Jev Guardrails

OpenClaw plugin that checks every outgoing agent message against its destination channel tier, running local key-format and term rules and then five Jev Noul questions in one call through OpenClaw's decisionModel, which can point at hosted Jev or a local Kev.

A DeepSeek Harness plugin where Jev Choice and Noul checks gate every non-read-only tool call - deny confident high-risk calls, escalate ambiguous ones, auto-approve clearly granted reversible ones - and optionally score session messages so valuable context survives instead of being dropped oldest-first.

Python sandbox and tutorial companion that turns Jev Choice and Noul judgments about issue category and reproduction into code-applied queue proposals, with confidence gates, a review fallback, synthetic offline fixtures and labeled replay. Live mode is opt-in.

pi coding-agent package that gates subagent dispatches: the task text is evaluated in one typed Jev call against configurable rules (per-rule probability questions with below/above thresholds), violations block the dispatch with per-rule reasons, and errors fail open to allow.

Supervisor for unattended coding agents: it tracks cross-turn state and asks one batched Jev Choice whether a trajectory should continue, pause or escalate, while deterministic code owns dangerous-command blocking, budget caps and Ed25519-signed receipts.

Permission gate extension for the pi coding agent: deterministic allow/deny rules and protected paths settle clear cases first, then gray-zone tool calls get one structured classify() verdict; Jev is an optional native classifier via TypeSafe direct, OpenRouter, or other gateways, failing closed on error.

OpenAI-compatible /v1/chat/completions proxy that answers an agent host's approval-guardian call with APPROVE, DENY or ESCALATE. Jev answers one Choice question with a calibrated distribution; missing keys, timeouts and malformed answers all fail closed to ESCALATE.

LiteLLM's proxy adds TypeSafe guardrail hooks and a Jev complexity classifier for policy judgments and request routing across 100+ model providers.

An MCP proxy and hooks adapter that screens an agent's tool calls before they run and the tool results before the agent reads them, with thresholds in a policy file.

OpenLayer's eval and guardrail library that replaces LLM judges with Jev-style decisions, packing every check for an agent trace into one request.

Vercel's Pranit reports benchmarking the fx auto-mode safety classifier with Jev: about 5 to 18 times faster and more accurate than GPT-5.6 Luna.

.NET SDK for the System One API with typed question sets, HttpClientFactory and dependency-injection wiring, plus Microsoft.Extensions.AI guardrail, routing, tool, and evaluator adapters.

An MCP server for Antigravity IDE, Cursor, and Claude Code that uses Jev (jev-latest) to locate code semantically and audit git diffs, replacing broad grep sweeps of entire files.

A code reviewer built as AI-powered software rather than an LLM agent: code slices the diff, Jev answers narrow typed questions per hunk, and code composes the verdict and findings.

A research-stage proposal-review contract in TypeScript: an LLM proposes one action, Jev answers four pinned questions, and code produces evidence the host can consider.

A Claude Code Stop hook that reads the turn transcript locally and spends one four-question Jev call only when files changed with no passing check since, blocking unverified done claims.

A pre-commit hook that answers five Jev questions about the commit message and staged diff together, warning on mismatches and blocking only on a credential-shaped line.

CLI and GitHub Action that turn project-defined semantic preferences from AGENTS.md into Jev Noul or Choice checks over code changes, then map each result to an advisory or blocking outcome. Checks can be tuned on labeled diffs.

Pi extension that supervises a coding agent with Jev judgments: it holds risky tool calls, checks writes against project Markdown rules, and feeds most issues back to the agent as a steer instead of interrupting you. The conscience is beta and off by default.

Python MCP server that exposes classify, score, check, match, and screen tools to MCP-compatible agents, returning typed judgments over a closed answer set with token usage and latency. Ships generic triage and routing question examples.

Agent-ergonomic CLI following the AXI conventions that gives coding agents Jev judgments for blocking risky tool calls, screening fetched content for prompt injection, triaging logs, flagging risky diffs, and ranking many items. Routine commands are decided locally at no cost.

A Node CLI that sends source, configuration, build, and CI files to Jev and reports suspicious files and lines with probabilities, as a second opinion before running unfamiliar code.

A security hook for coding agents that asks Jev three typed questions before every tool call and scans tool results for prompt injection, with adapters for Claude Code, Codex, Copilot CLI, Gemini CLI, Cursor, pi, OpenCode and ACP.

A self-hosted Discord bot that classifies messages with Jev in real time, deletes spam and scam links, escalates repeat offenses through warnings to timeouts, and profiles members from message history.

A CLI that turns Jev into Unix pipeline primitives: is, choose, score, filter, and guard commands return typed verdicts, stable exit codes, and versioned JSON without prompt or parsing glue.

A pytest plugin for semantic assertions: jev.expect sends many holds and lacks claims about one text to Jev in a single request and fails visibly when a probability is uncertain.

A hybrid coding harness for Pi where any LLM writes the code while Jev returns typed probabilities that deterministic policy code turns into allow, confirm, block, warn, or steer.

A warden for Claude Code and Codex CLI that keeps an append-only ledger of what an agent did, refuses a done claim when no check passed after the last edit, and uses optional Jev Noul judgments.

A TypeScript library that wraps Jev answers in a policy, a confidence gate, shadow mode, reusable recipes, and an offline eval CLI, so structured decisions can drive actions an application ships.

An experimental TypeScript harness that connects evidence to Jev judgments, permitted actions, and verified outcomes, with pluggable data sources and tools, local JSONL traces, and zero runtime dependencies.

A Pi extension that turns constraints stated in conversation into a scoped, replayable policy and checks every side-effecting tool call against it before it runs, with Jev only classifying how each message changes the policy.

A single-binary Rust gate that screens staged git diffs for unredacted secrets, prompt-injection attacks, and destructive commands with calibrated Jev probabilities in about 80 ms, invoked as git jev.

A tool-call gate for Claude Code, Codex, Cursor, and other harnesses backed by Jev: deterministic rules and a read-only allowlist run first, then one request answers allow, ask, or deny in about 400 ms.

An experimental Codex skill and JavaScript runtime that has Jev pick macOS Accessibility elements and actions from text candidates, while local policy gates hold sensitive operations for human confirmation.

Neon Function proxy and showcase site that puts request bodies to Jev: the judgment is the URL path, and bodies that pass can be forwarded unchanged to a caller-chosen HTTPS target.

Spring WebFlux starter that asks Jev whether HTTP 200 responses actually encode a failure, converting suspected silent failures into SemanticFailureException for existing error handling.

Private TypeScript proof of concept that flags Spanish profanity with Jev Boolean and Score questions and replaces the matching word intervals with a beep via ffmpeg, keeping the original duration.

A Chrome side panel that turns the current page's WebMCP tool schemas into typed Jev questions, predicting the tool call, its arguments, confidence, and latency as you type.

A Hermes Agent plugin that replaces the auxiliary approval reviewer with TypeSafe's Jev decision model: one request asks six typed questions and deterministic code returns APPROVE, DENY, or ESCALATE.

A test report that puts Jev in front of each agent action before it runs: most attacks were caught, with almost no false positives on legitimate tool calls.

Catch wrong and hallucinated citations against RFC 7519 with a string match plus one Choice question: four accurate citations verified at 0.93 or more and all four planted failures caught.

Score retrieved passages with four Noul questions each, then route in code: prompt injections dropped, false premises kept as conflicts, and evidence assembled separately for the generator.

A map of documented Jev use cases, routing, guardrail verification, citation checking, reranking, rubric scoring, bulk classification, and semantic linting in CI, each with a confidence policy.

Screen every message going into and out of an LLM app: a Noul describes possible hazards, a Score rates severity, and code thresholds the probabilities.