Submit a find
한국어
日本語
Español
Français
Deutsch
简体中文
繁體中文
Português
Bahasa Indonesia
ไทย
Italiano
Tiếng Việt
Türkçe
Polski
Русский
Back to the labJEVLAB / FIELD NOTES
Fresh from the lab
A dated reading list of newly added resources and JEVLAB NEWS. Come back for what changed, or follow the RSS feed in your own reader.
Saved collection · Automatic updates, chat and payments are paused.
25
awesome-trustworthy-jev
This repository serves as a comprehensive library of research papers, projects, and documentation focused on Jev foundations, trustworthiness, and cybersecurity applications. It provides curated resources for exploring System One decision-making models and their integration into various technical workflows.
github.comJevOut
JevOut provides tools to investigate how natural-looking context additions can influence decision models. The project demonstrates that subtle background details can redirect correct decisions toward specific incorrect outcomes without altering the original task or instructions.
github.comJEV-MLX
JEV-MLX is a Python library for local LLM action selection on Apple Silicon using MLX. It maps application state and user requests to specific actions without requiring additional model training or fine-tuning.
github.comCARVE
CARVE is an open-source Jev-like model family based on Qwen3.8-27B. It demonstrates a TypeSafe-compatible API that performs one-token decisions without free-form generation, offering various serving profiles for balanced accuracy and throughput in decision-making tasks.
github.comex-regex
ex-regex is a Python library that replaces traditional regex with semantic pattern matching using the Jev model. It allows developers to extract and classify text based on meaning rather than syntax, providing probability-based results without generating new text.
github.comjevotron
jevotron is a CLI tool for field-level anomaly detection in structured files and databases, powered by Jev. It enables users to preview data, scan for inconsistencies, and manage review queues using local caching and SQL-based filtering.
github.comJevTown
JevTown is a simulation engine where AI agents navigate a pixel-art environment. It demonstrates a hybrid architecture using Jev for structured decision-making and LLMs for natural language processing, allowing agents to interact autonomously within a browser-based game world.
github.comJev Instance Review Studio
This toolkit provides a structured, typed framework for auditing Roblox Studio instances. It enables developers to perform reproducible reviews by validating scene properties against defined criteria and generating detailed, citation-backed reports to ensure consistency across team projects.
github.comhelpme-Jev
HelpMe Jev is a self-hosted decision-making workbench that integrates Jev for structured analysis and LLMs for language tasks. It helps users evaluate tricky social or workplace situations by providing probability-based insights and drafting support while keeping the final decision-making power with the user.
github.comJev Cookbook
This collection of Jupyter notebooks provides a curriculum for building typed AI decisions using Jev. It teaches developers how to structure inputs and outputs for tasks like classification and scoring, emphasizing deterministic Python logic alongside model-based judgments.
github.comClinicalJev
ClinicalJev is an open-source model family designed for general-purpose clinical text classification. It allows users to define tasks and labels at inference time, enabling local execution for sensitive healthcare data without requiring task-specific fine-tuning.
github.comreflex-cu
reflex-cu is an MCP server for AI agents to perform computer use tasks on macOS and Windows. It improves efficiency by reading UI accessibility trees instead of relying solely on screenshots and uses Jev for rapid decision-making.
github.comjev-mail
jev-mail is a browser extension that uses the Jev AI model to classify, label, and clean up Gmail inboxes. It runs locally in your browser, using your own API key to provide privacy-focused email triage and bulk management tools.
github.comjev-max
jev-max is a browser agent library that enhances Jev-based automation by piercing shadow DOMs and iframes. It provides robust features like canvas fallback, occlusion-checked clicks, and multi-tab support, enabling agents to interact with complex web pages through a unified MCP interface.
github.comCloakBrowser-Agent
CloakBrowser-Agent is a stealth browser automation tool that integrates TypeSafe Jev for decision-making. It functions as an MCP server, CLI, or Python library, enabling human-like browser navigation by selecting operations based on calibrated probabilities rather than generating free-form code.
github.comLaser
Laser is an AI-powered distraction blocker for Omarchy and Hyprland. It uses TypeSafe Jev to evaluate active windows against your declared task, providing gentle, escalating nudges to keep you focused while maintaining strict local privacy controls.
github.comtypesafe-mcp
This MCP server integrates TypeSafe AI's Jev decision model into AI agents. It provides structured, deterministic answers including certainty scores and act/review/abstain verdicts, enabling agents to perform typed evaluations on states or files without generating conversational text.
