Jev AI GitHub projects
ranked and curated.

A browser agent uses Jev to choose an action and page element; a separate model writes text.

An Android chat companion that reads visible messages through Accessibility or OCR, asks Jev about intent and risk, and ranks draft replies without sending them.

A Claude Code plugin uses Jev to keep, shorten or remove tool calls and results without rewriting conversation text.

A Monad trading bot with optional Jev buy/sell decisions; the default run uses a mock rule.

A CLI and agent skill that uses Jev to follow relevant folders, files and declarations, returning source excerpts and reading leads for coding tasks.

A Windows WeChat helper using local screenshots and OCR, Jev judgments and ranked reply candidates; filling is explicit and sending remains manual.

Serves supported open decision checkpoints through vLLM with batching, multimodal input and a System One API for Choice, Noul and Score requests.

Jev monitors coding agents' progress and flags work needing verification or intervention.

Uses staged Jev judgments to review code diffs and repositories for risk, evidence and severity.

Installable Jev skills and editable agent workflows with scenario templates, recorded examples and tests.

Routes each Claude Code or Codex turn to a model tier using Jev while keeping the existing CLI workflow.

An MCP server offering Jev tools for verification, screening, ranking and code review.

Uses Jev to choose search sources and rank results returned through Search1API, showing links instead of generated answers.

Lets an authoring LLM build and optionally evolve a task-specific Jev harness, then freezes its code and questions for traced runtime decisions.

A macOS WeChat overlay using screen capture and OCR with local or Jev decisions to classify intent, flag risk and draft replies without automatic sending.

A mobile agent that uses Jev to choose actions on an Android device through Mobilerun, with a live studio and execution traces.

A browser extension that uses Jev to classify and hide page clutter, then reuses local rules for matching page templates.

Uses Jev to select actions in a headless browser, then returns a step trace, final page and screenshot.

A semantic linter that uses Jev to check code against custom rules.

Three paper-trading bots that ask Jev to choose valid market actions while code enforces budgets, cooldowns and execution limits.

An open benchmark suite for comparing Jev-class typed decision models on quality, cost, latency and reliability.

A CLI for improving Jev classifiers with human labels and GEPA.

The official Python SDK for typed Jev questions and structured results.

The official TypeSafe client for sending typed questions and reading structured Jev responses in JavaScript or TypeScript.

A self-hosted skills library for agents with optional Jev recommendations across authorized skills through TypeSafe, OpenRouter or Vercel AI Gateway.

A DeepSeek Harness decision plugin that uses Jev for bounded judgments in DSH and compatible coding-agent workflows.

Thirty runnable Jev examples across ten levels, from simple typed decisions to coding-agent hooks, tools and autonomous use.

Adds semantic filtering, extraction, ranking and verification operators to PostgreSQL, with Jev-backed execution, caching and explainable probability data.

Combines crypto market metrics and selected X posts, then asks Jev for bounded action, sentiment, squeeze and catalyst judgments without executing trades.

A multi-provider workflow toolkit whose optional decision-support addon sends bounded judgments to Jev while keeping project rules and artifacts reusable.

A DeepSeek Harness plugin that uses Jev for skill and file selection, supervision, result filtering, corrections and approval assistance.

A local-first social research app that uses Jev Choice to select read-only Instagram, TikTok and LinkedIn operations from observed targets; the socai CLI runs them in the user's Chrome and returns source-linked evidence.

A Windows and Chrome delegation prototype for Codex that uses Jev for semantic choices while local code executes steps, checks outcomes and handles recovery.

A TypeScript AI SDK whose decide interface supports TypeSafe Jev and other decision providers for typed routing, gating and relevance judgments.

A collaborative visual workflow editor that connects Jev and LLM nodes, exposes a REST API and previews test runs.

A semantic grep CLI that uses Jev to score source lines against meanings and combine filters with AND, OR and NOT operators.

An unofficial agent skill that teaches typed Jev decisions, confidence-aware workflows and reusable patterns drawn from community projects.

A durable news agent that gathers sources on a schedule, uses Jev to gate, deduplicate and score stories as Hype, Facts and Signal, and survives restarts.

A Rust CLI that uses Jev-compatible providers to rank agent skills against live session context, with abstention, hooks and local feedback.

A Claude Code skill that sends bulk filtering, classification and ranking work to Jev, returning shortlists and confidence for the agent to review.

