2026-09-30
- topics: agent, framework, harness, javascript, markdown, typescript
- use_cases: judge or verify an agent's output; benchmark or calibrate Jev; review code or pull requests
- rule: 011

### typesafe-ai/skills

- url: https://github.com/typesafe-ai/skills
- description: Agent skills for designing questions, building workflows, and evaluating them.
- stars: 2 469
- license: MIT
- official: true
- pushed_at: 2026-09-12
- use_cases: classify support tickets or messages; judge or verify an agent's output; pick which skill or prompt to load
- rule: 011

### vercel-labs/ai-cli

- url: https://github.com/vercel-labs/ai-cli
- description: The Vercel AI SDK in your terminal, with an evaluate path that runs on Jev.
- stars: 817
- language: TypeScript
- pushed_at: 2026-09-30
- use_cases: judge or verify an agent's output; review code or pull requests; call Jev from the command line
- rule: 011

### typesafe-ai/system-one-adapter-python

- url: https://github.com/typesafe-ai/system-one-adapter-python
- description: Same `TypeSafeClient` interface backed by an LLM API, so you can compare Jev against a chat model on identical questions.
- stars: 363
- language: Python
- license: MIT
- official: true
- pushed_at: 2026-09-22
- use_cases: compare Jev with an LLM; benchmark or calibrate Jev; judge or verify an agent's output
- rule: 011

### typesafe-ai/typesafe-sdk-js

- url: https://github.com/typesafe-ai/typesafe-sdk-js
- description: TypeScript and JavaScript client with answer types inferred from your questions.
- stars: 259
- language: TypeScript
- license: MIT
- official: true
- pushed_at: 2026-09-15
- use_cases: review code or pull requests; judge or verify an agent's output; extract structured fields from text
- rule: 011

### typesafe-ai/typesafe-sdk-python

- url: https://github.com/typesafe-ai/typesafe-sdk-python
- description: Python client, sync and async.
- stars: 256
- language: Python
- license: MIT
- official: true
- pushed_at: 2026-09-26
- use_cases: review code or pull requests; judge or verify an agent's output; moderate content or detect abuse
- rule: 011

### Agent skill

- url: https://docs.typesafe.ai/agent-skill
- description: How to install the official skill in Claude Code, Cursor, and friends.
- official: true
- use_cases: drive a browser or GUI with an agent; call Jev from a language SDK; pick which skill or prompt to load
- rule: 011

## 104 · Coding Agents

Routers, tool gates, reviewers, skills and MCP servers for agent harnesses.

### nicobailon/pi-mcp-adapter

- url: https://github.com/nicobailon/pi-mcp-adapter
- description: Opt-in typed evaluation and semantic search over MCP tool results, behind a per-server data-egress allowlist.
- stars: 1 568
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-30
- topics: ai, claude, coding-agent, extension, llm, mcp
- use_cases: compact or prune agent context; judge or verify an agent's output; route requests through a gateway or proxy
- rule: 104

### kerpopule/hermes-jev-skills

- url: https://github.com/kerpopule/hermes-jev-skills
- description: Hands an agent's small decisions to Jev: which model answers the turn, which skills to load, which passages matter, which turns survive compaction.
- stars: 918
- language: Python
- license: MIT
- pushed_at: 2026-09-29
- use_cases: compact or prune agent context; pick which skill or prompt to load; route between models by task difficulty
- rule: 104

### gargpratyush/jev-router

- url: https://github.com/gargpratyush/jev-router
- description: Routes each task to the cheapest Claude model that can handle it.
- stars: 499
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-19
- use_cases: route between models by task difficulty; call Jev from the command line; learn how Jev works
- rule: 104

### jkudish/jev-mcp

- url: https://github.com/jkudish/jev-mcp
- description: The first MCP server for Jev, and still the most linked.
- stars: 464
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-29
- use_cases: serve Jev over MCP to any agent; score or rank candidates in a pipeline; extract structured fields from text
- rule: 104

### notque/vexjoy-agent

- url: https://github.com/notque/vexjoy-agent
- description: Agent toolkit whose `/d` command picks the specialist agent, skill, and pipeline with one Jev call, plus an optional Jev auto-compact plugin.
- stars: 426
- language: Python
- license: MIT
- pushed_at: 2026-09-30
- topics: ai, ai-agents, ai-skills, claude, claude-code, ai-pipelines
- use_cases: gate or approve tool calls before they run; pick which skill or prompt to load; judge or verify an agent's output
- rule: 104

### TianyuCodings/JevHarness

- url: https://github.com/TianyuCodings/JevHarness
- description: Has an LLM write a task-specific harness that turns observations into Jev questions, then freezes it and improves it from rewards and full execution traces.
- stars: 400
- language: Python
- pushed_at: 2026-09-21
- use_cases: judge or verify an agent's output; benchmark or calibrate Jev; compare Jev with an LLM
- rule: 104

### itsmostafa/typesafe-mcp

- url: https://github.com/itsmostafa/typesafe-mcp
- description: Single-binary Go MCP server and CLI that exposes TypeSafe Jev judgments to Claude Desktop, Claude Code and Codex.
- stars: 337
- language: Go
- license: MIT
- pushed_at: 2026-09-30
- topics: mcp, mcp-tools
- use_cases: classify support tickets or messages; serve Jev over MCP to any agent; add Jev to a chat bot or Discord
- rule: 104

### miuuyy/Astra-Ares

- url: https://github.com/miuuyy/Astra-Ares
- description: Has Jev pick the reasoning effort and how long to hold it for a running Codex task, on a patched Codex CLI built from upstream source.
- stars: 293
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-23
- use_cases: benchmark or calibrate Jev; review code or pull requests; learn how Jev works
- rule: 104

### 0xNatoshi/jev-codex-router

- url: https://github.com/0xNatoshi/jev-codex-router
- description: Picks model, thinking depth, and speed mode for every Codex turn.
- stars: 278
- language: JavaScript
- license: MIT
- archived: true
- pushed_at: 2026-09-22
- topics: ai, codex, jev, llm, macos, routing
- use_cases: route between models by task difficulty; benchmark or calibrate Jev; review code or pull requests
- rule: 104

### kitze/skillbox

- url: https://github.com/kitze/skillbox
- description: Self-hosted, versioned skills library served over MCP, with Jev recommending which skill to load.
- stars: 255
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-19
- use_cases: pick which skill or prompt to load; serve Jev over MCP to any agent; run Jev on open models without the vendor
- rule: 104

### DevMortimer/pi-warden

- url: https://github.com/DevMortimer/pi-warden
- description: Guardrails that steer instead of interrupt: irreversible calls, off-task calls, stuck loops, unverified done claims, about 250 ms each.
- stars: 154
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-29
- topics: guardrails, pi-extension, pi-package, typesafe, agent-guardrails
- use_cases: stop an agent from finishing early; judge or verify an agent's output; detect prompt injection or risky commands
- rule: 104

### y0usaf/pi-jev

- url: https://github.com/y0usaf/pi-jev
- description: A measured tool-call gate plus a `jev_ask` tool for typed answers inside Pi.
- stars: 151
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-25
- topics: guardrails, jev, pi-extension, pi-package, typesafe
- use_cases: gate or approve tool calls before they run; judge or verify an agent's output; detect prompt injection or risky commands
- rule: 104

### dbreunig/building-with-jev-skill

- url: https://github.com/dbreunig/building-with-jev-skill
- description: Skill for writing and improving programs that call Jev.
- stars: 145
- pushed_at: 2026-09-17
- use_cases: score or rank candidates in a pipeline; add Jev to a chat bot or Discord; pick which skill or prompt to load
- rule: 104

### Dicklesworthstone/skillranker

- url: https://github.com/Dicklesworthstone/skillranker
- description: Rust CLI and hooks that rank installed skills for the next step using live session context, with abstention.
- stars: 125
- language: Rust
- license: NOASSERTION
- pushed_at: 2026-09-29
- topics: agent-skills, ai-agents, asupersync, claude-code, cli, developer-tools
- use_cases: pick which skill or prompt to load; benchmark or calibrate Jev; call Jev from the command line
- rule: 104

### devagrawal09/jev-code

- url: https://github.com/devagrawal09/jev-code
- description: Command-line toolkit that coding agents hand judgment-heavy work to, one typed Jev workflow per request.
- stars: 119
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-19
- use_cases: judge or verify an agent's output; review code or pull requests; call Jev from the command line
- rule: 104

### devagrawal09/stanley-code

- url: https://github.com/devagrawal09/stanley-code
- description: Coding CLI where Jev routes a plain-language request to one deterministic workflow, and that workflow asks Jev fixed-choice questions about the evidence it gathered.
- stars: 119
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-19
- use_cases: judge or verify an agent's output; review code or pull requests; call Jev from the command line
- rule: 104

### fabricioctelles/skills

- url: https://github.com/fabricioctelles/skills
- description: Agent-skill directory that can score subjective evaluation criteria with Jev.
- stars: 97
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-27
- topics: agentic-ai, agentic-skills, skills, steering, claude-code, claude-skills
- use_cases: judge or verify an agent's output; pick which skill or prompt to load; classify support tickets or messages
- rule: 104

### EliaAlberti/jev-rules

- url: https://github.com/EliaAlberti/jev-rules
- description: Scores your standing rules against each prompt and delivers only the ones that apply, once per session.
- stars: 63
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-27
- use_cases: compact or prune agent context; learn how Jev works; pick which skill or prompt to load
- rule: 104

### TheoOliveira/pi-jev

- url: https://github.com/TheoOliveira/pi-jev
- description: Semantic tool routing and typed decisions as Pi tools.
- stars: 58
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-24
- topics: jev, pi-coding-agent, pi-extension, pi-package, system-one, tool-routing
- use_cases: pick which skill or prompt to load; compact or prune agent context; judge or verify an agent's output
- rule: 104

### DevMortimer/pi-typesafe

- url: https://github.com/DevMortimer/pi-typesafe
- description: Batched evaluation tool, terminal playground, and a typed API for Pi extension authors.
- stars: 49
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-30
- use_cases: monitor or observe Jev usage and cost; call Jev from the command line; benchmark or calibrate Jev
- rule: 104

### tacticocc/Jevbridge

- url: https://github.com/tacticocc/Jevbridge
- description: ACP and MCP adapter that pairs Jev with any LLM for computer use and typed decisions.
- stars: 46
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-21
- topics: acp, agent, claude, computer-use, grok, jev
- use_cases: gate or approve tool calls before they run; run Jev on open models without the vendor; serve Jev over MCP to any agent
- rule: 104

### shantanugoel/ask-jev-skill

- url: https://github.com/shantanugoel/ask-jev-skill
- description: Lets Hermes and similar agents ask Jev directly.
- stars: 42
- language: Python
- license: MIT
- pushed_at: 2026-09-17
- use_cases: judge or verify an agent's output; score or rank candidates in a pipeline; compare Jev with an LLM
- rule: 104

### shitianfang/jev-use

- url: https://github.com/shitianfang/jev-use
- description: Hands the Claude Code, Codex and pi steps that need no text output to Jev, with a typed escalation contract for everything it should not decide.
- stars: 31
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-22
- topics: agent, ai-agents, claude-code, claude-code-plugin, codex, jev
- use_cases: gate or approve tool calls before they run; compact or prune agent context; drive a browser or GUI with an agent
- rule: 104

### jomatsu/pi-jev-auto-mode

- url: https://github.com/jomatsu/pi-jev-auto-mode
- description: Auto-approves bash, write, and edit calls semantically and fails closed when it cannot decide.
- stars: 31
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-24
- topics: auto-mode, coding-agent, guardrail, jev, permission, pi-coding-agent
- use_cases: gate or approve tool calls before they run; detect prompt injection or risky commands; benchmark or calibrate Jev
- rule: 104

### keeltrace/hermes-jev

- url: https://github.com/keeltrace/hermes-jev
- description: Typed decisions, ranking, verification, and an opt-in tool gate.
- stars: 31
- language: Python
- license: MIT
- pushed_at: 2026-09-28
- use_cases: gate or approve tool calls before they run; judge or verify an agent's output; run Jev on open models without the vendor
- rule: 104

### compozy/yoshi

- url: https://github.com/compozy/yoshi
- description: Context-pruning proxy for Claude Code and Codex, with the savings measured rather than claimed.
- stars: 27
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-18
- topics: bun, claude-code, codex, context-window, jev, llm-proxy
- use_cases: compact or prune agent context; route requests through a gateway or proxy; benchmark or calibrate Jev
- rule: 104

### blakestone-x/jev-mcp

- url: https://github.com/blakestone-x/jev-mcp
- description: Classify, score, check, match, and screen, with confidence on every answer.
- stars: 26
- language: Python
- license: MIT
- pushed_at: 2026-09-16
- use_cases: serve Jev over MCP to any agent; judge or verify an agent's output; score or rank candidates in a pipeline
- rule: 104

### GodsBoy/jev-agent-skill-router

- url: https://github.com/GodsBoy/jev-agent-skill-router
- description: Confidence-aware skill routing with an abstain path.
- stars: 24
- language: Python
- license: MIT
- pushed_at: 2026-09-16
- topics: agentic-ai, ai-agents, hermes-agent, jev, python, skill-routing
- use_cases: pick which skill or prompt to load; judge or verify an agent's output; benchmark or calibrate Jev
- rule: 104

### Brainwires/jevwire

- url: https://github.com/Brainwires/jevwire
- description: MCP server, embeddable decision model, and an escalate-only plugin that can make the harness stricter but never looser.
- stars: 22
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-21
- use_cases: gate or approve tool calls before they run; serve Jev over MCP to any agent; judge or verify an agent's output
- rule: 104

### valentynkit/jev-belay

- url: https://github.com/valentynkit/jev-belay
- description: Stop hook that blocks an unverified "done": reads the transcript for evidence and, only when files changed with no passing check since, spends one four-question Jev call; fails open on every error path.
- stars: 20
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-20
- use_cases: judge or verify an agent's output; stop an agent from finishing early; benchmark or calibrate Jev
- rule: 104

### anpicasso/hermes-jev-approvals

- url: https://github.com/anpicasso/hermes-jev-approvals
- description: Approves, denies, or escalates flagged shell commands before they run; vendor-reported speedups.
- stars: 20
- language: Python
- license: MIT
- pushed_at: 2026-09-22
- use_cases: gate or approve tool calls before they run; detect prompt injection or risky commands; benchmark or calibrate Jev
- rule: 104

### mejiasd3v/pi-jev-router

- url: https://github.com/mejiasd3v/pi-jev-router
- description: Automatic model routing for Pi through the Vercel AI Gateway.
- stars: 16
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-22
- use_cases: route requests through a gateway or proxy; route between models by task difficulty; call Jev from the command line
- rule: 104

### DECRUX9812/typesafe-skill-router

- url: https://github.com/DECRUX9812/typesafe-skill-router
- description: Names the one skill worth loading before the model call; stdlib only, about a tenth of a cent per turn.
- stars: 15
- language: Python
- license: MIT
- pushed_at: 2026-09-21
- topics: ai-agents, hermes-agent, hermes-plugin, llm-agents, plugin-catalog, prompt-caching
- use_cases: pick which skill or prompt to load; compact or prune agent context; review code or pull requests
- rule: 104

### kubet/azdaja

- url: https://github.com/kubet/azdaja
- description: Recursive language-model layer for Claude Code, Codex, Gemini and OpenCode that keeps sources local; Jev is an optional leaf for reranking, verification and semantic joins.
- stars: 14
- language: Python
- license: MIT
- pushed_at: 2026-09-21
- topics: agent-skills, jcode, recursive-language-models, rust, gemini-cli-extension, claude-code
- use_cases: judge or verify an agent's output; compact or prune agent context; extract structured fields from text
- rule: 104

### HyunjunJeon/pi-quiet-ask

- url: https://github.com/HyunjunJeon/pi-quiet-ask
- description: Jev as the Pi coding agent's quiet decision layer.
- stars: 12
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-18
- use_cases: gate or approve tool calls before they run; judge or verify an agent's output; detect prompt injection or risky commands
- rule: 104

### harshwasan/pi-jev-sentinel

- url: https://github.com/harshwasan/pi-jev-sentinel
- description: Checks Pi tool calls, tool outputs, and replies for risky actions and prompt injection, with user approvals, context re-checks, secret scrubbing, and optional task pinning.
- stars: 11
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-20
- use_cases: gate or approve tool calls before they run; detect prompt injection or risky commands; judge or verify an agent's output
- rule: 104