github.comsystem_one_client
This Elixir library provides a client for System One decision models, supporting TypeSafe Jev, Cloudflare Clef, and OpenRouter. It enables developers to retrieve typed answers with full probability distributions for state-based decision routing, including built-in telemetry and testing stubs.
github.compi-subagent-jev
This package integrates a JEV System One decision model into the pi coding agent to gate subagent dispatches. It provides compliance rules, direct model querying, and optional runtime monitoring to help users manage agent behavior and task execution.
github.comcodex-gemini-jev-bridge
This local MCP bridge enables Codex to delegate drafting tasks to Gemini while using TypeSafe Jev for automated quality checks. It leverages Google AI Pro quotas via Antigravity CLI to manage routine code proposals and analysis.
github.comoptim-jev
optim-jev is a Claude Code plugin that optimizes token usage through context compaction, model routing, and selective skill loading. It aims to reduce session overhead by managing tool outputs and model selection without disrupting existing developer workflows.
github.comjev-comment-triage
This WordPress plugin uses the Jev model to asynchronously moderate comments for spam, abuse, and relevance. It processes comments in the background to maintain submission speed, using configurable thresholds to decide whether to approve, hold, or mark comments as spam.
github.comJevvy
Jevvy provides a plugin collection for coding agents that uses TypeSafe System One models to automate shell permission approvals. It evaluates commands against risk-based thresholds, abstaining from decisions when uncertainty exists to maintain agent security.
github.comnocap
nocap is a System One evidence gate for RAG pipelines that evaluates retrieved context before generation. It routes queries to answer, retrieve more, abstain, or review conflicts, helping prevent hallucinations by ensuring LLMs only generate responses supported by provided evidence.
github.comjevr
jevr is an R client for JEV and TypeSafe System One models. It enables users to send structured states and typed questions to models, returning parsed, programmatic answers for decision-making tasks rather than conversational chat.
github.com
20
Awesome-jev-papers
This repository provides a curated collection of academic preprints and technical reports regarding Jev, a System One model designed for typed probabilistic decisions. It organizes research into core sources, benchmarks, and applications to help track the rapidly evolving literature on decision-focused AI.
github.comJevector
Jevector demonstrates a profile matching system using Clef or Jev to encode criteria into numeric vectors. It utilizes HNSW for retrieval and ranks candidates based on preference coverage, showing higher synthetic benchmark scores compared to BM25 and BGE-small.
github.comsocial-risk-qwen-jev
This project demonstrates a multimodal moderation pipeline using Qwen3-VL with LoRA to generate structured evidence for TypeSafe Jev policy decisions. It provides reproducible baselines and evaluation protocols for identifying misogynistic content, emphasizing auditability and the trade-offs between recall and manual review costs.
github.comjev-playground
This repository provides practical GTM workflow examples using Jev, including lead triage, deal risk assessment, and account management. It demonstrates how to integrate TypeSafe AI processes with Claude for automated business operations.
github.comjev-olympics
This experimental playground uses the Jev model to perform arithmetic, chat, and drawing tasks one decision at a time. It allows users to inspect probabilistic outputs and incident reports for every step, focusing on model behavior rather than task completion.
github.comNucleiSniper
NucleiSniper is a pipeline that uses TypeSafe Jev AI to profile targets and rank Nuclei templates by relevance. It aims to surface critical security findings faster by prioritizing high-value checks over generic scans.
github.comComfyUI-Jev
ComfyUI-Jev provides custom nodes for integrating Jev text interpretation and judgment capabilities directly into ComfyUI workflows. It enables users to perform tasks like text scoring, candidate selection, and conditional logic using natural-language instructions.
github.comJEV Safety Gateway
JEV Safety Gateway is a transparent reverse proxy that sits between Nginx and LLM backends. It intercepts requests to extract user input, evaluates them via the JEV model for safety, and blocks harmful content while passing legitimate requests through.
github.comJevspresso
Jevspresso demonstrates integrating Jev with XState to manage an espresso bar simulation. By restricting the model to only select valid state machine transitions, it ensures the AI agent operates within defined physical constraints rather than generating arbitrary tool calls.
github.comComfyUI-SystemOne
ComfyUI-SystemOne provides custom nodes that enable branching workflows based on natural-language judgments. It supports local Laya models or hosted Jev and Clef APIs to route, score, or filter image generation tasks through automated decision-making.