A browser extension demo that combines page tools exposed through WebMCP with Jev selection.

Applies Jev judgments to coding-agent tool calls and untrusted results before allowing, asking or denying an action.

A Pokémon Red harness that reads legal game state and lets Jev choose navigation, dialogue, team and battle actions without writing to game memory.

A macOS computer-use harness that reads accessible UI state and lets Jev select actions without a vision model.

A native Swift 6 bridge that combines Apple Foundation Models with Jev for typed, probability-bearing decisions.

A desktop browser where a planning model sets the goal, Jev chooses from observed page actions, and guarded local code executes and records the result.

An agent evaluation and guardrail toolkit that turns a trace and rubric into Jev decisions and can run compatible local models.

A CLI and local MCP server that sends authorized code fragments to Jev for relevance scoring, then returns exact excerpts with paths and line numbers.

A read-only local Gmail triage app that classifies messages into fixed folders and labels with Jev.

An MCP server and CLI that uses Jev to select links and passages while crawling a site, returning source-backed extracts for an agent to read.

Turns Jev choices, scores and probabilities into Home Assistant entities and automation actions.

A local gateway and visualizer that records Jev requests and answers in SQLite and maps calls live while an app runs.

A community Java client and Spring AI integration for the TypeSafe Jev API.

Routes each coding-agent step between fast and strong models with a Jev decision model, exposed as an OpenAI-compatible proxy with reproducible Terminal-Bench results.

A local workbench that combines OpenCV measurements, hard numeric rules and Laya or Jev decisions for agriculture and visual quality inspection.

An Android WeChat module that sends conversation context to Jev for emotion, implied-meaning and reply guidance.

A Rails search layer that uses Jev for index-time labels, query intent and streamed reranking alongside keyword and embedding search.

Runs a GraphRAG pipeline with swappable local Laya or cloud Jev decisions for graph edges, traversal and result filtering.

A self-hosted multi-bot chat where Jev selects which bot should reply and a local Pi daemon runs each turn.

An MCP server that exposes Jev-backed verification, screening, ranking and review tools, then maps probabilities to proceed, review or escalate policies.

A step-by-step Jev guide with Playground exercises, API calls, coding-agent setup and a runnable model-router example tested on Jev 1.14.

A Go SDK and CLI for calling Jev Choice, Score and Noul decisions from applications or UNIX pipelines.

Starts from one confirmed malicious entity, follows telemetry links and uses Jev to reconstruct an incident timeline, execution chain and evidence table.

A Python interface for Jev checks, labels, classifications and scores, with sync, async, batch and schema-based workflows.

Uses Jev to judge transcript segments, then deterministic code cuts highlight and condensed videos with timestamp-linked summaries.

A Rust command-line client that turns Jev answers into terminal output, JSON, exit codes and MCP tools for scripts, CI and agents.

A local evaluation lab for comparing decision models through inspectable test cases, recorded measurements, workflow replays and portable reports.

Fifteen tested Jev recipes covering support triage, file organization, PII checks, search reranking, browser actions and more.

Checks new arXiv papers against plain-language interests with Jev probabilities, then publishes a ranked page and RSS feed.

Tests Jev questions against labelled tune and holdout data, reporting errors, calibration and whether each question can safely gate or rank.

A Claude Code permission gate that combines shell-aware checks with Jev decisions to allow, deny or defer auto-mode tool calls.

A browser extension that filters an X feed with plain-language rules, using Jev or another compatible model chosen by the user.

A collection of Jupyter notebook recipes for Choice, Noul and Score workflows, separating shipped live or recorded examples from planned ones.

Pits local Laya against Jev in Snake races and a fighting-game arena, with every move made by a live model decision.

A Jev-first DeepSeek Harness agent loop with reusable typed actions, confidence thresholds, retries and execution traces.

A local MCP server for browser and macOS tasks where Jev selects bounded actions and local code executes them, verifies outcomes and keeps traces.

Adds Jev decision support to OmO and senpi agents for skill and tool selection, progress checks, loop detection and optional guarded actions.

A CLI, package manager and SDK where Jev selects a small installed program from a plain-language request and confidence gates its execution.

A hands-on Jev course with notebooks for agents, routing, safety checks and other fast-brain/slow-brain patterns.