### 24601/Augustus

- url: https://github.com/24601/Augustus
- description: Skill for deciding where a typed judgment belongs at all and what stays in code; a companion to the official skill, not a replacement.
- stars: 11
- language: Python
- license: MIT
- pushed_at: 2026-09-28
- topics: agent-skills, agent-workflows, ai-agents, calibrated-confidence, claude-code, decision-systems
- use_cases: score or rank candidates in a pipeline; benchmark or calibrate Jev; rerank search results
- rule: 104

### adarshmishra07/jcm-router

- url: https://github.com/adarshmishra07/jcm-router
- description: Local proxy that picks model and effort per message and leaves the cached main chat alone.
- stars: 9
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-17
- use_cases: route between models by task difficulty; route requests through a gateway or proxy; benchmark or calibrate Jev
- rule: 104

### AbdelStark/bicameral

- url: https://github.com/AbdelStark/bicameral
- description: Hybrid harness for Pi: an LLM writes the code, Jev reflexes gate every call as allow, confirm, block, warn, or steer.
- stars: 9
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-16
- topics: harness-engineering, hybrid-machine-learning, machine-learning
- use_cases: gate or approve tool calls before they run; judge or verify an agent's output; detect prompt injection or risky commands
- rule: 104

### ajensenwaud/hermes-jev-plugin

- url: https://github.com/ajensenwaud/hermes-jev-plugin
- description: Four Hermes tools for atomic checks, routing, and rubric scoring; listed in the Hermes plugin catalog.
- stars: 8
- language: Python
- license: MIT
- pushed_at: 2026-09-19
- topics: ai-agents, decision-making, hermes, hermes-agent, jev, typesafe
- use_cases: judge or verify an agent's output; classify support tickets or messages; score or rank candidates in a pipeline
- rule: 104

### HyunjunJeon/jev-judgment

- url: https://github.com/HyunjunJeon/jev-judgment
- description: Sends a coding agent's closed judgments to Jev instead of the chat model.
- stars: 7
- language: Python
- license: MIT
- pushed_at: 2026-09-17
- use_cases: gate or approve tool calls before they run; judge or verify an agent's output; voice or realtime decisions
- rule: 104

### 3clyp50/a0-typesafe-ai

- url: https://github.com/3clyp50/a0-typesafe-ai
- description: Typed tools and probability cards for Agent Zero.
- stars: 6
- language: Python
- license: MIT
- pushed_at: 2026-09-17
- use_cases: judge or verify an agent's output; score or rank candidates in a pipeline; add Jev to a chat bot or Discord
- rule: 104

### bestagentkits/jev-skillful

- url: https://github.com/bestagentkits/jev-skillful
- description: Per-prompt router over skills, MCP servers, agents, and commands, and it measures whether the injection helped.
- stars: 5
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-17
- topics: agent-tooling, ai-agents, claude-code, codex, coding-agents, developer-tools
- use_cases: compact or prune agent context; pick which skill or prompt to load; benchmark or calibrate Jev
- rule: 104

### noplan-inc/limpet

- url: https://github.com/noplan-inc/limpet
- description: A Stop hook that keeps the agent from stopping too early, judged against plain-language rules.
- stars: 5
- language: Python
- license: MIT
- pushed_at: 2026-09-17
- topics: ai-agents, claude-code, codex, hooks, jev
- use_cases: stop an agent from finishing early; judge or verify an agent's output; benchmark or calibrate Jev
- rule: 104

### suenot/codex-jev-router

- url: https://github.com/suenot/codex-jev-router
- description: Uses Jev to select the model and reasoning effort for Codex subagents, with confidence gates and a Sol fallback.
- stars: 5
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-27
- use_cases: judge or verify an agent's output; compact or prune agent context; compare Jev with an LLM
- rule: 104

### legacybridge-tech/pi-typesafe-jev

- url: https://github.com/legacybridge-tech/pi-typesafe-jev
- description: Pi extension exposing Jev judgments as five Pi tools.
- stars: 4
- language: TypeScript
- license: NOASSERTION
- pushed_at: 2026-09-17
- use_cases: judge or verify an agent's output; score or rank candidates in a pipeline; classify support tickets or messages
- rule: 104

### BYK/jev-mcp

- url: https://github.com/BYK/jev-mcp
- description: Eval-first MCP server: prototype a question, map it over many items, then measure variants against labeled examples with a threshold sweep.
- stars: 4
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-18
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; classify support tickets or messages
- rule: 104

### samtay32/jev-system-architect

- url: https://github.com/samtay32/jev-system-architect
- description: Finds the fuzzy judgment in a system and turns it into small Choice, Score, and Noul primitives.
- stars: 3
- license: MIT
- pushed_at: 2026-09-17
- use_cases: classify support tickets or messages; extract structured fields from text; score or rank candidates in a pipeline
- rule: 104

## 118 · Routing & Gateways

Per-turn model and tool selection, under a hard deadline.

### BillionsBobby/JevRouter

- url: https://github.com/BillionsBobby/JevRouter
- description: Models, subagents, skills, MCP tools, and CLIs as one candidate set; Jev picks, the router enforces permissions and risk; reports 44 percent first-five tool-call hits against 24 for DeepSeek on Toolathlon.
- stars: 303
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-26
- use_cases: gate or approve tool calls before they run; pick which skill or prompt to load; benchmark or calibrate Jev
- rule: 118

### vinilana/jev-gateway

- url: https://github.com/vinilana/jev-gateway
- description: Local gateway for Codex and Claude Code that sends the "which tool next" decision to Jev and everything else to your usual model.
- stars: 267
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-25
- use_cases: gate or approve tool calls before they run; route requests through a gateway or proxy; compare Jev with an LLM
- rule: 118

### juspay/neurolink

- url: https://github.com/juspay/neurolink
- description: TypeScript AI SDK where decide, via Jev, is a peer of generate and stream: one typed-judgment call routes model choice, prunes context, and picks MCP tools.
- stars: 144
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-30
- topics: ai, developer-tools, llm, mcp, model-context-protocol, agents
- use_cases: compact or prune agent context; route between models by task difficulty; gate or approve tool calls before they run
- rule: 118

### nidhi-singh02/agent-router

- url: https://github.com/nidhi-singh02/agent-router
- description: Picks Cursor, Claude Code, Codex, or OpenCode plus model and effort for a task, then launches it.
- stars: 99
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-27
- topics: agents, ai, cli, coding, developer-tools, herdr
- use_cases: route between models by task difficulty; call Jev from the command line; review code or pull requests
- rule: 118

### yusukebe/hono-jev-router

- url: https://github.com/yusukebe/hono-jev-router
- description: Route HTTP requests by meaning in Hono.
- stars: 51
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-18
- use_cases: benchmark or calibrate Jev; compare Jev with an LLM; review code or pull requests
- rule: 118

### xinyao27/jevonian

- url: https://github.com/xinyao27/jevonian
- description: Local OpenAI, Anthropic and Responses-compatible proxy that serves one Jev call per turn to pick both the model route and the thinking level, filtering candidates by context window and quota.
- stars: 16
- language: TypeScript
- license: AGPL-3.0
- pushed_at: 2026-09-29
- use_cases: route between models by task difficulty; route requests through a gateway or proxy; monitor or observe Jev usage and cost
- rule: 118

### prismhq/jev-router

- url: https://github.com/prismhq/jev-router
- description: LLM router on top of LiteLLM.
- stars: 15
- language: Python
- license: MIT
- pushed_at: 2026-09-17
- use_cases: route requests through a gateway or proxy; add Jev to a chat bot or Discord; compare Jev with an LLM
- rule: 118

### iamvatsalpatel/tiershift

- url: https://github.com/iamvatsalpatel/tiershift
- description: Shifts every LLM call to the cheapest model that can handle it, policy in YAML, decision in about 180 ms.
- stars: 4
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-28
- topics: agents, cost-optimization, jev, llm, model-routing, typesafe
- use_cases: route between models by task difficulty; route requests through a gateway or proxy; monitor or observe Jev usage and cost
- rule: 118

### FirasSX914/Janus

- url: https://github.com/FirasSX914/Janus
- description: Measures on your data when Jev beats other models, then routes accordingly.
- stars: 3
- language: Python
- license: MIT
- pushed_at: 2026-09-18
- topics: benchmark, calibration, confidence, deepseek, jev, llm
- use_cases: benchmark or calibrate Jev; compare Jev with an LLM; judge or verify an agent's output
- rule: 118

### daviddl9/jev-router

- url: https://github.com/daviddl9/jev-router
- description: Jev picks the worker tier for each step in OMP and Pi, keeping planning and review on a strong model and bounded work on cheaper ones.
- stars: 0
- language: TypeScript
- license: NOASSERTION
- pushed_at: 2026-09-21
- use_cases: route between models by task difficulty; compact or prune agent context; judge or verify an agent's output
- rule: 118

## 131 · Context & Compaction

Decide what stays in the window before the model ever reads it.

### tamaratran/fast-jev-compaction

- url: https://github.com/tamaratran/fast-jev-compaction
- description: Replaces the compaction summary with Jev decisions: every tool call and result scored in one request, stale ones dropped, everything kept stays verbatim.
- stars: 7 225
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-18
- use_cases: compact or prune agent context; judge or verify an agent's output; benchmark or calibrate Jev
- rule: 131

### tamaratran/jev-pruner

- url: https://github.com/tamaratran/jev-pruner
- description: Trims long Bash output with Jev after the command runs and before the model sees it; short output, errors, and structured formats pass untouched.
- stars: 153
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-30
- use_cases: compact or prune agent context; judge or verify an agent's output; review code or pull requests
- rule: 131

### GhalebDweikat/winnow

- url: https://github.com/GhalebDweikat/winnow
- description: Judges every tool result before it enters context, so the window fills slower instead of being cleaned later.
- stars: 100
- language: Python
- license: MIT
- pushed_at: 2026-09-30
- topics: claude-code, claude-code-plugin, context-management, jev, llm-agents, typesafe
- use_cases: compact or prune agent context; judge or verify an agent's output; compare Jev with an LLM
- rule: 131

### IAmUnbounded/save-token-jev-clean

- url: https://github.com/IAmUnbounded/save-token-jev-clean
- description: Compaction that asks Jev which tool calls still matter and keeps the rest verbatim, with adapters for Claude Code, Codex, OpenCode, and raw API transcripts.
- stars: 76
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-18
- use_cases: compact or prune agent context; judge or verify an agent's output; run Jev on open models without the vendor
- rule: 131

### joelhooks/pi-fast-jev-compaction

- url: https://github.com/joelhooks/pi-fast-jev-compaction
- description: The verbatim compaction idea, ported to Pi.
- stars: 14
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-18
- use_cases: compact or prune agent context; benchmark or calibrate Jev; compare Jev with an LLM
- rule: 131

### Nyarlathoteppppp/pi-heed

- url: https://github.com/Nyarlathoteppppp/pi-heed
- description: Checks every side-effecting tool call against what you said earlier in the session, so "review only" still holds after compaction.
- stars: 11
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-19
- topics: ai-agents, coding-agent, guardrails, jev, pi-coding-agent, pi-package
- use_cases: gate or approve tool calls before they run; compact or prune agent context; judge or verify an agent's output
- rule: 131

## 147 · Code Review & Quality

Judges, linters, coverage gates and review dashboards.

### devagrawal09/jev-review

- url: https://github.com/devagrawal09/jev-review
- description: Staged code-review workflow with a local dashboard.
- stars: 643
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-17
- topics: ai, code-review, jev, typesafe-ai, typescript
- use_cases: review code or pull requests; judge or verify an agent's output; learn how Jev works
- rule: 147

### thruwire/foreman

- url: https://github.com/thruwire/foreman
- description: Supervises a software factory of agents, with Jev making the go and no-go calls.
- stars: 619
- language: Python
- license: MIT
- pushed_at: 2026-09-28
- use_cases: judge or verify an agent's output; review code or pull requests; benchmark or calibrate Jev
- rule: 147

### lakeday-org/perch

- url: https://github.com/lakeday-org/perch
- description: Semantic linting: rules in plain language, each file judged by Jev, run locally or in CI.
- stars: 316
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-30
- topics: ai, cli, code-quality, code-review, devtools, linter
- use_cases: review code or pull requests; call Jev from the command line; judge or verify an agent's output
- rule: 147

### NiazMorshed2007/jev-review

- url: https://github.com/NiazMorshed2007/jev-review
- description: Local-first MCP plugin for continuous quality review by coding agents.
- stars: 231
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-17
- topics: agent-plugin, ai-agents, claude-code, code-review, codex, coding-agents
- use_cases: judge or verify an agent's output; serve Jev over MCP to any agent; review code or pull requests
- rule: 147

### supercorp-ai/supercov

- url: https://github.com/supercorp-ai/supercov
- description: Code quality and coverage signals for coding agents.
- stars: 141
- language: Rust
- license: MIT
- pushed_at: 2026-09-27
- topics: coverage, jev, typesafe, gemini-cli-extension, claude-code, code-coverage
- use_cases: judge or verify an agent's output; review code or pull requests; benchmark or calibrate Jev
- rule: 147

### kyu1204/jgrep

- url: https://github.com/kyu1204/jgrep
- description: `--diff` gates a PR in CI on a rule written in English, `--tests` lists the test files a diff can affect, and plain `jgrep` greps code by what it does; one Noul per chunk, 16 chunks per request.
- stars: 58
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-30
- use_cases: review code or pull requests; call Jev from the command line; judge or verify an agent's output
- rule: 147

### Alurith/jeff

- url: https://github.com/Alurith/jeff
- description: Read-only Go CLI that checks files against coded rules such as hidden side effects and weak error handling.
- stars: 38
- language: Go
- license: Apache-2.0
- pushed_at: 2026-09-20
- topics: golang, jev
- use_cases: review code or pull requests; call Jev from the command line; judge or verify an agent's output
- rule: 147

### devanshbatham/commit-miner

- url: https://github.com/devanshbatham/commit-miner
- description: Classifies commit diffs and messages: bug fixes, security fixes with CWEs, change types.
- stars: 36
- language: Rust
- pushed_at: 2026-09-17
- use_cases: review code or pull requests; label data or build a dataset; judge or verify an agent's output
- rule: 147

### lukstei/slop-grader

- url: https://github.com/lukstei/slop-grader
- description: Grades text and markdown files for AI slop, grammar, and technical documentation quality, and guides an AI agent to auto-fix violations.
- stars: 32
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-24
- topics: ai-agents, jev, no-ai-slop, system-one, ai-slop, ai-writing
- use_cases: judge or verify an agent's output; call Jev from the command line; compare Jev with an LLM
- rule: 147

### valentynkit/jev-commit

- url: https://github.com/valentynkit/jev-commit
- description: Pre-commit hook: one Jev call judges whether the commit message matches the staged diff, plus debug leftovers, unmentioned work, and a credential belt; warns except on a secret, which it blocks.
- stars: 13
- language: Python
- license: MIT
- pushed_at: 2026-09-19
- use_cases: judge or verify an agent's output; call Jev from the command line; compare Jev with an LLM
- rule: 147

### frostney/clean-code-review

- url: https://github.com/frostney/clean-code-review
- description: Every file in a PR judged against Clean Code rules, then reviewed by an LLM.
- stars: 12
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-23
- topics: agents, ai, ai-gateway, ai-sdk, clean-code, code-quality
- use_cases: review code or pull requests; judge or verify an agent's output; compare Jev with an LLM
- rule: 147

### huntedman/JevLint

- url: https://github.com/huntedman/JevLint
- description: Configurable semantic linting with file-level Noul judgments.
- stars: 12
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-20
- use_cases: judge or verify an agent's output; review code or pull requests; call Jev from the command line
- rule: 147

### doeixd/jev-pref

- url: https://github.com/doeixd/jev-pref
- description: Turns the preferences in your AGENTS.md into a linter that runs on code changes and reports back to the agent.
- stars: 11
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-18
- topics: agent-skills, claude-code, code-review, jev, skills-sh, typesafe
- use_cases: judge or verify an agent's output; review code or pull requests; pick which skill or prompt to load
- rule: 147

### nozomi-koborinai/jev-spec

- url: https://github.com/nozomi-koborinai/jev-spec
- description: Checks the code against the requirements in a Markdown spec on every commit and fails the build when the two drift apart.
- stars: 11
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-29
- topics: ai, jev, linter, typesafe-ai, typescript, agent-skills
- use_cases: call Jev from the command line; judge or verify an agent's output; review code or pull requests
- rule: 147

### HexyeDEV/JevPR

- url: https://github.com/HexyeDEV/JevPR
- description: GitHub App that asks Jev whether a pull request is safe to approve or needs a specialist, then maps the verdict to a check run.
- stars: 10
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-25
- topics: classification, jev, pull-requests, python
- use_cases: judge or verify an agent's output; review code or pull requests; route between models by task difficulty
- rule: 147

### stratonext/software-factory

- url: https://github.com/stratonext/software-factory
- description: Runs several coding agents locally with Jev as judge and orchestrator.
- stars: 9
- language: Python
- license: MIT
- pushed_at: 2026-09-30
- use_cases: judge or verify an agent's output; run Jev on open models without the vendor; benchmark or calibrate Jev
- rule: 147

### raihankhan-rk/diffjury

- url: https://github.com/raihankhan-rk/diffjury
- description: PR risk router and review coach.
- stars: 8
- language: TypeScript
- pushed_at: 2026-09-22
- use_cases: review code or pull requests; judge or verify an agent's output; compact or prune agent context
- rule: 147

### cephalization/jev-triage

- url: https://github.com/cephalization/jev-triage
- description: Pulls large repositories and triages their issues with typed Jev questions.
- stars: 5
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-29
- use_cases: review code or pull requests; judge or verify an agent's output; voice or realtime decisions
- rule: 147

### Ramneet-Singh/jevopt

- url: https://github.com/Ramneet-Singh/jevopt
- description: C/C++ compiler driver that asks Jev whether to inline each discretionary call site, from the LLVM IR and the original source.
- stars: 4
- language: Python
- license: GPL-3.0
- pushed_at: 2026-09-21
- use_cases: benchmark or calibrate Jev; compare Jev with an LLM; judge or verify an agent's output
- rule: 147

### allebee/pytest-jev

- url: https://github.com/allebee/pytest-jev
- description: Pytest plugin that asks Jev whether plain-English claims about a test's text hold, all in one request, and fails the test with each claim's probability unless Jev is at least 80 percent sure.
- stars: 3
- language: Python
- license: MIT
- pushed_at: 2026-09-21
- topics: jev, llm, llm-evaluation, pytest, pytest-plugin, python
- use_cases: judge or verify an agent's output; compare Jev with an LLM; review code or pull requests
- rule: 147

### fatwang2/jev-review-action

- url: https://github.com/fatwang2/jev-review-action
- description: GitHub Action for submission review and PR classification with Jev, no text-generation model in the loop.
- stars: 2
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-20
- topics: automation, awesome-list, classification, github-action, jev, pull-request
- use_cases: review code or pull requests; score or rank candidates in a pipeline; route requests through a gateway or proxy
- rule: 147

## 152 · Browser & Computer Use

Jev picks the operation and the DOM element; a small LLM only writes the text.