github.comnf-jev
This Nextflow plugin integrates TypeSafe System One judgments into bioinformatics pipelines. It allows developers to define typed questions as pipeline values, enabling branching logic based on probability distributions rather than generated text.
github.comJevSceneMiner
JevSceneMiner is a tool for extracting driving maneuvers from vehicle logs using Jev. It converts motion and sensor data into structured, timestamped scenes, enabling users to curate datasets, validate driving performance, and analyze maneuver patterns through a synchronized review interface.
github.comJevLight
JevLight is a framework for traffic signal control using the Jev decision model within SUMO simulations. It demonstrates how to map intersection states and phase decisions into structured API requests, supporting both per-intersection and coordinated network-level control strategies.
github.comHunch
Hunch is a browser-based playground for the Jev decision model. It allows users to visually construct typed queries, execute them against the OpenRouter API, and inspect probability outputs, latency, and cost metrics without requiring a backend server.
github.comjev-browser-wingman
This tool integrates the TypeSafe Jev model into browser automation workflows to handle element selection. It functions as a wrapper for existing MCP browser servers, aiming to improve task completion and cost efficiency during automated web interactions.
github.comjevify
jevify identifies AI call-sites suitable for migration to JEV by auditing code for structured decision patterns. It provides a shadow-mode migration workflow to validate performance and accuracy against existing LLM implementations before committing to production changes.
github.comjev-lint
jev-lint is a command-line tool that scans Markdown knowledge bases for contradictions, stale claims, and broken links. It uses TypeSafe Jev to perform semantic analysis on content, generating a self-contained HTML report to help users identify and resolve documentation issues.
github.comcould-jev
This Claude skill provides a design gut-check to determine if TypeSafe's Jev model is suitable for specific development tasks. It offers architectural patterns and feasibility assessments to help developers identify potential use cases for Jev's typed-judgment capabilities.
github.comopencode-jev-router
This OpenCode plugin uses Jev to dynamically adjust reasoning effort for AI models. It aims to optimize token usage and latency by selecting appropriate effort levels per request, rather than relying on fixed settings for all tasks.
github.comfilament
Vibefilter is a Filament plugin that enables semantic filtering of table data using natural language. It utilizes the TypeSafe Jev model to score rows based on user-provided statements, allowing for zero-shot classification without requiring manual training data or keyword lists.
github.com
27
awesome-system-one
This repository provides a curated, auto-updated list of System One decision models. It documents various open-source alternatives to proprietary Jev interfaces, detailing their architectures, licensing, and compatibility with standard decision-making API patterns for typed inputs.
github.comawesome-jev-survey
This repository provides a curated evidence survey of Jev and Jev-like typed decision models. It synthesizes research on calibration, failure modes, and system integration, tracing reported performance metrics back to their original academic sources.
github.comagent-workflow-typesafe-ai
This tool provides a standalone adapter for integrating TypeSafe AI semantic evidence into Agent-Workflow. It enables bounded decision-making and skill evaluation while ensuring the host system retains full control over policy, execution, and lifecycle management.
github.comtypesafe-eval
typesafe-eval is a CLI tool and CI gate for automated document evaluation using the TypeSafe System One API. It provides regression detection, checklist verification, and credential masking to help maintain document quality and safety in development workflows.
github.comCodex-Jev
Codex-Jev is a VS Code plugin that filters repetitive tool output before processing by Codex. It aims to improve clarity by highlighting relevant evidence while providing links to original logs for verification.
github.comvellum
Vellum is an experimental tool that maps Markdown content into Sanity documents using a classifier rather than an LLM. It demonstrates how to structure data by asking a classifier to assign text blocks to specific fields based on provided schema descriptors.
github.comDecidealot
Decidealot provides a unified HTTP and MCP interface for running local models like Laya, Von, and CLM, or hosted Jev services. It enables typed classification and scoring, allowing applications to programmatically handle decisions based on model-provided probabilities and confidence metrics.
github.comTypeWright
TypeWright is a declarative compiler and runtime for Jev decisions. It allows users to define state and decision contracts, compile them into checksummed JSON programs using labeled data, and execute these programs without requiring external dependencies like DSPy at runtime.
