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.
Remember this visit on this browser to highlight new additions next time. No sign-in or cross-device tracking.
Dates show when an item became available in JEVLAB, in UTC, not when its original source was published. Translation-only edits do not count as new resources.
36
jev-bfs
This project implements a Wikipedia link racing game using Jev ranking. It demonstrates a breadth-first search approach to finding paths between articles, featuring a live terminal interface for tracking progress during the race.
github.comjevegis
jevegis provides a unified API for implementing guardrails in LLM applications. It is designed to detect prompt injections, jailbreaks, data leaks, and unsafe content, leveraging the TypeSafe Jev framework to enhance security for language model interactions.
github.comJev-Driven Pac-Man Gameplay
This demonstration showcases a Pac-Man session where every movement is dictated by Jev. It serves as an experiment in applying Jev-based decision-making logic to classic arcade mechanics to observe strategic navigation patterns.
x.comDiogo Almeida Tech Talk
This presentation by Jev founder Diogo Almeida explores the platform's architecture. The creator claims the system achieves high operational speed and cost efficiency while maintaining accuracy, though these performance metrics remain independent of third-party verification.
x.comDoom with branching futures
This experiment demonstrates using Jev alongside an LLM to play Doom. It explores the capability of simulating multiple potential future outcomes simultaneously during gameplay to inform decision-making processes within the game environment.
x.comCodebase Complexity Classifier
This project introduces a Jev-based classifier designed to evaluate codebase complexity. It aims to help developers identify and mitigate instances of overengineered code through automated structural analysis.
x.comTypeSafe AI Documentation
This resource provides the foundational documentation for TypeSafe AI. It covers core primitives, implementation patterns, and practical cookbooks, alongside comprehensive technical references for the HTTP API and available software development kits.
docs.typesafe.aijev-use
This plugin integrates Jev with AI agents like Claude Code to handle non-text output steps. It demonstrates a workflow for typed escalation back to the LLM, with the creator reporting p50 latency of 230ms and costs around $0.02 per 1,000 judgments.
github.comBuilding the fastest agent brain in 10 steps
This guide outlines a ten-step process for constructing an agent brain using Jev. The author claims this approach optimizes for high-speed performance, providing a structured workflow for developers looking to build responsive autonomous systems.
x.comConfidence-gated routing
This pattern demonstrates a method for routing requests based on model confidence scores. It teaches how to implement conditional logic to ensure higher quality outputs by directing tasks to appropriate processing paths.
docs.typesafe.aiA run at the Pokémon League
This project demonstrates Jev attempting to defeat the Pokémon League in Pokémon Gold. It showcases the use of basic control inputs to navigate the game environment and progress through competitive battles.
x.comtypesafe-go
This repository provides an idiomatic Go SDK designed for interacting with the TypeSafe AI API. It serves as a developer tool to facilitate integration and communication with the service using standard Go programming patterns.
github.comGround Truth news-framing extension
This browser extension utilizes Jev to analyze and classify online articles. It demonstrates how to automatically identify news framing, content topics, and the presence of loaded language within web-based media.
x.comvon
von is an open-source System One decision model designed as a local, non-autoregressive alternative to TypeSafe Jev. The creator claims the model achieves sub-15ms inference speeds for decision-making tasks.
github.comtypesafe-client
This Rust library provides an unofficial asynchronous client for interacting with the TypeSafe System One API. It demonstrates how to implement type-safe communication patterns when integrating with the platform's backend services.
github.comjev-mcp
This MCP server integrates TypeSafe Jev into Claude Code, providing structured tools for classification, scoring, and batch processing. It demonstrates how to expose calibrated judgment capabilities as standardized functions for automated agent workflows.
github.comjev-t-rex-runner
This project demonstrates the application of the Typesafe AI Jev model to automate gameplay in the classic Chrome T-Rex runner. It serves as a practical experiment in integrating AI agents with browser-based game environments.
github.comVoice control over Fusion 360
This project demonstrates a voice-controlled interface for Fusion 360. It uses Jev to distinguish between actionable commands and casual conversation, aiming to streamline CAD workflows through natural language input.
x.comtaste-lint
This command-line tool utilizes Jev to identify potential AI-generated low-quality content within user interfaces, copy, and agent instructions. It provides semantic taste checks to help developers maintain quality standards before deploying their projects.
github.comrh-guard
rh-guard provides a reward-hack detection mechanism for coding agents. It utilizes structural denies and a TypeSafe Jev System One sidecar to monitor hooks within Claude Code and Cursor, aiming to identify potential reward-hacking behaviors during automated development tasks.
github.comfast-jev-compaction
This Claude Code plugin replaces standard compaction summaries with Jev-based decision-making. It scores tool calls and results in a single request, automatically pruning or truncating stale data while preserving essential information in its original format.
github.comjevscape
jevscape provides a RuneBench harness for Jev, featuring a bounded action catalog and a tick-mode controller. It includes a live dashboard designed to help developers monitor and manage agentic workflows within the Jev ecosystem.
github.comspendbrake
spendbrake provides a mechanism for managing AI agent expenditures. It demonstrates how to implement budget controls by enabling users to continue, downgrade, or stop agent operations using TypeSafe Jev.
github.comjev-codes
This tool enables automated auditing of git diffs against defined YAML coding standards. It utilizes the Jev model to provide consistency checks directly through a command-line interface or integrated AI agent workflows.
github.comJev Room
Jev Room is a web-based tool that transforms a single descriptive sentence into six distinct interior room visualizations. The application uses Jev to interpret user input and generate corresponding spatial designs for creative inspiration.
jev-room.moe136231.chatgpt.sitejev-trader
jev-trader demonstrates an automated trading bot that utilizes Jev to analyze asset price feeds. The project showcases how the model makes buy and sell decisions and executes real-world financial transactions based on its internal logic.
x.comjev-mail-classifier
This project demonstrates an automated email management system using Jev. It provides a configuration-driven approach to classify, tag, move, and flag incoming messages, showcasing how TypeSafe AI models can be applied to streamline personal inbox organization workflows.
github.comFourteen checks for every pull request
This tool automates code quality by running fourteen distinct checks on pull requests. It aims to identify potential security vulnerabilities like hardcoded secrets and problematic SQL patterns during the development workflow.
x.comTypeSafe Introduction
This official documentation provides a foundational overview of the TypeSafe ecosystem. It serves as the primary entry point for understanding core concepts and architectural principles governing the platform's approach to safe and reliable AI integration.
docs.typesafe.aijev-crawlers
This repository provides automated crawlers built with Unix-style primitives for system analysis. It demonstrates a modular approach to bug-hunting by utilizing lightweight, composable tools to perform systematic data collection and vulnerability discovery across target environments.
github.comtypesafe-local
This project demonstrates a method for querying local LLMs on Apple Silicon to receive calibrated probabilities rather than raw text. It aims to provide structured output directly, bypassing the need for traditional text generation or subsequent parsing steps.
github.comCheshi
Cheshi is a macOS workspace integrating Jev-powered conversation memory with OpenAI Codex. It enables users to manage AI agents, navigate codebases via CodeGraph, and synchronize workflows across Git, Ghostty terminals, and Apple Notes within a unified environment.
github.comgame-coach
This project integrates Stockfish engine analysis with Jev-based decision logic to provide real-time strategic guidance for chess players. It demonstrates a method for combining traditional chess engines with modern AI judgment systems.
github.comSDE cascade
This official TypeSafe reference outlines the SDE cascade methodology. It demonstrates how to structure multi-stage AI workflows to improve output reliability and logical consistency during complex software development tasks.
docs.typesafe.aiDGP (Decision Graph Protocol)
This experimental protocol uses a Jev adapter to evaluate evidence through typed choices. It demonstrates how application code can validate permissions and execute simulated demo actions within a structured decision-making framework.
github.comkyotsu-ai-bench
This repository provides a static dashboard comparing Jev, luna-none, and luna-low models on Japan's 2026 Common Test. It serves as an experimental benchmark to evaluate how these specific AI architectures perform on standardized academic examination questions.
github.com