### browser-use/jev-ultrafast

- url: https://github.com/browser-use/jev-ultrafast
- description: One request picks both the operation and the target element from an indexed DOM table; a small LLM only writes typed text. Zürich to London booked in 7.1 seconds.
- stars: 21 495
- language: Python
- license: MIT
- pushed_at: 2026-09-30
- use_cases: drive a browser or GUI with an agent; benchmark or calibrate Jev; learn how Jev works
- rule: 152

### awlevin/typesafe-computer-use

- url: https://github.com/awlevin/typesafe-computer-use
- description: OCR the screen, classify the next action, click; about $0.0002 a step on macOS.
- stars: 1 097
- language: Python
- license: MIT
- pushed_at: 2026-09-29
- topics: ai-agents, automation, computer-use, macos, ocr, typesafe
- use_cases: drive a browser or GUI with an agent; learn how Jev works; benchmark or calibrate Jev
- rule: 152

### wy-coliney/jev-browser-use

- url: https://github.com/wy-coliney/jev-browser-use
- description: Codex skill and plugin where Jev handles navigation, clicks, and scrolling and Codex keeps typing and verification; reports browser steps 5 to 10 times faster.
- stars: 707
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-23
- use_cases: drive a browser or GUI with an agent; judge or verify an agent's output; learn how Jev works
- rule: 152

### Sac-Y/Jev-cu

- url: https://github.com/Sac-Y/Jev-cu
- description: Codex computer use where Jev picks the element, action, completion, and risk from on-screen text, no screenshots sent; Chinese readme.
- stars: 613
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-22
- use_cases: gate or approve tool calls before they run; drive a browser or GUI with an agent; learn how Jev works
- rule: 152

### moritzkremb/jev-voice-browser

- url: https://github.com/moritzkremb/jev-voice-browser
- description: Intent and target decided per spoken word in about 300 ms, often before the sentence ends.
- stars: 371
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-21
- use_cases: drive a browser or GUI with an agent; voice or realtime decisions; compare Jev with an LLM
- rule: 152

### jkudish/jev-browser

- url: https://github.com/jkudish/jev-browser
- description: The first community browser agent on Jev, with a demo GIF.
- stars: 297
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-29
- use_cases: call Jev from the command line; drive a browser or GUI with an agent; benchmark or calibrate Jev
- rule: 152

### YUTA-fywoo/jev-gui-delegate

- url: https://github.com/YUTA-fywoo/jev-gui-delegate
- description: Runs a delegated GUI task for Codex through the real Chrome session or Windows UI Automation, with Jev picking the control at each step.
- stars: 131
- language: Python
- pushed_at: 2026-09-27
- use_cases: drive a browser or GUI with an agent; judge or verify an agent's output; monitor or observe Jev usage and cost
- rule: 152

### socai-io/jev-social

- url: https://github.com/socai-io/jev-social
- description: Lets Jev choose each read-only social research step while socai runs it in a real Chrome session and streams inspectable Instagram, TikTok, or LinkedIn evidence into a report.
- stars: 128
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-30
- topics: jev, social-media, ai-agents, browser-automation, instagram, research-agent
- use_cases: judge or verify an agent's output; drive a browser or GUI with an agent; gate or approve tool calls before they run
- rule: 152

### savka777/jev-use

- url: https://github.com/savka777/jev-use
- description: Voice and typed computer use for macOS: Jev picks the next on-screen action from the Accessibility tree, with no screenshots.
- stars: 112
- language: Swift
- license: MIT
- pushed_at: 2026-09-21
- topics: ai, browser-use, computer-use, harness, jev, typesafe-ai
- use_cases: drive a browser or GUI with an agent; call Jev from the command line; voice or realtime decisions
- rule: 152

### Ying-Kai-Liao/jev-browser

- url: https://github.com/Ying-Kai-Liao/jev-browser
- description: An LLM plans, Jev decides; library, CLI, and MCP server.
- stars: 94
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-29
- topics: browser-automation, llm-agents, mcp, mcp-server, playwright
- use_cases: drive a browser or GUI with an agent; call Jev from the command line; gate or approve tool calls before they run
- rule: 152

### hqman/JevScout

- url: https://github.com/hqman/JevScout
- description: Job-hunting skill that drives Chrome over CDP and has Jev score every link and listing.
- stars: 38
- language: Python
- pushed_at: 2026-09-18
- use_cases: drive a browser or GUI with an agent; call Jev from the command line; compare Jev with an LLM
- rule: 152

### romaluev/jev-ego

- url: https://github.com/romaluev/jev-ego
- description: Browser agent that spends one Jev request per step to pick the action.
- stars: 17
- language: TypeScript
- license: NOASSERTION
- pushed_at: 2026-09-17
- topics: ai-agent, browser-agent, cdp, cli, ego-browser, ego-lite
- use_cases: drive a browser or GUI with an agent; benchmark or calibrate Jev; learn how Jev works
- rule: 152

### tontoko/jev-browser

- url: https://github.com/tontoko/jev-browser
- description: One grounded Jev and Playwright core behind a typed SDK, a persistent CLI, and an MCP server.
- stars: 11
- language: JavaScript
- license: Apache-2.0
- pushed_at: 2026-09-30
- topics: browser-automation, cli, jev, mcp, playwright, typescript
- use_cases: drive a browser or GUI with an agent; call Jev from the command line; judge or verify an agent's output
- rule: 152

### imanshu03/jev-browser-use

- url: https://github.com/imanshu03/jev-browser-use
- description: Runs browser tasks from a plain-language instruction over CDP or Vercel's agent-browser, with Jev selecting operations and targets and code checking confidence before it acts.
- stars: 0
- language: TypeScript
- pushed_at: 2026-09-27
- use_cases: call Jev from the command line; drive a browser or GUI with an agent; serve Jev over MCP to any agent
- rule: 152

## 158 · Mobile & Desktop Automation

Driving phones, IM clients and native UIs without hooking or patching.

### droidrun/mobile-jev

- url: https://github.com/droidrun/mobile-jev
- description: The same loop on a real Android phone; nine Uber actions in 21 seconds in the demo.
- stars: 426
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-17
- use_cases: control a phone or mobile app; benchmark or calibrate Jev; drive a browser or GUI with an agent
- rule: 158

### ainame/swift-typesafe

- url: https://github.com/ainame/swift-typesafe
- description: Swift 6.4 SDK following the Python SDK's API, on Apple platforms and Linux.
- stars: 16
- language: Swift
- license: MIT
- pushed_at: 2026-09-23
- use_cases: judge or verify an agent's output; extract structured fields from text; score or rank candidates in a pipeline
- rule: 158

### friedjof/jev-mobile

- url: https://github.com/friedjof/jev-mobile
- description: Android sub-agent over USB running observe, normalize, decide, mutate, verify, with Jev deciding.
- stars: 8
- language: Python
- license: MIT
- pushed_at: 2026-09-18
- topics: agents, android, automation, jev, mcp, mobile
- use_cases: control a phone or mobile app; judge or verify an agent's output; drive a browser or GUI with an agent
- rule: 158

### xinwang-nwpu/jev-mobile

- url: https://github.com/xinwang-nwpu/jev-mobile
- description: Android automation where one Jev request picks both the action and the target element from the accessibility tree, executed over ADB.
- stars: 3
- language: Python
- license: MIT
- pushed_at: 2026-09-21
- topics: mobile-agent, jev
- use_cases: control a phone or mobile app; compare Jev with an LLM; drive a browser or GUI with an agent
- rule: 158

## 163 · Search, Reranking & RAG

Query understanding, source selection, reranking and semantic SQL.

### superagents-lab/jev-search

- url: https://github.com/superagents-lab/jev-search
- description: Source selection, query understanding, and relevance ranking for web search.
- stars: 493
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-20
- use_cases: rerank search results; learn how Jev works; compare Jev with an LLM
- rule: 163

### jexp/neo4jev

- url: https://github.com/jexp/neo4jev
- description: Walks a Neo4j graph by classifying neighbouring relationships.
- stars: 154
- language: Jupyter Notebook
- license: MIT
- pushed_at: 2026-09-18
- use_cases: compare Jev with an LLM; learn how Jev works; benchmark or calibrate Jev
- rule: 163

### uehaj/jev-semgrep

- url: https://github.com/uehaj/jev-semgrep
- description: Greps by meaning instead of by regex: every line gets a probability from Jev, and meanings combine with AND, OR, and NOT.
- stars: 145
- language: JavaScript
- license: NOASSERTION
- pushed_at: 2026-09-30
- use_cases: call Jev from the command line; moderate content or detect abuse; rerank search results
- rule: 163

### ellipsis-dev/blink

- url: https://github.com/ellipsis-dev/blink
- description: Codebase search where Jev scores the candidates.
- stars: 93
- language: TypeScript
- pushed_at: 2026-09-16
- use_cases: call Jev from the command line; score or rank candidates in a pipeline; judge or verify an agent's output
- rule: 163

### kbhuw/jev-sift

- url: https://github.com/kbhuw/jev-sift
- description: MCP tool that scores a batch of files, URLs, or snippets for relevance so the agent opens only what matters.
- stars: 48
- language: JavaScript
- pushed_at: 2026-09-18
- use_cases: compact or prune agent context; rerank search results; score or rank candidates in a pipeline
- rule: 163

### hev/reranker

- url: https://github.com/hev/reranker
- description: Jev as a calibrated reranker: one call, up to 30 documents, a probability per document.
- stars: 15
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-17
- use_cases: benchmark or calibrate Jev; score or rank candidates in a pipeline; compare Jev with an LLM
- rule: 163

### reachjalil/jev-tree

- url: https://github.com/reachjalil/jev-tree
- description: Recursive choice over a taxonomy, past the 255-option cap.
- stars: 10
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-18
- topics: classification, jev, taxonomy, typesafe, typescript
- use_cases: classify support tickets or messages; review code or pull requests; add Jev to a chat bot or Discord
- rule: 163

### WiktorB2004/llama-index-jev

- url: https://github.com/WiktorB2004/llama-index-jev
- description: LlamaIndex reranker and router, cheaper than an LLM judge.
- stars: 9
- language: Python
- license: MIT
- pushed_at: 2026-09-25
- topics: llamaindex, rag, reranker, jev, postprocessor, python
- use_cases: score or rank candidates in a pipeline; pick which skill or prompt to load; compare Jev with an LLM
- rule: 163

### kylemclaren/jev-search

- url: https://github.com/kylemclaren/jev-search
- description: A shadcn/ui registry block: keyword hits on the first keystroke, re-ranked by Jev a moment later, and keyword order stands if the call fails.
- stars: 9
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-23
- topics: jev, typesafe, typesafe-ai
- use_cases: rerank search results; score or rank candidates in a pipeline; review code or pull requests
- rule: 163

### AkashPriyadarshii/jev-scout

- url: https://github.com/AkashPriyadarshii/jev-scout
- description: Rust CLI and MCP server that finds real, maintained repos and crates for a plain-language request, with Jev scoring the candidates.
- stars: 6
- language: Rust
- license: MIT
- pushed_at: 2026-09-30
- topics: ai-agents, claude-code, cli, code-search, coding-agents, crates-io
- use_cases: score or rank candidates in a pipeline; rerank search results; call Jev from the command line
- rule: 163

### kylemclaren/jevpdf

- url: https://github.com/kylemclaren/jevpdf
- description: Searches a PDF by meaning in the browser: pdf.js extracts the lines locally, Jev answers one Noul per line in batches of 16, and matching lines light up page by page, ranked by probability.
- stars: 5
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-23
- topics: jev, pdf
- use_cases: rerank search results; compare Jev with an LLM; score or rank candidates in a pipeline
- rule: 163

### mttrbrts/jev-folio-recursive-classifier

- url: https://github.com/mttrbrts/jev-folio-recursive-classifier
- description: Classifies OCR'd legal agreements through the FOLIO Document Types ontology with recursive Jev Choices, beam search, confidence-gated leaf stopping, and context-length benchmarking.
- stars: 1
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-19
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; compare Jev with an LLM
- rule: 163

### liou666/senseek

- url: https://github.com/liou666/senseek
- description: Browser extension that searches the page you are reading by meaning, in a Ctrl+F style box, with your own key and no backend.
- stars: 0
- language: JavaScript
- pushed_at: 2026-09-21
- use_cases: compare Jev with an LLM; rerank search results; judge or verify an agent's output
- rule: 163

## 179 · Safety, Moderation & Verification

Guardrails, prompt-injection checks, judges that abstain.

### leepokai/jev-guard

- url: https://github.com/leepokai/jev-guard
- description: Auto mode for Claude Code, Codex, Cursor, Gemini CLI, Pi, and OpenCode: risk-scores each tool call as deny, ask, or allow and flags prompt injection in results.
- stars: 48
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-28
- topics: ai-agents, claude-code, codex, cursor, gemini-cli, hooks
- use_cases: gate or approve tool calls before they run; detect prompt injection or risky commands; judge or verify an agent's output
- rule: 179

### brainstormity/Jev-Moderation-Bot

- url: https://github.com/brainstormity/Jev-Moderation-Bot
- description: Chat moderation with editable rules.
- stars: 48
- language: Python
- license: MIT
- pushed_at: 2026-09-22
- use_cases: moderate content or detect abuse; add Jev to a chat bot or Discord; voice or realtime decisions
- rule: 179

### DanRWilloughby/snifftest

- url: https://github.com/DanRWilloughby/snifftest
- description: Prose linter for AI writing tells: countable rules plus one judgment model.
- stars: 34
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-18
- topics: ai-writing, claude-code, claude-code-plugin, claude-skills, github-action, jev
- use_cases: judge or verify an agent's output; moderate content or detect abuse; review code or pull requests
- rule: 179

### luantak/is-malicious

- url: https://github.com/luantak/is-malicious
- description: Scans a codebase for hidden or data-stealing behavior before you run it; a clean report is not proof, and it says so.
- stars: 33
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-23
- topics: jev, security-scanner, typesafe-ai, static-analysis, static-analysis-tools, supply-chain-security
- use_cases: call Jev from the command line; review code or pull requests; judge or verify an agent's output
- rule: 179

### MarissaFamularo/citation-verifier

- url: https://github.com/MarissaFamularo/citation-verifier
- description: Does the cited paper support the sentence citing it? Claude finds the quote, Jev scores it, a human decides.
- stars: 11
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-17
- topics: citation-verification, claude, peer-review, typesafe
- use_cases: judge or verify an agent's output; label data or build a dataset; compare Jev with an LLM
- rule: 179

### scale-venture-partners/riff

- url: https://github.com/scale-venture-partners/riff
- description: Ruff-style rule codes for writing.
- stars: 7
- language: Python
- license: MIT
- pushed_at: 2026-09-29
- use_cases: judge or verify an agent's output; compare Jev with an LLM; review code or pull requests
- rule: 179

### caiovicentino/jev-shield

- url: https://github.com/caiovicentino/jev-shield
- description: Semantic MCP firewall that screens every tool call, result, and description; reports 94 percent block recall at about $0.00002 a check.
- stars: 4
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-17
- topics: ai-safety, mcp, prompt-injection, security, vercel-ai-gateway
- use_cases: gate or approve tool calls before they run; detect prompt injection or risky commands; route requests through a gateway or proxy
- rule: 179

### noelzappy/tripwire

- url: https://github.com/noelzappy/tripwire
- description: AI SDK middleware and proxy that runs seven Jev checks on every LLM response before the user sees it; no accuracy numbers yet, and it says so.
- stars: 4
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-18
- topics: ai-sdk, guardrails, llm, pii, prompt-injection
- use_cases: route requests through a gateway or proxy; detect prompt injection or risky commands; moderate content or detect abuse
- rule: 179

### teyhouse/jev-secret-detection

- url: https://github.com/teyhouse/jev-secret-detection
- description: Measures how well Jev spots real credentials in file snippets, with the hard config-shaped cases scored separately.
- stars: 3
- language: Python
- pushed_at: 2026-09-18
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; compare Jev with an LLM
- rule: 179

### santos-sanz/jev-audio-beeper

- url: https://github.com/santos-sanz/jev-audio-beeper
- description: Low-latency audio censorship proof of concept: Jev typed decisions drive ffmpeg.
- stars: 3
- language: TypeScript
- pushed_at: 2026-09-17
- use_cases: judge or verify an agent's output; benchmark or calibrate Jev; voice or realtime decisions
- rule: 179

### asfarsadewa/human-compiler

- url: https://github.com/asfarsadewa/human-compiler
- description: Paste text, get diagnostics, like a compiler for prose.
- stars: 2
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-17
- use_cases: judge or verify an agent's output; benchmark or calibrate Jev; moderate content or detect abuse
- rule: 179

### DansiDanutz/fake-real-jev

- url: https://github.com/DansiDanutz/fake-real-jev
- description: Checks claims against cited excerpts with Jev in an English and Romanian fact-checking site, with a public integration example and a closed-source full application.
- stars: 1
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-26
- topics: evidence-verification, fact-checking, javascript, jev, typesafe
- use_cases: judge or verify an agent's output; learn how Jev works; moderate content or detect abuse
- rule: 179

### paulgoodchild/SkillsCheck

- url: https://github.com/paulgoodchild/SkillsCheck
- description: Sends an agent skill's text to Jev before installation and returns a verdict with category scores, without loading the skill into the agent's context; the submitted text is not redacted for secrets.
- stars: 0
- language: JavaScript
- pushed_at: 2026-09-21
- use_cases: detect prompt injection or risky commands; moderate content or detect abuse; judge or verify an agent's output
- rule: 179

### hteariH/stopspam-jev-bot

- url: https://github.com/hteariH/stopspam-jev-bot
- description: Telegram bot that removes spam and scam messages from group chats on calibrated-confidence classification.
- stars: 0
- language: Python
- pushed_at: 2026-09-21
- use_cases: moderate content or detect abuse; run Jev on open models without the vendor; benchmark or calibrate Jev
- rule: 179

## 185 · Data & Ops

Postgres extensions, semantic SQL and telemetry pipelines that call Jev.

### kyotofin/tax-doc-classifier

- url: https://github.com/kyotofin/tax-doc-classifier
- description: One request per page picks among 261 IRS forms and seven page kinds; reports 100 percent on its corpus at $0.001 a page, 34 times cheaper than the LLM pipeline it replaced.
- stars: 484
- language: TypeScript
- license: Apache-2.0
- pushed_at: 2026-09-29
- use_cases: compare Jev with an LLM; review code or pull requests; label data or build a dataset
- rule: 185

### realZachi/pg-jev

- url: https://github.com/realZachi/pg-jev
- description: PostgreSQL extension that answers plain-language questions about your tables.
- stars: 381
- language: Shell
- license: NOASSERTION
- pushed_at: 2026-09-18
- use_cases: run Jev inside a database or SQL; score or rank candidates in a pipeline; classify support tickets or messages
- rule: 185

### AgriciDaniel/jev-seo

- url: https://github.com/AgriciDaniel/jev-seo
- description: Crawls a site, checks it against 52 SEO rules, has Jev judge every page, and writes PDF, spreadsheet, and Markdown reports.
- stars: 266
- language: Python
- license: MIT
- pushed_at: 2026-09-22
- use_cases: call Jev from the command line; judge or verify an agent's output; benchmark or calibrate Jev
- rule: 185

### AkashPriyadarshii/jev-curate

- url: https://github.com/AkashPriyadarshii/jev-curate
- description: Sifts Parquet and JSONL training data at more than 1,500 rows a second.
- stars: 92
- language: Rust
- license: MIT
- pushed_at: 2026-09-30
- topics: arrow, cli, data-cleaning, data-engineering, dataset-curation, eval-harness
- use_cases: label data or build a dataset; benchmark or calibrate Jev; score or rank candidates in a pipeline
- rule: 185

### giuliosmall/pg_typesafe

- url: https://github.com/giuliosmall/pg_typesafe
- description: Pre-alpha PostgreSQL extension for categorical classification with Jev.
- stars: 88
- language: C
- license: MIT
- pushed_at: 2026-09-24
- use_cases: run Jev inside a database or SQL; score or rank candidates in a pipeline; label data or build a dataset
- rule: 185

### AboveColin/HA-Jev

- url: https://github.com/AboveColin/HA-Jev
- description: Home Assistant integration: ask a question about your house, get a probability, choice, or score as an entity.
- stars: 69
- language: Python
- license: MIT
- pushed_at: 2026-09-30
- topics: ai, custom-components, hacs, home-assistant, home-automation, homeassistant
- use_cases: voice or realtime decisions; monitor or observe Jev usage and cost; review code or pull requests
- rule: 185

### choxos/jev-reviewer

- url: https://github.com/choxos/jev-reviewer
- description: Asks a clinical trial report for systematic-review data by voice, text, or a questions file; every answer is a verbatim quote with its file and place.
- stars: 38
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-19
- topics: clinical-trials, data-extraction, evidence-synthesis, jev, meta-analysis, pdf
- use_cases: judge or verify an agent's output; extract structured fields from text; compare Jev with an LLM
- rule: 185

### chenmingtang830/jevgraph

- url: https://github.com/chenmingtang830/jevgraph
- description: Parses PDF, DOCX, PPTX, or text locally, then asks Jev one closed-set relation question per candidate entity pair and exports a graph with per-edge probabilities and page evidence to JSON, CSV, or Neo4j.
- stars: 31
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-20
- topics: information-extraction, jev, knowledge-graph, llm, python, relation-extraction
- use_cases: judge or verify an agent's output; benchmark or calibrate Jev; compare Jev with an LLM
- rule: 185

### colliber/duckdb-jev

- url: https://github.com/colliber/duckdb-jev
- description: DuckDB extension that asks a question of every row and returns a real SQL type.
- stars: 28
- language: C++
- license: MIT
- pushed_at: 2026-09-18
- use_cases: run Jev inside a database or SQL; extract structured fields from text; label data or build a dataset
- rule: 185

### chopratejas/invalidate

- url: https://github.com/chopratejas/invalidate
- description: Gives every stored agent memory a lease and asks Jev whether new evidence ends it; [live demo](https://invalidate-playground.vercel.app).
- stars: 23
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-21
- use_cases: compact or prune agent context; judge or verify an agent's output; add Jev to a chat bot or Discord
- rule: 185

### reachjalil/jevlogs

- url: https://github.com/reachjalil/jevlogs
- description: Scores OpenTelemetry log signal before paying for LLM analysis.
- stars: 17
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-22
- use_cases: monitor or observe Jev usage and cost; compare Jev with an LLM; benchmark or calibrate Jev
- rule: 185

### kylemclaren/jevql

- url: https://github.com/kylemclaren/jevql
- description: Semantic SQL for PostgreSQL, with Jev answering the predicates.
- stars: 15
- language: Go
- license: MIT
- pushed_at: 2026-09-19
- topics: jev, postgres
- use_cases: call Jev from the command line; call Jev from a language SDK; serve Jev over MCP to any agent
- rule: 185

### collapseindex/jev-ultralightspeed

- url: https://github.com/collapseindex/jev-ultralightspeed
- description: Packs 32 items into one request for bulk classification and calibrates the confidence cut that sends the least-sure rows to a person, reporting 533 items a second at 89.2 percent agreement with human labels.
- stars: 12
- language: Python
- license: NOASSERTION
- pushed_at: 2026-09-22
- topics: batching, classification, http2, jev, llm, throughput
- use_cases: benchmark or calibrate Jev; classify support tickets or messages; label data or build a dataset
- rule: 185

### keltokhy/jlink

- url: https://github.com/keltokhy/jlink
- description: Links records across two datasets from a match rule written in plain English, from Python, the shell, Stata, or R, and reports F1 0.73 against 0.69 for tuned string matching on NBER patent assignees to Compustat.
- stars: 7
- language: Python
- license: MIT
- pushed_at: 2026-09-28
- topics: economics, entity-resolution, fuzzy-matching, jev, record-linkage, stata
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; call Jev from the command line
- rule: 185

### EugeneBoondock/jevsql

- url: https://github.com/EugeneBoondock/jevsql
- description: SQL with natural-language predicates over SQLite: filter, rank, and classify rows by meaning, batched and cost-guarded.
- stars: 6
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-19
- topics: ai, jev, llm, semantic-search, sql, sqlite
- use_cases: judge or verify an agent's output; rerank search results; score or rank candidates in a pipeline
- rule: 185

### Foadsf/jev-for-engineers

- url: https://github.com/Foadsf/jev-for-engineers
- description: Eight small examples from mechanical and electrical engineering: CAD routing, FEM triage, DFM screening, BOM alignment.
- stars: 6
- language: Python
- license: MIT
- pushed_at: 2026-09-16
- topics: cad, cae, calibrated-confidence, classification, jev, llm
- use_cases: extract structured fields from text; benchmark or calibrate Jev; learn how Jev works
- rule: 185

### Query-farm/vgi-typesafe

- url: https://github.com/Query-farm/vgi-typesafe
- description: DuckDB worker that exposes choice, noul, and score as lateral-joinable table functions in SQL.
- stars: 5
- language: Python
- license: MIT
- pushed_at: 2026-09-19
- topics: ai, apache-arrow, arrow, classification, data-engineering, duckdb
- use_cases: score or rank candidates in a pipeline; classify support tickets or messages; label data or build a dataset
- rule: 185

### mgaitan/sqlite-jev

- url: https://github.com/mgaitan/sqlite-jev
- description: Adds Jev Noul, Choice, and Score judgments to SQLite through a loadable C extension and Python wrapper, with scalar functions and batched virtual-table queries.
- stars: 4
- language: C
- pushed_at: 2026-09-18
- use_cases: run Jev inside a database or SQL; label data or build a dataset; classify support tickets or messages
- rule: 185

### opaielsheikh/typesafe-migration-guard

- url: https://github.com/opaielsheikh/typesafe-migration-guard
- description: Reviews database migrations for safety before they run.
- stars: 3
- language: TypeScript
- pushed_at: 2026-09-17
- use_cases: review code or pull requests; judge or verify an agent's output; gate or approve tool calls before they run
- rule: 185

### ddfeyes/jev-mode

- url: https://github.com/ddfeyes/jev-mode
- description: Ticket triage and file tagging on a typed-judgment model; reports 78 percent fewer tokens and 96.1 percent accuracy against a 93.7 percent baseline.
- stars: 3
- language: Python
- license: MIT
- pushed_at: 2026-09-18
- topics: agents, ai, classification, cli, jev, llm
- use_cases: compact or prune agent context; classify support tickets or messages; compare Jev with an LLM
- rule: 185

## 191 · Applications & Extensions

End-user tools people actually open every day.

### anishfn/shapeshift

- url: https://github.com/anishfn/shapeshift
- description: Turns one text box into the matching card as you type, judged by a single parallel Jev call, with a keyword fallback when no key is set.
- stars: 752
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-23
- use_cases: extract structured fields from text; compare Jev with an LLM; review code or pull requests
- rule: 191

### kitze/unclutter

- url: https://github.com/kitze/unclutter
- description: Browser extension that removes page clutter with reusable template rules.
- stars: 341
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-18
- use_cases: review code or pull requests; learn how Jev works; compare Jev with an LLM
- rule: 191

### FerryCorleone/crush-monitor

- url: https://github.com/FerryCorleone/crush-monitor
- description: Reads a chat log locally and labels every message with the three likeliest of twelve emotions and thirty-five intents.
- stars: 262
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-25
- use_cases: extract structured fields from text; route requests through a gateway or proxy; label data or build a dataset
- rule: 191

### monteduro/killmyidea

- url: https://github.com/monteduro/killmyidea
- description: Describe your startup idea; Jev says kill it, fix it, or ship it.
- stars: 244
- language: TypeScript
- pushed_at: 2026-09-24
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; compare Jev with an LLM
- rule: 191

### usenotra/notra

- url: https://github.com/usenotra/notra
- description: Turns work into content, with Jev deciding what is worth posting.
- stars: 226
- language: TypeScript
- license: AGPL-3.0
- pushed_at: 2026-09-30
- topics: aeo, engine, generative, generative-engine-optimization, geo, optimization
- use_cases: judge or verify an agent's output; benchmark or calibrate Jev; rerank search results
- rule: 191

### wquguru/dasheng

- url: https://github.com/wquguru/dasheng
- description: Read English aloud and see it scored: streaming ASR listens, Jev judges word by word, and the total is arithmetic in code.
- stars: 144
- language: JavaScript
- pushed_at: 2026-09-20
- use_cases: voice or realtime decisions; benchmark or calibrate Jev; review code or pull requests
- rule: 191

### trungdq88/youtube-sponsor-detection

- url: https://github.com/trungdq88/youtube-sponsor-detection
- description: Chrome extension that finds sponsor reads from the transcript or live audio and jumps past them; code owns every timestamp, under a cent an hour in transcript mode.
- stars: 109
- language: JavaScript
- pushed_at: 2026-09-18
- use_cases: voice or realtime decisions; monitor or observe Jev usage and cost; review code or pull requests
- rule: 191

### kevinbadi/jev-voice

- url: https://github.com/kevinbadi/jev-voice
- description: Talk to macOS: local whisper.cpp transcribes, one fan-out Jev call picks the action and its typed arguments, and code runs it.
- stars: 106
- language: Python
- license: MIT
- pushed_at: 2026-09-23
- use_cases: drive a browser or GUI with an agent; voice or realtime decisions; compare Jev with an LLM
- rule: 191

### w3cj/jev-chat

- url: https://github.com/w3cj/jev-chat
- description: Chat-shaped command bar where Jev picks the tool, the arguments, and the reply shape, and code builds the answer from the tools' own data, so no model writes the text.
- stars: 105
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-18
- use_cases: gate or approve tool calls before they run; judge or verify an agent's output; compare Jev with an LLM
- rule: 191

### ChetasLua/jevmeter

- url: https://github.com/ChetasLua/jevmeter
- description: Puts a live meter on any video: every sentence scored on five questions, rendered as a 16:9 edit, a whole debate for about two cents.
- stars: 103
- language: Python
- license: MIT
- pushed_at: 2026-09-17
- topics: debate, jev, typesafe, video-editing, whisper
- use_cases: judge or verify an agent's output; call Jev from the command line; compare Jev with an LLM
- rule: 191

### fazlerocks/jevmail

- url: https://github.com/fazlerocks/jevmail
- description: Read-only Gmail triage: three Jev questions per message sort it into one of five trays with an urgency score.
- stars: 92
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-19
- topics: ai-email-assistant, ai-sdk, email-classification, email-sorter, email-triage, gmail
- use_cases: classify support tickets or messages; review code or pull requests; compare Jev with an LLM
- rule: 191

### realZachi/typesafe-adblock

- url: https://github.com/realZachi/typesafe-adblock
- description: Chrome extension that asks "is this element an ad?" per DOM node; a toy, and it says so.
- stars: 88
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-17
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; compare Jev with an LLM
- rule: 191

### AkashPriyadarshii/jev-seo

- url: https://github.com/AkashPriyadarshii/jev-seo
- description: Rust CLI and MCP server for SEO and GEO checks over DuckDuckGo results, scored by Jev.
- stars: 86
- language: Rust
- license: MIT
- pushed_at: 2026-09-30
- topics: ahrefs-alternative, claude-code, cli, generative-engine-optimization, geo, mcp
- use_cases: call Jev from the command line; serve Jev over MCP to any agent; judge or verify an agent's output
- rule: 191

### RafalWilinski/vibecheck

- url: https://github.com/RafalWilinski/vibecheck
- description: Vibe-check your X post before you hit publish.
- stars: 49
- language: JavaScript
- pushed_at: 2026-09-20
- use_cases: judge or verify an agent's output; compare Jev with an LLM; moderate content or detect abuse
- rule: 191

### parth-kp/jev-mail-classifier

- url: https://github.com/parth-kp/jev-mail-classifier
- description: Config-driven inbox classification that tags, moves, flags, and notifies, through TypeSafe directly or OpenRouter.
- stars: 19
- language: Python
- license: MIT
- pushed_at: 2026-09-20
- use_cases: compare Jev with an LLM; classify support tickets or messages; moderate content or detect abuse
- rule: 191

### kevthetech143/super-jev

- url: https://github.com/kevthetech143/super-jev
- description: Small harness connecting evidence, Jev judgments, permitted actions, and verified outcomes.
- stars: 14
- language: Python
- license: MIT
- pushed_at: 2026-09-30
- use_cases: gate or approve tool calls before they run; judge or verify an agent's output; add Jev to a chat bot or Discord
- rule: 191

### manifoldor/xtags

- url: https://github.com/manifoldor/xtags
- description: Labels every post in your X timeline with what it wants you to do.
- stars: 12
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-27
- topics: ai, chrome-extension, jev, typesafe, userscript, x-twitter
- use_cases: detect prompt injection or risky commands; compare Jev with an LLM; judge or verify an agent's output
- rule: 191

### gtaras7/typesafe-jev

- url: https://github.com/gtaras7/typesafe-jev
- description: Jev experiments starting with a local CV-screening workbench, each with its own measured results.
- stars: 10
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-27
- use_cases: benchmark or calibrate Jev; score or rank candidates in a pipeline; learn how Jev works
- rule: 191

### harshil1712/slidepilot

- url: https://github.com/harshil1712/slidepilot
- description: Voice-driven auto-advance for Slidev on Cloudflare Agents.
- stars: 9
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-22
- topics: cloudflare-workers, jev, slidev, slidev-addon, typescript, voice
- use_cases: voice or realtime decisions; review code or pull requests; compare Jev with an LLM
- rule: 191

### valentynkit/jev.nvim

- url: https://github.com/valentynkit/jev.nvim
- description: Neovim plugin: ask the buffer a plain-language question, Treesitter splits it into functions, Jev scores each one, and the answers land in quickfix ranked by probability.
- stars: 9
- language: Lua
- license: MIT
- pushed_at: 2026-09-19
- use_cases: rerank search results; judge or verify an agent's output; learn how Jev works
- rule: 191

### andrelandgraf/safer-with-jev

- url: https://github.com/andrelandgraf/safer-with-jev
- description: Neon Function proxy for the Neon AI Gateway with Jev routing in front.
- stars: 6
- language: TypeScript
- pushed_at: 2026-09-18
- use_cases: detect prompt injection or risky commands; moderate content or detect abuse; judge or verify an agent's output
- rule: 191

### valentynkit/jev-skip

- url: https://github.com/valentynkit/jev-skip
- description: Browser extension that reads the caption track and paints a per-segment sponsor probability on the seek bar, with no crowd database.
- stars: 6
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-19
- use_cases: benchmark or calibrate Jev; compare Jev with an LLM; learn how Jev works
- rule: 191

### chris-wozniczek/jev-voice-control

- url: https://github.com/chris-wozniczek/jev-voice-control
- description: Menu-bar Swift app turning spoken commands into Jev typed decisions and macOS actions.
- stars: 5
- language: Swift
- license: MIT
- pushed_at: 2026-09-21
- topics: computer-use, cua, jev, jev-ai, jev-api, jev-model
- use_cases: drive a browser or GUI with an agent; voice or realtime decisions; gate or approve tool calls before they run
- rule: 191

### hellogumbo/should-ai-kill-us-all

- url: https://github.com/hellogumbo/should-ai-kill-us-all
- description: Asks Jev the question every ten minutes, using the actual headlines.
- stars: 4
- language: JavaScript
- license: CC0-1.0
- pushed_at: 2026-09-18
- use_cases: benchmark or calibrate Jev; compare Jev with an LLM; call Jev from the command line
- rule: 191

### sriganesh/jevibe-check

- url: https://github.com/sriganesh/jevibe-check
- description: Live tone labels for Bluesky posts and drafts.
- stars: 3
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-17
- topics: jev, typesafe, atproto, bluesky
- use_cases: monitor or observe Jev usage and cost; moderate content or detect abuse; benchmark or calibrate Jev
- rule: 191

### phureewat29/jev-got

- url: https://github.com/phureewat29/jev-got
- description: Game of Thrones roleplay where a story model writes each scene and Jev answers five typed questions that drive the header, soundtrack, art, and next prompt.
- stars: 3
- language: TypeScript
- pushed_at: 2026-09-19
- use_cases: judge or verify an agent's output; learn how Jev works; extract structured fields from text
- rule: 191

### hazlema/jev-riffs

- url: https://github.com/hazlema/jev-riffs
- description: Mines repeating motifs from a MIDI melody, scores every candidate in one batched Jev call, and highlights and plays the winner on a piano roll; on an orchestral Swan Lake the top motif is the swan theme.
- stars: 3
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-25
- use_cases: score or rank candidates in a pipeline; benchmark or calibrate Jev; review code or pull requests
- rule: 191

### thenewpotato/privacy-facts

- url: https://github.com/thenewpotato/privacy-facts
- description: Turns privacy policies into nutrition-style labels with plain-language answers, Jev confidence scores, and suggested source clauses.
- stars: 2
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-18
- topics: ai, jev, privacy
- use_cases: judge or verify an agent's output; compare Jev with an LLM; extract structured fields from text
- rule: 191

### 0xShin0221/openpoke-meets-jev

- url: https://github.com/0xShin0221/openpoke-meets-jev
- description: OpenPoke fork that moves email screening, a tool-call guardrail, and search reranking onto Jev, with an A/B against the Sonnet call it replaced and an adversarial run on the injection gate.
- stars: 1
- language: Python
- license: MIT
- pushed_at: 2026-09-20
- use_cases: gate or approve tool calls before they run; compare Jev with an LLM; detect prompt injection or risky commands
- rule: 191

## 205 · SDKs & Community Clients

Unofficial clients for the languages without a first-party SDK.

### pithings/advocaat

- url: https://github.com/pithings/advocaat
- description: Small TypeScript client for asking questions about your own data.
- stars: 97
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-18
- use_cases: score or rank candidates in a pipeline; extract structured fields from text; call Jev from a language SDK
- rule: 205

### milvus-io/milvus-model

- url: https://github.com/milvus-io/milvus-model
- description: Reranker package whose Jev adapter sends every candidate document as a Noul question in one request, sorts by the returned probabilities, and keeps original indices; shipped in the v0.3.4 release.
- stars: 61
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-23
- topics: ai, embedding, nlp, rag, reranker
- use_cases: rerank search results; score or rank candidates in a pipeline; benchmark or calibrate Jev
- rule: 205

### obie/ruby_decision_model

- url: https://github.com/obie/ruby_decision_model
- description: Ruby client for decision models with OpenRouter and TypeSafe providers behind one interface, stdlib only.
- stars: 52
- language: Ruby
- license: MIT
- pushed_at: 2026-09-18
- use_cases: call Jev from a language SDK; monitor or observe Jev usage and cost; benchmark or calibrate Jev
- rule: 205

### dannote/jev

- url: https://github.com/dannote/jev
- description: Elixir client built for OTP: reply to Jev from a GenServer and pattern match on the answer.
- stars: 35
- language: Elixir
- license: MIT
- pushed_at: 2026-09-26
- use_cases: run Jev on open models without the vendor; learn how Jev works; extract structured fields from text
- rule: 205

### kieranklaassen/ruby_llm-typesafe

- url: https://github.com/kieranklaassen/ruby_llm-typesafe
- description: TypeSafe as a structured-output provider for RubyLLM 2.
- stars: 19
- language: Ruby
- license: MIT
- pushed_at: 2026-09-29
- use_cases: classify support tickets or messages; call Jev from a language SDK; score or rank candidates in a pipeline
- rule: 205

### Twister915/typesafe-ai

- url: https://github.com/Twister915/typesafe-ai
- description: Rust client with async and blocking backends and observable retries.
- stars: 14
- language: Rust
- license: Apache-2.0
- pushed_at: 2026-09-16
- use_cases: classify support tickets or messages; score or rank candidates in a pipeline; judge or verify an agent's output
- rule: 205

### saibimajdi/typesafeai-dotnet-sdk

- url: https://github.com/saibimajdi/typesafeai-dotnet-sdk
- description: Community .NET SDK with typed Noul, Choice, and Score questions.
- stars: 13
- language: C#
- license: MIT
- pushed_at: 2026-09-29
- topics: ai, classification, csharp, dotnet, llm, sdk
- use_cases: extract structured fields from text; score or rank candidates in a pipeline; judge or verify an agent's output
- rule: 205

### Premo-Cloud/typesafe-sdk-java

- url: https://github.com/Premo-Cloud/typesafe-sdk-java
- description: Unofficial Java client for the TypeSafe System One API, calling Choice, Score and Noul questions over HTTP.
- stars: 11
- language: Java
- license: MIT
- pushed_at: 2026-09-25
- topics: java, sdk, spring-boot, typesafe, jev, system-one
- use_cases: call Jev from a language SDK; review code or pull requests; add Jev to a chat bot or Discord
- rule: 205

### pambrose/jev4k

- url: https://github.com/pambrose/jev4k
- description: Kotlin DSL and client.
- stars: 11
- language: Kotlin
- license: Apache-2.0
- pushed_at: 2026-09-28
- topics: jev, jev-api, kotlin, java, dsl, llm
- use_cases: call Jev from a language SDK; classify support tickets or messages; add Jev to a chat bot or Discord
- rule: 205

### Tangerg/typesafe-sdk-go

- url: https://github.com/Tangerg/typesafe-sdk-go
- description: Go SDK with no third-party dependencies.
- stars: 10
- language: Go
- license: MIT
- pushed_at: 2026-09-19
- use_cases: judge or verify an agent's output; review code or pull requests
- rule: 205

### jomatsu/zod-jev

- url: https://github.com/jomatsu/zod-jev
- description: Zod 4 schemas with semantic rules: shape checks stay in Zod, meaning checks go to Jev in one request and come back as Zod issues.
- stars: 9
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-17
- use_cases: moderate content or detect abuse; classify support tickets or messages; benchmark or calibrate Jev
- rule: 205

### joshmn/typesafe-sdk

- url: https://github.com/joshmn/typesafe-sdk
- description: Unofficial Ruby client for the TypeSafe AI API, returning typed answers with probabilities instead of parsed prose.
- stars: 8
- language: Ruby
- license: MIT
- pushed_at: 2026-09-26
- use_cases: call Jev from a language SDK; score or rank candidates in a pipeline; classify support tickets or messages
- rule: 205

### inanna-malick/jev-dsl

- url: https://github.com/inanna-malick/jev-dsl
- description: Haskell DSL with typed packets and inferred answer types.
- stars: 8
- language: Haskell
- license: MIT
- pushed_at: 2026-09-18
- use_cases: judge or verify an agent's output; review code or pull requests; compare Jev with an LLM
- rule: 205

### gilljon/typesafe-ai-rs

- url: https://github.com/gilljon/typesafe-ai-rs
- description: Independent async and blocking Rust SDK.
- stars: 7
- language: Rust
- license: MIT
- pushed_at: 2026-09-17
- topics: ai, api-client, rust, rust-sdk, typesafe
- use_cases: score or rank candidates in a pipeline; review code or pull requests; judge or verify an agent's output
- rule: 205

### Gaurav-Gosain/jev-go

- url: https://github.com/Gaurav-Gosain/jev-go
- description: Go client that returns typed judgments and probabilities.
- stars: 6
- language: Go
- license: MIT
- pushed_at: 2026-09-16
- use_cases: score or rank candidates in a pipeline; call Jev from a language SDK; judge or verify an agent's output
- rule: 205

### jamesward/zio-typesafe-ai

- url: https://github.com/jamesward/zio-typesafe-ai
- description: Scala client on ZIO.
- stars: 6
- language: Scala
- license: Apache-2.0
- pushed_at: 2026-09-23
- use_cases: call Jev from a language SDK; score or rank candidates in a pipeline; judge or verify an agent's output
- rule: 205

### Stumble/jev-go

- url: https://github.com/Stumble/jev-go
- description: Community Go SDK for TypeSafe Jev / System One, covering typed Choice, Score and Noul questions.
- stars: 6
- language: Go
- license: MIT
- pushed_at: 2026-09-18
- use_cases: call Jev from a language SDK; call Jev from the command line; review code or pull requests
- rule: 205

### nshkrdotcom/typesafe_sdk

- url: https://github.com/nshkrdotcom/typesafe_sdk
- description: Elixir port of the TypeScript AI SDK with a TypeSafe provider.
- stars: 6
- language: Elixir
- license: MIT
- pushed_at: 2026-09-20
- topics: agents, ai, ai-sdk, artificial-intelligence, elixir, function-calling
- use_cases: extract structured fields from text; review code or pull requests; judge or verify an agent's output
- rule: 205

### alterhq/typesafe-sdk-swift

- url: https://github.com/alterhq/typesafe-sdk-swift
- description: Unofficial Swift SDK for the TypeSafe AI API, wrapping typed questions and answers in Swift types.
- stars: 5
- language: Swift
- license: MIT
- pushed_at: 2026-09-15
- use_cases: route requests through a gateway or proxy; review code or pull requests; control a phone or mobile app
- rule: 205

### zhirschtritt/typesafe-go

- url: https://github.com/zhirschtritt/typesafe-go
- description: Idiomatic Go SDK for the TypeSafe API.
- stars: 4
- language: Go
- license: MIT
- pushed_at: 2026-09-29
- topics: ai, jev, sdk, systemone, typesafe
- use_cases: route requests through a gateway or proxy; review code or pull requests; pick which skill or prompt to load
- rule: 205

### Butochnikov/laravel-typesafe-jev

- url: https://github.com/Butochnikov/laravel-typesafe-jev
- description: Laravel integration with typed responses, async requests, and testing fakes.
- stars: 4
- language: PHP
- license: MIT
- pushed_at: 2026-09-17
- use_cases: score or rank candidates in a pipeline; judge or verify an agent's output; review code or pull requests
- rule: 205

### Hawxy/TypeSafeAI.Net

- url: https://github.com/Hawxy/TypeSafeAI.Net
- description: Unofficial .NET SDK for the TypeSafe AI platform, usable from any .NET application.
- stars: 4
- language: C#
- license: Apache-2.0
- pushed_at: 2026-09-19
- topics: csharp, typesafe-ai, dotnet
- use_cases: judge or verify an agent's output; classify support tickets or messages; score or rank candidates in a pipeline
- rule: 205

### mateonunez/jod

- url: https://github.com/mateonunez/jod
- description: Zod-style schemas over Jev: validate the state locally, then project typed answers.
- stars: 4
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-17
- topics: ai, jev, typesafe, zod
- use_cases: judge or verify an agent's output; classify support tickets or messages; extract structured fields from text
- rule: 205

### GenieRobot/typesafe-ai-rails

- url: https://github.com/GenieRobot/typesafe-ai-rails
- description: Rails integration built on the community Ruby gem.
- stars: 4
- language: Ruby
- license: MIT
- pushed_at: 2026-09-16
- use_cases: monitor or observe Jev usage and cost; judge or verify an agent's output; score or rank candidates in a pipeline
- rule: 205

### steven-shoemaker/hunch

- url: https://github.com/steven-shoemaker/hunch
- description: Turns Choice, Score, and Noul questions into Python functions over lists and DataFrames, with deduplication, caching, escalation of unsure rows to an LLM held to the same labels, and a TypeScript port on npm as hunch-jev.
- stars: 4
- language: Python
- license: MIT
- pushed_at: 2026-09-22
- topics: classification, jev, jev-ai, llm, python
- use_cases: score or rank candidates in a pipeline; extract structured fields from text; judge or verify an agent's output
- rule: 205

