Jev projects & practices

production GEO platform whose NOTRA_JEV_CLASSIFIERS flag routes brand-visibility classifiers off an LLM and onto Jev Boolean decisions at a 0.5 threshold, targeting 300 ms p50.

routes Claude Code tasks to the cheapest capable model by asking Jev to choose among candidates.

open-source LiteLLM-based router where a Jev decision picks which model serves each request.

adds automatic per-request model routing to the Pi coding agent through Jev decisions on Vercel AI Gateway.

local proxy that picks the Claude model and reasoning effort per message with a Jev decision while leaving the cached main chat untouched.

routes agent skill selection through typed, confidence-aware Jev decisions so weak matches are declined instead of guessed.

screens a folder of CVs with Jev typed judgments against an editable policy, re-scoring candidates for free when the policy changes.

async LangGraph workflow gets a typed Jev Choice (invoice or general) and routes each inbound email to the matching handler.

WXT extension where Jev decides per page element whether it is clutter, removing it under reusable template rules.

staged code-review workflow and local dashboard where Jev gates each review stage before a change advances.

adds a measured tool-call gate to the Pi coding agent so risky calls are checked by Jev before execution.

wires Jev into its eval testkit as a verification judge so agent-produced work is gated by typed verdicts rather than a text model.

prompt-injection and dangerous-action guard for Claude Code, Codex, Pi, and ACP agents, with Jev deciding what to block.

scores every file of a pull request on 31 boolean Clean Code smells plus function size and nesting, then hands the verdicts to a writing model for the review prose.

checks whether each cited paper actually supports the sentence citing it, with Claude locating the quote, Jev scoring the support, and a human making the final call.

finds link paths between English Wikipedia articles by having Jev rank each page's outgoing links while Python controls the search.

browser-use's ultrafast agent where Jev decides each next action and element to click, calling a language model only when text must be typed.

exposes System One judgments as five Pi tools so a model makes narrow semantic judgments while code and users keep control of thresholds, weights, and actions.

agent skill that sends closed coding-agent judgments to Jev so verdicts stay typed, cheap, and comparable across runs.

Stop hook that keeps an agent from finishing too early by judging plain-language completion rules with Jev.

SO-101 arm workbench where a Jev decision runner picks bounded joint steps from typed candidate actions under a spend budget.

multiplayer triage dashboard where Jev answers a fixed set of typed questions per issue — kind, severity, urgency, duplicate, and next step — and every human correction is kept and shown back to the model on later runs.

benchmarks Jev against Luna, Haiku, and Gemini at choosing validated legal moves in explicit-state games, scoring decision quality and consistency across a sequence of moves.

head-to-head test of Jev against Mistral Small and Gemini Flash-Lite at validating local event listings.

reproducible eval harness plus field note for Jev Ultrafast research-browser tasks, with QC'd cases, a suite runner, and a report generator.

reproduces the System One shape with a Qwen3.5-2B fine-tune that emits typed decisions with calibrated probabilities in one pass.

independent local preview that answers bilingual probability questions from context, questions, and candidate answers, inspired by TypeSafe Jev.

RLCD-trained Qwen2.5-1B demo exploring open-source parallel constrained decoding as an alternative to Jev.

Vercel's eve engine ships Jev as the default evaluation model (typesafe-ai/jev) in its experimental evaluate path.

Vercel Labs CLI that can run Jev as the evaluation model for its evaluate command.

proof-of-concept MCP server that puts Jev claim verification, content screening, and candidate ranking behind standard MCP tools.

MCP server exposing Jev classify, score, check, match, and screen as tools for any agent, with confidence on every answer.

ZIO client for TypeSafe AI with a typed DSL over Jev decisions.

unofficial Laravel integration for Jev with typed responses, async requests, scoped dependency injection, and testing fakes.

small type-safe client for asking Jev questions about a dataset.

async client for Jev published on PyPI.

Neon Function proxy for the Neon AI Gateway that routes decisions with Jev.

TypeSafe/Jev agent that plays Super Mario Bros. from structured emulator state, choosing each action from emulator-derived features.

camera-only autonomous drone in MuJoCo that puts a Jev judgment model in the control loop at 2.5 Hz.

drives original StarCraft shareware through keyboard and mouse with Jev action probabilities recorded per decision.

benchmark that asks Jev to predict federal motion-to-dismiss rulings from the judge's written record and scores the calibrated probabilities with claim-defendant micro-Brier metrics.

Discord bot that scores incoming messages for phishing, spam, and social engineering with Jev and drives a four-stage escalation ladder, injecting pardoned messages back into context as verified-safe precedent.

zero-shot spam classification with Jev Boolean questions, benchmarked against TF-IDF baselines.

1,800-point launch thread whose ~480 comments debate whether typed decisions replace LLM calls for classification, routing, and verification.

the 63k-like announcement from TypeSafe's founder arguing RLCD-trained decision models are a shorter path to economic value than chat models.

948-like demo where Jev decides which model should serve a request before it is forwarded.

builder reports that a small MLP trained on top of Qwen 4B already reproduces Jev-like decision behaviour.

attempt at running a Jev-style decision model locally, with speed noted as still improvable.

test report using Jev to check each agent action first, reportedly catching most attacks with almost no false blocks and much lower latency.

argues that purely engineering decisions in security work belong to Jev rather than a chat model.

public Convex-backed demo inviting one million judged questions instead of generated answers.

a shipped Mac app routes setup and troubleshooting questions to Jev when no language model is loaded.

explains the System One category to Chinese readers as a calibrated, typed decision layer for code.

launch thread framing Jev as a low-hallucination, low-cost decision model for software rather than chat.

builders describe using Jev as an agent tool-use safety layer and planning a prompt-complexity model router.

work-in-progress demo of Jev driving Minecraft, including fleeing zombies at night, as a test of fast structured decisions.