A batching proxy with Python and JavaScript clients that evaluates many Jev states concurrently and returns ordered Apache Arrow streams.

An OpenAI-compatible local server that pairs a small llama.cpp model with Jev decisions while keeping text generation on the user's machine.

Adds image, video and RGB-D inputs to Jev by turning detected objects and scene details into structured visual decisions.

A browser-testing tool where Jev finds controls during authoring, recordings handle later replays and application code determines pass or fail.

An OpenCode Magpie plugin that discovers free model catalogs from OpenCode Zen and Kilo Gateway and adds native Jev System One routing.

A reproducible multi-agent lab and coordination kernel that asks Jev whether concurrent Claude Code writes conflict, then allows, waits or escalates uncertain cases.

A comparison app that asks Jev and AnyJev whether a headline is satire or real news, then shows both decisions and probabilities.

Zero-shot named-entity recognition that uses Jev to select, verify and type text spans, with probabilities, traces and multilingual benchmarks.

A stealth game where Jev proposes guard judgments and deterministic game code validates and applies them.

Evaluates 20 Jev-like model families across eight benchmark suites through one CLI, with prepared datasets, probability metrics and downloadable leaderboards.

A macOS control surface for coding agents that uses Jev to route each request to a live session, folder or earlier conversation.

Indexes visible browser controls so Jev can choose a valid Playwright action, then records evidence and replayable tests.

An open 4B vision decision model with published weights, training and calibration code, served through the TypeSafe System One API.

A React inspector for tracing TypeSafe Jev requests, responses and multi-call decision flows while developing Node, Vite or Next.js applications.

A Python Jev toolkit with sync and async clients, batching, calibration, confidence gates, graphical-model utilities, CLI and MCP server.

Eight Python examples of Jev decisions in engineering tasks, including CAD routing, FEM triage and BOM alignment.

A browser extension that asks Jev to score readable page sections against ten clarity, writing and on-page SEO rubrics.

Combines asynchronous Jev Box detections with optical flow and a Kalman filter to maintain real-time object tracks without blocking each video frame on the network.

An Android job-listing assistant that combines Jev judgments with local rules to flag pay, benefits and clauses worth checking.

A cross-platform desktop workbench for Claude, Codex and Grok that optionally uses Jev for fast voice and agent-control decisions.

A Google Apps Script that uses Jev to label, prioritize and archive Gmail messages, create deadline tasks, flag phishing and compile a daily digest.

A local server that accelerates open decision models with exact state forking and serves many text and multimodal families through one interface.

Benchmarks 20 specialized decision models and three reasoning-off LLM references on 1,342 text decisions across 31 categories, reporting accuracy, cost and median/p95 latency.

A SeleniumBase example that reduces webpage HTML to relevant evidence, then asks Jev whether a CAPTCHA is present and which type it is.

Code, datasets and released results for a reproducible black-box red-team study of universal adversarial suffixes against typed Jev decisions.

An OpenCode plugin that exposes Jev-compatible System One decisions as agent tools and can automatically judge selected workflow events.

A local MCP gateway combining deterministic PII and PHI controls with bounded Jev decisions before coding agents can access sensitive files.

A reproducible counterfactual audit of 134,772 Jev answers measuring how identity-only prompt changes affect typed decisions.

Adds calibrated Jev classification, yes/no and passage-location subcalls to recursive language model loops, with spend caps, offline tests and reproducible experiments.

A typed Go layer for Jev and Cloudflare Clef that supports Choice, Noul and Score requests, provider comparisons, explicit policies and test fakes.

A governance extension for the pi coding agent that uses Jev for planning, model routing, permission review and outcome verification.

Runnable Python examples through OpenRouter, including support-ticket triage and offline tests.

A self-hosted fighting arena for decision models with configurable fighters, paired tournaments, deterministic replays and a bundled 21-model benchmark.

Benchmarks System One models and generative LLMs on 52,199 typed-decision requests and closed-loop environments, publishing datasets, runners and a leaderboard.

A multilingual benchmark comparing Jev, another decisions API and three LLMs across four real-time voice-call decisions in English, Hindi and Hinglish.

A goal-driven Playwright agent that uses Jev to choose UI actions and verify progress, recording walkthrough videos and generating user guides.

Community n8n nodes for typed yes/no, choice and score decisions through TypeSafe Jev, Venice or another /v1/systemone endpoint.