### AboveColin/jevclient

- url: https://github.com/AboveColin/jevclient
- description: Async Python client, probabilities and choices out, no prose to parse.
- stars: 3
- language: Python
- license: MIT
- pushed_at: 2026-09-21
- topics: async, asyncio, classification, jev, python, typesafe
- use_cases: score or rank candidates in a pipeline; build a game or simulation on Jev; benchmark or calibrate Jev
- rule: 205

### Butochnikov/typesafe-sdk-php

- url: https://github.com/Butochnikov/typesafe-sdk-php
- description: PHP client with sync calls, Guzzle promises, and PSR-3 logging.
- stars: 3
- language: PHP
- license: MIT
- pushed_at: 2026-09-17
- use_cases: classify support tickets or messages; score or rank candidates in a pipeline; label data or build a dataset
- rule: 205

### AbdelStark/s1-rs

- url: https://github.com/AbdelStark/s1-rs
- description: Turns Rust enums and structs into Choice, Score, and Noul questions with compile-time-checked, confidence-gated answers.
- stars: 2
- language: Rust
- license: MIT
- pushed_at: 2026-09-16
- topics: machine-learning, sdk, typesafe-ai
- use_cases: judge or verify an agent's output; classify support tickets or messages; score or rank candidates in a pipeline
- rule: 205

### AbdelStark/typesafe-rs

- url: https://github.com/AbdelStark/typesafe-rs
- description: Latency-first Rust client, on crates.io.
- stars: 2
- language: Rust
- license: MIT
- pushed_at: 2026-09-16
- topics: machine-learning, typesafe-ai
- use_cases: score or rank candidates in a pipeline; judge or verify an agent's output; extract structured fields from text
- rule: 205

### DomMonte/n8n-nodes-typesafe-ai

- url: https://github.com/DomMonte/n8n-nodes-typesafe-ai
- description: n8n community node for yes/no, choice, and score questions.
- stars: 1
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-21
- use_cases: score or rank candidates in a pipeline; classify support tickets or messages; benchmark or calibrate Jev
- rule: 205

### kunobi-ninja/kunobi-jev

- url: https://github.com/kunobi-ninja/kunobi-jev
- description: Rust client for the System One API.
- stars: 1
- language: Rust
- license: Apache-2.0
- pushed_at: 2026-09-24
- use_cases: score or rank candidates in a pipeline; call Jev from a language SDK; classify support tickets or messages
- rule: 205

### bariskisir/JevSharp

- url: https://github.com/bariskisir/JevSharp
- description: .NET 10 SDK for Jev decisions through TypeSafe, OpenRouter, Vercel AI Gateway, and compatible endpoints.
- stars: 0
- language: C#
- license: MIT
- pushed_at: 2026-09-24
- use_cases: route requests through a gateway or proxy; call Jev from a language SDK; run Jev on open models without the vendor
- rule: 205

## 217 · Command Line

Call Jev from a shell, no SDK required.

### dzhng/jevgrep

- url: https://github.com/dzhng/jevgrep
- description: Finds the files, declarations, and verbatim excerpts a coding agent needs by asking Jev which declarations answer a plain-language question about the repo.
- stars: 1 853
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-29
- topics: ai-sdk, claude-code, cli, code-search, codex, coding-agents
- use_cases: call Jev from the command line; route requests through a gateway or proxy; judge or verify an agent's output
- rule: 217

### dorkitude/webctl

- url: https://github.com/dorkitude/webctl
- description: Searches the web for an agent and has Jev score and dedupe results and page chunks, so only the relevant text reaches the model.
- stars: 150
- language: Go
- license: MIT
- pushed_at: 2026-09-23
- use_cases: compact or prune agent context; rerank search results; run Jev on open models without the vendor
- rule: 217

### keltokhy/jgrep

- url: https://github.com/keltokhy/jgrep
- description: Prints the lines that fit a plain-English description, streaming from `tail -f` under a spend cap, and reports F1 0.91 on SMS spam against 0.72 for a keyword grep.
- stars: 132
- language: Python
- license: MIT
- pushed_at: 2026-09-28
- topics: cli, grep, jev, openrouter, semantic-search, typesafe
- use_cases: moderate content or detect abuse; classify support tickets or messages; call Jev from the command line
- rule: 217

### mrnugget/jev-shell-history

- url: https://github.com/mrnugget/jev-shell-history
- description: Fish-style zsh history suggestions, ranked by Jev.
- stars: 116
- language: TypeScript
- pushed_at: 2026-09-18
- use_cases: compare Jev with an LLM; review code or pull requests; benchmark or calibrate Jev
- rule: 217

### sharziki/semdecide

- url: https://github.com/sharziki/semdecide
- description: Typed semantic decisions for Unix pipelines and CI.
- stars: 74
- language: Python
- license: MIT
- pushed_at: 2026-09-16
- topics: ai, ci, cli, jev, semantic, unix
- use_cases: call Jev from the command line; score or rank candidates in a pipeline; judge or verify an agent's output
- rule: 217

### shiftynick/jev-axi

- url: https://github.com/shiftynick/jev-axi
- description: Shell verbs for agents and humans: pick, rate, check, rank, triage, guard.
- stars: 26
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-24
- topics: ai-agents, axi, claude-code, cli, jev, typesafe
- use_cases: gate or approve tool calls before they run; detect prompt injection or risky commands; judge or verify an agent's output
- rule: 217

### Nasrallah-AL/jev-cli

- url: https://github.com/Nasrallah-AL/jev-cli
- description: npm CLI with the key in the OS keychain; typed judgments from the shell.
- stars: 23
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-27
- topics: ai, cli, fact-checking, guardrails, jev, llm-tools
- use_cases: call Jev from the command line; judge or verify an agent's output; moderate content or detect abuse
- rule: 217

### tumf/jev-cli

- url: https://github.com/tumf/jev-cli
- description: Dependency-free Python CLI wrapping Choice, Score, and Noul.
- stars: 14
- language: Python
- license: MIT
- pushed_at: 2026-09-23
- topics: ai, cli, jev, python, typesafe
- use_cases: call Jev from the command line; serve Jev over MCP to any agent; judge or verify an agent's output
- rule: 217

### cristianoliveira/jeq

- url: https://github.com/cristianoliveira/jeq
- description: Pipes and composes Jev judgments over JSON and NDJSON with map, reduce, rank, and rate, then applies policy through an explicit offline gate.
- stars: 10
- language: Go
- license: MIT
- pushed_at: 2026-09-27
- topics: ai-tools, jev, jq, terminal, unix, ai
- use_cases: call Jev from the command line; score or rank candidates in a pipeline; judge or verify an agent's output
- rule: 217

### sufianetaouil/every

- url: https://github.com/sufianetaouil/every
- description: Ask a yes/no question of every function in a codebase; grep whose pattern is a question.
- stars: 8
- language: Python
- license: MIT
- pushed_at: 2026-09-17
- use_cases: call Jev from the command line; benchmark or calibrate Jev; judge or verify an agent's output
- rule: 217

### y0usaf/typesafe-cli

- url: https://github.com/y0usaf/typesafe-cli
- description: Noul, choice, and score answers as numbers from the shell.
- stars: 5
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-19
- use_cases: call Jev from the command line; benchmark or calibrate Jev; judge or verify an agent's output
- rule: 217

### jtsang4/jev-cli

- url: https://github.com/jtsang4/jev-cli
- description: Typed questions in, structured JSON answers out.
- stars: 4
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-18
- topics: cli, jev, skills, typesafe
- use_cases: judge or verify an agent's output; call Jev from the command line; score or rank candidates in a pipeline
- rule: 217

### jexp/watfile

- url: https://github.com/jexp/watfile
- description: Sorts PDFs and text files into category folders, one Jev Choice per document, with a local Laya backend as the alternative.
- stars: 2
- language: Python
- pushed_at: 2026-09-21
- topics: categorization, cli, jev, terminal, tool, laya
- use_cases: call Jev from the command line; label data or build a dataset; compare Jev with an LLM
- rule: 217

### allebee/jevgrep

- url: https://github.com/allebee/jevgrep
- description: Filters log lines and other text streams by meaning, one Noul per line in micro-batches, streaming from `tail -f` with grep's flags and exit codes.
- stars: 1
- language: Python
- license: MIT
- pushed_at: 2026-09-21
- topics: cli, grep, jev, log-analysis, logs, python
- use_cases: compare Jev with an LLM; call Jev from the command line; benchmark or calibrate Jev
- rule: 217

## 223 · Benchmarks, Evals & Calibration

Measure it before you trust it.

### sutro-sh/jev-align

- url: https://github.com/sutro-sh/jev-align
- description: Finds the examples a Jev function is least sure about, asks you to label them, and improves the function with GEPA.
- stars: 301
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-20
- topics: active-learning, classification, cli, gepa, human-in-the-loop, jev
- use_cases: benchmark or calibrate Jev; label data or build a dataset; classify support tickets or messages
- rule: 223

### fstandhartinger/jevbench

- url: https://github.com/fstandhartinger/jevbench
- description: Benchmark for Jev-class decision models: bounded rubric in, typed answer with a probability per option out, with cascade and committee experiments reported separately.
- stars: 186
- language: Python
- license: MIT
- pushed_at: 2026-09-29
- topics: jev, systemone, decisionmodels
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; score or rank candidates in a pipeline
- rule: 223

### vinilana/jev-eval-agent

- url: https://github.com/vinilana/jev-eval-agent
- description: Personal-assistant agent with 100 mocked tools, measuring how many steps a Jev-gated agent needs.
- stars: 107
- language: HTML
- pushed_at: 2026-09-17
- use_cases: benchmark or calibrate Jev; compare Jev with an LLM; gate or approve tool calls before they run
- rule: 223

### openlayer-ai/jevals

- url: https://github.com/openlayer-ai/jevals
- description: Runs a trace's agent, quality, and security evals as one batch of typed Jev questions instead of separate LLM-judge calls.
- stars: 97
- language: Python
- license: MIT
- pushed_at: 2026-09-24
- topics: agents, evals, guardrails, jev, llm, llm-evaluation
- use_cases: judge or verify an agent's output; detect prompt injection or risky commands; benchmark or calibrate Jev
- rule: 223

### pinecone-io/cultivar

- url: https://github.com/pinecone-io/cultivar
- description: Pinecone's skill-testing CLI, with a Jev grading backend it reports at about 30 times cheaper than the LLM grader.
- stars: 41
- language: Python
- license: MIT
- pushed_at: 2026-09-18
- topics: agent, agent-skills, benchmarking, claude-code, codex, copilot
- use_cases: judge or verify an agent's output; benchmark or calibrate Jev; compare Jev with an LLM
- rule: 223

### AbdelStark/jev-benchmarks

- url: https://github.com/AbdelStark/jev-benchmarks
- description: Probability-aware evaluation for typed decision models: calibration, selective risk, latency, reproducible.
- stars: 21
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-17
- topics: benchmarking, calibration, evaluation, machine-learning, reproducibility, selective-classification
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; route between models by task difficulty
- rule: 223

### AntonioCoppe/jev-harness

- url: https://github.com/AntonioCoppe/jev-harness
- description: Confidence gates, shadow mode, recipes, and evals; reports Claude CLI at 48.9 s against Jev at 1.3 s on the same row-filter job.
- stars: 16
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-25
- topics: agents, confidence, decision, evals, harness, jev
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; gate or approve tool calls before they run
- rule: 223

### abhixhek/jevcal

- url: https://github.com/abhixhek/jevcal
- description: Stop guessing thresholds: calibrate, threshold, and drift-check against an LLM teacher.
- stars: 11
- language: Python
- license: MIT
- pushed_at: 2026-09-18
- topics: calibration, jev, llm-evals, system-one, typesafe, confidence-thresholds
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; route between models by task difficulty
- rule: 223

### jmanhype/jev-dspy-lab

- url: https://github.com/jmanhype/jev-dspy-lab
- description: Reproducible calibration and selective-risk benchmarks for Jev decisions in DSPy.
- stars: 11
- language: Python
- license: MIT
- pushed_at: 2026-09-20
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; monitor or observe Jev usage and cost
- rule: 223

### zhuyansen/jev-search-rerank-eval

- url: https://github.com/zhuyansen/jev-search-rerank-eval
- description: Does a Jev rerank beat embedding search? 9,831 graded pairs, with the judge-circularity bias measured.
- stars: 10
- language: Python
- license: MIT
- pushed_at: 2026-09-18
- topics: bge-m3, evaluation, information-retrieval, ndcg, reranking, typesafe-jev
- use_cases: benchmark or calibrate Jev; compare Jev with an LLM; rerank search results
- rule: 223

### anessbelbati/jev-rerank-bench

- url: https://github.com/anessbelbati/jev-rerank-bench
- description: Jev against Cohere Rerank, ZeroEntropy, and a chat baseline on 14 datasets, raw responses included.
- stars: 9
- language: Python
- license: MIT
- pushed_at: 2026-09-25
- topics: benchmark, cohere, information-retrieval, llm-evaluation, ndcg, rag
- use_cases: benchmark or calibrate Jev; compare Jev with an LLM; judge or verify an agent's output
- rule: 223

### anisselbd/jev-phishing-bench

- url: https://github.com/anisselbd/jev-phishing-bench
- description: Jev against Claude Haiku on 2,000 phishing emails: accuracy, calibration, latency, cost.
- stars: 8
- language: Python
- pushed_at: 2026-09-19
- use_cases: benchmark or calibrate Jev; compare Jev with an LLM; judge or verify an agent's output
- rule: 223

### mahlernim/jev-korean-benchmark

- url: https://github.com/mahlernim/jev-korean-benchmark
- description: Korean understanding and medical text, with runtime and cost evidence.
- stars: 7
- language: Python
- pushed_at: 2026-09-17
- use_cases: judge or verify an agent's output; benchmark or calibrate Jev; monitor or observe Jev usage and cost
- rule: 223

### wondertwins/jev-benchmark

- url: https://github.com/wondertwins/jev-benchmark
- description: Two benchmarks, chess and predator identification, one inside Jev's lane and one outside, both with results.
- stars: 7
- language: Python
- license: MIT
- pushed_at: 2026-09-16
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; learn how Jev works
- rule: 223

### Gaurav-Gosain/jev-sec-bench

- url: https://github.com/Gaurav-Gosain/jev-sec-bench
- description: Blind benchmarks for prompt injection and vulnerable-code detection.
- stars: 4
- language: Go
- license: MIT
- pushed_at: 2026-09-16
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; moderate content or detect abuse
- rule: 223

### TokenTrim/jev-agent-failure-benchmark

- url: https://github.com/TokenTrim/jev-agent-failure-benchmark
- description: Jev against a strong LLM on the Who and When agent-failure-attribution benchmark.
- stars: 4
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-17
- use_cases: benchmark or calibrate Jev; compare Jev with an LLM; judge or verify an agent's output
- rule: 223

### RINNECODER/jev-behavior-study

- url: https://github.com/RINNECODER/jev-behavior-study
- description: Controlled prompt experiments on jev-1.13.0, raw results and offline verification.
- stars: 4
- language: Python
- license: MIT
- pushed_at: 2026-09-17
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; learn how Jev works
- rule: 223

### chenmingtang830/jevarena

- url: https://github.com/chenmingtang830/jevarena
- description: Hosted arena that puts Jev against an opponent judge you connect and takes your vote before revealing which was which, with latency, cost provenance, and self-reported confidence; a vote records preference, not verified correctness.
- stars: 4
- language: TypeScript
- license: Apache-2.0
- pushed_at: 2026-09-20
- topics: byok, evaluation, jev, llm-as-a-judge, open-source
- use_cases: benchmark or calibrate Jev; compare Jev with an LLM; monitor or observe Jev usage and cost
- rule: 223

### bitnovus/jev-spam-eval

- url: https://github.com/bitnovus/jev-spam-eval
- description: Zero-shot spam filtering with Noul questions against TF-IDF baselines.
- stars: 3
- language: Jupyter Notebook
- license: MIT
- pushed_at: 2026-09-18
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; moderate content or detect abuse
- rule: 223

### PistachioAIHQ/jev-synergy-screening

- url: https://github.com/PistachioAIHQ/jev-synergy-screening
- description: Choice and Noul questions scored against ASReview SYNERGY gold labels for abstract screening.
- stars: 3
- language: Python
- pushed_at: 2026-09-16
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; score or rank candidates in a pipeline
- rule: 223

### jgridifier/jev-research-eval

- url: https://github.com/jgridifier/jev-research-eval
- description: Reproducible harness over a pinned jev-ultrafast commit, with baseline and stress suites.
- stars: 3
- language: HTML
- license: NOASSERTION
- pushed_at: 2026-09-17
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; compare Jev with an LLM
- rule: 223