github.commappity
mappity is a tool that uses TypeSafe Jev and Cactus Needle to answer natural language queries about geographic locations. It processes OpenStreetMap and Mapillary data to assign probability scores to places based on user requests.
github.comJEV-27B-Desktop
This resource provides a local GGUF implementation of the JEV-27B model for Windows. It demonstrates integration with Ollama, featuring a Qwen3.8-based backbone and optional vision-language capabilities for local AI processing tasks.
github.comjev-pr-judge
This tool demonstrates using TypeSafe System One to provide typed, parallel verdicts on GitHub pull requests. It replaces slow LLM chains with deterministic code-based policies and model-driven judgments to automate code review signals efficiently.
github.comjev-playground
This resource provides interactive simulations demonstrating Jev, a System One model designed for fast, typed decision-making. It illustrates how parallel sampling and calibrated confidence can improve architecture for routing, triage, and verification tasks within existing AI systems.
github.comjev-test
This repository provides a test harness for TypeSafe's Jev model, designed to evaluate confidence calibration, hallucination detection, and retrieval-augmented generation performance. It includes scripts for benchmarking accuracy, cost, and latency against various datasets and alternative LLM judges.
github.comJevline
Jevline automates incident reconstruction by starting from a single confirmed malicious process. It uses the Jev model to iteratively link related telemetry, generating a comprehensive timeline, execution chain, and evidence table from diverse, unmapped log formats.
github.comjev-risk-check-provider
This tool provides signed pre-payment risk assessments for wallets and agents. It evaluates counterparties using OFAC lists, phishing feeds, transaction simulation, and proprietary kit-watch monitoring to detect malicious contract patterns and address poisoning.
github.comVision-Jev
Vision-Jev is an open-source project providing vision-enabled models for judging and acting on dynamic candidate sets. It utilizes structured output heads for choice, evidence sufficiency, and quality scoring, focusing on auditable and calibratable decision-making rather than open-ended generation.
github.comnu_plugin_jev
This Nushell plugin integrates TypeSafe Jev structured decision-making directly into the shell environment. It allows users to define questions, annotate data streams, and retrieve model-based insights using native Nushell pipelines and commands.
github.comSOLO Decision
SOLO Decision optimizes relational data layouts to maximize prefix-caching efficiency for LLM-based decision models. By reordering rows and fields to expose shared input prefixes, it demonstrates improved throughput in synthetic workloads using the JEV-9B model on NVIDIA hardware.
github.combrowser-laya
This project enables running the Laya non-autoregressive decision model entirely within a web browser using WebGPU or WebAssembly. It demonstrates how to perform state-based decision tasks locally without servers or API keys by leveraging ONNX Runtime Web.
github.comjev-mcp
This MCP server allows AI agents to interface with the Jev API for structured decision-making. It enables agents to perform classification, routing, and verification tasks by returning typed values with confidence scores instead of prose.
github.comjeffort
jeffort is a Claude Code plugin that dynamically routes prompt effort levels using TypeSafe's Jev. It modifies turn steps to apply effort settings while including data redaction features to protect sensitive information during API requests.
github.comfast-jev-compaction-openrouter
This tool provides a Claude Code plugin and library for context compaction using Jev via OpenRouter. It prunes unnecessary tool calls and results while preserving verbatim text, enforcing zero data retention through mandatory provider routing configurations.
github.comjev-lint
jev-lint is a fuzzy linter for coding agents that provides real-time feedback on code quality. It helps agents identify and fix rule violations immediately after editing, reducing the need for costly post-review corrections.
github.comjev-skill-router
This project provides a reference implementation for using TypeSafe's Jev model as a skill router for Claude Code. It demonstrates how to log and evaluate skill suggestions, though the author found limited utility in this specific integration.
github.comjev-blindspot
jev-blindspot is a prompt review tool for Claude Code and Codex CLI that identifies potential knowledge gaps in your requests. It uses a classifier to decide if analysis is needed, providing feedback in a local web UI without interrupting your session.
github.comjev-gateway
jev-gateway is a local proxy that integrates Jev, a fast decision model, into coding agents like Codex and Claude Code. It intercepts tool-calling decisions to improve efficiency, while forwarding other requests to the original LLM provider without modification.
github.comjev4pg
jev4pg integrates JEV semantic intelligence into PostgreSQL, providing tools for natural language queries and semantic filtering. It features explainable embeddings and parallel execution to help users analyze existing database records through a unified workspace or native SQL extensions.
github.com