User repository on GitHub
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Agents & MCP
Grok Jev Guard
A typed preflight and approval layer for Grok Bot: local code owns hard boundaries like secrets, deletions and payments before the Jev SDK is imported, Jev judges ambiguity, and the guard returns one action such as proceed, ask_human or block.
- Stars
- 2
- Language
- Python
- License
- MIT
- Reading time
- 2 min
- Published by
- Community
- Author
- 0xwhrari
- Last verified
- Oct 2026
Highlights
- Returns one explicit action: proceed, scope_down, checkpoint_then_run, change_approach, ask_human or block.
- Hard facts such as publishes_external and deletes_data are evaluated locally before the Jev SDK import, so no model answer can downgrade them.
- Default shadow mode enforces local hard gates but keeps Jev recommendations advisory via recommended_action until you switch to active.
- Each CLI decision is appended to a JSONL audit log after redaction, and logging can be turned off when metadata should not be retained.
- The offline local-policy check needs no credentials and flags a non-reversible external publish as ask_human.
Quickstart
bash
grok-jev-guard --offline '{
"goal": "publish release 1.4.0",
"operation": "publish",
"reversible": false,
"publishes_external": true
}'Watch out
Grounded in the truncated repository README and GitHub metadata reviewed on 2026-10-15; the README was cut off in the Grok Bot installation section, and the GitHub metadata carries no description, so the framing comes from the README alone. Turning Jev off never disables local hard gates.
Questions about this entry
Short answers drawn from the linked project or page.
A typed preflight and approval layer that sits immediately before a meaningful Grok Bot tool sequence. It receives a compact description of the pending operation and returns one explicit action, from proceed through ask_human to block.
Local code owns secrets, deletion, publishing, payments, account changes, permissions and retry limits; Jev makes typed judgments about intent, approval expectation, destructive potential, scope fit and risk; Grok Bot executes within the returned limits and verifies; humans approve consequential or irreversible actions.
Because a model answer cannot downgrade explicit high-consequence facts here: publishes_external, financial_action, deletes_data, permission changes, account actions and secret-bearing state are evaluated locally before the Jev SDK is even imported.
The default. Local hard gates are always enforced; Jev recommendations come back as recommended_action while effective_action stays proceed for non-local recommendations. After reviewing representative decisions you can set mode: active so Jev-composed recommendations take effect.
Run the credential-free local-policy check: grok-jev-guard --offline with a JSON object describing the operation. A non-reversible external publish comes back as ask_human with source local_policy, approval_required true and jev_used false.
Each CLI decision is appended to the configured JSONL audit log after redaction. If the task metadata should not be retained at all, set logging.enabled: false in config.yaml.
Python 3.11 or newer, per the README badge. Install with pip install -e . from a clone, copy config.example.yaml to config.yaml, and set TYPESAFE_API_KEY for Jev-backed preflights.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Agent Handoff Gate: agents verify what they hand off
X: Jev as an LLM guardrail inside Agentgateway
X: Vercel's fx safety reviewer, measured at 18× faster
Reddit: Early experiments wiring Jev into Pi as a safety layer and model router.
More Jev resources like this
A guard for coding agents that puts Jev in front of every step: it judges tool calls for intent and risk, scans tool outputs for instructions aimed at AI agents, and checks replies, mapping probabilities to allow, ask or warn.
Jev Agents & MCP#community#catalog-import#logicrw
Community
harshwasan/pi-jev-sentinelTypeScript
harshwasan/pi-jev-sentinelTypeScriptA Claude Code plugin that keeps agents honest: Claude lays work out as phases with a definition of done and a check, and whenever it tries to stop, Jevflow runs the checks and asks Jev whether the work is really done. Several agents can share one plan.
Jev Agents & MCP#community#catalog-import#logicrw
Community
parth1811/jevflowPython
parth1811/jevflowPythonLets AI agents use Windows applications through MCP or the command line: hand off a bounded task and Jev observes windows, chooses controls, types supplied text and navigates locally, returning a result or a request for help.
Jev Agents & MCP#community#catalog-import#logicrw
Community
jacks3tr/jev-desktopPython
jacks3tr/jev-desktopPythonOne skill and one MCP server for Claude Code, Codex and Hermes: Jev gates large-model turns with gate_triage, gate_verdict, gate_decide and gate_pick, runs build-and-repair loops with independent review, routes model tiers and skills, and guards Git steps.
Jev Agents & MCP#community#catalog-import#logicrw
Community
king4s/jev-ai-skillPython
king4s/jev-ai-skillPythonA native Hermes Agent tool plugin - not a chat-model provider - where Jev supplies bounded semantic judgments and an explicit model recommendation while the main agent keeps planning, generation, execution and approvals. Backends: TypeSafe, Cloudflare and OpenRouter.
Jev Agents & MCP#community#catalog-import#logicrw
Community
ourines/hermes-jevPython
ourines/hermes-jevPythonOfficial MCP server for JevGuard and TypeSafe AI: zero-dependency guardrails over JSON-RPC 2.0 on stdio, gating shell commands and code patches, caching decisions in local SQLite and pruning state.
Jev Agents & MCP#community#catalog-import#cobanov
Community
seb4ez/jevguard-mcpPython
seb4ez/jevguard-mcpPythonFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.