How do I pick between candidate plans without another LLM call?

Generate 2 to 3 approaches with your large model, then call arbitrate() with context and candidates. It returns winner, confidence, probabilities and latency without generating text.

What happens when the safety check times out or loses network?

Call guard() with proposed action, current state and risk threshold. Fail-closed blocks on timeouts or network errors, while fail-open allows through to avoid halting workflows.

How does the stuck-loop breaker detect repeated failures?

Pass the recent tool-output history to check_stuck() with a confidence threshold. The digest example halts after three repeated connection-refused curl retries.

What setup does a fresh Python project need?

Install via pip install jev-pilot, resolve the key via argument, TYPESAFE_API_KEY, JEV_API_KEY or saved config, and read SKILL.md for Claude Code, Cursor or Codex usage.

Where does the key come from, and what are the client defaults?

Resolution runs in order: the api_key argument, TYPESAFE_API_KEY, JEV_API_KEY, then a saved file at ~/.jev_pilot/config.json written with chmod 600 when you construct with save=True. The endpoint defaults to https://api.typesafe.ai/v1/systemone, the model to jev-latest, the network timeout to 3.0 seconds, and max_retries to 2 with exponential backoff on 429 and 5xx.

What do the guard's returned numbers actually mean?

danger_score runs 0.0 for safe to 1.0 for critical, action_type categorizes the proposed command as safe_read, reversible_write, or destructive, and risk_threshold is your own float from 0.0 strict to 1.0 lenient. The default fail_closed raises a PermissionError from the guardrail decorator when the action exceeds the threshold or is categorized as destructive.

What else does it ship with besides those helpers?

A SKILL.md agent specification any coding agent can read to use Jev for fast safe decisions, a Persian README, and verify_fact for grounding a model claim against a reference document, returning is_hallucination and a risk_score. Published on PyPI as jev-pilot under MIT.

CommunityJev Agents & MCP

Jev Pilot

Python helper that wraps TypeSafe Jev for agent arbitration, pre-execution guard, stuck-loop checks, fact verification and intent routing with probabilities and confidence.

See highlights

Open github.com

Stars: 6
Language: Python
License: MIT
Published by: Community
Author: h0j5bz0adh0-stack
Last verified: Oct 2026

h0j5bz0adh0-stack/jev-pilot

User repository on GitHub

View on GitHub

Fast System-1 Decision, Arbitration & Safety Engine for Autonomous AI Agents (Powered by TypeSafe Jev)

Forks: 0
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

🇮🇷 راهنمای جامع فارسی (Persian Documentation) Modern LLMs (Claude 3.5, GPT-4o, DeepSeek) represent System 2 thinking: they are brilliant at deep reasoning, creative writing, and writing complex code. However, using an LLM for operational split-second decisions inside an autonomous agent creates huge bottlenecks: ⏳ Too Slow: Waiting 3 to 10 seconds for an LLM to stream a simple yes/no or routing choice. 💸 Too Expensive: Burning thousands of prompt tokens just to verify safety or pick a tool.

Quoted from the project's README. Read the full README.

Tagscommunity awesomejev catalog-import

Works forJev for Classification Jev for Moderation Jev for Tool calling

Used inJev for Software & DevTools

Languages: Python

Highlights

arbitrate() selects among candidate strategies with calibrated confidence.
guard() gates shell, SQL and mutations with fail-closed or fail-open on_error.
check_stuck(), verify_fact() and route() cover loops, hallucinations and intents.
Constructor sets endpoint, model, timeout, retries and secure key persistence.

Watch out

Latency and cost figures follow TypeSafe published pricing and live API tests; guard defaults to fail-closed. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

jev-decisions

Hermes plugin that puts an agent's plans, drafts and claimed completions up for review — plan_review, evidence_review, action_verify — plus local approval rules that make no model request.

Jev Agents & MCP#community#awesomejev#catalog-import

Community

bojansandhaus/jev-decisionsPython

jev-recruiter

Local LinkedIn recruiting workspace on Browser Use's Jev Ultrafast: Jev screens titles, drives navigation, and assesses profile excerpts against your brief while a live browser stays visible; matches and evidence export as JSON.

Jev Agents & MCP#community#jevbest#catalog-import

Community

skeptrunedev/jev-recruiterPython

jev-skill-suggester

Codex Skill plus stdlib-only Python CLI that recommends which installed skill to read next for a task, allowing no-skill or uncertain answers and never executing candidates.

Jev Agents & MCP#community#jevbest#catalog-import

Community

win4r/jev-skill-suggesterPython

jev-judge-mcp

MCP server over stdio that gives coding agents fourteen Jev judgment tools, turning evidence plus an enumerable question into auto, review, or escalate actions from calibrated probabilities.

Jev Agents & MCP#community#jevbest#catalog-import

Community

PyModel/jev-judge-mcpPython

jev-agent-design-with-topk-logits-choices

Research runtime that pages a large logical tool and context space through a small bounded Jev decision surface with explicit approvals.

Jev Agents & MCP#community#jevbest#catalog-import

Community

6Mikao9/jev-agent-design-with-topk-logits-choicesPython

muse-jev-playbook

Agent-oriented playbook for using Jev as a cheap, fast decision layer that triages, classifies, scores, and gates work before expensive browser, research, retry, or subagent steps.

Jev Agents & MCP#community#awesomejev#catalog-import

Community

Bodila51/muse-jev-playbookPython

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai