1
A policy engine for LLM decisions: declare routes, gates and actions once in TypeScript or Python, and every decision comes validated, explainable, replayable and audited.
Guardrails & safety — page 3
Permission checks, moderation and safety filters for agents. Jev answers an explicit policy question before the action runs.
147 active projects match this task, ordered by traction.
An autonomous risk-monitoring MCP server. Define a risk profile — a topic, the geographic locations to watch, and the policy triggers that matter — and the server runs an agentic sweep: live web search scoped by a judgment model (Jev), source scoring, and a Markdown briefing you can read from any MCP client.
Turn messy state into typed, auditable classifications, scores, and policy decisions over REST and MCP.
A Jev (TypeSafe System One) permission gate for the Pi coding agent: judges every bash / write / edit call before it runs
Skill Dash uses Jev to judge Claude Code and Codex skills and plugins: usefulness, redundancy, clarity, duplicates, safety. Local dashboard plus the corpus pipeline behind whichskills.dev. MIT.
Auto permission mode for Command Code: screens every tool call with TypeSafe Jev before it runs, and rejects anything out of scope.
DeepSeek Harness plugin: per-tool-call Auto-permission review powered by TypeSafe Jev, with account usage and API-key management inline on its settings page
Claude Code plugin: a 3-cent bouncer for your agent's shell. Jev typed probabilities auto-allow routine commands, deny destructive ones, and flag prompt injection in tool results.
Context compaction and safety gating for AI agents via TypeSafe Jev: keeps messages verbatim, no summarization. OpenAI, Anthropic, LangChain, CLI, MCP.
A safety boundary for AI-assisted Home Assistant decisions, with explicit policy checks and deterministic state verification.
Drop-in decision/routing layer powered by TypeSafe Jev (System One) — typed Choice/Noul/Score routes with production safety gates.
1
Delegation policy for Claude Code, cut down to what its own benchmark supports: two pinned roles, and a skill that argues against delegating by default.
A model-agnostic, Jev-first Responsible AI harness for assessing AI systems for prompt injection, secret/PII leakage, unsafe tool use, and policy bypass. Runs portable tests, captures evidence, scores risk with calibrated uncertainty, and routes serious or unclear findings to human review.
1
Jev-powered Judgment layer for Claude Code. Stops paying reasoning prices for if-statements: a PreToolUse hook scores every tool call against a YAML policy you own — allow / deny in ~100 ms, no LLM in the loop. TypeSafe Jev now, local models next. Dry-run by default, calibration table published.
Semantic firewall for LLM agents: tool calls gated by TypeSafe Jev (System One decision model via OpenRouter) + deterministic policy. PoC with corpus, stability eval, baseline, results.
1
Moderation for communities and apps, powered by Jev (TypeSafe): probabilities per category, thresholds you own. Discord/Telegram/Reddit bots, CLI, Python, npm, HTTP API, MCP.
Decides whether an AI agent step can finish on its own, retry, or escalate, with deterministic checks and a safety veto.
Guardrail + model router for LLM gateways on TypeSafe's Jev (System One model), with an independent accuracy/calibration/latency evaluation. Stdlib Python.
Jev-powered model-judged permission gate for OMP (TypeSafe System One)
1
Your terminal, with a coding agent inside: reads your code, edits files, runs commands — with permission gates on every risky step.
A safety-first personal AI orchestrator for coordinating heterogeneous coding agents through ACP, isolated Git worktrees, deterministic verification, and quota-aware routing.
Fail-closed Antigravity PreToolUse guardrail using deterministic rules and TypeSafe Jev to screen tool calls, block risky commands, and escalate ambiguous actions to manual approval.
Crop Steering, Jev edition: the HA crop-steering engine with TypeSafe Jev judging every decision across P0-P3, probes, shots, salt and alerts, inside a deterministic safety envelope.
Self-hosted AI moderation bot for Telegram groups. Catches spam, ads and insults with the open Laya model running locally on CPU, or with the Jev API. Escalating mutes, appeals, admin menu. RU/EN.