### HackSing/jev-report

- url: https://github.com/HackSing/jev-report
- description: Independent Chinese research report: 52 pages, 50 reproducible tests, 143 traceable data rows.
- stars: 2
- language: Python
- license: MIT
- pushed_at: 2026-09-17
- topics: ai-research, chinese, jev, llm-evaluation, typesafe
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; compare Jev with an LLM
- rule: 223

### hegargarcia/jev-playground

- url: https://github.com/hegargarcia/jev-playground
- description: Jev against other models in games with explicit states, legal actions, and a measurable outcome.
- stars: 2
- language: TypeScript
- pushed_at: 2026-09-17
- use_cases: benchmark or calibrate Jev; compare Jev with an LLM; judge or verify an agent's output
- rule: 223

### yodablocks/jev-orderby-bench

- url: https://github.com/yodablocks/jev-orderby-bench
- description: Measures whether ORDER BY over a Jev probability is defensible: pairwise inversion, ordinality against a human grade, calibration and wording invariants under a pre-registered gate.
- stars: 1
- language: Python
- license: MIT
- pushed_at: 2026-09-20
- topics: benchmark, calibration, duckdb, information-retrieval, jev, llm-evaluation
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; score or rank candidates in a pipeline
- rule: 223

### 4esv/jev-eval

- url: https://github.com/4esv/jev-eval
- description: Jev against GPT-5.6 Terra on three labeled tasks: equal on the easy ones, 6.7 points lower on 77-way routing, 5 times faster, 41 to 50 times cheaper.
- stars: 1
- language: Python
- pushed_at: 2026-09-23
- use_cases: benchmark or calibrate Jev; compare Jev with an LLM; judge or verify an agent's output
- rule: 223

### themsquared/jev-benchmark

- url: https://github.com/themsquared/jev-benchmark
- description: Tool-call risk classification with the run-to-run variance reported; every wrong answer came with hedged confidence.
- stars: 1
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-24
- topics: agentgateway, ai-agents, benchmark, calibration, llm, mcp
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; compare Jev with an LLM
- rule: 223

### blowxian/jev-fanout-bench

- url: https://github.com/blowxian/jev-fanout-bench
- description: Measures what a Jev request is billed and how its answers hold up under batching, translation, and rewording, from 3,455 billed requests with public raw logs.
- stars: 0
- language: Python
- license: MIT
- pushed_at: 2026-09-28
- topics: benchmark, jev, llm-cost, openrouter, system-one, typesafe
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; monitor or observe Jev usage and cost
- rule: 223

## 229 · Open Models & Replicas

Run System One semantics without the vendor — on a 3090 if you like.

### jaredpalmer/kev

- url: https://github.com/jaredpalmer/kev
- description: LoRA adapter and readout head on Qwen2.5-0.5B that answers many typed questions in one prefill; trains in under two hours on a MacBook, held-out ECE 0.065, speaks the TypeSafe wire format.
- stars: 7 998
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-30
- topics: decision-model, jev, qwen3
- use_cases: run Jev on open models without the vendor; call Jev from a language SDK; benchmark or calibrate Jev
- rule: 229

### TheoLeeCJ/SemIf

- url: https://github.com/TheoLeeCJ/SemIf
- description: Semantic ifs from open models on a single 3090; the most starred independent replica, formerly openjev.
- stars: 4 604
- language: Python
- license: MIT
- pushed_at: 2026-09-23
- use_cases: run Jev on open models without the vendor; compare Jev with an LLM; gate or approve tool calls before they run
- rule: 229

### TianyuCodings/NanoJev

- url: https://github.com/TianyuCodings/NanoJev
- description: 0.6B replica with parallel decisions, dynamic candidates, and an end-to-end training pipeline.
- stars: 2 445
- language: Python
- license: MIT
- pushed_at: 2026-09-21
- use_cases: build a game or simulation on Jev; run Jev on open models without the vendor; benchmark or calibrate Jev
- rule: 229

### vinnylarouge/jevlike

- url: https://github.com/vinnylarouge/jevlike
- description: Open option scorer that reads candidate logits instead of generating JSON.
- stars: 1 335
- language: Python
- license: MIT
- pushed_at: 2026-09-16
- use_cases: score or rank candidates in a pipeline; run Jev on open models without the vendor; judge or verify an agent's output
- rule: 229

### ollaya-dev/ollaya

- url: https://github.com/ollaya-dev/ollaya
- description: Pulls and serves open decision models such as Laya, kev, and JevK5 behind a TypeSafe-compatible local endpoint, the way Ollama serves LLMs.
- stars: 1 011
- language: Rust
- license: Apache-2.0
- pushed_at: 2026-09-29
- topics: classification, decision-models, jev, laya, llm-routing, local-inference
- use_cases: run Jev on open models without the vendor; classify support tickets or messages; benchmark or calibrate Jev
- rule: 229

### Mapika/decider

- url: https://github.com/Mapika/decider
- description: One-pass typed decisions fine-tuned from Qwen3.5-2B.
- stars: 988
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-30
- use_cases: score or rank candidates in a pipeline; classify support tickets or messages; judge or verify an agent's output
- rule: 229

### nokia-applied-research/AnyJev

- url: https://github.com/nokia-applied-research/AnyJev
- description: Turns an open Hugging Face model into a calibrated decision endpoint served through vLLM, with no fine-tuning at the first two levels.
- stars: 981
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-28
- topics: calibration, decision-model, jev, jev-model, llm, system-one
- use_cases: run Jev on open models without the vendor; benchmark or calibrate Jev; call Jev from the command line
- rule: 229

### wfzyx/von

- url: https://github.com/wfzyx/von
- description: Non-autoregressive open decision model reporting under 15 ms locally, as a drop-in alternative to Jev.
- stars: 781
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-30
- topics: decision-model, jev, machine-learning, python, rlcd, system-one
- use_cases: run Jev on open models without the vendor; score or rank candidates in a pipeline; call Jev from a language SDK
- rule: 229

### Rizzo-AI-Academy/rizzo-flow

- url: https://github.com/Rizzo-AI-Academy/rizzo-flow
- description: Local-first take on the System One idea: unstructured state in, typed probabilistic decisions out, without generating a token.
- stars: 763
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-25
- use_cases: run Jev on open models without the vendor; score or rank candidates in a pipeline; classify support tickets or messages
- rule: 229

### featherless-ai/simple-jev

- url: https://github.com/featherless-ai/simple-jev
- description: Reads next-token logits from any Hugging Face model for choice, rubric, and support questions; public demo API with no key.
- stars: 575
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-30
- use_cases: run Jev on open models without the vendor; judge or verify an agent's output; extract structured fields from text
- rule: 229

### Liuziyu77/Valen

- url: https://github.com/Liuziyu77/Valen
- description: Multimodal decision model on Qwen3.5 that scores candidates against images and video as well as text, with training code and open weights.
- stars: 551
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-30
- use_cases: score or rank candidates in a pipeline; judge or verify an agent's output; review code or pull requests
- rule: 229

### razorback16/openjev

- url: https://github.com/razorback16/openjev
- description: Jev-compatible decision server on DiffusionGemma 26B through vLLM, images included, hosted free on [Codiv](https://codiv.ai).
- stars: 541
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-29
- use_cases: run Jev on open models without the vendor; call Jev from a language SDK; compare Jev with an LLM
- rule: 229

### Yinsongxu/LLM2Jev

- url: https://github.com/Yinsongxu/LLM2Jev
- description: Adapts local language models to runtime-defined Choice, Score, and Noul questions and returns typed answers with probabilities.
- stars: 377
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-26
- topics: jev, llm, mllm
- use_cases: run Jev on open models without the vendor; compare Jev with an LLM; score or rank candidates in a pipeline
- rule: 229

### ekzhang/openjev-sglang

- url: https://github.com/ekzhang/openjev-sglang
- description: Jev-compatible API endpoint on SGLang, prefill only.
- stars: 336
- language: Python
- pushed_at: 2026-09-25
- topics: jev, llm, structured-generation, systemone
- use_cases: run Jev on open models without the vendor; extract structured fields from text; benchmark or calibrate Jev
- rule: 229

### malevrigns/agent-jev

- url: https://github.com/malevrigns/agent-jev
- description: Decision model on Qwen3-0.6B that answers without decoding tokens, reports 79.25 percent top-1 on the public Typed Decisions benchmark.
- stars: 328
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-23
- use_cases: gate or approve tool calls before they run; score or rank candidates in a pipeline; benchmark or calibrate Jev
- rule: 229

### hr98w/jev-visual

- url: https://github.com/hr98w/jev-visual
- description: Educational visual-inference variant on Apple Silicon: shared context, direct candidate scoring.
- stars: 303
- language: Python
- license: MIT
- pushed_at: 2026-09-21
- use_cases: run Jev on open models without the vendor; score or rank candidates in a pipeline; judge or verify an agent's output
- rule: 229

### Heman10x-NGU/openJev-verdict-2.0

- url: https://github.com/Heman10x-NGU/openJev-verdict-2.0
- description: Non-autoregressive 151M decision engine with a WebGPU browser runtime, reporting 77.1 percent top-1 and 1.44 percent calibration error on its own benchmark.
- stars: 293
- language: Python
- license: NOASSERTION
- pushed_at: 2026-09-20
- topics: agentic-ai, brier-score, calibration, decision-engine, deep-learning, machine-learning
- use_cases: judge or verify an agent's output; benchmark or calibrate Jev; run Jev on open models without the vendor
- rule: 229

### logan-markewich/jeff

- url: https://github.com/logan-markewich/jeff
- description: Self-hosted System One API on the 400M GLiFormer, with benchmarks that say where it trails Jev.
- stars: 273
- language: Python
- license: MIT
- pushed_at: 2026-09-20
- topics: classification, encoder, gliner, jev, typesafe
- use_cases: run Jev on open models without the vendor; score or rank candidates in a pipeline; compare Jev with an LLM
- rule: 229

### togethercomputer/tev1

- url: https://github.com/togethercomputer/tev1
- description: Data recipe and training code that fine-tune Qwen3.5-4B into an open-weight decision model on Together AI.
- stars: 177
- language: Python
- license: MIT
- pushed_at: 2026-09-24
- use_cases: classify support tickets or messages; benchmark or calibrate Jev; compare Jev with an LLM
- rule: 229

### kshetrajna12/reflex

- url: https://github.com/kshetrajna12/reflex
- description: Small open decision model on Qwen3.5: state plus typed questions to calibrated probabilities.
- stars: 160
- language: Python
- license: MIT
- pushed_at: 2026-09-27
- use_cases: run Jev on open models without the vendor; classify support tickets or messages; compare Jev with an LLM
- rule: 229

### allebee/jevk5

- url: https://github.com/allebee/jevk5
- description: Qwen3.5 replica with distilled LoRA weights, ranked second of 76 systems and first among open ones on JevBench v1.4.
- stars: 126
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-28
- topics: calibration, decision-model, jev, jevbench, lora, qwen
- use_cases: run Jev on open models without the vendor; benchmark or calibrate Jev; compare Jev with an LLM
- rule: 229

### daseinlabs/open-jev

- url: https://github.com/daseinlabs/open-jev
- description: Prefills once and scores every option in one padded pass on Gemma 3 4B with MLX; plays Doom from the terminal in the demo.
- stars: 121
- language: Python
- license: MIT
- pushed_at: 2026-09-30
- use_cases: run Jev on open models without the vendor; score or rank candidates in a pipeline; learn how Jev works
- rule: 229

### mmastrac/djev

- url: https://github.com/mmastrac/djev
- description: Serves `/v1/systemone` on DiffusionGemma by reading typed answers and text spans off a seeded, pinned canvas, with no generation.
- stars: 111
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-24
- use_cases: benchmark or calibrate Jev; run Jev on open models without the vendor; learn how Jev works
- rule: 229

### Heman10x-NGU/Verdict-open-jev

- url: https://github.com/Heman10x-NGU/Verdict-open-jev
- description: Non-autoregressive decision engine on ModernBERT with calibrated uncertainty and an in-browser WebGPU playground.
- stars: 110
- language: Python
- license: NOASSERTION
- pushed_at: 2026-09-28
- topics: brier-score, calibration, decision-engine, edge-ai, gliclass, jev
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; run Jev on open models without the vendor
- rule: 229

### zwliJay/jev-forge

- url: https://github.com/zwliJay/jev-forge
- description: End-to-end stack for auditable data construction, Qwen3.5-0.8B training, fixed Mind2Web and OOD evaluation, local serving, and a preliminary RLCD baseline.
- stars: 94
- language: Python
- license: NOASSERTION
- pushed_at: 2026-09-23
- use_cases: benchmark or calibrate Jev; score or rank candidates in a pipeline; run Jev on open models without the vendor
- rule: 229

### kikoncuo/jevfire

- url: https://github.com/kikoncuo/jevfire
- description: Parallel decisions for CUDA LLMs through a vLLM API, with game-agent examples and benchmarks.
- stars: 72
- language: JavaScript
- license: MIT
- pushed_at: 2026-09-18
- topics: cuda, game-ai, inference, jev, llm, parallel-decoding
- use_cases: benchmark or calibrate Jev; run Jev on open models without the vendor; build a game or simulation on Jev
- rule: 229

### bnsd55/jevmlx

- url: https://github.com/bnsd55/jevmlx
- description: Parallel constrained decisions for any MLX model on Apple Silicon, one forward pass.
- stars: 69
- language: Python
- license: MIT
- pushed_at: 2026-09-25
- topics: apple-silicon, jev, local-llm, local-models, mlx
- use_cases: extract structured fields from text; run Jev on open models without the vendor; score or rank candidates in a pipeline
- rule: 229

### r-ms/mini-jev

- url: https://github.com/r-ms/mini-jev
- description: Preregistered experiment on a frozen Qwen3-4B: read the option letter's logits, skip the JSON.
- stars: 58
- language: Python
- license: MIT
- pushed_at: 2026-09-18
- use_cases: compare Jev with an LLM; benchmark or calibrate Jev; judge or verify an agent's output
- rule: 229

### ikermoel/open-alternative-jev

- url: https://github.com/ikermoel/open-alternative-jev
- description: Typed, calibrated decisions from any open-weights model in one forward pass, on Hugging Face and vLLM.
- stars: 57
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-25
- topics: calibration, classification, llm, transformers, vllm, jev
- use_cases: run Jev on open models without the vendor; compare Jev with an LLM; judge or verify an agent's output
- rule: 229

### zhengxuyu/litjev

- url: https://github.com/zhengxuyu/litjev
- description: Turns any off-the-shelf LLM into a Jev-style decision layer.
- stars: 46
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-21
- use_cases: run Jev on open models without the vendor; score or rank candidates in a pipeline; compare Jev with an LLM
- rule: 229

### OmniJev/PlayJev

- url: https://github.com/OmniJev/PlayJev
- description: Plays ten browser games from the frame alone on a fine-tuned Qwen3.5-0.8B, one forward pass per move, open weights and a browser demo.
- stars: 45
- language: JavaScript
- license: Apache-2.0
- pushed_at: 2026-09-24
- topics: browser-games, game-ai, game-playing-agent, html5-games, imitation-learning, jev
- use_cases: run Jev on open models without the vendor; compare Jev with an LLM; drive a browser or GUI with an agent
- rule: 229

### JoshuaSP/open-jev

- url: https://github.com/JoshuaSP/open-jev
- description: Typed JSON inference with DiffusionGemma, benchmarked against Jev.
- stars: 43
- language: Python
- license: MIT
- pushed_at: 2026-09-16
- use_cases: compare Jev with an LLM; judge or verify an agent's output; extract structured fields from text
- rule: 229

### iammrduncan/typesafe-ai-benchmark

- url: https://github.com/iammrduncan/typesafe-ai-benchmark
- description: LLM gateway that mimics the TypeSafe response shape, useful as a stand-in while you wait for a key.
- stars: 40
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-19
- use_cases: compare Jev with an LLM; benchmark or calibrate Jev; judge or verify an agent's output
- rule: 229

### mithalouni/system-one-open

- url: https://github.com/mithalouni/system-one-open
- description: Typed calibrated decisions in one forward pass on Gemma 4 E2B and Gemma 3 270M.
- stars: 38
- language: Python
- license: NOASSERTION
- pushed_at: 2026-09-17
- use_cases: classify support tickets or messages; run Jev on open models without the vendor; compare Jev with an LLM
- rule: 229

### zhihz/openjev

- url: https://github.com/zhihz/openjev
- description: Bilingual local decisions from context, questions, and candidate answers.
- stars: 35
- language: Python
- license: NOASSERTION
- pushed_at: 2026-09-16
- use_cases: score or rank candidates in a pipeline; judge or verify an agent's output; run Jev on open models without the vendor
- rule: 229

### rorshopping/jev-on-a-laptop

- url: https://github.com/rorshopping/jev-on-a-laptop
- description: Study of Jev-style decisions on stock 1.5B to 8B models on a laptop, with a Hugging Face demo.
- stars: 25
- language: Python
- license: NOASSERTION
- pushed_at: 2026-09-17
- use_cases: run Jev on open models without the vendor; compare Jev with an LLM; benchmark or calibrate Jev
- rule: 229

### olanotolu/jevbetter

- url: https://github.com/olanotolu/jevbetter
- description: A stronger one-pass scorer with a head-to-head benchmark against the jevlike starter design.
- stars: 16
- language: Python
- license: MIT
- pushed_at: 2026-09-16
- use_cases: score or rank candidates in a pipeline; benchmark or calibrate Jev; pick which skill or prompt to load
- rule: 229

### genai-craft/openvons

- url: https://github.com/genai-craft/openvons
- description: Open decision layer answering a finite option set with probabilities split into execute, confirm, and reject. Independent replica, not TypeSafe weights.
- stars: 14
- language: Python
- license: NOASSERTION
- pushed_at: 2026-09-21
- topics: calibration, decision-model, japanese, speech-recognition, voice-commands, whisper
- use_cases: gate or approve tool calls before they run; benchmark or calibrate Jev; compare Jev with an LLM
- rule: 229

### amithgc/local-jev

- url: https://github.com/amithgc/local-jev
- description: Offline server wire-compatible with the System One endpoint, checked against the official SDK; reports 80.5 percent on JevBench's public items against 86.6 percent published for the hosted API.
- stars: 13
- language: Python
- license: MIT
- pushed_at: 2026-09-21
- use_cases: run Jev on open models without the vendor; classify support tickets or messages; compare Jev with an LLM
- rule: 229

### David-Lolly/Jev-Compatible

- url: https://github.com/David-Lolly/Jev-Compatible
- description: Gateway that turns an existing SGLang or vLLM deployment into a Jev-compatible decision service by scoring candidate tokens, with no training and no model changes.
- stars: 6
- language: Python
- pushed_at: 2026-09-21
- use_cases: score or rank candidates in a pipeline; classify support tickets or messages; run Jev on open models without the vendor
- rule: 229

### metask-ai/metask-jev

- url: https://github.com/metask-ai/metask-jev
- description: Open-weight typed-decision models on Qwen3.5 with calibrated per-option probabilities in one forward pass; reports 80.1 percent on JevBench against 75.3 for Jev 1.13.
- stars: 2
- language: Python
- pushed_at: 2026-09-22
- use_cases: benchmark or calibrate Jev; score or rank candidates in a pipeline; compare Jev with an LLM
- rule: 229

## 236 · Games, Robotics & Simulation

Decisions as game mechanics.

### rmalde/minecraft-agent

- url: https://github.com/rmalde/minecraft-agent
- description: Kills the dragon on a vanilla server with a frontier model planning and Jev selecting every player action; the logged run took 131 Jev decisions and 35 planner calls.
- stars: 569
- language: JavaScript
- pushed_at: 2026-09-20
- use_cases: judge or verify an agent's output; benchmark or calibrate Jev; compare Jev with an LLM
- rule: 236

### fhshaik/typesafe-mario

- url: https://github.com/fhshaik/typesafe-mario
- description: Plays Super Mario Bros. from structured emulator state; Jev picks the NES controller input directly.
- stars: 422
- language: Python
- pushed_at: 2026-09-16
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; build a game or simulation on Jev
- rule: 236

### FBddcz/embodied-jev

- url: https://github.com/FBddcz/embodied-jev
- description: Browser workbench for embodied experiments in MuJoCo where every robot step is a visible Jev decision, backed by a local model or a cloud API.
- stars: 250
- language: Python
- license: MIT
- pushed_at: 2026-09-22
- use_cases: compare Jev with an LLM; judge or verify an agent's output; run Jev on open models without the vendor
- rule: 236

### rokbenko/quackd

- url: https://github.com/rokbenko/quackd
- description: Command line for LLM-piloted robots across seven bodies, with an optional Jev stepper that picks among calls without ever writing a joint angle.
- stars: 245
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-30
- topics: mcp, microduck, physical-ai, claude, openai, local-llm
- use_cases: gate or approve tool calls before they run; call Jev from the command line; run Jev on open models without the vendor
- rule: 236

### RomanSlack/jev-drone

- url: https://github.com/RomanSlack/jev-drone
- description: Camera-only drone in MuJoCo with Jev in the loop at 2.5 Hz.
- stars: 229
- language: Python
- license: MIT
- pushed_at: 2026-09-24
- topics: autonomous-agents, drone, mujoco, robotics, typesafe
- use_cases: build a game or simulation on Jev; judge or verify an agent's output; review code or pull requests
- rule: 236

### standardagents/jevpilot

- url: https://github.com/standardagents/jevpilot
- description: Three.js driving simulator where Jev picks steering and speed from sampled paths up to four times a second; [drive it](https://jevpilot.standardagents.ai).
- stars: 203
- language: JavaScript
- pushed_at: 2026-09-17
- use_cases: build a game or simulation on Jev; voice or realtime decisions; compact or prune agent context
- rule: 236

### emrickgarrett/OneVOneJev

- url: https://github.com/emrickgarrett/OneVOneJev
- description: 1v1 quickscope arena in Three.js.
- stars: 41
- language: TypeScript
- pushed_at: 2026-09-18
- use_cases: review code or pull requests; build a game or simulation on Jev; benchmark or calibrate Jev
- rule: 236

### phyous/tsai-sc

- url: https://github.com/phyous/tsai-sc
- description: Plays the original StarCraft shareware through keyboard and mouse, action probabilities recorded.
- stars: 28
- language: Python
- license: MIT
- pushed_at: 2026-09-16
- use_cases: judge or verify an agent's output; benchmark or calibrate Jev; compare Jev with an LLM
- rule: 236

### lukaske/jev-doom-agent

- url: https://github.com/lukaske/jev-doom-agent
- description: Browser-native Doom agent with structured spatial state and live decision telemetry.
- stars: 28
- language: TypeScript
- pushed_at: 2026-09-17
- use_cases: judge or verify an agent's output; benchmark or calibrate Jev; drive a browser or GUI with an agent
- rule: 236

### sorrycc/typesafe-snake

- url: https://github.com/sorrycc/typesafe-snake
- description: One Choice per tick; legal moves and facts generated in code.
- stars: 24
- language: TypeScript
- pushed_at: 2026-09-17
- use_cases: build a game or simulation on Jev; judge or verify an agent's output; benchmark or calibrate Jev
- rule: 236

### vinilana/live-jev

- url: https://github.com/vinilana/live-jev
- description: Top-down car in the browser sending four typed questions every 200 ms, with confidence-gated overrides in code.
- stars: 21
- language: JavaScript
- pushed_at: 2026-09-18
- use_cases: build a game or simulation on Jev; compare Jev with an LLM; voice or realtime decisions
- rule: 236

### TarunTomar122/jev-askable-arm

- url: https://github.com/TarunTomar122/jev-askable-arm
- description: Zero-shot English goals on a simulated Franka arm; Jev chains hardcoded primitives.
- stars: 12
- language: Python
- license: MIT
- pushed_at: 2026-09-17
- use_cases: compare Jev with an LLM; benchmark or calibrate Jev; learn how Jev works
- rule: 236

### valentynkit/jev-plays-pokemon-red

- url: https://github.com/valentynkit/jev-plays-pokemon-red
- description: Pokemon Red on PyBoy: code owns the route and the arithmetic, Jev picks only at branches, and every battle turn logs a faint prediction scored by Brier against what the RAM says.
- stars: 10
- language: Python
- license: MIT
- pushed_at: 2026-09-19
- use_cases: call Jev from the command line; judge or verify an agent's output; learn how Jev works
- rule: 236

### AbdelStark/heist-one

- url: https://github.com/AbdelStark/heist-one
- description: Stealth game where Jev makes the guards' judgments and deterministic code owns the world.
- stars: 9
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-17
- topics: ai-game, jev, phaser, probabilistic-ai, remotion, stealth-game
- use_cases: judge or verify an agent's output; benchmark or calibrate Jev; build a game or simulation on Jev
- rule: 236

### anxkhn/JevPlaysPokemon

- url: https://github.com/anxkhn/JevPlaysPokemon
- description: Generation 3 Pokémon through Showdown and a real FireRed ROM.
- stars: 7
- language: HTML
- license: GPL-3.0
- pushed_at: 2026-09-18
- topics: firered, mgba, pokemon, showdown, typesafe
- use_cases: benchmark or calibrate Jev; learn how Jev works; compare Jev with an LLM
- rule: 236

### vishxrad/clashroyale-jev

- url: https://github.com/vishxrad/clashroyale-jev
- description: Plays Clash Royale with Jev choosing cards and placements from Qwen battlefield vision and local OpenCV hand and elixir recognition.
- stars: 4
- language: Python
- pushed_at: 2026-09-22
- use_cases: control a phone or mobile app; benchmark or calibrate Jev; learn how Jev works
- rule: 236

### phyous/tsai-civ2

- url: https://github.com/phyous/tsai-civ2
- description: Civilization II in a browser, full-game harness, live action probabilities.
- stars: 3
- language: Python
- license: NOASSERTION
- pushed_at: 2026-09-18
- use_cases: benchmark or calibrate Jev; drive a browser or GUI with an agent; compare Jev with an LLM
- rule: 236

### siroccomask/snake-jev

- url: https://github.com/siroccomask/snake-jev
- description: Snake controlled by parallel Jev assessments, one API call per tick.
- stars: 3
- language: Python
- license: MIT
- pushed_at: 2026-09-19
- use_cases: benchmark or calibrate Jev; compare Jev with an LLM; build a game or simulation on Jev
- rule: 236

### hazlema/jev-connect4

- url: https://github.com/hazlema/jev-connect4
- description: Connect Four with nine swappable query strategies for one model, every answer graded against engine ground truth in a live inspector.
- stars: 1
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-21
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; monitor or observe Jev usage and cost
- rule: 236

## 242 · Finance & Trading

Scoring financial signals with calibrated probabilities.

### jarrodwatts/jev-trader

- url: https://github.com/jarrodwatts/jev-trader
- description: One trade decision every Monad block, on Kuru MON-USDC, about 300 ms each.
- stars: 2 695
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-17
- use_cases: trade or score financial signals; review code or pull requests; benchmark or calibrate Jev
- rule: 242

### imikerussell/beebots

- url: https://github.com/imikerussell/beebots
- description: Runs three Jev-driven trading bots against OKX perpetuals with a live dashboard, paper trading by default and real money behind explicit flags.
- stars: 180
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-27
- use_cases: benchmark or calibrate Jev; trade or score financial signals; review code or pull requests
- rule: 242

### aowang-ai/jev-trade

- url: https://github.com/aowang-ai/jev-trade
- description: Live Jev trader on Hyperliquid.
- stars: 171
- language: TypeScript
- license: NOASSERTION
- pushed_at: 2026-09-21
- topics: hyperliquid, jev, quantitative-finance, typesafe
- use_cases: trade or score financial signals; benchmark or calibrate Jev; review code or pull requests
- rule: 242

### zadescoxp/Jev-Trades

- url: https://github.com/zadescoxp/Jev-Trades
- description: Crypto trading bot with backtesting.
- stars: 37
- language: Python
- license: Apache-2.0
- pushed_at: 2026-09-25
- topics: automated-testing, backtesting, crypto-trading-bot-latest, trading-bot-ai-automated
- use_cases: trade or score financial signals; benchmark or calibrate Jev; learn how Jev works
- rule: 242

### justinhe16/trade-jev

- url: https://github.com/justinhe16/trade-jev
- description: Backtests Jev as a buy, sell, or hold trader on NQ order-book data.
- stars: 11
- language: Python
- license: MIT
- pushed_at: 2026-09-17
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; trade or score financial signals
- rule: 242

## 248 · Playgrounds & Demos

Try it in thirty seconds.

### dabit3/jev-experiments

- url: https://github.com/dabit3/jev-experiments
- description: Nader Dabit's grab bag of small Jev experiments.
- stars: 397
- language: TypeScript
- pushed_at: 2026-09-21
- use_cases: benchmark or calibrate Jev; learn how Jev works; review code or pull requests
- rule: 248

### TypeSafeAI/typesafe-playground

- url: https://github.com/TypeSafeAI/typesafe-playground
- description: 110 use cases, games, and model challenges with editable prompts and A/B comparisons; a community org, not the vendor, formerly under BunsDev.
- stars: 22
- language: TypeScript
- license: MIT
- pushed_at: 2026-09-28
- use_cases: extract structured fields from text; pick which skill or prompt to load; build a game or simulation on Jev
- rule: 248

### kavehmz/typesafe-playground

- url: https://github.com/kavehmz/typesafe-playground
- description: From support routing to a 3D driving simulation with visible sensor inputs.
- stars: 14
- language: JavaScript
- pushed_at: 2026-09-29
- use_cases: learn how Jev works; judge or verify an agent's output; build a game or simulation on Jev
- rule: 248

### GiesN/typesafe-jev-workflow

- url: https://github.com/GiesN/typesafe-jev-workflow
- description: LangGraph demo that sends a mocked email to Jev and routes on the typed Choice it returns.
- stars: 12
- language: Python
- pushed_at: 2026-09-16
- use_cases: judge or verify an agent's output; classify support tickets or messages; learn how Jev works
- rule: 248

### lbotinelly/jev-little-airways

- url: https://github.com/lbotinelly/jev-little-airways
- description: Show-and-tell capability study for Jev.
- stars: 7
- language: HTML
- license: MIT
- pushed_at: 2026-09-17
- use_cases: benchmark or calibrate Jev; voice or realtime decisions; judge or verify an agent's output
- rule: 248

### haseeb-heaven/jev-system-one

- url: https://github.com/haseeb-heaven/jev-system-one
- description: Terminal interface where OpenAI answers and Jev separately scores relevance, reliability, and quality.
- stars: 6
- language: Python
- license: MIT
- pushed_at: 2026-09-17
- use_cases: judge or verify an agent's output; compare Jev with an LLM; benchmark or calibrate Jev
- rule: 248

### markjaquith/typesafe-ai-playground

- url: https://github.com/markjaquith/typesafe-ai-playground
- description: Rust CLI playground for experiments around Jev.
- stars: 5
- language: Rust
- license: MIT
- pushed_at: 2026-09-22
- use_cases: call Jev from the command line; classify support tickets or messages; benchmark or calibrate Jev
- rule: 248

### replynodes/jev-web-analyzer

- url: https://github.com/replynodes/jev-web-analyzer
- description: Turns a SaaS landing page into Markdown and asks Jev ten bounded Choice questions about what a first-time visitor understands, shown as a founder teardown.
- stars: 5
- language: TypeScript
- license: Apache-2.0
- pushed_at: 2026-09-28
- topics: ai-evaluation, jev, product-positioning, saas, typesafe-ai, vercel-ai-gateway
- use_cases: benchmark or calibrate Jev; learn how Jev works; judge or verify an agent's output
- rule: 248

### wustep/jev-playground

- url: https://github.com/wustep/jev-playground
- description: Can a System One model steer a music composition through typed classify, score, and pick decisions alone.
- stars: 2
- language: TypeScript
- pushed_at: 2026-09-26
- use_cases: benchmark or calibrate Jev; learn how Jev works; compare Jev with an LLM
- rule: 248

### willprout/magic-8-ball

- url: https://github.com/willprout/magic-8-ball
- description: Ask a question, one choice over twenty answers picks the reply and shows the click-to-answer latency; [live demo](https://willprout.github.io/magic-8-ball/).
- stars: 1
- language: TypeScript
- pushed_at: 2026-09-18
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; compare Jev with an LLM
- rule: 248

### bud-ro/jev-demos

- url: https://github.com/bud-ro/jev-demos
- description: Demos built to test what Jev is good at.
- stars: 1
- language: Dart
- pushed_at: 2026-09-18
- use_cases: benchmark or calibrate Jev; judge or verify an agent's output; pick which skill or prompt to load
- rule: 248

### rishi-raj-jain/hn-thread-judge

- url: https://github.com/rishi-raj-jain/hn-thread-judge
- description: Reads the most-discussed Hacker News threads comment by comment with Jev and reduces each one to a single verdict, served live from Postgres.
- stars: 0
- language: TypeScript
- pushed_at: 2026-09-21
- topics: ai, bm25, drizzle-orm, full-text-search, hacker-news, hackernews
- use_cases: extract structured fields from text; benchmark or calibrate Jev; judge or verify an agent's output
- rule: 248